2018

S. Schmidt, 2018, "Development of effective gearbox fault diagnosis methodologies utilising
various levels of prior knowledge"

Effective fault diagnosis techniques are important to ensure that expensive assets such as wind turbines can operate reliably. Vibration condition monitoring data are rich with information pertaining to the dynamics of the rotating machines and are therefore popular for rotating machine diagnostics. However, vibration data do not only contain diagnostic information, but operating condition information as well. The performance of many conventional fault diagnosis techniques is impeded by inherent varying operating conditions  ncountered in machines such as wind turbines and draglines. Hence, it is not only important to utilise fault diagnosis techniques that are sensitive to faults, but the techniques should also be robust to changes in operating conditions.

Much research has been conducted to address the many facets of gearbox fault diagnosis e.g. understanding the interactions of the components, the characteristics of the vibration signals and the development of good vibration analysis techniques. The aforementioned knowledge, as well as the availability of historical data, are regarded as prior knowledge (i.e. information that is available before inferring the condition of the machine) in this thesis.

The available prior knowledge can be utilised to ensure that effective gearbox fault diagnosis techniques are designed. Therefore, methodologies are proposed in this work which can utilise the available prior  knowledge to effectively perform fault diagnosis, i.e. detection, localisation and trending, under varying operating conditions. It is necessary to design different methodologies to accommodate the different kinds of historical data (e.g. healthy historical data or historical fault data) that can be encountered and the different signal analysis techniques that can be used.

More specifically, a methodology is developed to automatically detect localised gear damage under varying operating conditions without any historical data being available. The success of the methodology is attributed to the fact that the interaction between gear teeth in a similar condition results in data being generated which are statistically similar and this prior knowledge may be utilised. Therefore, a dissimilarity
measure between the probability density functions of two teeth can be used to detect a gear tooth with  localised gear damage.

Three methodologies are also developed to utilise the available historical data from a healthy machine for gearbox fault diagnosis. Firstly, discrepancy analysis, a powerful novelty detection technique which has been used for gear diagnostics under varying operating conditions, is extended for bearing diagnostics under varying operating conditions. The suitability of time-frequency analysis techniques and different models are compared for discrepancy analysis as well. Secondly, a methodology is developed where the spectral coherence, a powerful second-order cyclostationary technique, is supplemented with healthy historical data for fault detection, localisation and trending. Lastly, a methodology is proposed which utilises narrowband feature extraction methods such as the kurtogram to extract a signal rich with novel
information from a vibration signal. This is performed by attenuating the historical information in the signal. Sophisticated signal analysis techniques such as the squared envelope spectrum and the spectral coherence are also used on the novel signal to highlight the benefits of utilising the novel signal as opposed to raw vibration signal for fault diagnosis.

Even though a healthy state is the desired operating condition of rotating machines, fault data will become available during the operational life of the machine. Therefore, a methodology, centred around discrepancy analysis, is developed to utilise the available historical fault data and to accommodate fault data becoming available during the operation of the machine. In this investigation, it is recognised that the machine condition monitoring problem is in fact an open set recognition problem with continuous transitions between the healthy machine condition and the failure conditions. This is explicitly incorporated into the methodology and used to infer the condition of the gearbox in an open set recognition framework. This methodology
uses a different approach to the conventional supervised machine learning techniques found in the literature.

The methodologies are investigated on numerical and experimental datasets generated under varying operating conditions. The results indicate the benefits of incorporating prior knowledge into the fault diagnosis process: the fault diagnosis techniques can be more robust to varying operating conditions, more sensitive to damage and easier to interpret by a non-expert. In summary, fault diagnosis techniques are more effective when prior knowledge is utilised.

Supervisor: Prof. P.S. Heyns

---

H.O. Omoregbee, 2018, "Diagnosis and prognosis of rolling element bearings at low speeds and varying load conditions using higher order statistics and artificial intelligence"

Condition monitoring (CM) is commonly used in determining the operational states and health of rotating machines. The rise in the complexity of modern machines have led to advances in CM technologies to increase and improve product reliability and reduced downtime. Vibration and acoustic emission signals generated by these modern and complex machines are often immersed in background noise making it difficult to detect faults. Extracting signal features that are sensitive to faults still attracts considerable attention to detect and identify faults in rotating machines. This is especially so for low speed machinery
under varying load and speed conditions.

