Suresh Perinpanayagam

Battery management systems require models that are faster to run on a micro controller while retaining its fidelity. This paper presents a seven state electrochemical model reduced from 44 state isothermal implementation of the thermal single particle model with electrolyte dynamics (TSPMe) at 298 K. The model uses two conservative finite volume shell states in the negative particle, three in the positive particle and two voltage weighted electrolyte coordinates with an eleven-feature algebraic decoder projected into the zero-inventory subspace. The dynamic order is reduced by 84.1 % while exactly preserving the solid and liquid lithium inventories. Across 0.5C and 1C constant charge and discharge validation cases, the model achieves a mean terminal voltage RMSE of 2.726 mV with maximum absolute error of 49.826 mV at the initial current step compared to reference. further, model is evaluated on an experimental Stanford W8 cell UDDS drive cycle, terminal voltage RMSE is 23.277 mV, compared to 22.042 mV for the reference of 44-state model. The model is combined with a sparse near-equilibrium voltage correction for state of charge (SOC) estimation when validated on ten different cells on UDDS drive cycle it has lower SOC error than a 2RC extended Kalman filter. It is deployed in NXP i.MX RT1060 evaluation board and the longest measured update is 2.493 ms (2.493 % of the 100 ms task deadline).

Predictive maintenance aims to reduce operational costs by anticipating and preventing system failures or inefficiencies. While high-performance AI models such as neural networks offer accurate predictions, their lack of transparency limits their usefulness for guiding interventions. Conversely, explainable AI (XAI) models provide insight but often at the expense of accuracy. This paper proposes a framework for cost-based evaluation of interpretable AI models in predictive maintenance, using both classification and regression contexts. We establish criteria to determine when the benefit of interpretability outweighs any reduction in accuracy and show that the utility of XAI is bounded by the relative cost of maintenance versus failure. These findings offer practical tools for assessing the business case for interpretable models in predictive maintenance and related domains. In research, the criteria enable cost-based evaluation and comparison of alternative machine learning methods for regression and classification.

This review paper explores the pivotal role of digital twin (DT) technology in advancing nuclear fusion research and its potential to address key challenges in achieving fusion energy as a sustainable power source. Whilst DTs have been developed to tackle complex issues in data analysis, real-time control, optimisation, and simulation, existing efforts are largely fragmented and constrained to isolated applications due to the experimental and evolving nature of fusion reactors. To bridge this gap, there is a pressing need for an integrated, end-to-end DT framework that can unify these capabilities, enabling a more holistic approach to reactor design and operation. This paper systematically reviews the current state of DT development, focusing on its application to critical components such as the divertor, breeder blanket, and magnets, which serve as ideal candidates for modular DT instances. By leveraging high-performance computing and vast datasets generated by fusion experiments worldwide, we identify opportunities to create scalable, data-driven surrogate models that can continuously refine live DT systems. The findings highlight the potential for centralised storage of the data and models to accelerate simulation-based testing (in silico) and optimisation of reactor components, ultimately contributing to the faster realisation of commercially viable fusion reactors. This integration of DT technology promises to significantly reduce the need for costly, iterative physical testing, paving an efficient pathway to achieving fusion as a reliable energy source.

Electric motors find widespread application across various industrial fields. The pursuit of enhanced comprehensive electric motors performance has consistently drawn significant attention, prompting extensive research in this domain. With the rapid and profound integration of artificial intelligence (AI) technology into diverse sectors, an increasing array of AI-based methodologies is being employed in electric motors research. This review aims to comprehensively delineate the applications of AI within the realm of electric motors, specifically focusing into three pivotal stages: electric motors design, drive control, and health maintenance, covering a full-cycle research and development of electric motors. The review commences by outlining the elucidation of the fundamental principles and components of AI. Subsequently, the encompassed input signals, characteristic methodologies, and AI techniques are reviewed and analyzed as demonstrated by pertinent research cases. Furthermore, a synthesis of the distinctive attributes characterizing various techniques is presented. Conclusively, this article offers an unprecedented review of AI-based technology throughout the entirety of the electric motors research cycle. Considering the existing body of knowledge, this article stands as the pioneering endeavor to encapsulate the expansive domain of AI's influence in electric motors research.

The proliferation of Power Electronic Converters (PECs) has had a pervasive affect in a variety of industries including the power generation, automotive and aerospace sectors, where their use brings reliability to the forefront, especially in applications where safety critical and harsh environments are experienced. Continuous improvements in the power density and efficiency through extensive research into new semiconductor technologies, passive components, circuit topologies and control methodologies has seen the performance of PECs improve indubitably. However, the manifestation of stresses occurring from significant heating due to high currents and switching frequencies; which over time can cause degradation in the performance of PEC components is still a real concern. This paper outlines a methodology for monitoring the degradation of PECs over the operational lifetime by utilizing a component level, physics based Digital Twin (DT). As well as providing a methodology for the real time comparison of parameters to realize the onset of faults subjected to operational stresses, the DT also provides a novel method of training a classifier by simulating the faults within the PEC, a process that in reality, would be difficult to achieve from a physical device. Feature extraction is via Wavelet Scattering and classification is provided using a Support Vector Machine (SVM) approach. The overall approach is one that is novel and expandable to a wide range of PEC topologies and will be beneficial to the optimization and maintenance of PECs in a variety of platforms.

