Fault diagnosis framework for system and speed specific variations using Machine learning Techniques

Abstract

Machine fault diagnosis is crucial for ensuring the reliability, safety, and efficiency of machinery across various industries and applications. Physics-based and data-driven approaches are the most popular methods for fault diagnosis. Physics-based approaches are suitable when a mathematical model of the system is available, while the latter is suitable when the data for the healthy and fault conditions are available. Hybrid approaches, which combine physics-based and data-driven methods, are also popular for their improved performance. Conventional data-driven methods rely on machine-specific models customized for individual machine types, limiting their generalizability. To address this, machine-independent (system-independent) fault diagnosis was explored by developing models capable of diagnosing faults across different capacity machines without requiring separate fault models for each machine. In addition to system variations, speed variations also affect the performance of fault models. To address the performance deterioration, a speed-independent fault diagnosis model was then developed to ensure accurate fault detection across varying speeds. To evaluate the effectiveness of system-independent fault diagnosis, experiments were conducted using synchronous generators with capacities of 3kVA and 5kVA. Initially, statistical features were extracted from the time, frequency, and wavelet domains and used to train an SVM classifier. Among these, wavelet domain features provided moderate baseline accuracies of 61.23%, 78.00%, and 66.20% for R, Y, and B phase faults, respectively. To address system-specific variations, locality-constrained linear coding (LLC) was applied to map the extracted features into a higher-dimensional space where they become linearly separable. In the transformed space, system-specific variations and fault-specific variations are more easily distinguishable. This approach enabled a linear SVM to capture fault-specific patterns........