An Antinoise Feature Extraction and Improved Harris Hawks Optimization for On-Load Tap Changer Mechanical Fault Diagnosis

Traditional on-load tap changer (OLTC) mechanical fault diagnosis methods often focus on vibration burst data in the diverter switch moving stage but neglect the entire vibration signal of the shifting process. This limitation results in insufficient mining of equipment failure information and difficult to diagnose gear transmission system faults. Additionally, laboratory-based fault simulation experiments commonly overlook the influence of noise on fault diagnosis, while the huge computational demand makes deep learning difficult to process the entire OLTC vibration signal. To solve the above problems, a novel OLTC mechanical fault diagnosis method is proposed. First, the multichannel vibration signal of the entire shifting process is transformed into the Euclidean norm of short-time Fourier transform (STFT) matrix elements. Subsequently, novel vibration frequency component amplitude entropy (VFCAE) and frequency statistical feature (FSF) are extracted from the matrix. Following this, limited-patience Harris Hawks optimization (LPHHO) is proposed to obtain better-performing parameters of support vector machine (SVM), by forcing the optimization algorithm to jump out of the local optimal. Thereafter, fault simulation experiments with vibration noise and Gaussian white noise prove the strong antinoise capabilities of the proposed VFCAE and FSF. Furthermore, the stable performances of the proposed LPHHO in OLTC and cross-disciplinary open datasets prove the robustness of LPHHO. The sensitivity of LPHHO is proved by the high optimization accuracies on OLTC datasets with white Gaussian noise. Finally, the proposed method can diagnose transmission gear faults that are confused with diverter switch faults in other existing OLTC fault diagnosis methods.