SRSayanjit Singha Roy

Sayanjit Singha Roy

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Sayanjit Singha Roy has a total of 38 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Image Visibility Patch-Aided Partial Discharge Recognition Framework for Identifying Defects in XLPE Cables

    by Sayanjit Singha Roy, Ashish Paramane, Jiwanjot Singh, Soumya Chatterjee

    Partial discharge (PD) is a critical degradation phenomenon in cross-linked polyethylene (XLPE)-insulated polymeric power cables, which is responsible for premature failure if left unattended. Therefore, accurately identifying PD defects is essential to prevent such incidents in the XLPE cable. This study proposes a novel image visibility graph (IVG) theory-aided phase-resolved partial discharge (PRPD) pattern analysis and recognition framework employing an optimally tuned bi-directional long short-term memory (bi-LSTM) classifier for automated PD detection. To this end, several PD defects have been synthetically emulated inside an 11 kV XLPE cable, and the PD signals corresponding to each type of defect are measured using an HFCT sensor. From the obtained HFCT data, PRPD patterns were generated, which were converted into connected graphs using image visibility graph (IVG). Moreover, image visibility patches (VPs) were computed from the graph-converted PRPD plots to quantify the intricate pixel-level changes due to altering discharge patterns. Following that, the frequency of occurrences (FOCs) of the unique visibility codes was computed from the extracted VPs. The visibility features were further employed to train the bi-LSTM classifier for PD defect identification, which yielded high accuracy. Ablation studies with classical CNN models and comparison with previously reported state-of-the-art methods also revealed superior efficiency of the proposed PD detection methodology, suggesting its potential application for automated health monitoring of XLPE cable insulation.


  • Condition Monitoring of Polymeric Insulators: A Remote Diagnostic Framework With Improved Intermediate Hydrophobicity Grade Detection

    by Sayanjit Singha Roy, Ashish Paramane, Biswarup Ganguly, Arup Kumar Das, Subho Paul, Soumya Chatterjee

    Polymeric insulators deliver superior pollution performance owing to their excellent hydrophobic characteristics. However, due to aging, temporary loss of hydrophobicity occurs and continuous water films start to form, leading to dry-band arcing and subsequently, flashover of the insulators. Hence, accurate identification of different hydrophobicity grades (HGs) is essential. Considering this issue, a novel image visibility theory-aided remote HG detection framework with improved intermediate HG identification has been proposed in this contribution. To achieve this, image visibility patches (VPs) were extracted to quantify the intricate pixel-level alterations taking place particularly for the intermediate HGs. In addition, the concept of uniform visibility patches is proposed to reduce computational burden of the conventional VP extraction process along with providing additional noise resiliency. The employed HG detection methodology infused with a transformer encoder-based classifier module was trained on experimental HG images acquired from an 11kV silicone rubber (SiR) insulator and tested on benchmark data utilizing a cross-dataset learning strategy, which yielded satisfactorily high HG detection performance, Ablation study in addition to comparison with existing studies revealed superior efficiency of the proposed framework with highly impressive reliability and robustness in identifying different HGs, suggesting it's potential application for remote condition monitoring of outdoor polymeric insulators.


  • Characterization and Identification of Electrical Tree Growth Stages Inside High-Voltage Cable Insulation

    by Sayanjit Singha Roy, Ashish Paramane, Jiwanjot Singh, Soumya Chatterjee, Zelin Hong, Xiangrong Chen

    Formation of micro-cavities takes place inside high-voltage cross-linked polyethylene (XLPE) cable insulation due to severe electrical and thermal aging, which further act as sources of void discharges and lead to the development of dendritic carbonyl channels, commonly referred to as electrical trees. Accompanied by intermittent partial discharge (PD) activities, the electrical tree growth bridges the entire insulation over time and results in insulation breakdown, which exposes the cable to fatal safety incidents. Therefore, accurately identifying various stages of electrical tree growths is crucial to prevent such safety instances as well as ensure reliable and extended operation of the XLPE cable. To this end, the present study proposes an automated electrical tree growth monitoring scheme employing a customized 19-layer residual neural network (ResNet) model. For this purpose, microscopic images of different tree growth stages inside XLPE insulation samples were acquired at different temperatures (30, 50, and 70 °C) and augmented using a deep convolutional generative adversarial network (DCGAN). Further, using fractal dimension ( df ) of the tree structures, different electrical tree propagation levels were characterized. Finally, different tree growth stages were identified using the proposed 19-layer ResNet model. This approach yielded satisfactorily high recognition accuracies irrespective of varying temperatures, engaging a significantly reduced computational time compared to classical convolutional neural network (CNN) models. This infers its potential application for automated health monitoring of XLPE cable.


