George van Schoor

The Tennessee Eastman Process is a widely used benchmark for process control and fault detection and diagnosis, yet earlier implementations in FORTRAN and MATLAB C-mex restricted access to internal variables, limiting flexibility for advanced studies. The recently developed complete and open Simulink model of the TEP (COSTEP), removes these barriers by providing full access to all process parameters and expanding the number of available measurements to 138. This open architecture makes COSTEP particularly valuable for customised dataset generation, process monitoring, and machine learning applications aligned with the goals of Industry 4.0 and Big Data.This paper demonstrates the advantages of COSTEP through two key contributions. First, a review of Tennessee Eastman Process fault conditions highlights how the model enables a clearer distinction between true process faults and normal operating variations, improving the accuracy of fault detection and diagnosis systems. Second, exergy analysis is introduced as a novel approach to fault detection, made feasible by COSTEP’s comprehensive access to process variables. The results show that exergy-based methods can reduce data dimensionality while enhancing diagnostic performance. Together, these contributions establish COSTEP as a powerful and flexible benchmark framework for advancingresearch in process monitoring, optimisation, and fault detection and diagnosis.

Simultaneous Localization and Mapping (SLAM) is a crucial component to the push towards full autonomy of robotic systems, yet it is computationally expensive and can rarely achieve real-time execution speeds on embedded platforms. Therefore, a need exists to evaluate the performance of SLAM algorithms in practical embedded environments – this paper addresses this need by creating prediction models to estimate the performance that ORB-SLAM3 can achieve on embedded platforms. The paper uses three embedded platforms: Nvidia Jetson TX2, Raspberry Pi 3B+ and the Raspberry Pi 4B, to generate a dataset that is used in training and testing performance prediction models. The process of profiling ORB-SLAM3 aids in the selection of inputs to the prediction model as well as benchmarking the embedded platforms’ performances by using PassMark. The EuRoC micro aerial vehicle (MAV) dataset is used to generate the average tracking time that the embedded platforms can achieve when executing ORB-SLAM3, which is the target of the prediction model. The best-performing model has the following results 2.84%, 3.93%, and 0.95 for MAE, RMSE and R2 score respectively. The results show the feasibility of predicting the performance that SLAM applications can achieve on embedded platforms.

While South Africa is deemed one of the countries with the highest irradiation levels, it still utilises coal as its primary energy source due to its abundance. Due to the world-wide drive towards carbon neutrality, residential, commercial, agricultural, and industrial consumers are considering small-scale embedded generation systems. The National Rationalised Specifications 097-2-3 document specifies the scale of the embedded generation capacity a consumer is allowed to install. However, specifications do not yet make the required provisions for the addition of energy storage. The effective collective management of the grouped small-scale embedded generation systems could provide a high level of energy security and increase the percentage of renewable energy generation in the total energy mix. Potential challenges come into play when considering the stochastic nature of photovoltaic generation and its effect on the storage capacity and the dispersion in load profiles of the residential units typically present on a low-voltage network. This paper contributes by investigating the utilisation of photovoltaic generation in conjunction with storage as the basis for virtual power plant control, with the aim to safely increase renewable energy penetration and improve energy security, all while remaining within the South African low-voltage regulatory limits. A two-level virtual power plant controller is proposed with the dispersed energy storage units as the primary controllable resources and the dispersed photovoltaic generation as the secondary controllable resources. The objective of the controller is to achieve nodal energy management, energy sharing, and ancillary service provision and finally to increase renewable energy penetration. A representative single-feeder low-voltage network is simulated, and test cases of 50% and 75% renewable energy penetration are investigated as the basis for evaluation. The proposed controller architecture proved to maintain network integrity for both test cases. The adaptability of the controller architecture was also confirmed for a changed feeder topology; in this case, it was a multi-feeder topology. Future work is warranted to inform policy on the allowed levels of renewable energy penetration to be based not only on demand but also on the level of energy storage present in a network.

This paper presents a methodology for fault detection and isolation (FDI) in transient conditions using a multiway principal component analysis (MPCA) approach where practical data have been augmented with simulated data to conduct FDI when there are insufficient practical data. The motivation for using a heated two-tank system is due to the fact that it resembles a basic process in terms of controllable variables, noise, disturbances, and changes in operating points. Normal and faulty condition data of the practical heated two-tank system as well as a Simulink^® model of the heated two-tank system were used. The MPCA technique has enhanced ability to detect and isolate faults in transient conditions compared to classic principal component analysis (PCA). MPCA, however, requires a vast amount of normal process transient conditions data to train the model to then enable meaningful fault detection and isolation. In this study, the practical normal transient conditions data are augmented with simulated normal transient conditions data to meet the requirement of a large amount of data. Utilising different datasets for the training of the MPCA model, the fault detection and isolation performance was evaluated with various metrics. This paper presents positive results towards the implementation of MPCA for fault detection in transient conditions.

Industrial process monitoring plays an essential part in the effective operation of industrial processes. With the ever increasing push to autonomous factories

When applying graph-based fault detection and isolation (FDI) methods to the attributed graph data of large and complex industrial processes, the computational abilities and speed of these methods are adversely affected by the increased complexity. This paper proposes and evaluates five reduction techniques for the exergy-graph-based FDI method. Unlike the graph reduction techniques available in literature, the reduction techniques proposed in this paper can easily be applied to the type of attributed graph used by graph-based FDI methods. The attributed graph data of the Tennessee Eastman process are used in this paper since it is a popular process to use for the evaluation of fault diagnostic methods and is both large and complex. To evaluate the proposed reduction techniques, three FDI methods are applied to the original attributed graph data of the process and the performance of these FDI methods used as control data. Each proposed reduction technique is applied to the attributed graph data of the process, after which all three FDI methods are applied to the reduced graph data to evaluate their performance. The FDI performance obtained with reduced graph data is compared to the FDI performance using the control data. This paper shows that, using the proposed graph reduction techniques, it is possible to significantly reduce the size and complexity of the attributed graph of a system while maintaining a level of FDI performance similar to that achieved prior to any graph reduction.

In order to achieve effective coffee roasting control, there is a need for coffee roasting process models that can solve in real-time. This paper contributes by evaluating the roasting process model performance for different roasting time periods and bean moisture contents. The method firstly involved selecting roasting profile models from literature on the basis of their computational efficiency and simplified geometry definition of the coffee bean structure. Three models were considered constituting different bean temperature and moisture content sub-models. Next the modeled results were compared with experimental results obtained from an instrumented commercial artisan roaster. It was found that all three models could sufficiently predict the roast profile from the modeled bean temperature and moisture content. However, all three consistently overestimated the moisture content during roasting. For longer roasting times, the final moisture content prediction was closer to the experimental value.