Masters Degrees (Electrical and Electronic Engineering)

Permanent URI for this collection

https://scholar.sun.ac.za/handle/10019.1/766

Browse

Now showing 1 - 7 of 7

Automated remote industrial inspection platform using spot
(Stellenbosch : Stellenbosch University, 2023-03) Roux, Dominic; Booysen, Thinus; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.
ENGLISH ABSTRACT: In an industrial environment, good housekeeping practices are essential to ensuring safety and efficiency on site. In this context, housekeeping refers to keeping an industrial site clean of loose equipment and debris, which present slip, trip and fall hazards. Managing these hazards often falls upon highly skilled managerial staff, with numerous other urgent responsibilities. As a result, the laborious and time-consuming discipline of regular housekeeping inspections is easily neglected. Enabled by advancements in mobile robotics, autonomous legged robots are increasingly applied to the automation of industrial inspections. However, the automation of on-site housekeeping is an unexplored area in this field of research. In this project, an automated remote inspection platform is developed for the novel application of automating housekeeping inspections in a mining processing plant. It uses a Boston Dynamics Spot for autonomous data collection along a pre-determined inspection route. To enable data collection and transfer from the robot, a custom Raspberry Pi payload is developed that facilitates data transfer from the robot to a GPU-enabled computation endpoint. The remote inspection platform implements a vision-based hazard detection and reporting system, based on a Mask R-CNN computer vision model. While real-time performance is desirable, it is not possible with cloud computing in lieu of dedicated on-site computation. Experimental analysis revealed a 0.442 FPS lower bound on the system performance with cloud computing, due to a significant network overhead. The system uses a novel hazard risk estimation algorithm to classify detected hazards as high or low risk. It evaluates hazard detection and walkway segmentation masks generated by the Mask R-CNN model. The masks are used to determine the relative placement of the hazards on the walkway, by which the risk is estimated. The final system was shown to classify hazard risks accurately 93.22% of the time. The available input sensors with Spot were considered, namely the robot’s built-in stereo grayscale cameras, the latter with an additional depth channel, and the Spot CAM+IR fisheye cameras. Three separate image data sets were collected on site at the Rosh Pinah zinc and lead mine for this evaluation. It was found that the Spot CAM+IR fisheye camera results in the best performing system, achieving a Mask R-CNN precision and recall of 84.4% and 76.4%, respectively. The final system was shown to have an accuracy of 58.79% when trained on a very small data set. The system accuracy could easily be increased to 89.53% and above by significantly increasing the amount of training data. The accuracy could even be increased to meet industrial safety standards, thereby making the system feasible for real-life operations.
A cashless payment platform for minibus taxis
(Stellenbosch : Stellenbosch University, 2023-03) Tenderere, Kudzai; Booysen, Thinus; Visagie, Lourens; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.
ENGLISH ABSTRACT: The South African Taxi Recapitalisation Program determined that electronic/cashless fare collection systems must be installed in every minibus taxi. However, the past and present initiatives to implement cashless fare collection systems have been focused on using Euro Master Visa (EMV) cards. The problem with these cards is that they exclude part of the population which has been historically marginalised and excluded by financial i nstitutions, m ainly b anks. T his h as h ighlighted t he n eed f or a more inclusive cashless fare collection system in the minibus taxi industry. The cashless fare collection system is also meant to gather data on passenger movement for infrastructure and planning purposes. It was determined that mobile phones have enjoyed a significant p enetration rate into the African and South African population and, as such, mobile money is the best financial technology, particularly for the historically marginalised and financially excluded populations. It was assumed that minibus taxi passengers have Bluetooth devices switched on when they travel. It was determined that social media applications, particularly WhatsApp and Telegram, have become the most used platforms to communicate. Therefore, a decision was made to implement a cashless fare collection system in the form of a Telegram Chatbot: the system would add passenger detection and tracking using Bluetooth. The system for passenger detection and tracking using Bluetooth was designed, implemented, and tested on a couple of local minibus taxi trips. The data gathered from the local trips was analysed to determine the reliability of using Bluetooth for passenger detection and tracking. Generally, Bluetooth use in minibus taxi passengers was neither available nor reliable enough to be used for passenger tracking and identification. The cashless fare collection system was developed without the Bluetooth passenger tracking and identification subsystem.
Illumination-invariant face skin pigmentation prediction
(Stellenbosch : Stellenbosch University, 2024-03) Mbatha, Success Katlego; Theart, Rensu; Booysen, Thinus; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.
