Master's completed 2018


T. Kumirai, 2018. "Development of a Design Tool for PCM based Free Comfort Cooling System in Office Buildings in South Africa​​"

Space cooling energy demand is projected to increase due to climate changes. For example, the South African climate change model projected warming to reach around 3 to 4°C along the coast, and 6 to 7°C in the interior. Such temperature increases will significantly increase the energy demand by building cooling applications. Thus, there is an urgent need to improve the energy efficiency in buildings and to reduce the peak cooling loads.

Various studies for building free cooling using phase change materials have shown to reduce or avoid the need for mechanical space cooling.  Very few of these studies covered Southern African climatic conditions and no research was found reporting a comparison of free cooling thermal performance of different PCM types for an individual climate scenario. The purpose of this study was to experimentally evaluate and compare the cooling performance of three PCM materials in plate-air heat exchanger modules subjected to Southern African climatic conditions and to use the data to deduce empirical correlations that can be used by thermal designers to determine the number of modules required to maintain an objective cooling load within the range of operating conditions.

In this experimental investigation the cooling (discharging) performance of plate encapsulated Phase Change Materials (PCMs) for passive cooling applications were evaluated as measured by its average effectiveness, cooling power, energy absorption and phase transformation duration. A test facility that mimics a PCM-air heat exchanger module installed in a ventilation duct was used to consider the impact of varying air flow rate and inlet air temperature. PCM plate encapsulations with a thickness of 10 mm orientated vertically and spaced at a pitch of 15 mm were investigated. The thermal storage characteristics of three commercial PCMs were considered.  Two paraffin type PCMs with melting temperature ranges of 25 °C to 28 °C and 22 °C to 26 °C and one type salt hydrate with a phase change temperature range 24 °C to 25 °C were used in air flows ranging in temperature from 30 °C to 35 °C and duct air velocities ranging from 0.4 m/s to 0.9 m/s.

The results indicated that average effectiveness of the PCM modules decreased with increasing convective air mass flow rate. Increasing air mass flow rate (at constant inlet air temperature) or increasing the inlet air temperature (at constant air mass flow rate) increased the average cooling power. The phase transformation durations of the PCMs decreased as both the air flow rate and inlet air temperature increased. The salt hydrate (SP24E) module had the highest energy absorption capacity for all experimental conditions. The rate of energy absorption increased with inlet air temperature. From a design standpoint the desirable thermal performance of PCM is to have a high instantaneous heat absorption capacity and also extended over a longer period. Paraffinic PCMs met the first condition of high instantaneous heat absorption but did not meet the second condition of extended heat absorption duration. SP24E met the condition for extended heat absorption duration but had a lower instantaneous heat absorption capacity than the paraffin.

Empirically-based correlations for determining the number of modules to maintain an objective cooling load were developed using a multiple regression analysis technique. From this, air conditioning system designers can determine the number of modules (installed in parallel) required to maintain an objective cooling load within the range of operating conditions tested.

Supervisor: Prof J Dirker

Co-Supervisor:Prof J.P. Meyer  


N. A. van der Merwe, 2018. "ABS braking on rough terrain​​"

Anti-lock braking systems (ABS) have revolutionised vehicle safety by providing a mechanism for braking effectively whilst maintaining stability control.

ABS works very well on smooth terrains, where there are little disturbances from the terrain and thus has been wildly adopted on commercial vehicles. The ability of ABS to perform on rough terrains is however deteriorated by the fact that rough terrains cause fluctuations in the wheel speeds and in the loading conditions of the tyre, which negatively affects the ABS algorithm as well as the tyre-terrain interaction. The adoption of ABS on off-road vehicles have been very slow due to these challenges.

This study investigates the performance of ABS on rough terrain through performing experimental and simulated ABS tests using a testing trailer. A MSC ADAMS model of the trailer was built and validated. The ABS tests show good correlation between the tyre forces measured and the forces simulated using an FTire (Flexible Structure Tire Model) model in MSC ADAMS. The ABS algorithm’s performance deteriorates significantly due to the oscillation of wheel speeds. The work done in this study paves the way towards future ABS investigation and development using both the experimental and a simulated platform.

Supervisor: Prof P.S. Els


M Slootweg, 2018. "Numerical performance analysis of novel solar tower receiver​"

Concern over the altering climate due to the release of anthropogenic greenhouse gases has caused a major shift in the developments of ways to minimise human impact on the climate. Solar energy is seen as one of the most promising sources to transform the energy market for low-carbon energy generation. Currently, solar power is generated via photovoltaic (PV) and concentrating solar power (CSP) technologies. The advantage of CSPs to scale up renewable energy to utility level, as well as to store thermal energy for electrical power generation when the sun is not available (after sunset or during cloudy periods) makes this technology an attractive option for sustainable clean energy. CSP development, however, is still in its infancy, and for it to be a competitive form of energy-generation technology, techno-economic developments in this field need to improve the efficiency and decrease the costs of this technology. A policy report by the European Academies’ Science Advisory Council (EASAC) (2011) indicated that central receiver (solar tower) CSP systems show the greatest margin for technological improvements (40% to 65% is estimated), and that an improvement in receiver technology could make the greatest contribution to increase efficiency.

This study therefore focused on analysing the optical and thermal performance of a new proposed solar cavity molten salt receiver design for a central receiver CSP system using a numerical approach. In this study, the receiver’s performance was analysed by first selecting an existing heliostat field, Planta Solar 10 (PS-10). For the numerical analysis to reflect conditions that are as realistic as possible, numerical models for different aspects were selected and validated. For modelling the sun, the solar tracking numerical model proposed by Iqbal (1983) was selected and implemented after literature and comparison showed adequate results. The direct normal irradiation (DNI) was modelled by applying a clear sky model, with the parameterisation model C proposed by Iqbal (1983) as the chosen model. The variables in this model that were subject to temperature, and humidity values were more accurately presented by adding numerical approximations of the region’s actual weather data. The DNI model reflected realistic fluctuations. For the thermal modelling, a validation study was conducted on impingement flow heat transfer to select an appropriate Reynolds-averaged Navier-Stokes (RANS) model that would provide accurate results when conducting the thermal performance test on the receiver. The study concluded that the transitional Shear Stress Transport (SST) turbulence model performed the best.

A new method was also developed and validated that allows one to not only simulate complex geometries within the Monte Carlo ray tracing environment SolTrace, but also to apply the results obtained by simulating this model as a heat source within the computational fluid dynamics (CFD) environment ANSYS Fluent. This allows SolTrace modelling to be more accurate, since models do not need to be approximated to simple geometries. It also provides an alternative for solar modelling in ANSYS Fluent.

The optical analysis was conducted by first performing an analysis on the receiver aperture and studying its sensitivity on the captured flux. This was followed by analysing the optics of the proposed receiver, the flux distributions on a simplified absorber surface area, and how these distributions are altered by changing some parameters. An in-depth analysis was finally done on the absorber area by applying the aforementioned model to simulate complex geometries within SolTrace, with the results illustrating the difference of the detailed geometry on optical modelling. An alternative receiver design with improved optical features was proposed, with an initial study providing promising results. The thermal analysis was done within the CFD environment, with only a section of the absorber surface area considered, and by applying the solar flux simulated during the optical analysis as heat source within the geometry model. This allowed the model to simulate the effects of re-radiation at the surface of the absorber while simulating the heat transfer at the fluid molten salt side simultaneously. The results showed that, for the current design and requirements, the absorber surface temperature reaches impractical temperatures. Altering the design or being more lenient on the requirements has, however, shown dramatic improvements in terms of thermal performance. Sensitivity studies for both the optical and thermal analyses have shown that changes in design can dramatically improve the performance of the design, making it a possible feasible receiver design for central receiver systems.

Supervisor: Prof K.J. Craig

Co-Supervisor:Prof J.P. Meyer  


I. Sethsedi, 2018. "Technical and Economic Feasibility of a Regenerative Braking System with On-board Energy Storage for Freight Trains"

This dissertation presents the technical and economic feasibility of a novel regenerative braking system (RBS) for the freight rail industry. A concept for a distributed RBS, integrated into the bogies of freight rail wagons, is proposed in a patent by Transnet SOC Ltd. The system allows for numerous RBSs to be installed on a single freight train, in a distributed manner, which collectively functions together to perform regenerative braking on the train with the goal of reducing the energy consumption of the train. The proposed system would, if implemented successfully, alleviate challenges and limitations with current RBS on diesel-powered freight trains. The patent also proposes that the RBS utilise mechanical energy storage by means of a high-speed flywheel which is connected to the train axles by a continuously variable transmission (CVT).

The proposed RBS is conceptualised in this study by first establishing the requirements of the system from in-service train data, followed by the development of the subsystems to deliver workable concepts that would meet the requirements identified. A multi-domain, physical system simulation model is subsequently developed to establish the energy savings performance of each of the system concepts for typical freight train routes. The simulation results show that energy savings of between 10% and 24% can be realised by the feasible system concepts, depending on the configuration of the RBS concept and the duty-cycle of the specific train route. This proves the technical feasibility of the proposed system.

Next, the proposed system and the candidate concepts are evaluated in economic terms. A cost-benefit analysis (CBA) is performed in which the cost and benefits over the life cycle of the RBS were combined into a single distribution and analysed. The decision criteria calculated in the CBA provide unanimous results as to which of the candidate concepts are economically feasible. It is shown that four of the candidate concepts, all utilising the same transmission topology incorporating a CVT with different flywheel configurations, are economically feasible. The RBS concepts show good return on investment and provide an internal rate of return (IRR) of 17% and a benefit-cost ratio (BCR) of 2.13. These results therefore indicate that the proposed distributed RBS for freight trains is economically feasible and would deliver favourable financial returns if pursued.

Supervisor: Prof P.S. Heyns


A. Beneke, 2018. "The Simultaneous Optimization of the Nose and Tail Geometry of a High-Speed Train for Drag and Crosswind Stability"

The South African Government requires that a high-speed rail network is available to commuters by the year 2050. One of the key role players in achieving the aforementioned is Transnet Engineering (TE), which at present has neither experience in the field of high speed rail, nor a prototype train that would perform well aerodynamically travelling at speeds of approximately 350km/h. As such the final deliverable of this study was a nose and tail geometry for a train travelling at 350km/h, which had been optimized for total drag under windless conditions and drag as well as crosswind stability when subjected to crosswinds. In this study attention was thus given to the flow surrounding high-speed trains so as to address the knowledge deficit within TE, as well as the challenges typically faced in high speed rail applications with the purpose of ensuring the validity of the optimization goals set out by TE. This study furthermore identified appropriate geometric variables from literature which are important for an efficient aerodynamic nose and tail shape, i.e., the nose length (L), nose-tip height (Z0) and the inflection point height (H). Another objective was to identify an appropriate turbulence model that was able to accurately analyze the flow surrounding the train body, with attention also being given to the choice of a suitable optimization algorithm. The validation case completed on the Ahmed’s body not only revealed an appropriate grid resolution to use for the optimization study on the train geometry, but further showed the aptness of the linear pressure-strain Reynold’s stress turbulence model and the SHERPA optimization algorithm. For the optimization of the nose and tail geometry under windless conditions, the design space was sampled by making use of a 5-level full factorial, while the radial basis function with thin spline method was used to connect the data points obtained from simulation. The surrogate model was found to be highly predictive with the largest discrepancy between its results and simulation values being -1.6%. The SHERPA algorithm was used for the optimization study itself and identified an optimal geometry with L=7.7m, H=2.73m and Z0=1.364m. The associated minimized total drag force is 13.6kN which is 30.4% less than the maximum drag force that can be actualized by the geometric parameters within their respective ranges. In the case of a train subjected to crosswinds, the Latin Hypercube sampling method along with 27 designs points was made use of in order to sample the design space. The surrogate model was obtained by making use of the same fitting method as above and was once again found to be predictive with the greatest discrepancy reported being 1.6%. The same search algorithm as above was also used in order to identify the Pareto front, with the recommended geometry displaying the following features; L=7.7m, H=2.71m and Z0=1.364m. This yielded a minimum drag force of 14.6kN and rolling moment of 93.5kN, which corresponds to a reduction of 23.5% and 12.6%, respectively, from the maximum values that can be actualized by the geometric parameters within their respective ranges. Finally, it was found that either of the aforementioned optimized geometries is able to perform well when exposed to the other’s load case.  

