The Data-driven methods for predicting the State of Health (SOH) under diverse ageing scenarios provide a robust and effective solution. This work proposes novel SOH estimation frameworks that utilize parallel pathway fusion to enhance feature integration. Specifically, the Parallel LSTM-CNN Network (PLCN) and Parallel CNN-GRU Network (PCGN) architectures are designed to improve forecasting accuracy by leveraging both spatial and temporal features, including long-term dependencies, from the data. The performance of the PLCN and PCGN frameworks are evaluated using battery datasets with an input feature selection method. The study is structured in two phases. First, Grey Relational Analysis (GRA) is employed to evaluate the relevance of various input features, selecting those with the highest GRA scores to develop the base model. Second, the proposed parallel models are optimized with these inputs, demonstrating superior generality and efficacy compared to existing methods. Notably, the parallel models outperforming single and sequential hybrid models due to their simultaneous processing capabilities. By combining CNN-based feature extraction with the long-term temporal analysis of LSTM and GRU, these architectures are providing a more comprehensive understanding of the battery input data, leading to robust and enhanced SOH predictions.

Monitoring and diagnostics of inverter-fed induction motors is important since it is an inevitable part of smart grid, renewable energy systems and other critical applications. After the advent of deep learning-based techniques, in which the feature descriptors are extracted directly from the raw signals, these methods are finding place in such applications, in recent times. Thus, the merits of machine learning and deep learning techniques have been exploited in this study for early diagnosis of inter turn short circuit fault in inverter-fed induction machine, in presence of different speed and load conditions. A novel deep trihybrid DeepCGS structure combining 1D convolutional neural network (1DCNN), gated recurrent unit (GRU) and support vector machine (SVM) has been proposed for early fault detection in drive-fed induction motor. The useful spatial and temporal features are extracted by the trained 1DCNN-GRU model in the first stage. Then, these features are fed to SVM for classification. The proposed scheme has been compared to state-of-art techniques, and the experimental evaluation reveals a better performance as per the performance parameters such as accuracy, specificity and sensitivity. To prove the effectiveness of the proposed trihybrid DeepCGS architecture, the comparable performances of the classifiers have been verified using t-test and Friedman test. The results establish that the proposed trihybrid model “deepCGS” is statistically significant, and the results are not due to mere chance.

Line-fed induction motors (IMs) are essential in transportation systems, powering traction systems, hoists, elevators, escalators, and conveyor belts. Stator inter-turn faults (SITFs) pose a significant threat, potentially causing severe motor damage and operational failures. Diagnosing SITFs is further complicated under practical conditions where IMs are subjected to power quality (PQ) disturbances, deviating from ideal balanced sine wave operation. This article presents two hybrid diagnostic machine learning (ML) pipelines, developed through architectural searches, for fault diagnosis in IMs operating under PQ disturbances. The first pipeline predicts the severity of SITFs using stator voltage and current signals, while the second estimates the type of PQ event-based solely on stator voltages. A dedicated experimental test-bed was developed to emulate various fault severities under various PQ disturbances and load conditions. The diagnostics pipelines achieved an $R<^>{2}$ score of 0.9699 on training and 0.9533 on testing for severity estimation. The threshold-based fault classification achieved accuracies of 96.92% (training) and 93.42% (testing). PQ event classification accuracies reached 98.87% and 96.71%, respectively. In addition, comparisons with models trained on raw time-series data were conducted to demonstrate the effectiveness of the proposed diagnostic approach. This research uniquely integrates domain knowledge with ML techniques to diagnose incipient SITFs in IMs under PQ disturbances-an area largely unexplored in existing literature.

