Real-Time Fault Diagnosis in High-Frequency VFDs for EM Systems Using Recurrent Neural Networks

The reliability of high-frequency Variable Frequency Drives (VFDs), which are integral to modern electromechanical systems, is paramount for ensuring operational safety and minimizing costly downtime. This paper presents a novel, data-driven framework for the real-time diagnosis of incipient electrical faults within VFDs, leveraging the power of Recurrent Neural Networks (RNNs). Unlike traditional methods that rely on manual feature extraction from spectral analysis, our approach utilizes a Long Short-Term Memory (LSTM) network to directly process the raw, high-frequency time-series data of the VFD’s output current and voltage waveforms. The LSTM architecture is specifically chosen for its ability to learn and remember long-term temporal dependencies within the signal, which are characteristic of subtle fault signatures such as IGBT open-circuit faults, DC-link capacitor degradation, and phase imbalances. A high-fidelity simulation environment is used to generate a comprehensive dataset of VFD operational data under both healthy and various faulty conditions. The trained LSTM model demonstrates a high degree of accuracy in not only detecting the presence of a fault but, more importantly, in classifying its specific type with minimal latency. The results establish the superiority of this end-to-end learning approach, which bypasses the need for domain-specific signal processing expertise and provides a robust, scalable pathway for implementing intelligent, on-line predictive maintenance systems in critical industrial and energy applications.

کلیدواژه ها:

Fault diagnosis ، Variable Frequency Drive (VFD) ، Recurrent Neural Network (RNN) ، LSTM ، time-series analysis ، predictive maintenance ، power electronics