ENGLISH ABSTRACT: Skin tone estimation is a critical task with a wide range of applications in fields such as cosmetic science, dermatology, image processing, and facial recognition. Accurate skin tone estimation plays a significant role in improving the inclusivity and fairness of these systems. As machine learning and artificial intelligence continue to advance and finds application in widely used systems, addressing the challenges of skin tone estimation has become increasingly important to ensure that these technologies perform consistently across diverse skin tones. This study focuses on contributing to this field of study by developing a CNN based skin tone classification/estimation model capable of delivering consistent accuracy for individuals with varying skin tones and under diverse lighting conditions. Early in 2022, Google introduced a new skin tone classification system known as the “Monk Skin Tone (MST)” which offers a broader range of skin tones compared to the commonly used “Fitzpatrick Skin Type (FST)” system. This more comprehensive scale is designed to be inclusive and representative of the diverse global population and has been adopted in this study. A data collection campaign was hosted at Stellenbosch University with an aim to develop a dataset with a diverse range of skin tones, classified into the MST, addressing the shortcomings of existing openly accessible datasets. This effort resulted in the acquisition of 21 375 images from 285 participants. Furthermore, the model development process involves exploring various CNN architectures, model configurations, and data pre-processing techniques to maximise the accuracy of skin tone estimation. Experimental findings on various model configurations are summarised as follows: The regression model trained on LAB images, which was selected as the best performing model, demonstrated the highest accuracy at 58.12%. In contrast, a pre-trained CNN model showed limited accuracy, achieving a modest 36.85%. When colour balancing techniques were applied to the dataset images, the resulting accuracy was 55.05%, falling short of the performance of the regression model using LAB images. Moreover, an attempt to increase both image and model dimensionality by converting images to RGB-LAB-HSV led to overfitting issues, resulting in an accuracy of 45.76%. Furthermore, the study explored the model’s performance under different lighting conditions, with the highest accuracy recorded under warmer lighting conditions such as “Halogen warm white” (62.07%), and “Florescent warm white” (58.37%). The study also discusses the impact of spectral characteristics on lighting conditions, particularly noting that “LED warm white” exhibited lower accuracy at 55.73%. This reported accuracy is of samples at distance ≤ 0.5 units from the actual targets. However, when the margin distance is increased to 1 and 2 units, the average accuracy across all light types becomes (85.45 }2.01) % and (97.16 }1.00) %. In the evaluation of skin tone estimation accuracy based on Monk skin tones in the LAB space, the study found that Group 1, representing individuals with lighter skin tones (Monk skin tone 1-3), exhibited strong accuracy. Group 2, encompassing individuals with middle-range skin tones (Monk skin tones 4-7), displayed comparatively lower accuracy in the skin tone estimation. Group 3, consisting of individuals with higher pigmentation skin tones, fell between the performance levels of Groups 1 and 2 in terms of accuracy. Despite the non-linear performance observed across all these skin tone groups, the overall skin tone estimation performance is satisfactory, with an average predicted-to-target error distance value of 16.40 } 20.62 in the LAB space for all samples. Overall, this research contributes to the advancement of skin tone estimation, with practical implications for enhancing the performance of facial analysis algorithms in real-world applications.
Residential load modelling to predict household consumption for design of photovoltaic systems
(Stellenbosch : Stellenbosch University, 2022-12) Avenant, Jason; Booysen, Thinus; Rix, Arnold J. ; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.
ENGLISH ABSTRACT: South Africa’s electrical network is restricted, and its supply is variable. Adoption of solar systems is a potential solution. Despite the abundance of sunshine in South Africa, the adoption of domestic rooftop solar has remained limited. The unequal distribution of wealth and linked usage, a legacy of apartheid, and the uncertain cost/benefit of large-scale solar installations play a significant role in this. To mitigate the uncertainty of large-scale solar deployment, PV system design may be predicted using simulation. This provides users with a better understanding of the energy savings and partial independence that PV system adoption can offer from the volatile grid . Domestic electricity usage and solar insolation are key components in the design of a PV system. Because usage patterns vary and solar insolation fluctuates, these variations must be considered when properly analysing the viability of PV adoption. In South Africa, a lack of load data hinders large-scale solar PV system planning and implementation. Although high-resolution data on individual residential energy use is scarce in South Africa, the DELS (Domestic Electrical Load Study) dataset, which contains data for 14 945 households sampled hourly, is an exception. Given the dataset’s geographical and demographic representation, it provides an intriguing approach of assessing large-scale solar PV implementation. To use the DELS dataset in this way, a model needs to be developed to statistically describe the measured usage profiles for each household. Thereafter, the model can be used to simulate household usage, the resultant profiles of which can be compared to simulated solar PV profiles and used to size solar installations for each household. In this thesis, we presented