Such conditions are found in many industrial applications and include draglines in the mining industry and large rolling mills in many materials processing environments. Condition monitoring techniques for  stationary systems are inadequate at accurately detecting and diagnosing faults under such conditions. While skewness and kurtosis have been used extensively in the condition monitoring of bearings and gears, higher order statistical (HOS) techniques have not found wide application in machine condition monitoring. This is because if a process is Gaussian then HOS provide no additional information that can be obtained
from the second or higher order statistics.

HOS techniques like the 6 th order statistical moment (hyperflatness) could play an important role in its condition monitoring (CM). By applying this method, processing of acoustic emission signal at low speed and varying load condition could be very useful by providing details about the signal which the conventional second order statistics cannot. HOS techniques are extensions of the better-known concepts of correlation (in time or space) power spectra. Higher order spectra are higher order Fourier spectral representations of third and higher order correlations or moments.

Other empirical models which could also be used include the fields of regression and classification which are also collectively referred to as supervised learning and are dependent on its construction or optimization on large sets of representative data. In classifying faults in rolling element bearings (REBs) at low speeds and under varying load conditions, support vector machines for regression and genetic algorithm (SVMGA) which is a supervised machine learning algorithm can be used for its classification or regression problems. The Bayesian Robust New Hidden Markov Model (BRNHMM) was also used here to diagnose faults of two different categories: debris induced fault on roller bearing and a fault induced on the outer race of a roller bearing.

Fault prognosis was achieved with the use of eXtended Takagi-Sugeno (xTS) fuzzy and recursive least square algorithm (exTSFRLSA) and support vector data descriptive (SVDD) method in this work. The (exTSFRLSA) has many applications of which one is the prediction of the sequence of state change, based on the sequence of observations. SVDD belong to the statistical learning theory class which is used here in this work to show the remaining useful life (RUL) of the bearing under study.

Supervisor: Prof PS Heyns

---

M.J. Stallmann, 2018 "Cost effective FTire parameterisation methods for ride simulations with large off-the-road tyres"

All forces acting on a road vehicle are either generated in the tyre-road interface or are due to aerodynamic effects, where at low speeds the latter can be ignored. The accuracy of the tyre model, describing the forces at the tyre-road interface, is thus of utmost importance in vehicle dynamics simulations. Accurate tyre models are essential to ensure that vehicle dynamics models are an accurate representation of the actual vehicle so that it can be used with confidence to study and improve the vehicle’s safety, durability, ride comfort and handling capabilities.

Various tyre models have been developed to describe the forces and moments that are generated in the tyre contact patch. FTire is one of the well-known and widely used nonlinear physics-based models that has been developed for vehicle comfort and handling simulations and the prediction of road loads due to short wave-lengths obstacles. The parameterisation of the FTire model is done by extracting model parameters from experimental results that cover the operational conditions of the tyre. Acquiring these parameterisation datasets, with an acceptable accuracy, is however a challenge. This is especially true for large off-the-road
(OTR) tyres. Expensive test equipment is typically required to conduct these tests for passenger car tyres but virtually no test equipment is available to test OTR tyres. Due to a lack of standardisation, the test equipment is often custom built and adapted to the required test conditions. The situation is further complicated due to the large range of possible tyre perating and loading conditions of OTR tyres, as well as the differences in tyre and rim sizes, that the test equipment should be able to accommodate. Alternative methods of obtaining parameterisation data for large OTR tyres thus need to be investigated. This thesis
describes two alternative methods that can be used to parameterise an FTire model of a large OTR tyre. Parameterisation methods were investigated that can be applied irrespective of the tyre size and operating load.

A baseline parameterisation process, using an experimental data set, was performed. The obtained FTire model was validated against experimental results. The validation process showed that the parameterised model can be used to accurately predict the tyre forces and moments, of large OTR tyres, while driving over uneven terrain. This method can thus successfully be used if the tyre forces and moments, and the applied boundary, as well as environmental conditions, can accurately be measured. Conducting these asurements was however found to be challenging, limited by the available test equipment and problematic to expand to higher loads and larger tyres.