Supervisor: Prof K.J. Craig


A. Pyper, 2018. "Technical and Economic Feasibility of a Regenerative Braking System with On-board Energy Storage for Freight Trains​​"

This dissertation presents the technical and economic feasibility of a novel regenerative braking system (RBS) for the freight rail industry. A concept for a distributed RBS, integrated into the bogies of freight rail wagons, is proposed in a patent by Transnet SOC Ltd. The system allows for numerous RBSs to be installed on a single freight train, in a distributed manner, which collectively functions together to perform regenerative braking on the train with the goal of reducing the energy consumption of the train. The proposed system would, if implemented successfully, alleviate challenges and limitations with current RBS on diesel-powered freight trains. The patent also proposes that the RBS utilise mechanical energy storage by means of a high-speed flywheel which is connected to the train axles by a continuously variable transmission (CVT).

The proposed RBS is conceptualised in this study by first establishing the requirements of the system from in-service train data, followed by the development of the subsystems to deliver workable concepts that would meet the requirements identified. A multi-domain, physical system simulation model is subsequently developed to establish the energy savings performance of each of the system concepts for typical freight train routes. The simulation results show that energy savings of between 10% and 24% can be realised by the feasible system concepts, depending on the configuration of the RBS concept and the duty-cycle of the specific train route. This proves the technical feasibility of the proposed system.

Next, the proposed system and the candidate concepts are evaluated in economic terms. A cost-benefit analysis (CBA) is performed in which the cost and benefits over the life cycle of the RBS were combined into a single distribution and analysed. The decision criteria calculated in the CBA provide unanimous results as to which of the candidate concepts are economically feasible. It is shown that four of the candidate concepts, all utilising the same transmission topology incorporating a CVT with different flywheel configurations, are economically feasible. The RBS concepts show good return on investment and provide an internal rate of return (IRR) of 17% and a benefit-cost ratio (BCR) of 2.13. These results therefore indicate that the proposed distributed RBS for freight trains is economically feasible and would deliver favourable financial returns if pursued.

Supervisor: Prof P.S. Heyns


K Poovendran, 2018. "Integrated Brake based Torque Vectoring Control of Vehicle Yaw Rate and Side-Slip Angle​​"

Sport Utility Vehicle (SUV) sales are increasing globally, even surpassing sedan vehicle sales worldwide. Their increasing popularity is termed a continuous trend that is expected to last. SUVs are known to offer a higher ground clearance which makes them more susceptible to rollover and directional instability during emergency manoeuvres. This dissertation proposes an integrated controller which controls two vehicle states, namely yaw-rate and side-slip angle to improve handling while reducing rollover propensity and improving rollover stability. The control system employs brake based torque vectoring to control the vehicle states, torque vectoring control improves lateral stability by maintaining consistent handling characteristics over all driving conditions and the lateral stability is maintained whilst adhering to a rollover index. The desired vehicle states are obtained from a reference linear two degree of freedom model with tyre characteristics obtained from the linear region of the tyre. A coordinated control strategy is investigated with respect to Direct Yaw Moment Control (DYC) acting on a vehicle through individual brake torques. Two types of controllers are investigated, namely a Linear Quadratic Regulator (LQR) and a Linear Model Predictive Controller (LMPC). It is shown that yaw rate control together with side-slip angle control and the inclusion of a roll index limit allows for better vehicle handling. Simulation tests are done using Simulink/ADAMS and verified experimentally with a SUV undergoing evasive manoeuvres where the vehicle is near its performance limit. The vehicle managed to be successfully navigated through manoeuvres not possible prior to yaw rate and side-slip angle control, with a notable decrease in the vehicle roll.

SupervisorDr. T.R. Botha 

Co-Supervisor: P.S. Els 

R.A. Kapp, 2018. "Wireless Vehicle-in-the-Loop Platform for Supervisory Control​"

The development of Hardware-in-the Loop (HIL) and subsequent network-distributed HIL systems as testing methodologies have assisted in the rapid realisation of new safety technologies such as Advanced Driver Assistance Systems (ADAS). As an expansion of ADAS a supervisory accident mitigation system could be the culmination of research in the fields of vehicle dynamics, real-time testing, control systems and wireless communication networks. This study focuses on proving the concept of such a system by autonomously controlling the steering of a Land Rover Defender 110 Tdi test vehicle from a base station through a wireless communication link. The entire vehicle forms the hardware component of the hardware-in-the-loop system with all control intelligence situated at a road-side base station thus coining the term “wireless vehicle-in-the-loop”. Introducing a wireless network to a closed-loop control system unavoidably results in communication delays which are known to detrimentally affect system response. The effects of these delays are investigated by evaluating the path following ability of the test vehicle when performing autonomously-steered, pre-recorded ISO3888-1 double lane changes to simulate an emergency steering manoeuvre that a supervisory system may have to perform. Cross-track error and overall stability were used as performance metrics. The manoeuvre was first simulated in an MSC.ADAMS and MATLAB Simulink co-simulation using a validated multibody dynamics model of the test vehicle and a validated steering controller. Different magnitudes of simulated constant delay were then introduced between the two components based on round-trip delay data captured when testing the wireless link. The simulation was verified experimentally by adding artificial delays to the wireless link to determining the level of delay at which the vehicle would lose stability. Results indicated that the simulation closely resembled experimental results at 60 km/h (restricted for safety concerns) up to a round-trip delay of 50ms where both cases showed a cross-track error of around 0.3 m. At higher delays experimental results showed much higher degradation vehicle response as compared to simulation. It was found, however, that the concept of supervisory vehicle control is feasible and that an emergency steering manoeuvre can be successfully performed when exposed to relatively large communication delays.

SupervisorDr. T.R. Botha 

Co-Supervisor: P.S. Els 


D. K. Johnson, 2018. "Real-Time vehicle measurement using digital image correlation​"

The tyre-road interface is one of the most important research topics in the field of vehicle dynamics. This is largely due to all the vehicle excitation forces (besides aerodynamic forces) being generated at this interface. There are many parameters which govern the generation of tyre forces, of which the side-slip angle is of utmost importance.               

                Vehicle side-slip angle can be used as a measure of vehicle stability. Stability control schemes require side-slip angle and typically estimate this parameter instead of using a direct measurement. The relationship between tyre lateral force and tyre side-slip allows the lateral force generated by the tyre to be determined from the tyre side-slip angle. Therefore, real-time measurement of side-slip angle is important in tyre research and vehicle stability. Solutions exist to measure the side-slip angle, however, do not perform well at low speeds or over rough terrain and are prohibitively expensive.         

                In terramechanics, tyre soil deformation in the form of rut depth is a widely researched topic as it can be used as a measure of the vehicle's ability to traverse the terrain, estimate soil characteristics and for vehicle environmental impact studies. Currently, these measurements are labour intensive and are typically conducted by hand. Other solutions exist however they are developed for road use and are prohibitively expensive. The research field would, therefore, benefit largely from online rut depth measurements.     

                Digital Image Correlation is the mathematical process of tracking changes in digital images. The development of robust algorithms and ease of implementation has allowed many fields to be adapt this non-contact based, optical technique for application-specific measurements. Previous studies \citep{BothaPHD2015} have proved DIC to be a viable candidate for measuring the side-slip angle and rut depth that overcome current measuring hurdles. However, the analysis was conducted in post-processing from pre-recorded footage due to the large computational expense of the image processing. This opens the opportunity to adapt and optimise these techniques to achieve real-time processing speeds required for these camera-based sensors.

                This study builds on \citet{BothaPHD2015} with a real-time implementation which allows for online measurements to be made using inexpensive, off-the-shelf cameras with dedicated software. This will eventually provide systems such as ABS, stability control schemes and semi-active suspension with real time vehicle side-slip angle and rut depth with a cost-effective camera-based sensor. The aim of the present study is to develop and test two systems that can measure the side-slip angle and rut depth in real-time.

               The side-slip angle is measured using a single camera pointing down on the terrain and digital image correlation. It is shown to measure accurately and in real-time. The sensor is tested on a flat surface using a rig that allows for validation.     

                The rut depth is measured using multiple cameras pointing at the terrain and digital image correlation to create a 3D map of the terrain. Three methods for determining the rut depth from the 3D map is investigated, with varying degree of accuracy and processing speed.

Supervisor: Dr. T.R. Botha 

Co-Supervisor: P.S. Els 

J.S. Jooste, 2018. "Physiological Responses to Whole-Body Vibration​"

The research done in this study investigates physiological responses to vertical whole-body vibration. The aim is to determine whether or not quantifiable responses can be found when evaluating changes in breathing rate, heart rate and heart rate variability. Such a relationship could potentially be used in vehicle dynamics industries to improve suspension system designs. This would be done by supplementing subjective testing techniques with a more objective physiological response when evaluating ride comfort.

A group of 60 volunteers were subjected to vertical whole-body vibration using a single seat actuator. The physiological parameters mentioned were measured during three different states, and the changes from state 1-2 and state 2-3 were recorded. The three states were each measured at different stages during the test procedure with stage 1 corresponding to the physiological state 1. Stage 1 consisted of baseline measurements, during this stage the test participant was not exposed to any vibrations at all. During stage 2 the participant was exposed to a reference vibration signal which is identical for all participants, and during stage 3 each participant was exposed to one of 4 alternative signals. The 4 alternative signals are all variants of the reference signal with increased amplitudes. The weighted amplitudes of each alternative signal were increased by 6.47%, 9.57%, 14.64%, and 20% respectively.

After evaluating the recorded data, it was found that the physiological change from state 1-2 was statistically significant for heart rate variability indicators. Unfortunately when evaluating the changes from state 2-3, there had been no statistically significant change.  This suggests that while there is a clear and measurable physiological response to the initial vertical whole-body vibration, a change in this vibration is not reflected in the participant’s physiological state.

