A fault recognition method based on acoustic pulse signal change

Abstract

This invention relates to the field of signal analysis technology, and particularly to a fault identification method based on changes in acoustic pulse signals. The method involves establishing a basic database; performing difference analysis on faulty operating acoustic waves and faulty loop currents in the basic database compared to baseline operating acoustic waves and baseline loop currents to obtain significant feature thresholds; identifying suspicious operating acoustic waves and suspicious loop currents based on real-time operating acoustic waves, real-time loop currents, and significant feature thresholds; building a fault identifier based on the attention mechanism in machine learning, pre-training the fault identifier until convergence, and then adjusting the attention weights based on suspicious operating acoustic waves and suspicious loop currents; adjusting the significant feature thresholds until the confidence level of the fault identifier is greater than or equal to a preset threshold; and using the fault identifier to identify faults in target equipment. This invention can significantly improve the ability to distinguish between acoustic pulse signals of similar faults.

Images