  • Automated Space Charge Classification Inside ±500-kV HVDC Cable Insulation Using Fusion of Superpixel and Deep Features for Remote Condition Assessment

    by Sayanjit Singha Roy, Ashish Paramane, Jiwanjot Singh, Arup Kumar Das, Soumya Chatterjee, Xiangrong Chen

    The accumulated space charges cause electrical field distortion, which is fatal to the safe and reliable operation of polymeric high-voltage direct current (HVDC) cables. Hence, this paper aims to detect and classify the space charges to ensure reliability and a longer operating life of HVDC cables. To achieve this, experiments were carried out on cross-linked polyethylene (XLPE) insulation samples and space charge distributions were recorded under altering electric fields (10–50 kV/mm) and at different temperatures (30–70 °C). Subsequently, super-pixel color features were extracted from the space charge images using the simple linear iterative clustering (SLIC) algorithm. In addition, deep features were extracted using the AlexNet convolutional neural network (CNN) model. The fusion of the handcrafted and deep features was fed to three benchmark machine-learning classifiers for the recognition of different space charge accumulation categories. The method delivered high recognition performance in spite of altering electric fields and varying temperatures. As a result, the proposed framework can detect space charges in HVDC cable insulation in real time.


  • Accurate Sensing of Insulator Surface Contamination Using Customized Convolutional Neural Network

    by Sayanjit Singha Roy, Ashish Paramane, Jiwanjot Singh, Soumya Chatterjee, Arup Kumar Das

    Accurate sensing of contamination on the insulator surface is vital for the reliable operation of transmission lines. Hence, the present work aims to develop a deep learning framework for remote and accurate contamination sensing on the surface of outdoor insulators. The experiment is conducted on an 11 kV porcelain disc insulator to generate an extensive database of images representing different insulator surface conditions, i.e., clean surface and surface with sand, mud, and ash contaminations. The captured images were fed to a customized convolutional neural network (CNN) architecture for automated feature extraction and recognition. The proposed CNN model delivers appreciably high recognition accuracy at significantly reduced training time for sensing various insulator contaminations compared to other benchmark CNN models (AlexNet, VGGNet16, and ResNet50). Moreover, the proposed framework delivers an excellent recognition performance in sensing surface contamination under different lighting conditions. Therefore, the proposed methodology can be implemented for the condition monitoring of real-life insulators.


  • Remote Condition Monitoring of HVDC Cable Insulation Using Deep Learning-Aided Space Charge Classification

    by Sayanjit Singha Roy, Ashish Paramane, Jiwanjot Singh, Fanbo Meng, Chao Dai, Arup Kumar Das, Soumya Chatterjee, Xiangrong Chen, Yasuhiro Tanaka

    The homo-charges accumulated inside the polymeric insulation of high voltage dc (HVdc) cables increase the electrical field beyond the uniform value, whereas the hetero-charges reduce it. Such uneven electrical field distribution is detrimental to HVdc cables’ longevity and thus, needs to be identified accurately. To this end, this article proposes a deep learning (DL) framework for the automated detection of space charges in HVdc cable insulation. The space charges inside the cross-linked polyethylene (XLPE) insulation samples were measured under varying electric fields (10–50 kV/mm) as well as at different temperatures (30 °C–70 °C) conditions. The obtained space charge profiles corresponding to no charge, hetero-charge, and homo-charge categories were fed to two benchmark convolutional neural network (CNN) models, i.e., AlexNet and VGGNet16, for automated feature extraction and classification purpose. The convolutional blocks of the two CNN models were sequentially fine-tuned using transfer learning (TL) to observe the variation in classification accuracies. The proposed approach delivers appreciably high classification accuracies irrespective of varying electric field and temperature conditions with reduced computational time toward distinguishing different categories of space charge accumulations compared to other CNN models. Hence, it can be potentially used for real-time HVdc insulation diagnostic purposes.