Supervisor: Dr C. Kat 

Co-Supervisors: Dr C. Grant, Prof S. Els 


J.R. Wolmarans, 2018. "Forecasting spare parts demand using condition monitoring information"

Renewable energy, particularly solar energy is at the forefront of the fight against fossil fuels. Concentrated Solar Power plants utilizing heliostats, large reflecting mirrors, to concentrate the sun’s solar energy onto a central tower are one are the main solar technologies in use today. These plants consist of hundreds to hundreds of thousands of heliostats. The heliostats in most cases make up the largest portion of initial capital expenditure of a solar plant. Consequently, the design of these heliostats is an important area of research to enable Concentrated Solar Power to be a viable competitor to not only fossil fuels but also photovoltaic solar technologies. Vortex shedding and the resultant transient loadings on a medium sized heliostat are investigated in this paper. Reynolds-Averaged Navier-Stokes (RANS) Computational Fluid Dynamics (CFD) of an operational heliostat is used as a validation case. The Atmospheric Boundary Layer (ABL) is characterized and optimized. The ABL is implemented as the inlet flow boundary condition for both the RANS and Stress-Blended Eddy Simulation (SBES) simulations. The SBES Scale-Resolving Simulation (SRS) model was used which provided the transient peak wind loadings necessary to investigate the structural response of the heliostat. The synthetic turbulence technique used at the inlet of the SBES simulation was the Vortex Method. This method appears to produce unphysical pressure spikes in the flow but their effect appears to be negligible. The SBES results show a strong likeness to the experimental results of Peterka (1986) with a comparable mean and peak loading distribution. The SBES results couldn’t be accurately compared to the experimental results of Huss et al (2011) due to the uncertainty of the turbulence intensity in the experimental values. The transient SBES CFD pressure was implemented in a one-way FSI simulation. These simulations shed light on the structural response of the heliostat to the transient wind loading. The results showed that the response of the heliostat conformed to and depended on the mode shapes and frequencies of the heliostat structure more so than the vortex shedding frequencies. The results from the transient structural analysis using the temporal SBES heliostat surface pressure fields as input indicate that the method holds promise in predicting the transient response of heliostats. Importantly it can be concluded that due to the difference in frequencies between the vortex shedding and modal frequencies, the structure is safe from self-excitation.

Supervisor: Prof K.J. Craig


N. A Lelo, 2018. "Forecasting spare parts demand using condition monitoring information"

The control of an inventory where spare parts demand is infrequent has always been complex to manage because of the randomness of the demand, as well as the existence of a large proportion of zero values in the demand pattern. However, considering the importance of spare parts demand forecasting in production manufacturing and inventory management, several forecasting methods have been developed over the years to allow decision makers in industry to optimize the management of inventory where the demand pattern is infrequent. The Croston method is one of the traditional forecasting method, known because of its ability to take into consideration periods with zero demands. Yet, despite the Croston method’s advantage over other traditional methods, there are still shortcomings in the method because it does not consider the condition of the components to be replaced.

This dissertation proposes an alternative forecasting method to the traditional methods, by means of condition monitoring. This method overcomes the Croston method’s shortcomings by considering the condition information of the component under operation. A statistical model, the so-called proportional hazards model (PHM), which is a regression model, blending event and condition monitoring data, is used to estimate the risk of failure for the component under analysis, while subjected to condition monitoring. To obtain optimal decision making on spare parts demand, a blending of the hazard or risk with the economics is performed, and an optimal risk point is specified. The optimal risk point guides optimal decision making on spare parts policy for the component under analysis.

To generate the data needed to construct the proportional hazards model, a numerical investigation was performed on a fan axial bade where a crack was inserted and propagated to estimate the fatigue crack life and corresponding natural frequencies. The simulation was run using MSC.MARC/MENTAT 2016 software. To validate the finite element model, an experiment was run by using a 50kN Spectral Dynamics electrodynamics shaker to apply base excitation to the fan axial blade specimens. The treatment and computation of data generated from experimental and numerical approaches allowed the construction of the proportional hazards model, with the fatigue lifetime as event data and the blade natural frequencies as covariates or condition monitoring information. The baseline Weibull parameters were estimated by maximizing the likelihood function using the Newton Raphson method and the MATLAB package. This allowed the computation of an objective function to determine the shape, scale and location parameters. Instead of defining the covariate behaviour needed to build the cost function by means of the Markov process, a simulation procedure was utilized to define the cost function and determine the optimal risk which minimizes the cost. Furthermore, as the proportional hazards model depends on both, time and covariates, it was also shown how the PHM behaves when time or covariates carry more weight.

The added value of the proportional hazard model as forecasting spare parts method lies in the fact that it allows one to proactively gather failure information which enables a ‘just in time’ supply of spare parts as well as an optimal maintenance plan.

Forecasting spare parts demand, using condition information, performs better than traditional methods because it reduces an overly large spare parts stock pile. By allowing a ‘just in time’ part availability, the spare parts management becomes more related to the condition of the asset. Additionally, the supply chain management and maintenance cost are optimized, and the preventive replacement of components is optimized compared to the time-based method where a component can be replaced while still having a useful life.

Supervisor:Prof  P.S. Heyns

Co-Supervisor:Prof J Wannenburg  


C.F. Maré, 2018. "An investigation of computational fluid dynamics simulation for estimation of turbine remaining useful life "

Turbines encounter blade failures due to fatigue and creep. It has been shown in the literature that the primary cause of steam turbine blade failures worldwide can be ascribed to fatigue in low pressure (LP) turbine blades. The failure and damage to these blades can lead to catastrophic consequences. Some utilities use empirical methods to determine the forces experienced by turbine blades, but desire more accurate methods. The inaccurate prediction of high-cycle fatigue (HCF), thermal durability and stage performance is introduced when one does not consider blade row interaction. Blade row interactions can however be accounted for by means of computational fluid dynamics (CFD). Furthermore, modern high-fidelity CFD tools would be able to contribute greatly in predicting

the forces experienced by turbine blades.

Numerical tools such as CFD and Finite element analysis (FEA) can greatly contribute in the estimation of the remaining useful life (RUL) of turbine blades. However, in this estimation process there are various uncertainties and aspects that affect the estimated RUL. Understanding the sensitivity of the estimated RUL to these various uncertainties and aspects is of great importance, if RUL is to be estimated as accurately as possible.

In this dissertation, a sensitivity analysis is performed with the purpose of establishing the sensitivity of the estimated RUL of the last stage rotor of a LP steam turbine, to the number of harmonics used in a nonlinear harmonic (NLH) CFD simulation. The sensitivity of the estimated RUL is evaluated in the HCF regime, where the cyclic stresses occur below the yield strength of the turbine blade. A CFD model, FE model, and fatigue model were therefore developed in such a manner that would suffice, regarding the purpose of the sensitivity analysis. The CFD model is validated by comparing the predicted CFD power to that of actual generated power of a dual 100MW LP steam turbine.

The sensitivity analysis is performed for 3 operation conditions, and for each operational condition the aerodynamic forces were computed using 1, 2, and 3 harmonics in an NLH simulation. The estimation process considers a weak coupling between the CFD model and FE model. NLH simulations are firstly performed to calculate the unsteady static surface pressure distributions on the last stage rotor. This is followed by the mapping thereof to the FE model, for which a transient structural analysis is performed. Finally, the RUL is estimated by performing a fatigue analysis on the stress history obtained from the transient structural analysis.

Based on the results of the sensitivity analysis, the following recommendations were made, from a conservative point of view. Firstly, in general if the RUL is to be estimated with reasonable accuracy, just using 1 harmonic in a NLH simulation will not be sufficient and 2 harmonics should be used. Secondly, if the RUL has to be estimated with high accuracy, 3 harmonics should be used.

Supervisor:Prof P.S. Heyns 

Co-Supervisor: Dr D Dunn 

V. Ramnath, 2018. "Mathematical Modelling of Nanofluid Thermophysical Properties Using Copulas "

In this dissertation, mathematical research is performed to model nanofluid thermophysical properties in terms of multivariate probability density functions utilizing copulas from known verified and validated experimental data for water/alumina nanofluid mixtures.

A comprehensive review of the available data from the open scientific literature is undertaken to first understand the accuracy limits of the combination of available experimental and theoretical data for nanofluids. The nanofluid data is then processed using multivariate statistical analysis techniques in order to mathematically incorporate the input process parameter’s intrinsic measurement uncertainties. Having analysed the verified data, optimal functional expressions for the effective thermal conductivity are then determined. This mathematical analysis is inclusive of estimates of the process parameter’s respective experimental statistical uncertainties through stochastic based Monte Carlo simulations by incorporating information of the nanoparticle morphology such as the nanoparticle size and volume fraction, and the nanofluid temperature.

Numerical simulations are performed for the resulting copula-based PDF’s with custom developed multivariate sampling strategies which are derived and tested. These model predictions were verified and validated by comparing them to a MLP-NN scheme to check for consistency. Quantitative results from these simulations indicate that the copula mathematical model is able to achieve an  AARD=3.0953% accuracy for predicted behaviours of the developed thermal conductivity database compared to an AARD=4.2376%  accuracy for a conventional MLP neural network. The proposed mathematical modelling approach is a new novel original research technique that has been developed which is able to incorporate physical experimental measurement uncertainties such that the model is able to adaptively refine the predicted nanofluid model quantitative uncertainties in sub-domains of the input meta-parameters which is not presently mathematically possible with existing neural network modelling approaches.

Supervisor:Prof M. Sharifpur  

Co-Supervisor:Prof J.P. Meyer  


R.E. Browne, 2018. "Exploring the measurement of maintenance productivity using the Törnqvist index number approach"

Maintenance management serves the business through the strategy that it chooses.  The strategy largely determines the resourcing and cost of maintenance and should be chosen such that the output of maintenance is increased.  Maintenance productivity, comparing the outputs and inputs of the maintenance function, may be related to the maintenance strategy and therefore serves as a strategic performance indicator.  Measuring maintenance productivity performance is critical for any maintenance organisation to manage, monitor, control and take appropriate and timely strategic decisions. 

The measurement of maintenance productivity using the Törnqvist index number approach is explored.  It is proposed that measuring maintenance productivity using the Törnqvist index as a strategic performance indicator would be valid if a relationship between maintenance strategy and productivity exists in practice.  This relationship is tested and indeed proven by applying rigorous statistical tests to empirical data from very large manufacturing plants in South Africa.  The relative ease of measurement and broad application of this indicator may be exploited by senior management engaged with strategic decisions.

Supervisor:Prof J Wannenburg  

Co-Supervisor:Prof P.S. Heyns 


R. Deetlefs, 2018. "Rail surface anomaly detection: a deep learning approach for computer vision"

Rail surface defects have become more of an issue in recent years due to new manufacturing techniques which produce head-hardened rails and as industry demands higher speeds, heavier loads and increased traffic. These defects can cause catastrophic accidents, which have consequences such as death, injury, huge cost implications and loss of public confidence. Computer vision systems have become popular, as cameras are non-contact full-field sensors which are low in cost, have high sampling rates and provide appealing performance. However, accurate inspection remains challenging due to dynamic non-linear environmental and rail surface conditions in which images are captured, which result in a heterogeneous image dataset. It is also difficult to select useful features which satisfy the variations due to different failure modes. In addition, there is a class imbalance issue, as most captured images do not contain any defects.