One alternative, to experimentally obtained data for the parameterisation process, is to use finite element models to generate the relevant parameterisation data. The required input data of a finite element tyre model can be obtained with the same effort irrespective of the tyre size or operating condition. Geometric and material tests were conducted, and a nonlinear finite element model was created and validated against experimental test. The finite element tyre model was used to replicate the parameterisation tests that are required to parameterise a FTire model without tuning any model parameters. A FTire parameterisation was
conducted using FEA data and the model was validated.

A second alternative is the use of carcass deformation results to parameterise FTire models. The carcass deformation data can be obtained from experimental testing or from FE simulation results. To obtain the measured carcass deformation data, a set of cameras were mounted to the inside of the test tyre. The stereo vision principle is applied to determine the deformation of the tyre during static tyre tests. Initial results of using the carcass deformation data to parameterise a FTire model is also presented. Carcass deformation measurements can further be used to validate FE tyre models if force and moment  easurements are limited or non-existent to improve the confidence in the model.

The proposed methods create new possibilities to parameterise FTire models of large off- the-road tyres. These methods are especially useful if experimental results of the tyre are limited or non-existent. This thesis presents results that show that these methods can be used successfully to extract model parameters. More research is however required to extend these capabilities and to standardize these parameterisation methods.

Supervisor: Prof. P.S. Els

---

S. Matadin, 2018 "High Agility 6 Degree OF Freedom Hybrid Serial-Parallel Kinematic Robotic Arm"

Robotic manipulators are the cornerstone of modern automated manufacturing production lines. They
are also deployed in specialised applications where they offer advantages over the traditional human
workforce. Robotic manipulators most commonly take the form of serial kinematic machines, also
known as robotic arms due to the advantages they offer in terms of reach, large working envelopes,
agility and the ability to manipulate large payloads with large masses. This architecture is prone to
overall poor dynamic performance at high velocities and accelerations in the form of unwanted
vibrations.

The parallel kinematic machine architecture, although stiffer and more stable at high velocities and
accelerations, has the disadvantage of being less agile and offering a relatively small working
envelope. It is thus clear that each architecture offers advantages and disadvantage and in order to
benefit from the advantages while minimising the disadvantages a serial-parallel hybrid architecture
will need to be developed and assessed. A prominent design feature in a serial kinematic machine is
the distribution of actuators along the structure of the system. These actuators move in space as the
system moves. In contrast, parallel kinematic machines are commonly designed to have actuators
located at the base of the system and remain stationary as the system moves.

It was therefore hypothesised that if a hybrid system was developed it should take the form of a serial
kinematic machine in terms of the link architecture while housing all the actuators that power these
links at the base of the system. This hybrid system would aim at realising the advantages of both
traditional kinematic machines while minimising the inherent disadvantages of each. The focus of this
study was to propose a concept for a hybrid manipulator architecture and study the kinematic
performance of such a system.

Supervisor: Prof. N. Theron

---

J. Crous, 2018 "Development of an Aeroprediction Method for Slender Bodies Including Aeroelastic Effects Using Euler-Based Local Piston Theory "

Accelerated Fatigue Testing (AFT) is the process of performing fatigue testing in order to ascertain the
characteristics of a system’s degradation when subjected to repeated events that induce fatigue. In this
work the focus is on extreme events in mechanical systems which induce high levels of fatigue. In this
work the mechanical systems considered are automotive-suspension models. Due to the nature of these
events the system’s behaviour is highly nonlinear. To conduct an accurate fatigue test on a system the
measured high fatigue incidents must be reproduced in the system in a laboratory environment. This is
typically done using servo hydraulic actuators. The amount of data that can be harvested, however, is
limited since the testing procedure induces fatigue which compromises the integrity of the AFT test.
Performing AFT can be broken down into three steps: sampling the system to harvest data, then
nonlinear system identification is used to construct a plant model of the system which is then finally
used to perform control of the system to trace the target response. This work focuses specifically on the
first two items: sampling and nonlinear system identification. In this work an alternative approach to
nonlinear system identification is developed based on statistical learning techniques. The approach aims
to identify important events in the input and output spaces that characterise the system’s response and
construct a mapping between these events by using a high dimensional regression technique developed
in this work. This identified regression technique used is Multivariate Principal Component Regression
(MPCR).