Financial costs and energy savings are considered to be more critical on average for computationally intensive workflows, as such workflows which generally require extended execution times, and thus, require efficient energy consumption and entail a high financial cost. Through the effective utilization of scheduled gaps, the total execution time in a workflow can be decreased by placing uncompleted tasks in the gaps through approximate computations. In the current research, a novel approach based on multi-objective optimization is utilized with CloudSim as the underlying simulator in order to evaluate the VM (virtual machine) allocation performance. In this study, we determine the energy consumption, CPU utilization, and number of executed instructions in each scheduling interval for complex VM scheduling solutions to improve the energy efficiency and reduce the execution time. Finally, based on the simulation results and analyses, all of the tested parameters are simulated and evaluated with a proper validation in CloudSim. Based on the results, multi-objective PSO (particle swarm optimization) optimization can achieve better and more efficient effects for different parameters than multi-objective GA (genetic algorithm) optimization can.

Fault detection for railway door systems based on data-driven approaches has been investigated in recent years due to the massive amount of available monitoring data. Despite much attention to its application, the major challenge is the lack of available faulty datasets to build a reliable model since railway maintenance is usually conducted regularly to avoid significant defects from economic and safety points of view. We aimed to tackle the issue by employing transfer learning. Firstly, we built a long-short term memory-based deep learning model using linear actuator experimental datasets. Then, we employed a transfer learning technique to adjust the deep learning model to be available to real-world railway door systems using a small amount of faulty data. As a result, high fault detection accuracy can be obtained at 0.979 as F1 score. The result reveals that an accurate fault detection model can be built even though a large amount of labelled datasets is unavailable. In addition, the proposed method is applicable to other door systems or electro-mechanical actuators since the method is unspecific to physical mechanisms and fault modes, and the only motor current signal is used in this research. The signal is primarily available from the controller or motor drive without additional sensors.

Fault detection based on deep learning has been intensively investigated in the recent decade due to increasing availability of data and its ability to engineer features with deep neural network architectures. Despite much attention to its application, the major challenge is the lack of available labelled datasets to build the models since maintenance is usually conducted regularly to avoid significant defects. This paper aims to propose a successful real-time fault detection framework based on unsupervised deep learning using only healthy normal data. The approach is based on autoencoder architecture and a one-class support vector machine as a classifier. As a case study, large real-world datasets acquired from railway door systems have been employed. The five different types of deep learning models and a one-class classifier are trained and comprehensively validated based on performance metrics and sensitivity analysis. In addition, two experiments have been carried out to verify the model’s adaptability and robustness to variational time-series data. The result shows a typical autoencoder is the least sensitive to a decision boundary set by the one-class classifier. However, the two experiments show that the fault detection accuracy for a bidirectional long short-term memory-based autoencoder is considerably higher than other autoencoder-based models at 0.970 and 0.966 as F1 score, meaning only this model is adaptable and robust to variational data. The experimental result allows us to obtain the understandability of the deep learning models. Furthermore, the regions of anomalies are localised with unsupervised models, which enables diagnosing the cause of failure.

Digital twin (DT), primarily a virtual replica of any conceivable physical entity, is a highly transformative technology with profound implications. Whether it be a product development, design optimisation, performance improvement, or predictive maintenance, digital twins are changing the ways work is undertaken in various industries with multifarious business applications. Aerospace industry, including its manufacturing base, is one such keen adopter of digital twins with an unprecedented interest in their bespoke design, development, and implementation across wider operations and critical functions. This, however, comes with some misconceptions about the digital twin technology and lack of understanding with respect to its optimal implementation. For instance, equating digital twin to an intelligent model while ignoring the essential components of data acquisition and visualisation, misleads the creators into building digital shadow or digital models, instead of the actual digital twin. This paper unfolds such intricacies of digital twin technology for the aerospace community in particular and others in general so as to remove the fallacies that affect their effective realisation for safety-critical systems. It comprises a comprehensive survey of digital twins and their constituent elements. Elaborating their characteristic state-of-the-art composition along with corresponding limitations, three dimensions of the future digital twins for aerospace sector, termed as aero-Digital Twins (aero-DTs), are proposed as an outcome of this survey. These include the interactive, standardisation, and cognitive dimensions of digital twins, which if leveraged diligently could help the aero-DT research and development community quadruple the efficiency of existing and future aerospace systems as well as their associated processes.