In this dissertation, we develop deep generative models that are trained exclusively using healthy images of a rail surface so that we learn useful features to capture the complex nature of the images which are acquired. We propose multiple models which operate with images at different resolutions. We present a new dataset which will be made publicly available. Experimental results demonstrate that our proposed models can perform accurate detection using our dataset. The proposed algorithms are highly parallel and computationally efficient, which enables real-time inspection at speeds that exceed the world's fastest railway trains: Fuxing Hao CR400AF/BF that has a continuous operation speed of approximately 400 km/h.

Keywords: Computer vision, deep learning, rail surface anomaly detection, real-time inspection, unsupervised segmentation.

Supervisor:Prof PS Heyns 


D. Armfield, 2018. "Optimised Implementation of Physics-based, Strain-rate Dependent Material Model for the Improved Simulation of the Laser Shock Peening Process"

This dissertation details the investigation into – and subsequent implementation of – a physics-derived, history-dependent constitutive plasticity model for the improved numerical simulation of the laser shock peening (LSP) process.

Laser shock peening (LSP) is a surface engineering process, used to impart beneficial compressive residual stresses into the material. These residual stresses improve a processed component's resistance to surface-related failures.

This research endeavour is divided into two complementary studies that aim to address the need for an improved numerical simulation regime for the LSP process.

The first study focuses on analysis of the most popular LSP simulation regimes within the research community. LSP, to date, has seen the most extensive application in the aerospace industry. For this reason, the first study was performed on a high strength aluminium alloy – AA7075-T6. This specific alloy has seen wide use in aerospace applications, and subsequently been the focus of a number of LSP related research endeavours.

An in-depth literature review includes details on the constitutive equations typically used in the simulation of the LSP process. The popular use of the Johnson-Cook plasticity model as a constitutive equation leaves current simulation capabilities with little or no residual stress predictive capability when LSP input parameters (such as laser power density or laser shape) are varied.

The phenomenon of shock-wave propagation through solid media is also researched – and subsequently demonstrates the necessity of an equation of state in the modelling regime. Commonly, implementations of the equations of state (EOS) are neglected, but as will be demonstrated, hold significant influence over the resulting residual stresses produced during the LSP process.

To emphasise these arguments, a single laser shot is performed on a sample of AA7075-T6. The resulting surface deformation is measured. Using the deformation as a benchmark, multiple analyses are performed using ANSYS Explicit Dynamics, and other numerical codes, to investigate the influence of the constitutive model, the equation of state and optimisation schemes on the LSP simulation regimes.

Building on this foundation, a second study is developed that aims to improve upon the shortfalls of the current LSP simulation regimes within a directly-applicable South African context. The second study focuses on creating an LSP simulation with residual stress predictive capability for use in the power generation industry – and more specifically, on steam generator turbine blades and shafts. The turbine components of interest are made from high-strength, high-temperature-resistant FV566 stainless steel.

The plastic behaviour of FV-566 stainless steel is characterised through a set of equations that define the fundamental MTS model parameters. Using these base parameters and a set of Gleeble experimental results, the suitability of the MTS model to characterise FV-566 stainless steel is evaluated – while comparatively benchmarking a Johnson-Cook plasticity representation. 

Moving into the LSP domain, experimental work focused on the application of a single, controlled laser shot on an FV-566 sample. A set of validation samples are also processed, in which the LSP process input parameters (namely changing laser power density) were varied. The residual stresses developed by the single shot on the samples are then evaluated using X-ray diffraction at the Elettra synchrotron in Trieste, Italy.

A single set of residual stress results are then replicated using an LSP simulation regime that incorporates both a physics-derived, history-dependent constitutive plasticity model - the mechanical threshold stress (MTS) model – as well as a fully-developed equation of state. The outcome of this simulation yields a 17.3% symmetric mean absolute percentage error (SMAPE) in residual stress prediction, and a 7.08% symmetric mean absolute percentage error in deformation prediction for a 5.25GW/cm2 laser impact in comparison to the experimental results.

A second verification simulation is successfully performed, in which the laser power density is altered to 4GW/cm2 and the calibrated MTS material parameters are kept constant. The simulation prediction demonstrates a good correlation relative to the experimental results – within a 15MPa confidence interval range. This corresponds to an 85% confidence in residual stress prediction. The ability for the proposed simulation regime to predict the residual stress fields in samples processed using varied LSP input parameters is therefore successfully demonstrated.

Supervisor:Prof S. Kock 

Co-Supervisor:Prof P.S. Heyns  


CJ Louw, 2018. "Online Prognostics Strategies with Deep Learning Models for Fleets of Engineering Asset"

The accurate prediction of remaining useful life for fleets of engineering assets is an increasingly important task in prognostics and health management (PHM). This is because the accurate prediction of time to failure allows for improved planning, scheduling and decision-making of maintenance tasks for fleets of engineering assets. Accurate prediction of remaining useful life therefore has high potential to increase the reliability, availability, production output, profitability and safety, and to decrease downtime, unnecessary maintenance and operating costs for fleets of engineering assets. This work proposes general and convenient prognostics strategies with data-driven deep learning models for online remaining useful life prediction for fleets of engineering assets, where historical run-to-failure condition monitoring measurements with trendable exponential degradation trajectories are available.

The modeling of long-term sequence information in condition monitoring measurements has previously been shown to be very challenging and crucial for effective data-driven prognostics. Long short-term memory (LSTM) and gated recurrent unit (GRU) recurrent neural network (RNN) deep learning models are currently the state-of-the-art sequence modeling techniques and can effectively model long-term sequence information. These gated recurrent neural networks have however to date not been comprehensively investigated and compared for data-driven prognostics for fleets of engineering assets. In this work we investigate data sets which include a general asset degradation data set, turbofan engine degradation data set and turbofan engine degradation benchmarking data sets. The investigated data sets all simulate the exponential degradation trajectories and condition monitoring measurements for fleets of engineering assets that were run to failure, where each asset had either univariate or multivariate condition monitoring sensor measurements performed at fixed time intervals over its lifetime. The data sets investigated include training set examples and testing set examples. The training set examples represent previously seen engineering assets with condition monitoring measurements that were run to failure. The testing set examples represent completely unseen general engineering assets with condition monitoring sensor measurements that were run to failure. The objective and challenge is therefore to propose a prognostics strategy and train a model on the condition monitoring measurements of the training set examples offline. The proposed prognostics strategy and trained model must then predict the remaining useful life from the condition monitoring measurements of the testing set examples fully online. The turbofan engine degradation benchmarking data sets allow for simple and effective prognostics performance comparisons between publications with different prognostics strategies and models.

The proposed general prognostics strategies for this work include a prognostics classification strategy and prognostics regression strategy. The proposed prognostics classification strategy is to structure the remaining useful life modeling problem as a sequence-to-sequence classification deep learning problem, where the input sequence is the univariate or multivariate condition monitoring measurement time series and the target sequence is the remaining useful life classification time series. The remaining useful life classification time series for each individual training and testing set example consists of remaining useful life classes with different degradation levels that are based on its linearly decreasing remaining useful life time series with an applied threshold. The prognostics regression strategy is to structure the remaining useful life modeling problem as a sequence-to-sequence regression deep learning problem, where the input sequence is the univariate or multivariate condition monitoring measurement time series and the target sequence is the remaining useful life regression time series. The remaining useful life regression time series for each individual training and testing set example consists of remaining useful life values that are based on its linearly decreasing remaining useful life time series with an applied threshold. The sequence-to-sequence classification and regression deep learning models then learns and generalizes the mapping between the condition monitoring measurement time series and the remaining useful life classification and regression time series for all the training set examples offline. The trained sequence-to-sequence classification and regression deep learning models is then used to predict the remaining useful life classification and regression time series from the condition monitoring measurement time series for all the testing set examples fully online.

The proposed sequence-to-sequence deep learning classification and regression model architectures that are investigated and compared for the proposed prognostics classification and regression strategies include a feedforward neural network (FNN), simple recurrent neural network (S-RNN), long short-term memory recurrent neural network (LSTM-RNN) and gated recurrent unit recurrent neural network (GRU-RNN). The sequence-to-sequence deep learning classification and regression model architectures are trained on the training sets of the investigated data sets with the new and effective Adam algorithm. The deep learning models are also regularized with a combination of the early stopping, weight decay and dropout regularization techniques in order to reduce overfitting and improve generalization and prediction performance.

The prognostics classification and regression strategies were successfully applied on the investigated data sets with the FNN, S-RNN, LSTM-RNN and GRU-RNN classification and regression model architectures. The LSTM-RNN and GRU-RNN models drastically outperformed the FNN and S-RNN models as expected. The GRU-RNN models slightly outperformed the LSTM-RNN models and the S-RNN models significantly outperformed the FNN models on average. The prognostics regression strategy and LSTM-RNN and GRU-RNN regression models achieved very competitive results when compared with other state-of-the-art publications on the turbofan engine degradation benchmarking data sets.

Supervisor:Prof PS Heyns 


J Marsberg, 2018. "Development of numerical techniques for evaluation of point-focus solar cavity receiver performance"

Solar receiver cavities, which are designed to absorb large amounts of concentrated solar irradiation, form the central component of a solar collection plant. Since this receiver’s efficiency is directly proportional to the plant’s overall performance, the optimum design of these receivers is an important research field, as it is key to the maximisation of electricity output, while maintaining reasonable costs as an alternative to the high costs of fossil fuel energy generation technologies.

Due to the high temperatures that are reached inside a solar receiver, the prediction of heat flux distribution and the subsequent effects on conjugate heat transfer have been key areas of research in the solar field. Initially dominated by experimental studies, research has trended towards numerical prediction using finite volume methods (FVM), due to the low turnaround time and cost-effective nature of this type of analysis.

Owing to the need to accurately predict these heat flux distributions, a methodology to numerically simulate concentrated heat flux on complex surfaces of a solar receiver is developed. A combination of Monte Carlo ray tracing (MCRT) methods and computational fluid dynamics (CFD) is implemented to estimate system performance, while minimising computational time and expense, with limited sacrifice of accuracy.

After successful validation of this method with experimental data, iterative performance simulations on a candidate geometry, implemented in a realistic solar-concentrating field, are performed to showcase the ability of the methodology to accurately predict system performance. The sample geometry is based on a number of implementations from various case studies and receivers that are used nowadays, with each iteration allowing for parameter adjustment to maximise optical and thermal efficiency.

Key result outputs include the prediction of heat flux distributions and subsequent thermal stress raisers, such as hot spots, convective and re-radiation heat losses, and operating temperatures. Determining which of these thermal stress raisers from the implementation of this model can further improve and streamline designs.

