A product envelope spectrum generated from spectral correlation/coherence for railway axle box bearing fault diagnosis

The (pseudo-) cyclostationarity-based spectral analysis tools can be generated by arithmetic averaging or weighted averaging of spectral correlation/coherence for rotating machinery fault diagnosis. However, conventional arithmetic or weighted averaging can hardly adequately eliminate broadband interfering noise under harsh operating conditions, thus compromising fault diagnosis capability. To address this problem, motivated by the properties of the convolution theorem, a new cyclic spectral analysis tool called (pseudo-) cyclostationarity-based product envelope spectrum is developed via the product of multiple spectral frequency components of spectral correlation/coherence (each spectral frequency component can be regarded as a special envelope spectrum containing certain useful information). A generalized construction framework is established and a specific construction methodology is proposed, which incorporates a novel estimation and discrimination approach of frequency domain fault information distribution. The frequency domain product enables the proposed method to both adaptively extract fault features distributed in multiple frequency bands and effectively eliminate interference from in-band and noise-dominated frequency bands. The performance of the developed approach is verified using simulation signals, experimental signals of different railway axle-box bearing faults and a field test signal of rolling element fault of railway locomotive axle-box bearing and finally by comparison with state-of-the-art methods. The results demonstrate that the proposed method has the capability to reduce acyclic-stationary noise and enhance (pseudo-) cyclostationary components, and thus is capable of effectively diagnosing different faults of railway axle-box bearings.