Next, a sampling procedure is developed to find data that are important in order to model the target
responses. Therefore, in this work the interface between the sampling and system identification parts of
the AFT procedure is investigated. New approaches in terms of sampling and system identification are
developed, implemented and compared with current techniques as far as possible. The alternative
approach to AFT developed in this work is shown to produce more accurate models and leads to a more
accurate AFT test with higher integrity.

Supervisor : Prof. S. Kok
Co-supervisor: Prof P.S. Heyns

---

M.C. Meijer, 2018 "Development of an Aeroprediction Method for Slender Bodies Including Aeroelastic Effects Using Euler-Based Local Piston Theory "

Euler-based local piston theory (LPT) has received significant interest in recent literature. The method utilizes a simple, algebraic relation to predict between perturbation pressures directly from local surface  eformation and from the local fluid conditions obtained from a steady Euler solution. Early applications of Euler-based LPT to simple, non-interfering geometries and flows yielded the accurate (<5% error) prediction of unsteady pressures at orders-of-magnitude lower computational expense compared to unsteady CFD. These successes led to the broader application of Euler-based LPT to more complex scenarios, such as full-vehicle geometries and interference flows. However, a degradation in the prediction accuracy was noted. This motivated the present work, in which the suitability of Euler-based LPT as an aeroprediction method for slender bodies with aeroelastic effects is assessed.

An extensive and thorough review of the literature revealed that no investigation into higher-order terms in the pressure equation of LPT had been made. More significantly, the mathematical basis for LPT had yet been developed. Finally, no controlled numerical investigation into the application of Euler-based LPT under aerodynamic interference associated with cruciform control surfaces on slender bodies could be found in the literature. 

The present work addresses the above gaps in the literature. The first is addressed analytically, and shows that second-order LPT provides a non-zero contribution to aerodynamic stiffness. To address the second gap, a derivation of LPT from the 3D unsteady Euler equations is presented, with an in-depth discussion of the required assumptions. A number of significant conclusions regarding the validity of Euler-based LPT are drawn. It is argued that the method will be in significant error when applied in regions involving, amongst others, viscous boundary-layers, concentrated vorticity, transonic or embedded subsonic flows, sharp curvature, wing-body junctions, subsonic leading-edges, wing-tips, and trailing-edges. Furthermore, it is argued that Euler-based LPT will be in error when applied to mode-shapes of deformation involving localized bending and camber or point-local deformations. Finally, it is stressed that an algebraic pressure equation in LPT cannot account for flowfield interaction, which may be significant in the aforementioned scenarios. These conclusions are supported by a numerical investigation performed in the present work, which addresses the third gap in the literature.

Supervisor: Prof. L. Dala

 

I. F.  Okafor, 2018 "Influence of Circumferential Spans of Heat Flux Distributions on Secondary Flow, Heat Transfer and Friction Factors for a Linear Focusing Solar Collector Type Absorber Tube "

Solar collector absorber tubes play a critical role in converting incident solar heat flux into absorbed thermal energy and transferring it to a heat transfer fluid. In this study a single horizontally orientated absorber tube was investigated numerically in terms of the influence of different circumferential spans of symmetrical and asymmetrical heat flux distributions on buoyancy-driven secondary flow, internal heat transfer and friction factor characteristics. Three types of circumferential heat flux boundaries were considered, namely fully uniform, partial uniform and sinusoidal non-uniform heat flux distributions. Both gravitational symmetry and asymmetry for non-uniform heat flux distributions were investigated to cover symmetry angles in terms of the gravitational field (g) of 0° (symmetrical case), 20°, 30°, 40° and 60°. 