Supervisor:Prof K.J. Craig 

Co-Supervisor:Prof J.P. Meyer  


JG van Zyl, 2018. "A Predictive Method that Allows a Condition-Based Maintenance Implementation Based on Failure Statistics and Partial Knowledge of Failure Mechanisms"

Over the past decades, the progression from a reactive maintenance approach, to a time/use-based preventative approach, to a predictive approach, or Condition-Based Maintenance (CBM), for components subjected to ageing failure mechanisms such as fatigue, corrosion and wear, has led to significant savings on downtime and expenditures.

In this study, a spectrum of the level of insight and information available when embarking on this progression, is considered. On the one side of the spectrum is the case where a quantitative physical failure model is not available and/or the measurement of condition parameters is not feasible, but statistical failure data is available. This enables the use of Reliability Theory (RT) to implement a time/use-based Preventative Maintenance (PM) approach. On the other side is the ideal case for CBM, which entails feasible implementation of Condition Monitoring (CM) and where a physical failure model with all its parameters is known and the measured condition parameter enables the accurate calculation of the Remaining Useful Life (RUL). A bridge between CBM and time/use-based PM is represented by the Proportional Hazard Model (PHM) technique, which does not take a physical failure model into account, but where CM is feasible and relies on the fact that historic condition and failure data is available.

The main research question that is addressed during this study is the lack of an approach to implement CBM on equipment when historic condition monitoring data is not available, which may often be the case. On the spectrum, this would be placed between the ideal CBM case and the PHM technique. A new methodology is therefore developed that combines partial insight into the physical failure model with some form of measurable condition, as well as failure statistics, in order to develop degradation functions, or PF curves, to resemble component condition which may be used for CBM decisions. The newly developed method enables the implementation of CBM, which is initially based only on failure statistics and assumptions regarding the physical failure models, without the need for historic CM data. When the newly developed method is implemented, CM data is assembled, and this data may be used to continuously update the failure model assumptions, to progressively develop a full, economic CBM implementation.

The development of the new method is based on a numerical experiment, simulating components prone to fatigue failure, with various chosen initial conditions and operating conditions, to produce failure statistics. It is then assumed that, in practice, only these failure statistics would be known, as well as the form of the failure mechanism. The new method, to establish PF curves for a component with any given life based on this information, then entails arbitrarily choosing initial conditions, or defect sizes, and then calculating the operating condition parameter in the crack growth equation, to yield the required life. Using these arbitrarily chosen and calculated parameters, estimated PF curves may be derived, which would be used to base RUL and CBM decisions on.

With the “true” PF curve known from the numerically generated data, the accuracy of such decisions can be evaluated. This is done in the form of a sensitivity study, where the sensitivity of the accuracy of RUL decisions as a function of the arbitrary choice of initial conditions, can be tested for a wide range of component types. This sensitivity study yields promising results, as the error for all component types are low. The practical application of the new method is also demonstrated for bearings, where a fatigue related ageing mechanism is assumed and vibration CM provides indirect measurement of the condition. It is shown that the method provides sufficiently accurate predictions of RUL to enable implementation of CBM, without initial availability of historic CM data.

A further benefit of the new method is showcased, through its enablement of numerical simulation of the outcomes of the application of different maintenance tactics on a complex system. The simulated illustrative system consists of four component types, with ten of each component type with randomised initial and operating conditions. A time-based simulation is made possible, since the estimated PF curves for each component are known, using the newly developed method. The model simulates a period of ten years and replacements are made according to the applied maintenance tactic. CBM, which forms part of a predictive approach and would be enabled by the method developed in this study, is compared to a reactive approach and a preventative approach. Compared to a reactive approach, the predictive approach resulted in 78% less downtime and 67% less expenditure. Compared to a preventative approach, the predictive approach resulted in 56% less downtime and 57% less expenditure. These promising results would assist in making a business case for the implementation of CBM in practical applications.

Supervisor: Prof J Wannenburg 

Co-Supervisor: Prof PS Heyns  


M. Tikam, 2018 "Posture control of a low-cost commercially available hexapod robot for

uneven terrain locomotion"

Legged robots hold the advantage on uneven and irregular terrain, where they exhibit superior mobility over other terrestrial, mobile robots. One of the fundamental ingredients in achieving this exceptional mobility on uneven terrain is posture control, also referred to as attitude control. Many approaches to posture control for multi-legged robots have been taken in the literature; however, the majority of this research has been conducted on custom designed platforms, with sophisticated hardware and, often, fully custom software. Commercially available robots hardly feature in research on uneven terrain locomotion of legged robots, despite the significant advantages they pose over custom designed robots, including drastically lower costs, reusability of parts, and reduced development time, giving them the serious potential to be employed as low-cost research and development platforms. Hence, the aim of this study was to design and implement a posture control system on a low-cost, commercially available hexapod robot – the PhantomX MK-II – overcoming the limitations presented by the lower cost hardware and open source software, while still achieving performance comparable to that exhibited by custom designed robots.

For the initial controller development, only the case of the robot standing on all six legs was considered, without accounting for walking motion. This Standing Posture Controller made use of the Virtual Model Control (VMC) strategy, along with a simple foot force distribution rule and a direct force control method for each of the legs, the joints of which can only be position controlled (i.e. they do not have torque control capabilities). The Standing Posture Controller was experimentally tested on level and uneven terrain, as well as on a dynamic balance board. Ground truth measurements of the posture during testing exhibited satisfactory performance, which compared favourably to results of similar tests performed on custom designed platforms.

Thereafter, the control system was modified for the more general case of walking. The Walking Posture Controller still made use of VMC for the high-level posture control, but the foot force distribution was expanded to also account for a tripod of ground contact legs during walking. Additionally, the foot force control structure was modified to achieve compliance control of the legs during the swing phase, while still providing direct force control during the stance phase, using the same overall control structure, with a simple switching strategy, all without the need for torque control or modification of the motion control system of the legs, resulting in a novel foot force control system for low-cost, legged robots. Experimental testing of the Walking Posture Controller, with ground truth measurements, revealed that it improved the robot’s posture response by a small amount when walking on flat terrain, while on an uneven terrain setup the maximum roll and pitch angle deviations were reduced by up to 28.6% and 28.1%, respectively, as compared to the uncompensated case. In addition to reducing the maximum deviations on uneven terrain, the overall posture response was significantly improved, resulting in a response much closer to that observed on flat terrain, throughout much of the uneven terrain locomotion.

Comparing these results to those obtained in similar tests performed with more sophisticated, custom designed robots, it is evident that the Walking Posture Controller exhibits favourable performance, thus fulfilling the aim of this study.

Keywords:           Legged robots, hexapod robot, posture control, Virtual Model Control, force control, uneven terrain locomotion

Supervisor: Prof. N.J. Theron

Co-Supervisor: Dr D. Withey 


Tamsin Purkis, 2018, "Development and Validation of Pre- and Post-Processing Algorithms for Quantitative Gait Analysis using a prototype Wearable Sensor System"

Walking is the most common form of human locomotion and the systematic study thereof is known as gait analysis. Measurement and assessment thereof have application in many fields including clinical diagnosis, rehabilitation and biomechanics. The process of gait evaluation is typically done using an optical motion analysis system combined with stationary force platforms. This is considered the gold standard, but unfortunately, has several drawbacks. It is expensive, requires dedicated laboratories with spatial restrictions, calls for lengthy set up and post-processing times and cannot be used in 'real-world' environments. Alternative systems based on wearable sensors have been developed to overcome these limitations.

The Council for Scientific and Industrial Research (CSIR) has therefore developed a prototype wearable sensor unit consisting of an inertial measurement unit (IMU). The objective of the current study is, therefore, to advance the prototype to a wearable multi-sensor system for quantitative gait analysis. The focus is on the development of the pre- and post-processing algorithms and methods used to transform the measurements into interpretable information.

The focus outlined includes establishing techniques for synchronising the data from the sensors offline, pre-processing the signals, developing algorithms for stride and gait event detection, selecting an appropriate gait model and defining methods for estimating gait parameters. The determined parameters were the spatio-temporal and joint kinematics (hip, knee and ankle). The algorithms and new system were validated against the Vicon motion capture system through gait analyses. The twenty able-bodied volunteers that took part were required to walk across the laboratory six times at three self-selected walking speeds (slow, normal and fast). For the sake of simplicity and due to various limitations, only data in the sagittal plane of the right lower limb of each volunteer was used to validate the wearable system and associated algorithms.

The results obtained were then evaluated against several validation criteria. The absolute mean difference between the estimated timing of detected gait events of the two systems was consistently small (between 0.021 and 7.25% of the gait cycle overall). The spatially dependent parameters, stride length and walking speed, had significant maximum mean absolute percentage errors (31.9 and 34.5% respectively), but with little variation. Excluding outliers, that of the temporal parameters, stride time and cadence, was significantly lower (5.7 and 5.6% respectively). The kinematic results were substantially comparable with a minimum correlation co-efficient of 0.86 and a maximum RMSE of 7.8 degrees with little variation implying repeatability.

Although there were some discrepancies between the outputs, the wearable sensor system and its corresponding algorithms were considered feasible and potentially beneficial to developing countries like South Africa. Recommendations for future work include synchronising data between the wearable and reference system for stride-to-stride comparisons and validating algorithms using a known reliable wearable system.

Supervisor: Prof. N.J. Theron

Co-Supervisor: Ms. M. Conning

W.A. Roos, 2018,  "In-belt vibration monitoring of conveyor idler bearings by using wavelet package decomposition and artificial intelligenceIn-belt vibration monitoring of conveyor idler bearings by using wavelet package decomposition and artificial intelligence"

Conveyor systems make use of idlers that support the belt and its payload as it is circulated. These idlers have bearings to ensure lower friction between the idlers and the belt. These bearings do become contaminated with dust and dirt and bearings tend to fail or even seize, adding unwanted strain and stress on the belt. These idlers are monitored and replaced when needed to minimize the damage to the belt. There are several methods used to monitor the condition of the idlers. Thermal cameras are used to identify failing bearings that tend to run hotter than healthy bearings. Acoustic equipment exist that can capture the sound produced by the idler and processes it to indicate whether an idler is still working properly or when it is failing. These methods require an operator to travel the length of the belt, monitoring the idlers along the way. Vibrations have been used, with great success, to monitor idlers. An accelerometer is attached to the structure of the conveyor and the vibration signals are processed and from this a possible failing idler can be identified, either by an operator or an automated artificial intelligence system. However, the sensor can only monitor a few idlers close by and the cost of installing accelerometers along the entire length of a conveyor does make such a system infeasible. A method of using an accelerometer attached to the moving belt that travels over the idlers is investigated in this study. The vibration signals of the idler are captured as the accelerometer passes it and are then analyzed and used in a decision making system to identify and classify the idler bearing conditions. The accelerometer is attached at different positions across the width of the belt to investigate the possibility of only using one or two sensors to monitor all the bearings of the idlers across the width of the conveyor. Healthy bearings are tested against bearings with inner raceway, outer raceway and rolling element defects. Contaminated bearings are tested as well. Wavelet package decomposition is used to extract the bearing features and presents it to the intelligent decision making system. Neural networks and support vector machines are used with great success to identify and classify faulty bearings. The support vector machine monitoring system has a 100% accuracy in identifying and classifying faulty bearings, regardless of the sensor position and even when a localized payload is added. The system could not only identify a faulty bearing, but also classify the fault with 100% accuracy. These accuracies were obtained in a controlled experimental environment with a simplified test setup. The self-developed data acquisitioning system costs as much as one meter of steel reinforced rubber belt. There are some improvements needed before it could be implemented into a working conveyor, adding to the cost. A working in-belt idler monitoring system is not only plausible, but will be affordable as well.  

