Rapid fault localisation based on acoustic-vibration heterogeneous feature fusion under non-stationary operating conditions

Noise source localisation is a key prerequisite for accurate fault diagnosis and predictive maintenance of industrial rotating machinery. However, under non-stationary conditions like variable speed, equipment noise exhibits complex, time-varying characteristics. Traditional methods struggle to capture dynamic features and accurately locate faulty components, hindering diagnostic efficiency. This paper proposes a novel acoustic-vibration fusion-based method for localising noise sources in rotating machinery under variable conditions. First, a short-time analysis method for heterogeneous acoustic and vibration characteristics is proposed, defining four real-time indicators to generate corresponding acoustic-vibration cross-correlation matrices. Next, a dynamic weighted fusion mechanism fuses various indicator matrices by utilising the dispersion characteristics of correlation coefficients. Then, discrimination of fault characteristics under variable operating conditions is improved by optimising window length and overlap rate. Experimental results show the method achieves accurate noise source localisation under variable speed conditions across different equipment and fault types. Moreover, it maintains robust performance under acoustic signal energy attenuation and propagation path deviations. Compared with mainstream technologies, it exhibits superior localisation accuracy and computational efficiency in environments with strong noise and reverberation. The method provides a real-time, robust technical solution with significant engineering application value for industrial equipment fault warning and health management.