Different sized stainless steel absorber tubes having a length of 10 m, and inner diameters of  62.7 mm, 52.5 mm, 40.9 mm and 35.1 mm were considered. Three dimensional steady-state simulations were performed for water as working fluid, covering laminar flow Reynolds numbers ranging from 130 to 2200, as well as for turbulent flow Reynolds numbers ranging from 3030 to 202 600. Buoyancy effects, temperature dependent fluid thermal properties, tube-wall heat conduction and the external wall heat losses by convection and radiation were taken into consideration. Average internal heat transfer coefficients, local internal heat transfer coefficients, Richardson numbers and overall friction factors were obtained for different angular spans of incident heat flux, inlet fluid temperatures, heat flux intensities and outer wall thermal conditions

Laminar flow results indicated that the angular span, annular position, and intensity of the applied external heat flux all have significant influences on the buoyancy induced mixed convection inside the tube. This resulted in significant variations in the internal heat transfer coefficients and the friction factor which are not well described by classical empirical correlations. Buoyancy induced secondary flow significantly enhanced the internal heat transfer coefficient and significantly increased the friction factor compared to forced convection cases. Higher heat transfer coefficients and friction factors were obtained for non-uniform heat flux distributions compared to uniform heat flux distributions and were found to be dependent on the angle span and position of the heat flux. Higher inlet temperatures resulted in increased Nusselt numbers and lower friction factors, while higher external heat loss resulted in lower Nusselt numbers and lower friction factors. An increase in the asymmetry of the heat flux distribution resulted in a reduction of the Nusselt number and friction factor.

Even though turbulent flow cases with a Reynolds number range of approximately 3000 to 9000 were also influenced by buoyancy driven secondary flow, and followed the same parameter trends, it occurred to a lesser extent compared to the laminar flow cases. Turbulent flow cases with Reynolds numbers higher than 9100, exhibited little dependence on secondary flow effects and indicates the suitability of classical fully uniform heat flux heat transfer and friction factor correlations for highly turbulent flow irrespective of the distribution or intensity of the heat flux.

Supervisor:      Prof. J. Dirker

Co-supervisor: Prof. J. P. Meyer

---

D.D. Ndenguma, 2018 "Heat transfer and pressure drop in annuli with non-uniform internal wall temperatures in the transition flow regime"

It is common practice to design heat exchangers that are operated in either the laminar or the turbulent flow regime and not in the transition flow regime. This could mainly be due to a lack of knowledge on the behaviour in the transition flow regime for several reasons. However, due to factors such as design constraints, heat exchangers may indeed operate in the transition flow regime.

An experimental study was conducted to determine the lower and upper Reynolds number limits of the transition flow regime, and the characteristics of the heat transfer coefficients and friction factors for annular passages with different geometric dimensions. The inner wall of the annular passage had different degrees of non-uniform temperature, while the outer wall was insulated. Both heated and cooled flow applications with water as fluid (cold fluid and hot fluid respectively) were investigated. The isothermal condition investigation was also conducted for pressure drop. Four horizontal concentric counter-flow tube-in-tube heat exchangers with conventional inlet geometries were considered to obtain the required data. The hydraulic diameters of the test sections were 26.2 mm, 23 mm, 20.2 mm and 17 mm, their respective annular diameter ratios were 0.327, 0.409, 0.386 and 0.483 and their length-to-hydraulic ratios were 193, 221, 251 and 299 respectively. The flow was both hydrodynamic and thermally developing. Test data of laminar, transition and turbulent flow regimes was collected. However, the transition flow regime was the main area of interest for this study. The transition flow regime was found to exhibit either mixed or forced convection types. Average heat transfer coefficients were obtained for both heating and cooling cases, while friction factors were obtained for heating, cooling and isothermal conditions. Uncertainties in the friction factor and Nusselt number were on average below 5.6% and 10.5%, respectively.

The geometric size of the annular passage, degree of wall temperature uniformity and direction of the heat flux (heating and cooling cases of annular fluid) had a significant influence on the heat transfer coefficients, friction factors and Reynolds number span of the transition flow regime. New correlations for predicting the transition flow regime Reynolds number spans, Nusselt numbers and friction factors were developed for the transition flow regime and predicted most of the data to within ±10%.