Supervisor: Prof. P.S. Heyns

NM van der Merwe, 2018, "Fully developed forced convection heat transfer and pressure drop in a smooth tube in the transitional flow regime"

Extensive work has been done on characterising convective heat transfer and pressure drop in smooth tubes in the laminar and turbulent flow regimes. However, little work was completed in the transitional flow regime. In all previous transitional studies, experiments that were conducted between the laminar and turbulent flow regimes were with mixed convection in the laminar flow regime and not in the forced convection flow regime. The secondary flow that occurs during mixed convection should most probably influence the characteristics in the transitional flow regime. It can therefore be expected that the transitional flow characteristics of forced convection and mixed convection will be different. However, the transitional characteristics of forced convection flow have not yet been determined. The purpose of this study was therefore to determine the heat transfer and pressure drop transitional characteristics specifically in the forced convection flow regime. Furthermore, to focus on determining these factors for a circular, horizontal smooth tube for fully developed flow. The characteristics were determined in an experimental set-up through which flow occurred through a test section consisting of a horizontal and circular smooth tube. The test-section inside diameter was 4.04 mm, and the tube length was 8.4 m. Water was used as the test fluid and was circulated through the test section which was heated at a constant heat flux. A calming section with a square edge inlet was upstream of the test section. Temperatures at the tube inlet, outlet and outer surface of the test section were measured with a total of 58 thermocouples. Two pressure taps was also installed on the test section and was connected to a pressure transducer for pressure drop measurements. Experiments were conducted mainly on the last part of the test section where fully developed flow occurred. Experiments were conducted between Reynolds numbers of 1 000 to 10 000, Prandtl numbers of 3 to 8, and Rayleigh numbers of 330 and 11 000 (heat fluxes of 0.89 kW/m2 to 3.26 kW/m2). It was found that the heat transfer transitional range coincided with the friction factor transition range with a Reynolds number range of 2 484 to 2 849. Forced convection results in the laminar regime was achieved and compared well to literature. The results were mapped on published flow regime maps. This was inconclusive as the published flow regime maps have been specifically developed for fixed parameters that did not match the parameters of this study. 

Supervisor:         Prof JP Meyer

R.P. Gräbe, 2018, "Difference thresholds for a vehicle on a 4-poster test rig"

To improve ride comfort a reduction in the acceleration experienced by occupants is required. Simulation software and test equipment are able to measure reductions in acceleration that are too small for humans to perceive. It is therefore important to know how large the reduction in vibration should be for occupants to perceive an improvement in comfort. This study determined difference thresholds (DTs) for ten automotive engineers seated in a vehicle on a 4-poster test rig. Participants were exposed to all six axes of vibration. DTs were determined for two road profiles using vertical acceleration measured on the seat and seat rail. The two road profiles were obtained by scaling the magnitude of the vertical displacements of a test track used for ride comfort evaluations. The two roads had different magnitudes, but the same spectral shape, and were therefore used to investigate the validity of Weber's Law. The BS 6841 weighted r.m.s. magnitude of the vertical acceleration measured on the seat were 0.58 and 1.01 m/s for the two roads. An up-down-transformed-response (UDTR) test procedure was used with a three-down-one-up rule to determine DTs. There was no statistically significant difference found in the medians of the relative difference thresholds (RDTs), calculated from the vertical seat acceleration, for the two roads. The median RDT for the two roads were 10.1 % and 8.6 % respectively. Results were consistent with Weber's law.

Supervisor:         Dr C. Kat

Co-Supervisor:     Prof P.S. Els

W.J.Reid, 2018, "Experimental Investigation of Circumferentially Non-Uniform Heat Flux on the Heat Transfer Coefficient in a Smooth Horizontal Tube with Buoyancy Driven Secondary Flow"

Most heat transfer tubes are designed for either fully uniform wall temperature or fully uniform wall heat flux boundary conditions under forced convection. Several applications, including but not limited to the solar collectors of renewable energy systems, do however operate with non-uniform boundary conditions. Limited research has been conducted on non-uniform wall heat flux heat transfer coefficients in circular tubes, especially for mixed convection conditions. Such works are normally numerical in nature and little experimental work is available. In this experimental investigation the effects of the circumferential heat flux distribution and heat flux intensity on the single phase (liquid) internal heat transfer coefficient were considered for a horizontal circular tube. Focus was placed on the laminar flow regime of water within a stainless steel tube with an inner diameter of 27.8 mm and a length to diameter ratio of 72. Different outer wall heat flux conditions, including fully uniform and partially uniform heat fluxes were studied for Reynolds numbers ranging from 650 to 2 600 and a Prandtl number range of 4 to 7. The heat flux conditions included 360˚ (uniform) heating, lower 180˚ heating, upper 180˚ heating, 180˚ left and right hemispherical heating, lower 90˚ heating, upper 90˚ heating and slanted 180˚ heating. Depending on the angle span of the heating, local heat fluxes of 6 631 W/m2, 4 421 W/m2, 3 316 W/m2, 2 210 W/m2 and 1 658 W/m2 were applied. Results indicate that the local and average steady state Nusselt numbers are greatly influenced by the applied heat flux position and intensity. Highest average heat transfer coefficients were achieved for case where the applied heat flux was positioned on the lower half (in terms of gravity) of the tubes circumference, while the lowest heat transfer coefficients were achieved when the heating was applied to the upper half of the tube. Variations in the heat transfer coefficient were found to be due to the secondary buoyancy induced flow effect. The relative thermal performance of the different heating scenarios where characterised and described by means of newly developed heat transfer coefficient correlations for fully uniform heating, lower 180° heating, and upper 180° heating.

Supervisor:        Prof. J. Dirker

Co-Supervisor:  Prof. J.P. Meyer

JL van Niekerk, 2018, "Degradation Estimation of High Energy Steam Piping Using Hybrid Recurrent Neural Networks"

This dissertation is a study on estimating degradation of high energy steam pipework using modern machine learning techniques. High energy piping systems are very complex to simulate due to the many variables that could influence the useful life of a component. In this research a hybrid recurrent neural network is created that consists of a combined recurrent neural network and a feed forward neural network. The machine learning model is trained on historical data that has been captured over a six-year time period and is applied to a test dataset to see if any usable patterns exist within the training data.

In this research the following variables of the piping system components are used as input to the machine learning model: the operating temperature and pressure time sequence, the distance to the closest anchor point, the distances to neighbouring supports as well as their elevation survey readings and the last known creep damage of the component. The model is created in Python using the Tensorflow library. Two types of recurrent neural networks (RNN) are tested, gated recurrent unit (GRU) and long short term memory (LSTM). The standard gradient descent (GD) algorithm, as well as adaptive gradient descent (ADAGRAD) and adaptive movement estimation (ADAM) are tested. The model was able to predict the classification of a component with an accuracy of up to 91% on the training dataset and 56% on the test data set, which is considered to be high given the complexity of the problem.

The model is successful in recognising patterns within the data and offers an automated way to parse large data sets that consist of a temporal and static data mixture. This offers an approach to make an objective decision on similar complex data driven problems and its application is not constrained to this single problem. The methods applied in this research is expected to perform even better on problems where the frequency of data collection is higher than what is used in this research.

Supervisor:    Prof PS Heyns

Co-Supervisor:Dr MP Hindley

PJ van Niekerk, 2018, "Development of a wind turbine condition monitoring facility for drivetrain torsional dynamics investigations"

Maintenance can be performed according to one of two strategies, failure based or condition based. In most cases, where large and expensive assets such as wind turbines are operated on a continuous basis, condition based maintenance is preferred. However, condition based maintenance relies on the continuous and accurate gathering of condition-information of the particular machine and its various components.

This dissertation reports the experimental and numerical work performed as part of the development of an experimental facility that will allow the development of condition monitoring techniques for wind turbines. This work is focused on the torsional dynamics of a wind turbine setup. A physical setup, consisting of a 1.6 m diameter turbine, a 1:1.2 ̇ speed-ratio gearbox, and a 24 Volt direct current generator is built.

All of it is mounted within an open-return wind tunnel, which is also designed and built as part of this work. The following two cost-effective experimental techniques are used to measure the torsional natural frequencies: a shaft encoder tachometer from which instantaneous rotational frequency is obtained, and power signal analysis, where the generated voltage is recorded and analysed. It is shown how an algorithm developed by Diamond et al. (2016) is used for the shaft encoder geometry compensation. Frequency spectra based on Fourier transforms and short time Fourier transforms are used to identify harmonic frequencies. Both measurement techniques proves useful to identify not only natural frequencies of torsional vibration, but also various characteristic frequencies of the drivetrain such as shaft rotation, blade pass, gear mesh and generator armature. It is found that power signal analysis is more useful to identify the characteristic frequencies.

Torsional dynamics of the drivetrain and its components are also investigated with the following two numerical methods: an eight-degree-of-freedom torsional Lumped Mass Model (LMM), and a three-dimensional Finite Element Model (FEM). Torsional mode shapes and frequencies are calculated with both methods and a good agreement is found in the lower four modes. Numerical results are then compared with the experimental results, where there is also good agreement in the lower four modes. Model updating is performed on the FEM and by changing the torsional stiffness of the flexible couplings, the difference between measured and calculated natural frequencies are reduced to less than 6 %. It is concluded that future models should address lateral vibration of the drivetrain and the support structure.

From this study the following is contributed to the wind turbine condition monitoring field: considerations for the design and a working example of an experimental facility for investigating torsional dynamics, illustration of two measurement techniques, and two types of validated numerical models.

Supervisor:     Prof PS Heyns


JC van der Walt, 2018, A comparison between machine learning techniques to find leaks in pipe networks

In 2012, the National Non-Revenue Water assessment revealed that South Africa has 37% of non-revenue water. With the steadily growing demand for this scarce resource, the detection of leaks in pipe networks is becoming more important. Currently, in South Africa the primary method of detecting leaks is to install pressure management systems and monitoring minimum night time flows.

The pressure-flow deviation method, can be used to formulate an inverse analysis model based leak detection problem. This problem can then be solved using Artificial Neural Networks, Support Vector Machines and other optimization methods.

With EPANET, different networks were tested to compare these methods to finding leaks, using an inverse analysis formulated problem. Four different numerical networks were modeled and tested, a simple single pipe network, a small agricultural site, a distribution network and the simulated model of the experimental network that was designed and commissioned during the study in our laboratory.

From the numerical investigation, it was found that the optimization methods struggled to find solutions for simple networks with infinite number of solutions for the problem. For more complex numerical networks, it was seen that the Support Vector machine and the Artificial Neural Networks trained to the averages of their respective data sets.

