Optimization of digital twin model for switchgear vibration monitoring and computer-aided fault diagnosis
TL;DR
This paper builds a multi-measurement point and multi-condition vibration data set based on the existing vibration digital twin model, and proposes a joint optimization method that integrates error evaluation, swarm intelligence global search, gradient and surrogate model local optimization, and online update of time series residuals.
Abstract
To address the issue of concealed vibration faults in distribution network switch cabinets and the difficulty of timely early warning through traditional inspections, this paper builds a multi-measurement point and multi-condition vibration data set based on the existing vibration digital twin model, and proposes a joint optimization method that integrates error evaluation, swarm intelligence global search, gradient and surrogate model local optimization, and online update of time series residuals. At the fault diagnosis end, a one-dimensional convolution and time series encoding fusion network with multiple-channel vibration feature input is designed, and transfer learning and digital twin output are introduced to construct posterior probability of fault patterns and health indices. Experimental results show that the optimized model achieves a reduction of more than 40% in RMSE with actual measured vibrations under multiple conditions, the fault identification accuracy increases from 88.4% to 97.3%, and the average inference delay when 20 switch cabinets are simultaneously connected is less than 40 ms, and the early warning lead time is approximately 5-7 hours. This verifies the effectiveness of the proposed method in digital twin modeling and computer-aided fault diagnosis for distribution equipment, providing an engineering-implementation technology path for state maintenance and intelligent operation and maintenance of distribution equipment.