Synchronous Reluctance Motors (SynRMs) are emerging as strong contenders to Permanent Magnet Synchronous Motors (PMSMs) in industrial and vehicular applications due to their non-magnetic design, high power density, robustness, and superior efficiency. Despite a robust design, the stator windings of the SynRMs are susceptible to stator inter-turn faults (SITFs). Diagnosing these faults at an incipient stage is essential to prevent unexpected downtimes and mishaps thereby enhancing the safety of the system and the operator. This work proposes a semi-automated machine learning architectural search (SAMLAS), to find an optimal Machine Learning (ML) pipeline to assess the severity of the SITF, thereby enabling early detection in SynRM drives. The SAMLAS leverages domain knowledge, ML models and optimisation techniques to generate the optimal ML pipeline. A dedicated testbed was developed to emulate the faults. The optimal pipeline achieved a testing R-2 score of 0.8633 for the fault severity. By establishing an appropriate threshold, faults were classified with 100% confidence on both the training and testing datasets. These results that have not been reported in prior art. The optimal pipeline enhances interpretability by eliminating irrelevant harmonics and retaining only those essential for diagnosis. An interpretability metric is also proposed to evaluate the contribution of each harmonic to the model both locally and globally. Additionally, the pipeline is compared with multiple ML models and data structures for validation.

Reliable fault diagnosis is vital for maintaining the safety and performance of industrial machinery. However, collecting fault data by inducing faults in high-capacity or highspeed machines is often unsafe and impractical, which limits the application of traditional supervised learning approaches. Zero-shot diagnosis addresses this by enabling fault detection without fault data from the target machine, but current approaches struggle with scalability and generalization across different machine types, capacities, and operating conditions due to restrictive domain transfer assumptions that limit adaptability. Two different scalable fault modeling approaches are introduced to enable accurate diagnosis in target high-capacity or high-speed systems without requiring any fault data from those systems, thereby achieving zero-sample fault diagnosis under practical operational constraints. The first method uses constrained maximum likelihood linear regression (CMLLR) on latent features to synthesize representative fault data for the target machine from healthy-condition data. The second method maps source and target features into a domain-invariant feature space (DIFS), allowing a single model to generalize across machines with varying capacity. Fault modeling in the DIFS is further extended to address scalable speed modeling, using source data from the machine operating at low RPM to detect faults at high RPM, without requiring any fault condition data at higher RPM. Experimental validation on generator, gearbox systems and a publicly available benchmark dataset demonstrates that the proposed methods achieve robust and generalizable fault diagnosis under zero-sample conditions.

This paper presents a real-time energy management system (EMS) for a commercial electric vehicle charging station (EVCS), integrating a hybrid inverter to manage power supply from photovoltaic (PV), the grid, and the energy storage system (ESS). The presented EMS can minimize operating costs while accounting for system dynamics such as constant current (CC)- constant voltage (CV) charging, ramping constraints of electric vehicle (EV) and ESS. Additionally, the EMS proposes the ESS degradation model to reduce the battery wear. The proposed EMS is modeled as a three-layer optimization problem to schedule power among the grid, PV, and ESS while addressing PV and EV load forecasting errors. It utilizes the adaptive receding horizon (ARH) for real-time energy management without considering the availability of prior information on EV parameters, making it realistic and robust. The proposed method is evaluated under two pricing schemes namely (i) real-time pricing and (ii) time of use (ToU) tariffs. The effectiveness of the proposed method is compared with several existing real-time EMS algorithms and its performance is found to be better. The sensitivity of the proposed method is analyzed with respect to variations in the optimization horizon and the weight settings of the objective functions. In addition, the scalability in terms of computational time is evaluated to ensure its effectiveness and feasibility for real-time applications.

Stator interturn fault (ITF) is the most common failure in electrical machines; if no prompt detection is implemented, it can cause catastrophic results. This work proposes a novel method in permanent magnet synchronous machine (PMSM) drives to detect the ITF, which is insular to speed and load variations. The proposed ITF technique is based on negative-sequence instantaneous reactive power (IRP) distortions. The sensorless control of the PMSM drive, while using field-oriented technique, uses the voltage and current information for rotor position estimation. This serves the dual purpose of controlling the drive and also in developing the diagnostic technique. The IRP distortion is calculated from dq-reference frame voltage distortions, which are estimated using Luenberger observer and dq-reference frame current distortions. The novel fault indicator is calculated based on the vector magnitude of dc components obtained from negative-sequence IRP distortions, which is insular to various speed and load conditions of the drive. The proposed ITF detection technique is experimentally validated under varying load and speed conditions of the sensorless field-oriented controlled (FOC) PMSM drive. Further, a comparison of the proposed ITF detection scheme with the dq-reference frame current residuals technique shows the superiority of the proposed ITF detection scheme under various speed and load conditions of the PMSM drive scheme; furthermore, the reliability of the proposed ITF detection technique under various noise conditions is also verified.