Errors to ensure an accurate solution found by these algorithms were calculated as 2.6% for the numerical experimental network. The experimental network consisted of six possible leaking pipes, each having a length of 3m and a diameter of 10mm. Three leak cases were tested with diameters of 3mm and 2mm. Overall, the Support Vector machine could locate the leaking pipe with the best accuracy, while the minimizing of non-regularized error could calculate the size and location of the leak the most accurately.

Multiple leak cases were measured with the experimental network. The Support Vector machine was tested on these measurements, where it was found that two of the three leak cases could be solved with relative accuracies. Sensor usage optimization was completed on the measurements for the experimental network, where it was found that the leaks could be classified correctly with probabilities higher than 98% if only two sensors were used in the training of the SVM instead of all twelve.

Overall this method of leak detection shows promise for certain applications in the future. With practical applications on water distribution, transportation, and agricultural networks.

Supervisor:     Prof PS Heyns

Co-Supervisor:    Dr DN Wilke

AJJ Hayes, 2018, Characterisation of the core and winding vibrations of power transformers with regulator windings

This dissertation presents research and experimental work done to characterize the core and winding vibrations of power transformers with regulator windings by measuring the tank vibrations.

The experimental tests were performed in the manufacturing plant whilst the transformers under investigation were subjected to the standard factory acceptance tests, specifically the no load loss test (open circuit test) and the full load loss test (short circuit test). The vibration measuring sensors that were used included a laser Doppler vibrometer and a tri-axial accelerometer and the vibrations were recorded with a CoCo-80 data logger.

The test results show that the characteristics of the core and winding vibrations of transformers with and without regulator windings are very similar, but in the case of transformers with regulator windings, the winding vibrations have a few more dependencies.

Thus this research and experimental work provide key insights into how the core and winding vibrations of power transformers with regulator windings are influenced by the regulator windings, how the tank vibrations of transformers with regulator windings should be measured and the difference between the vibrations of transformers with and without regulator windings. The importance of this is that most of the research that has been done on transformer vibrations, have been done on transformers without regulator windings, but most practical transformers do have regulator windings. Thus there is a shortage of practical transformer vibration information, which this study aims to address.

Supervisor:    Prof PS Heyns

JF Grobler, 2018, "Multi-state hydro-pneumatic suspension system through the use of magneto-rheological (MR) valves"

This study is focused on modifying an existing solenoid valve based semi-active hydropneumatic spring-damper system using Magneto-Rheological (MR) fluid. The MR fluid’s effective viscosity can be altered by application of a magnetic field. Therefore, using a magnetic/MR valve makes it possible to change the state of the system by simply changing the applied magnetic field.

A prototype MR valve was developed to determine whether a unit small enough for installation was possible. This prototype valve was designed from first principles and properties such as pressure drop over the valve (damping) and flow blocking (for switching between spring characteristics) were measured. The measured pressure drop over the valve was higher than what was design for which was due to an incorrect assumption for the viscosity of the thixotropic MR Fluid. The flow blocking ability of the valve was determined by constant force tests. Results showed that the valve could virtually block the flow of fluid for approximately a quarter of the vehicles weight.

With the second prototype, the valve design and magnetic circuit design were improved. Two valves were constructed and implemented on a prototype suspension system. The damping characteristics of the system were lower than expected, however they can be improved by changing the valve geometry. The base spring characteristics are acceptable, however the higher spring characteristics fail when a high force is exerted on the strut that exceeds the valves flow blocking capability. The response time of the valve is not yet sufficient to make the system viable for real world implementation, especially under extreme conditions that can change more rapidly than the current valves.

Supervisor: Prof P.S. Els

L.M. Cramer, 2018, "Enhancement of the thermal performance of solar heat exchangers with porous inserts"

For high thermal performance and effectiveness, the flat plate heat exchangers and cooling channels are designed based on the three basic criteria: (i) small heat transfer area or large surface area to volume ratio, (ii) high heat transfer rate, and (iii) small pumping power. Numerous amounts of research have been dedicated to the notion of decreasing the size of a heat exchanger by enhancing the convective heat transfer inside the channels of a heat exchanger. Recently, the internal porous fins and porous foams of high thermal conductivity have gained considerable attentions in the research and development for their light weight, reduced fluid pumping power requirements, and high heat transfer characteristics. The results from the investigations show the enhancement of heat transfer coefficients and friction factors with the wavy screens relative to those in a smooth channel. This experimental research project aims to investigate the effects of the geometrical properties such the amplitude, period, and porosity of wavy porous mesh screen insert may have on the thermal performance of a heat exchanger and quantify the thermal performance of the channel employing the wavy porous screens for a wide range of applications at low to high Reynolds numbers. The friction factors, and heat transfer were measured in a rectangular channel when sinusoidal screen inserts were employed as turbulence promoters. The screen was made from porous mesh of flat metal screen available commercially. Two mesh screens were employed; one with a 68% porosity and one with a 48% porosity. Both mesh screens had a square shape pore and is delivered as a spool of material. The period of the screen was bent into the wavy mesh screen using a jig with two jaws. The screen wave vector was placed normal to the mean flow of the channel and allowed the peaks of the wave to make only line contact with the two larger side walls of the rectangular channel. The inlet Reynolds number for the experiments covered all three flow regimes: laminar, transition and turbulent. The measurements include the static pressure drop and wall temperature distributions along the channel. For the heat transfer experiments, the parallel walls of the channel touching the screen peaks are heated with a constant heat flux to simulate the channels in a flat plate heat exchanger. Heat transfer experiments were also obtained with one heated wall with a constant heat flux to simulate the conditions of a single channel heat exchanger employed in solar heaters and electronic cooling. Baseline data in a smooth channel without the screen inserts were also measured for comparisons with the data obtained in the same channel with the screen insert. The results on friction factors and heat transfer coefficients were then presented as ratios of data from the screen channel to the smooth channel to provide the performance of the screen channel relative to the smooth channel. The data and ratios were also presented in such a manner that the effect of change in porosity, period and amplitude of the screen insert could be studied. The sinusoidal screen inserts in the channels of a flat plate heat exchanger can provide desirable effects on the heat transfer enhancements (Nu/Nu0 > 1.0) only for the range of Reynolds number tested. The wire diameter of the mesh screen can significantly influence the thermal performance and pressure penalty provided by the wavy screen based on the present investigations and Mahmood et al. The present results are thus beneficial to the design of porous inserts for the heat exchangers operating over a wide range of flow rates. The effects of screen porosity and wave period are strong only on the efficiency index. The present results thus indicate the viability of the wavy porous inserts for the heat exchanger.

Supervisor:            Dr G. I. Mahmood

Co-supervisor:        Prof J. P. Meyer

RG du Toit, 2018, "A Stochastic Hybrid Blade Tip Timing Approach for the Identification and Classification of Turbomachine Blade Damage"

Industry is increasingly confronted by ageing turbomachines nearing their decommissioning dates. These turbomachines are especially prone to unexpected and catastrophic failure, which is often the consequence of rotating blade failures. Thus the failure of a single blade may result in immense safety and financial impacts. The aforementioned points raise further questions with regards to the optimal outage planning of turbomachines.  The distinct need to monitor the conditions of turbomachine blades during operation was therefore identified.  It was further identified that monitoring of the blade conditions should provide sufficient evidence as to when a blade damage threshold has been reached, therefore providing early warning of imminent blade failure.

Blade Tip Timing (BTT) has existed for many decades as an attractive vibration based condition monitoring technique for turbomachine blades. The technique is non-intrusive and online monitoring is possible. For these reasons, BTT may be regarded as a feasible technique to monitor the conditions of turbine blades. The processing of BTT data to find the associated vibration characteristics is, however, a non-trivial task. In addition, these vibration characteristics are difficult to validate, therefore resulting in questionable reliability of the various BTT techniques. The use of a hybrid blade condition monitoring approach is therefore proposed in this research project. This hybrid approach incorporates a stochastic Finite Element Model (FEM) modal analysis to supplement the BTT results, therefore creating a basis of comparison as the BTT results become available. The aim of this research is to test the ability of the proposed hybrid approach to perform the following processes: blade damage identification and damage classification.

The use of a new BTT technique based on Bayesian Linear Regression (BLR) was tested on an experimental setup, where the first bending mode of the blades in the rotor assembly were excited. BTT based on BLR assumes a Single-Degree of Freedom (SDOF) model to describe the blade tip displacements. The advantage of the BLR curve-fitting is that it solves for the parameters in this SDOF model as multivariate probabilistic quantities. More so, this technique solves these parameters for each revolution during the measured blade resonance conditions. The use of the multivariate probabilistic quantities in a Monte-Carlo Simulation (MCS) enables the amplitude and phase values of the blades to be derived (also as statistical quantities). The natural frequencies of the blades can then be determined by extracting features from the corresponding amplitude and phase results.

Incremental discrete damage was introduced to a particular blade to test the ability of the proposed technique to track the changes in the derived natural frequencies. Discrete damage was also introduced in the Finite Element Analysis (FEA). Slight variations in the material properties, operational conditions (centrifugal loads) and the geometry of the discrete damage were introduced in the FEA for each damage increment. This ensured that this analysis was stochastic rather than deterministic, thus enabling uncertainty to be modelled. The proposed damage identification requires that the change in natural frequencies of the BTT and FEA results as tracked as relative quantities. The changes in blade natural frequencies were inherently due to the increase in discrete blade damage. The effects of varying blade temperatures and how this would affect the performance of the proposed hybrid approach was also tested. Experimentally, this required the heating of the blades to the desired temperatures, before and during the BTT tests. The FEA incorporated temperature effects by modelling uncertainty in the material properties. The proposed BTT technique was able to detect the decrease in the natural frequencies of the blades due to the increase in temperature. More importantly, the hybrid approach demonstrated that it is general enough to still be applicable with regards to the relative natural frequency tracking (blade damage identification) with varying temperature effects. The relative changes in the natural frequencies from the undamaged, or reference states, of the relevant blades were computed to infer the degree discrete damage as part of a probabilistic damage identification procedure. A probabilistic damage threshold for the damage identification procedure was proposed based on the following question: What is the probability that the relative change in the natural frequency of the test blade is as large as what the FEM modal analysis {at a chosen discrete damage size) projected it to be? This probabilistic damage identification procedure was demonstrated for various scenarios, therefore demonstrating the ability of the hybrid approach to infer the degree of blade damage for various scenarios.

The blade damage classification process relies on the use of K-means clustering. The clustering implementation offers the advantage of a single BTT measurement being sufficient as an indication of blade specific conditions. The FEM natural frequency results were used to initialize cluster centroids and the individual BTT points were assigned to clusters with the closest centroid (based on the amplitude and natural frequencies). The classification of a point to a certain cluster thus provides an indication of the severity of the blade damage. This was done for the BTT tests with and without the effects of varying the blade temperatures. The accuracy of the damage classification implementation seems promising. The decision of whether a damage threshold has been reached for this implementation is purely based on the damage classification of the individual points.

