Fault Diagnosis of Voltage Source Inverter using Quadrant Transformation and Pattern Recognition

Abstract

This work purposes to address the need for open switch fault diagnosis in 3-phase voltage source inverters used in industry, with the goal of enhancing consistency, minimizing downtime, reducing maintenance costs, and ensuring optimal performance. Open switch faults in switching devices can considerably reduce the efficacy as well as consistency of a voltage source inverter. This leads to costly downtimes and increases security risks. This paper proposes a novel technique for open switch fault diagnosis that consists of feature extraction, rule-based modeling, as well as machine learning and pattern recognition in fault diagnosis with the combination of direct-quadrant transformation and convolutional neural networks. This technique successfully detects faulty switches by integrating direct-quadrant transformation, feature selection, as well as fault classification. The direct-quadrant transformation aids in distinguishing between healthy as well as defective states by mapping signals into different patterns. These patterns are then improved after feature selection, increased diagnostic accuracy, and machine learning-based fault identification, resulting in reliable and efficient diagnoses. Two methods for open switch fault diagnosis utilized in the paper are pattern recognition and fault diagnosis using a feature-based fault diagnosis system. A feature-based fault diagnosis system uses rule-based as well as machine learning models for fault classification, subsequently extracting features from direct-quadrant transformed data as well as selecting appropriate features. Convolutional neural networks with pattern recognition methods are applied in fault diagnosis to transform direct-quadrant transformed input images. These methods were compared on the basis of robustness, computational efficiency, as well as diagnostic accuracy in many operating conditions.