Early detection and diagnosis of inter-turn fault in electric drive system (EDS) is a vital topic to investigate in manufacturing systems. Also, identification the failure at an early stage can save the maintenance cost and invaluable time. This work proposes a methodology which is orthogonal deep convolution neural network (ODCNN) with soft orthogonality constraints (SOC) for stator turn fault detection at an incipient stage in EDS, in presence of high harmonics content and for a wide range of speed and load variations. The proposed technique has been examined on two different applications, viz: the dataset of stator turn fault in electric drive system and the CIFAR-10 dataset, to validate the robustness and effectiveness of the proposed method. The results of the proposed ODCNN have been compared to the normal convolution neural network (NCNN) without any orthogonality constraints, and the comparisons show that the proposed ODCNN has achieved a significant improvement in the field of fault detection in EDS at an early stage and image classification as well.

Stator interturn faults (ITFs) are the most common failure in electrical machines and, if not detected quickly, can have disastrous results. This work proposes a robust model-based ITF detection technique in direct torque control with space vector modulator (DTC-SVM) permanent magnet machine (PMSM) drive to detect the ITF, which is insular for speed and load variations, based on negative-sequence instantaneous reactive power (IRP) distortions. The IRP distortion is obtained from dq-reference frame voltage distortions and dq-reference frame current distortions. The obtained IRP distortions are transformed to negative-sequence IRP distortions. Based upon the vector magnitude of the dc components, which are presented in the negative-sequence IRP distortions, the novel fault indicator (FI) is proposed. The proposed FI obtained from negative-sequence reference frame IRP distortions shows invariant signature for DTC-SVM PMSM drive speed and load variations. The robustness of the proposed model-based ITF detection technique has been validated under various operating speed and load conditions on a DTC-SVM PMSM drive system. The proposed model-based ITF detection technique ensures improved reliability and availability of the motor drive system in critical applications.

In electrical machines, a stator interturn fault (ITF) is one of the most common failures and can cause catastrophic results if not detected promptly. This work proposes a novel adaptive-model-based ITF detection technique in sensorless permanent magnet synchronous machine (PMSM) drives to detect the ITF. In the proposed adaptive-model-based scheme, a machine model is developed, which adapts according to voltage distortions under the ITF, and the estimated rotor speed and position from the model reference adaptive scheme. This approach minimizes rotor position estimation errors caused by the ITF. As a result, fault index (FI) estimation errors are minimized under ITFs, ensuring reliable ITF detection in sensorless PMSM drives. In the proposed scheme for ITF detection, dq-reference frame currents are estimated using the adaptive machine model, and actual currents are measured using the faulty PMSM. Residual current vectors (RCVs) are generated from these two quantities. These generated RCVs are used to extract the FI under ITF conditions. The proposed adaptive-model-based ITF detection technique is experimentally validated under various operating speed and load conditions of the sensorless PMSM drive. Furthermore, the reliability of the proposed scheme is also validated under machine parameter variations.

This study presents a detailed investigation of field-oriented control (FOC) applied to surface-mounted permanent magnet synchronous motors (SPMSMs) driven by inverters based on Si-IGBTs, SiC MOSFETs, and GaN devices. A comparative analysis is carried out focusing on total power losses including both conduction and switching losses for silicon, SiC, and GaN-based inverter-fed PMSM drives. FOC has recently gained significant attention due to its ability to deliver precise control of motor performance, making it highly suitable for various industrial applications. Moreover, employing wide-bandgap semiconductors like GaN in the inverter enhances performance due to their compact size, high power density, and capability for high switching frequencies, all contributing to improved efficiency. Simulation results indicate that GaN-based inverter-fed PMSM drives exhibit the lowest losses among the three, resulting in superior efficiency. The overall system performance and reliability were evaluated using PLECS simulation software.