  • Accurate Detection of Bearing Faults Using Difference Visibility Graph and Bi-Directional Long Short-Term Memory Network Classifier

    by Sayanjit Singha Roy, Soumya Chatterjee, Saptarshi Roy, Pradip Bamane, Ashish Paramane, U. Mohan Rao, Muhammad Tariq Nazir

    This paper proposes a novel bearing fault detection framework for the real-time condition monitoring of induction motors based on difference visibility graph (DVG) theory. In this regard, the vibration signals of healthy as well as different rolling bearing defects were acquired from both fan-end and drive-end accelerometers. These data were recorded for three different bearing defects and under four loading conditions. The acquired vibration time series were converted to a topological network using DVG. From the transformed vibration data in the graph domain, degree distribution (DD) was selected as features to discriminate different fault networks. Using analysis of variance (ANOVA) test and false discovery rate (FDR) correction, most discriminative DD features were selected. These features were subsequently fed as inputs to a deep learning model, i.e., a bi-directional long short-term network (bi-LSTM) classifier for fault classification. In this study, 112 classification problems have been addressed, and for all of them, the proposed approach delivered very high fault detection accuracy. Finally, the classification performance of the proposed framework is compared with other well-known machine-learning classifiers all of which delivered satisfactory results.


  • Recognition of Hydrophobicity Class of Polymeric Insulators Employing Residual Morphological Neural Network and Granulometry-Based Image Analysis

    by Soumya Chatterjee, Sayanjit Singha Roy, Arpan Chatterjee, Biswarup Ganguly, Subho Paul

    Hydrophobicity is an important property of polymeric insulators which allow formation of discrete water droplets on the insulator surface. However, with aging or under extreme humidification, the insulator surface loses its hydrophobic property, resulting in continuous water channel formation, which may lead to dry band arcing and subsequently flashover, thereby affecting the long-term performance of the insulator. Therefore, accurate recognition of hydrophobic class (HC) is important as it provides information regarding the surface condition of the insulator. Considering the above-said facts, this article investigates the feasibility of employing residual morphological neural network (Res-Morph-NN) to accurately recognize different HC of polymeric insulators. In addition, for better understanding of the distribution of water droplets on the insulator surface, a novel approach based on multiscale mathematical morphology viz. granulometry is used in this article. Our investigations revealed significant alterations in the granulometric patterns with the change in HC of polymeric insulators. The efficacy of the proposed framework employing Res-Morph-NN has been validated on HC images obtained from experiment as well as on large set of images obtained from an online database, which returned satisfactory results. The proposed method can be efficiently used to discriminate different hydrophobicity classes which can be implemented for monitoring insulator surface condition in real time.


  • Focal Epileptic Area Recognition Employing Cross EEG Rhythm Spectrum Images and Convolutional Neural Network

    by Sudip Modak, Sayanjit Singha Roy, Rohit Bose, Soumya Chatterjee

    Accurate detection of focal seizure area through EEG screening is important to remove the affected regions of human brain, prior to surgery. Considering the aforesaid fact, in this paper, we propose a novel approach for automated detection and classification of focal electroencephalography (EEG) signals. In this contribution, we procured the focal and non focal EEG signals recorded from the temporal lobe epileptic patients and decomposed the recorded EEG data into several brain rhythms to obtain their time variation of different neural oscillations. Then, we selected a random EEG signal from each rhythm as a reference and performed cross wavelet transform (XWT) of the rest of the EEG rhythms with their respective chosen reference rhythms to examine their nature of correlation in both time scale and time-frequency frame. Finally, the obtained cross-rhythm spectrum time-frequency image plots were fed to a customized CNN model for the purpose of automated feature extraction and classification of EEG signals. The performance of the proposed CNN model was also compared with two pre-trained benchmark CNN models namely AlexNet and VGGNet16, respectively. We observed that the proposed EEG signal classification framework has delivered 100% accuracy for the delta (δ) rhythm and that too at a significantly reduced training time compared to the existing state-of-the-art CNN models. Interestingly, the detection accuracy was also found to be the highest for a particular EEG rhythm, which can be proven useful in real time analysis for accurate diagnosis of focal epilepsy.