Fault Diagnosis of Rotational Machine Using Remote Sensor Based on Neural Network for Operational Transfer Path Analysis

Abstract

This study introduces an innovative approach for enhanced vibration data acquisition and fault diagnosis in rotational machinery systems, particularly relevant in real-world industrial settings. Due to the operational environment, vibration data are gathered from locations distant from the area of interest, leading to challenges in extracting robust fault features and heightened susceptibility to external noise. Addressing this challenge, this paper employs Operational Transfer Path Analysis (OTPA) to derive the transfer function between the vibration excitation source and the measurement point. This method facilitates the inverse estimation of the signal of the excited source from the receiver. Subsequently, this estimated excitation signal is fed into a diagnostic model to identify system failures. The transfer function and the fault diagnosis model are constructed using neural network architectures, enabling better adaptation to operational conditions and system-induced nonlinearities. The efficacy of the proposed approach is demonstrated through case studies focusing on hydraulic piston pumps in engineering equipment and drivetrains of electric vehicles.