The self-excited synchronous reluctance generator (SE-SynRG) is gaining importance for its potential use in standalone renewable energy applications, especially in wind energy applications. Absence of rotor windings, simple design, reliability and lack of rare-earth minerals are few of its key features. Stator winding faults are the most common failures occurring in the electrical machines. The occurrence of stator inter-turn faults (SITF) can have a significant impact on its performance and reliability. This work proposes a novel analytic modeling for the SynRG which will aid in the diagnosis of inter-turn faults. The detailed mathematical development of a dq model for SynRG, which allows for the incorporation of SITF, to understand and analyze the fault current behavior at various fault severities and operating conditions has been envisaged. The conclusions of this study will be helpful for developing diagnosis techniques for SynRG, and to plan the post-fault operations. The developed model has been simulated in MATLAB environment for investigating the behavior of the machine at different fault severities and operating conditions.

This work introduces a machine learning approach for developing Digital Twins (DTs) for DC-DC converters, focusing on in-situ implementation in real-world operational conditions. A system based on a boost converter has been developed in MATLAB Simulink. To mirror real-world scenarios, commercial datasheets along with a range of input parameters, health degradation elements, temperature influence, and random noises have been considered. The study employs Multi-Layer Perceptron (MLP), Convolutional Neural Network (CNN), and Recurrent Neural Network (RNN) for predicting critical circuit responses of the boost converter, including inductor current, output voltage, and efficiency. Investigations show that MLP performs relatively poorly in the presence of noise. The CNN and RNN outperform the MLP under various noise levels, with the RNN exhibiting the best performance. This work advances DTs technology in power electronics, aiming to improve converter system optimization and enable predictive maintenance.

Synchronous Reluctance Motors (SynRM) are becoming popular in electric vehicles, renewable energy, and domestic applications over PMSM and Induction machines due to their advantages, such as the absence of rare earth materials, less copper losses, robust construction, and high efficiency. The performance of a SynRM is enhanced by its rotor geometry through improving the saliency ratio, along with its control strategies. This work focuses on the performance evaluation of control strategies, comparing two distinct methods applied to Synchronous Reluctance motors: Field oriented control (FOC) and Direct torque control (DTC), targeting EV applications. These control techniques are prominent strategies for SynRM, offering precise control over speed and torque. The features of both methods are illustrated and simulated for various EV torque and speed testing profiles using MATLAB/Simulink software.

Interturn fault (ITF) is a commonly occurring fault in permanent magnet synchronous machine (PMSM). The propagation of the ITF could cause damage to the entire electrical machine if it is not detected in time. In this proposed technique, second order harmonic components in dq-reference frame impedance distortions are used for ITF detection in field-oriented controlled PMSM drive. The dq-reference frame impedance distortions are obtained from the voltage distortions and current distortions by using the Luenberger observer, and also by healthy and faulty analytical models of PMSM. These dq-reference frame impedance distortions are transformed to negative-sequence reference frame. Then the magnitude of dc components obtained from negative-sequence dq-reference frame impedance distortions is defined as fault index. Further, the comparison with the dq-reference frame current residuals ITF detection technique shows the superiority of the proposed technique.