a load model and subsequent load synthesiser for South African residential households. Because data quality and availability are difficult to obtain in developing countries such as South Africa, we designed the model to function with minimal training data. In addition to the synthesiser, we conducted a case study to evaluate the synthesiser’s suitability for the sizing and design of fixed-axis rooftop PV systems based on the synthetic profiles. A data reduction framework to extract key features from representative daily load profiles was employed. The developed models accounted for the effects of seasonal and day-of-week trends on household consumption patterns. A sum-of-Gaussian (SOG) model was used to describe load profile shapes. The model made use of the assumption that residential households’ load behaviour would have two distinct peaks, one in the morning and one in the afternoon, representing their work/school schedule. A probabilistic model is used to simulate the distribution of peak amplitudes in households. Employing hypothesis tests we fitted probability distributions to the measured peaks amplitude values. A load synthesiser was created by combining the two models. The synthesiser simulates different seasons and days (weekdays, Saturdays and Sundays). Depending on the season and day type being synthesised, we generate load profiles using a SOG-based model that was trained on a relating seasonal and day type subset. To scale the amplitude of the synthetic load profile, we created synthetic amplitude values by simulating the amplitude distribution using the probabilistic model. In this way, the synthesiser can simulate load profiles with similar statistical characteristics and shapes to the measured daily load profile. The proposed synthesiser was used to assess its suitability for the design of fixed-axis rooftop PV systems. We designed a generic PV system that was used to produce PV simulations using SAM (System Advisory Model). The synthesised days were used to scale the system’s size to conform to a design requirement that the PV system export no more than 15% of the energy it produces. The developed probabilistic model was assessed by comparing the measured peak amplitudes, and shown to accurately portray the measured values’ statistical properties. The synthetic and measured distributions were virtually identical, with the first quantile of values showing the biggest difference, which showed that the poor performance was limited to very low consumption levels. Synthetic days generated for one calendar year of days were used to evaluate the sum-of-Gaussian based model. Utilising the SMAPE (symmetric mean absolute percentage error) and MAE (mean absolute error), we compared the synthetic profiles shape to the measured profiles. We discovered that even though the model performed well in terms of reproducing the shape of the RDLPs, the level of accuracy varied when simulating a year’s worth of dates. This was most likely brought on by the assumptions and approximations made regarding the shape of residential load profiles. To evaluate the performance of the synthesiser, 900 households that had more than a year’s worth of measured days were compared with a year’s worth of synthetic data. The synthesiser was tested to determine whether it captured the shape of the daily load profiles throughout the year. The tests determined if it captured the statistical characteristics of an individual household’s load as well as all households and if it captured the statistical characteristics of the aggregated normalised grid of 900 households. We discovered that the synthesiser tended to overestimate peak loads while underestimating off-peak loads. Rooftop solar PV systems were sized using measured data and compared with ones sized using synthesised data. Those sized using synthetic data proved to be oversized by a small amount – less than a PV module per household. The model behaved as expected, the synthesiser created slightly inflated peak loads, but proved to be a useful tool for assessing fixed axis rooftop PV systems. As a result, the project goals were achieved and objectives fulfilled. Additional adjustments to the synthesiser and models for future work were suggested. The project code is available online1.
Towards a paper-based electrochemical glucose biosensor
(Stellenbosch : Stellenbosch University,, 2023-03) Francois, Snyders; Booysen, Thinus; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.
ENDGLISH ABSTRACT: Glucose homeostasis forms a critical part of the development of a healthy neonate. The complex adaptation from fetal to neonatal life requires well-coordinated hormonal and metabolic adaptive changes to maintain glucose homeostasis. At-risk infants face high morbidity- and mortality risks and are also at larger risk of developing glucose metabolism disorders during their first week of e xtra-uterine l ife. Hyperglycemia i s a common glucose metabolism disorder developed in at-risk infants and is frequently accompanied by a urinary loss of glucose. The concentration of glucose excreted through neonatal urine can give an indication of the hyperglycemic state of the neonate. Glycosuria monitoring can therefore be used as an additional screening method to detect hyperglycemia - an alternative to frequently measuring blood glucose. This project aimed to develop a paper-based glucose sensor that uses electrochemistry and glucose-specific e nzyme r eactions t o d etect g lucose c oncentrations relevant to neonatal hyperglycemic glycosuria thresholds in sample solutions. The sensor was manufactured on photo paper in the form of a printed three-electrode sensor using a desktop ink-jet printer as the manufacturing tool. A compatible PEDOT:PSS ink dispersion was formulated