Supervisor:         Associate Professor, J. Dirker

Co-Supervisor: Professor, J.P. Meyer (co-supervisor)

---

M. Everts,2018  "Single-phase mixed convection of developing and fully developed flow in smooth horizontal tubes in the laminar, transitional, quasi-turbulent and turbulent flow regimes"

The laminar and turbulent flow regimes have been extensively investigated from as early as 1883, and research has been devoted to the transitional flow regime since the 1990s.  However, there are several gaps in the mixed convection literature, especially when the flow is still developing.  The purpose of the study was to experimentally investigate the heat transfer and pressure drop characteristics of developing and fully developed flow of low Prandtl number fluids in smooth horizontal tubes for forced and mixed convection conditions. 

An experimental set-up was designed and built, and results were validated against literature.  Two smooth circular test sections with inner diameters of 4 mm and 11.5 mm were used, and the maximum length-to-diameter ratios were 1 373 and 872 respectively.  Heat transfer measurements were taken at Reynolds numbers between 500 and 10 000 at different constant heat fluxes.  A total of 648 mass flow rate measurements, 70 301 temperature measurements and 2 536 pressure drop measurements were taken.  Water was used as the test fluid and the Prandtl number ranged between 3 and 7. 

It was found that a longer thermal entrance length was required for simultaneously hydrodynamically and thermally developing flow. Therefore, a coefficient of at least 0.12 (and not 0.05 as advised in most literature) was suggested.   Because free convection effects decreased the thermal entrance length, correlations were also developed to calculate the thermal entrance length for mixed convection conditions.  The boundaries between the flow regimes were defined mathematically, and terminology to define transitional flow characteristics was presented.  For laminar flow, three different regions (forced convection developing, mixed convection developing and fully developed) were identified in the local heat transfer results and nomenclature and correlations were developed to define and quantify the boundaries of these regions.   Correlations were also developed to calculate the local and average laminar Nusselt numbers of mixed convection developing flow.  The laminar-turbulent transition along the tube length occurred faster with increasing Reynolds number, and was also influenced by free convection effects.  As free convection effects became significant, the effect was first to disrupt the fluctuations inside the test section, leading to a slower laminar-turbulent transition along the tube length compared with forced convection conditions.  However, as free convection effects were increased, the fluctuations inside the test section increased and caused the laminar-turbulent transition along the tube length to occur faster. 

The Reynolds number at which transition started was found to be independent of axial position for both developing and fully developed flow. However, the end of transition occurred earlier as the flow approached fully developed flow.  When the flow was fully developed, the end of transition became independent of axial position.  Furthermore, free convection effects affected both the start and end of the transitional flow regime, and caused the Reynolds number range of the transitional flow regime to decrease.  Correlations were therefore developed to determine the start and end of the transitional flow regime for developing and fully developed flow in mixed convection conditions.  The transitional flow regime across the tube length was divided into three regions.  In the first region, the width of the transitional flow regime decreased significantly with axial position as the thermal boundary layer thickness increased, and free convection effects were negligible.  In Region 2, the width of the transitional flow regime decreased with axial position, due to the development of the thermal boundary layer, as well as with increasing free convection effects.  In the fully developed region (Region 3), the width of the transitional flow regime was independent of axial position, but decreased significantly with increasing free convection effects.  At high Grashof numbers, free convection effects even caused the transitional flow regime of fully developed flow to become negligible. 

It was found that the boundaries of the different flow regimes were the same for pressure drop and heat transfer, and a relationship between pressure drop and heat transfer existed in all four flow regimes.  In the laminar flow regime, this relationship was a function of Grashof number (thus free convection effects), while it was a function of Reynolds number in the other three flow regimes.  Correlations to predict the average Nusselt numbers, as well as the friction factors as a function of average Nusselt number, for developing and fully developed flow in all flow regimes were developed. 

Finally, flow regime maps were developed to predict the convection flow regime for developing and fully developed flow for a wide range of tube diameters and Prandtl numbers, and these flow regime maps were unique for four reasons.  Firstly, they contained contour lines that showed the Nusselt number enhancements due to the free convection effects.  Secondly, they were valid for a wide range of tube diameters and Prandtl numbers.  Thirdly, the flow regime maps were developed as a function of temperature difference (Grashof number) and heat flux (modified Grashof number).  Finally, four of the six flow regime maps were not only valid for fully developed flow, but also for developing flow.

Supervisor:  Prof J.P. Meyer