This work deals with development of analytical models for synchronous reluctance motor (SynRM), under stator interturn fault (SITF), considering series and parallel connected windings. The developed models are first of its kind for SynRM, which considers cross flux coupling interactions between healthy and faulty coils in the same winding, along with other phase windings. By adopting the proposed modeling approach, consistent equations in abc and dq coordinates suitable for analyzing performance metrics such as phase currents, electromagnetic torque, and dq voltages have been derived, which allows studying the post fault operations and fault diagnosis features, at different fault conditions. The key fault features of SITF such as the fault current and third harmonic in the faulty phase current have been considered, in an attempt to evaluate the motor's ability to tolerate interturn fault, in both series and parallel winding configurations. The experimental results agree well with the theoretical analysis and it demonstrates the effectiveness of the developed models for analyzing the machine behavior under SITF in the SynRM, with series and parallel connected windings. The investigation also points out that parallel winding SynRMs (P-SynRMs) are more challenging to diagnose and more fault tolerant to the SITF than the series winding SynRMs (S-SynRMs).

The proposed work presents a hybrid signal processing - three level deep learning framework leveraging transfer learning to detect incipient stator inter turn faults in induction motor drives. An experimental test rig is developed using a three phase squirrel cage induction motor to emulate various fault levels. This non-invasive methodology solely relies on the stator currents extracted from the developed test rig to detect the stator inter turn faults under its incipient condition under various conditions of load torque and speed. The framework integrates wavelet transform for initial feature extraction and a three stage deep learning architecture, comprising a convolution layer, a ResNet-50 block with pre-trained weights, and a shallow neural network of classification. The proposed strategy achieved a classification accuracy of 94.37% on the testing dataset. The model is also compared with other existing neural network architectures and displayed superior performance for the SITF diagnosis.

Recently, Synchronous Reluctance Motors (SynRM) are finding place in several applications such as electric vehicles, aerospace, renewable and marine due to their inherent advantages such as the absence of rare earth materials and rotor windings, high performance, no demagnetization effects, and low cost. Therefore, the need to study, analyze and diagnose the fault at an early stage is becoming crucial for these machines. This work aims to develop analytical models necessary for incipient Stator InterTurn Faults (SITF) diagnosis on SynRM drives. A detailed analysis of SynRM with SITF has been attempted to comprehend the fault current behavior at various fault severities and operating conditions. Furthermore, steady state analysis has been carried out, by developing the sequence component models, which are critical for designing model-based diagnosis techniques. A method for the estimation of the fault currents is developed, which could help in post-fault operations of the drive, for bringing the fault current to a safe value. The conclusions of this study will be helpful for developing diagnosis techniques for SynRM drives, and to plan the post-fault operations. The experimental validation has been performed in a laboratory prototype using 3.7kW SynRM drive.

This work introduces a novel online signal processing and machine learning (ML) framework designed for the incipient diagnosis of stator inter-turn faults (SITF) in threephase squirrel cage induction motors. Addressing the critical need for incipient fault detection to prevent severe motor damage, the framework focuses on motor speed estimation, incipient fault detection, fault severity estimation, and faulty phase identification using only stator currents. A distinctive contribution lies in the proposed interacting multiple model (IMM) framework that leverages carefully selected motor current signatures as features, offering a comprehensive strategy for stator fault diagnosis not explored previously. The paper pioneers the use of the selected harmonics with ML models to estimate a fault severity indicator, which is developed based on insights from the motor’s physics of failure. Experimental validation showcases the fault indicator’s effectiveness under diverse operating conditions, demonstrating its utility in fault severity assessment. Suitable standalone ML model is selected, or an ensemble is constructed from a pool of ML models at each stage of the IMM framework. Further, a feature relevance analysis is also performed to garner insights into the contributions of each handpicked feature in predicting the fault indicator.

This paper presents the comparative analysis of two different methods of Phased locked loop (PLL) with Sliding Mode Observer (SMO) for position estimation in a Permanent Magnet Synchronous Motor (PMSM) drive. The objective of PLL is to improve the position estimation in drive and the conventional Quadrature PLL (QPLL) unable to track accurate position and shows double frequency oscillations. In order to overcome this, and to remove the harmonics, Dual second order generalized integrator (DSOGI) is proposed with SMO and its performance is analyzed and compared with conventional QPLL at various speed and torque operating conditions. Simulation results shows the better harmonics rejection capability of DSOGI.