Maximum correlated Kurtosis deconvolution and application on gear tooth chip fault detection

In this paper a new deconvolution method is presented for the detection of gear and bearing faults from vibration data. The proposed maximum correlated Kurtosis deconvolution method takes advantage of the periodic nature of the faults as well as the impulse-like vibration behaviour associated with most types of faults. The results are compared to the standard minimum entropy deconvolution method on both simulated and experimental data. The experimental data is from a gearbox with gear chip fault, and the results are compared between healthy and faulty vibrations. The results indicate that the proposed maximum correlated Kurtosis deconvolution method performs considerably better than the traditional minimum entropy deconvolution method, and often performs several times better at fault detection. In addition to this improved performance, deconvolution of separate fault periods is possible; allowing for concurrent fault detection. Finally, an online implementation is proposed and shown to perform well and be computationally achievable on a personal computer.