Cascading effects detection and multiple components faults classification in power converters based on multi-sensor fusion approach and 1D-convolutional neural network

Power converters are essential in various applications such as aerospace, photovoltaic systems, smart grids, and electric vehicles etc. Component failures in these converters pose significant risks to overall system reliability, especially in safety-critical applications. Existing fault classification methods focus mostly on single faults within individual components but overlook the complex interactions of multiple faults in real-world scenarios. This paper addresses the challenge of identifying and classifying multiple faults in power converters using cascading failures. By analyzing the interconnections between components, we identify components that are prone to degrade simultaneously. This approach simplifies the fault classification process by highlighting these dependencies. Leveraging a multi-sensor fusion approach, incorporating data like RMS output voltage, input current, peak-to-peak voltage ripple, efficiency, and frequency domain analysis to construct a comprehensive dataset tailored for fault classification. A 1D-CNN is used to classify multiple faults, demonstrating its efficacy across various converter circuits including boost, flyback, and SEPIC converters. Visualization analysis further elucidates the inner workings of the proposed methodology 1D-CNN model using t-SNE. Classification accuracy reached 96.80%, 98.00%, and 95.00% for the respective converter circuits, demonstrating the potential of proposed approach in fault diagnosis of converters. The proposed methodology enables early detection of fault propagation within the converter circuits, efficient maintenance scheduling, minimizing downtime and extending the operational lifespan of critical industrial systems.