for this printer by optimising the concentration ratios of its constituents. Thin polymer films were p rinted i n s uccessive l ayers a nd were treated with organic solvents and annealing procedures to achieve optimal levels of film conductivity. An enzyme solution containing glucose oxidase enzymes and artificial mediators was prepared and validated for its ability to oxidise glucose. By combining the printed electrodes with the prepared enzyme solution, cyclical voltammetric methods were applied with the addition of glucose to realise an electrochemical glucose biosensor. Glucose concentrations ranging from 0 􀀀 32 mg=ml were tested using the fabricated sensors. The sensor proved its capability by consistently detecting all glucose concentrations relevant to hyperglycemic glycosuria thresholds in neonatal urine. A shortcoming of the proposed sensor solution included the incapability to differentiate between glucose concentrations of 4 􀀀16 mg=ml.
Using the internet of things for greenhouse temperature prediction, management, and statistical analysis
(Stellenbosch : Stellenbosch University, 2023-12) Hull, Keegan; Booysen, Thinus; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.
ENGLISH ABSTRACT: Unpredictable weather patterns caused by climate change are impacting agricultural productivity worldwide. This threatens sustainability and may lead to food insecurity, especially in developing regions. Affluent countries can afford costly investments toward mitigating the effects of climate change on food production. However, poorer countries tend to lag behind due to the lack of resources. To improve climate resilience, evolving technologies, such as the Internet of Things (IoT), have been proposed and developed for climate-smart farming. In this thesis, a technological solution was presented in the form of a digital twin for temperature monitoring and control of a greenhouse tunnel. Further, an aeroponics trial in the tunnel is statistically analysed for temperature variations due to the fan and wet wall temperature regulatory systems. A greenhouse tunnel in Stellenbosch, South Africa was instrumented with a prototype system to monitor temperatures inside the tunnel and control the cooling fan and wet wall. A generic hardware solution was then designed, assembled, and used in place of the prototype system to prove the feasibility of such a system in agriculture in South Africa. An analytical model was derived using the measurements as validation of the model. An empirical model using the Support Vector Regression algorithm was then developed and was used as comparison to the analytical model. The study was successful in producing an analytical model that was accurate to an acceptable degree. This physics-based model produced an RMSE of 2.93°C and an R2 value of 0.8. An empirical model was also produced that can simulate internal temperatures to an RMSE value of 1.76°C and an R2 value of 0.9 for a one-hour ahead simulation, outperforming the analytical model. The statistical analysis of the aeroponics system also showed a strong relationship between the temperature inside the system and the distance from the fan and wet wall. Finally, two analytical models of the irrigation process in a single container in the aeroponics system produced accurate results for the cooling effect of the irrigation (R2=0.901), while the unexpected heating effect when the fan was off produced less accurate results (R2=0.718). Future work stemming from this research includes improved data-driven modelling, cloud integration of the generic hardware, and fewer assumptions in analytical thermal modelling of the tunnel and container systems.
What the eye doesn’t see : using infrared to improve face recognition of individuals with highly pigmented skin
(Stellenbosch : Stellenbosch University, 2022-12) Muthua, Alex; Theart, Rensu; Booysen, Thinus; Stellenbosch University. Faculty of Engineering. Dept. of Electrical and Electronic Engineering.
ENGLISH ABSTRACT: Face recognition technology has become commonplace in security and access control applications. However, their performance leaves a lot to be desired when working with highly pigmented skin tones. One reason for this is the training bias introduced by under-representation in existing datasets. The other is inherent to pigmentation – darker skins absorb more light and therefore could reflect l ess d iscernible d etail i n t he v isible s pectrum. We s how how this can be enhanced by incorporating the infrared spectrum, which electronic sensors can perceive. We collect a database with images of highly pigmented individuals, captured using the visible, infrared and full spectra We fine-tune state-of-the-art face recognition systems and compare the performance of these three spectra. We also assess the impact of narrow and wide cropping, different facial orientations, and sunlit and shaded conditions. We find a marked improvement in the accuracy and in the AUC values of the ROC curves when including the infrared spectrum, with performance increasing from 97.5% to 99.1% for highly pigmented faces. Including different facial orientations and narrow cropping also improves the performance, and can therefore be deemed as recommended best practices. Analysis of the activation maps of the CNNs finds t hat fi ne-tuning mo dels ac tivate mo re ge nerally ov er al l re gions of the face while models with pre-trained weights, focus on fewer features with higher activation intensity values over those regions. In both cases, the nose region appears as the most important feature for face recognition for highly pigmented faces.

Browse

Browsing Masters Degrees (Electrical and Electronic Engineering) by browse.metadata.advisor "Booysen, Thinus"

Results Per Page

Sort Options