The BTT methodology incorporating BLR proved to be reliable when used as part of a hybrid approach. Furthermore, the advantages of the stochastic nature of the hybrid approach are highlighted in terms of quantifying uncertainty. The proposed hybrid methodology demonstrates the ability to identify and classify blade damage. In doing so, it was possible to determine that a blade damage threshold had been reached. It was therefore shown that the proposed stochastic hybrid approach may offer many short- and long-term benefits for practical implementation. The proposed method therefore offers a feasible turbomachine blade monitoring solution that provides early warning of imminent blade failure.

Supervisor:     Prof PS Heyns

Co-Supervisor: Dr DH Diamond

JJF Wiid, 2018, The experimental and numerical investigation of the effect of shaft rotation on leakage rate of non-contacting seals found in turbine applications

This project was initiated by ESKOM power generation. ESKOM loses up to 22% of their steam energy in the HP turbines due to leakage at the turbine labyrinth seals. Therefore the need was expressed to implement improved sealing configurations. The aim of this study is to investigate the effect that shaft rotation has on the leakage rate of labyrinth and brush seals. This is done by means of experimental and numerical methods.

For many decades it was assumed that the shaft rotation has no or little effect on seal performance and therefore it was neglected in seal design. It was decided to investigate this subject, in order to assist and improve in future seal design and operation.

Both labyrinth and brush seals were investigated experimentally on a test rig. A real life application of the labyrinth or brush seals can be found in the power generation industry where a turbine shaft has a diameter of 300 mm and rotates at 3 000 rpm. The test rig was designed to assist in this application. Therefor the test rig had a shaft diameter of 150 mm with shaft speeds ranging between 0-10 000 rpm and with five different upstream pressures ranging from 1-5 bar. The same seals were then simulated using the commercial Computational Fluid Dynamics (CFD) package STAR-CCM+ with the bristle pack of the brush seal modelled as a porous medium. The coefficients of resistance for the porous medium were experimentally obtained. The two investigation methods are compared for the labyrinth and brush seals. The labyrinth and brush seals are also compared against each other.

The results show that the experimental leakage rates have a good correlation with those predicted by CFD. The CFD simulation provided detailed leakage flow fields and pressure distributions of both seals. It was found that shaft rotation has an influence on the leakage rate of both seals. The leakage rate increased at higher shaft speeds, with the brush seal performing better than the labyrinth seal. An increase of up to 1.7% was found at 10 000 rpm for the labyrinth seal and 1.45% for the brush seal at 10 000 rpm.

Guidelines were created based on these results to assist with advanced seal design. It is recommended that these guidelines are used in future seal design and other research aspects of non-contacting seals in turbo machinery.

Supervisor:    Prof. K.J. Craig

Co-Supervisor:    Dr. C.J.H. Thiart

HJ Breedt, 2018 , Atmospheric Boundary Layer Stability and its Application to Computational Fluid Dynamics

In the wind resource and wind turbine suitability industry Computational Fluid Dynamics has gained widespread use to model the airflow at proposed wind farm locations. These models typically focus on the neutrally stratified surface layer and ignore physical process such as buoyancy and the Coriolis force. These physical processes are integral to the accurate description of the atmospheric boundary layer and reductions in uncertainties of turbine suitability and power production calculations can be achieved if these processes are included. The present work focuses on atmospheric flows in which atmospheric stability and the Coriolis force are included.

The study uses Monin-Obukhov Similarity Theory to analyse time series data output from a proposed wind farm location to determine the prevalence and impact of stability at the location. The output provides the necessary site data required for the CFD model as well as stability-dependent wind profiles from measurements. The results show non-neutral stratification to be the dominant condition onsite with impactful windfield changes between stability conditions.

The wind flows considered in this work are classified as high Reynolds number flows and are based on numerical solutions of the Reynolds-Averaged Navier-Stokes equations. A two-equation closure method for turbulence based on the k-ε turbulence model is utilized. Modifications are introduced to standard CFD model equations to account for the impact of atmospheric stability and ground roughness effects. The modifications are introduced by User Defined Functions that describe the profiles, source terms and wall functions required for the ABL CFD model. Two MOST models and two wall-function methods are investigated.

The modifications are successfully validated using the horizontal homogeneity test in which the modifications are proved to be in equilibrium by the model's ability to maintain inlet profiles of velocity and turbulence in an empty domain. The ABL model is applied to the complex terrain of the proposed wind farm location used in the data analysis study. The inputs required for the stability modifications are generated using the available measured data. Mesoscale data are used to describe the inlet boundary conditions. The model is successfully validated by cross prediction of the stability-dependent wind velocity profiles between the two onsite masts.

The advantage of the developed model is the applicability into standard wind industry loading and power production calculations using outputs from typical onsite measurement campaigns. The model is tuning-free and the site-specific modifications are input directly into the developed User Defined Functions. In summary, the results show that the implemented modifications and developed methods are applicable and reproduce the main wind flow characteristics in neutral and non-neutral flows over complex wind farm terrains. In additions, the developed method reduce modelling uncertainties compared against models and measurements that neglect non-neutral stratification.

Supervisor: Prof KJ. Craig

V.H.Wehrmeyer, 2018, Untripped Manoeuvre Induced Rollover Prevention for Sport Utility Vehicles

Rollover accidents account for a high number of serious injuries and fatalities and thus it is greatly important to reduce the number of occurrences. Although a large number of rollovers result from factors external to the vehicle design such as environmental obstacles there is a significant portion of rollover accidents which are preventable. On-road untripped rollovers are directly related to the vehicle design. It is possible to do work in this area to improve vehicle-related safety factors.

During rollover, lateral acceleration acts on the centre of gravity of the vehicle over turning it about the outer wheels. Thus the method to reduce or prevent rollover of this study stemmed from decreasing the overturning moment by reducing the movement arm though which the lateral acceleration acts. This is achieved by lower the ride height of the test vehicle using slow active suspension control of the test vehicle (Land Rover Defender 110) fitted with a hydro-pneumatic suspension system.

An experimental validated mathematical model representing the test vehicle is created to develop a rollover prevention control system that reduces the vehicle’s ride height and reduces the propensity to rollover. The control system applies one of three discrete suspension settings depending on the severity of the manoeuvre as well as lowering the ride height. The model is used to simulate the Fishhook 1B and the ISO 3888 Double Lane Change manoeuvres to evaluate the roll prevention system.

The rollover prevention control system improved the two wheel lift off speed of the vehicle through a Fishhook 1 B manoeuvre by 64% and the body roll angle of the vehicle through the Double Lane Change manoeuvre by 13% and the body roll rate by 25.7%. The rollover prevention control system significantly improved the vehicle’s response with regard to smooth flat on-road untripped rollover. Further improvements could be possible with the use of the proposed rollover prevention control system in conjunction with a fully active suspension system allowing for faster corrective action.

Supervisor:    Prof. P.S. Els

T.S. Mokobodi, 2018,"Designing and developing a free fall absolute gravity measuring system, using pneumatic actuators"

A gravimeter is an instrument that measures gravitational force Fg (N) or acceleration g (m/s-2). Absolute gravity measurements are preferred in metrology, due to the shortest traceability links to the SI base units of length and time, realising acceleration. The investigation on the suitable method of gravity measurement was performed at the National Metrology Institute of South Africa (NMISA). The free fall gravity measuring system was adopted for development.

The metrological need on redefining the kilogramme standard using the watt balance, supported the decision to mandate this project. Free fall gravimeters were researched. The new concept of fully pneumatic controlled vacuum chamber was invented and manufactured. Pneumatic actuators were used on the vacuum chamber to align, launch, capture and reposition the falling test mass. Laser interferometer and high-speed digitiser with embedded accurate clock module, were used in realising displacement and time, through numerical computations. Using stabilised He-Ne Laser red with wavelength =633 nm interferometer, free-falling test mass displacements were traceable to length standard.

Interference intensity signal produced from experimental free fall drops were converted to A digital voltage signal enabling processing. Post signal processing algorithms were applied to the signal to extract the displacement and time coordinates of the free-falling object, using a zero-crossing detection method in a LabView environment. The final prototype setup measured the value 9.786041 m/s2 with uncertainty of 0.0000705 m/s2 at the vacuum pressure of 0.05 Pa. It was validated and compared with the Council of Geophysics’ measured value of the site of 9.7860985 m/s2.

Supervisor:        Prof. Nico Theron

Co-supervisor:        Mr Pieter Greeff


G.S.Heymans, 2018, "Design and development of a responsive magneto-rheological (MR) equipped semi-active suspension system for off-road vehicles"

The aim of this study is to design, implement and investigate the use of a Magneto-Rheological (MR) equipped Hydro-Pneumatic suspension system to solve the ride versus handling compromise of off-road vehicles. This suspension technology makes use of MR fluid viscosity changes which are induced by a varying magnetic field which serves as the basis for changing the suspension system’s damping as well as the stiffness characteristics.  The primary focus of the study aimed to improve the response time characteristics of an existing prototype suspension system developed at the University of Pretoria. This improvement would be achieved through comprehensive magnetic optimisation of the MR Valve which characterises and controls the system. Additionally, work was done to understand and model the complex physical conditions which create the output characteristics of the semi-active suspension system and ultimately to understand how they will influence a vehicle during dynamic situations. This modelling platform was developed to account for the interrelated non-linear elements within the system as well as capture the MR valve characteristics observed experimentally. Further quarter-car simulation-based experiments where used to determine the feasibility and future contribution of this technology platform to solve ride verses handling compromise of off-road vehicles.

Supervisor:        Prof P.S. Els

Co-supervisor:        Dr. T.R. Botha


R.E. Goliada, 2018, "An Investigation on Inefficiencies in Spare Part Management Processes in South African Power Plants"

Inefficiencies in spare parts management can, and do, have great repercussions for the execution of maintenance in Eskom power stations. Despite this, the sub-processes of spare parts management in power stations had not yet been subjected to rigorous analysis to identify inefficiencies. Consequently, the purpose of this research was to establish the inefficiencies that exist in the management of spare parts in South African power stations, and also to determine the causes of the inefficiencies. All this was done with the intent of recommending a solution to improve spare parts management. It was for this reason that this research was conducted in 13 Eskom coal-fired power stations in South Africa. The first phase of the research was concerned with uncovering and documenting the spare parts management model presently used in Eskom. The second phase included the use of a modified process failure mode and effect analysis (PFMEA) and the Delphi method to identify inefficiencies, their sources, and their significance. The research found that the inefficiencies included unsuitable maintenance and inventory management strategies, as a result of inadequate analysis of plant history. Furthermore, the study established that inadequate analysis of history was a result of poor maintenance records, incompetence, and the lack of access to computerised maintenance management systems (CMMSs). With the inefficiencies and their causes identified, the third phase of the research was then devoted to developing a methodology for improving spare parts management. This led to the development of a framework for improving spare parts management practices in power stations. The framework was validated and verified through a Delphi process. This framework was then recommended for adoption in an Eskom standard procedure for improving spare parts management practices. The research was thus successful in recommending a solution to improve the operational effectiveness and efficiency of spare parts management in South African power stations.

Supervisor: Professor J.L. Coetzee

Published by Bradley Bock

Copyright © University of Pretoria 2021. All rights reserved.

FAQ's Email Us Virtual Campus Share Cookie Preferences