Application of a Statistical-Based Feature Extraction Method for Harbor Crane Bearings in Fault Diagnosis

In response to the weak early fault features of harbor crane bearings that are difficult to extract and monitor online, this article proposes a signal feature extraction method based on maximum likelihood estimation and a generalized likelihood ratio index based on frequency domain statistical features, used for the identification of early bearing faults. The normalized envelope spectrum and sample labels of the signal are taken as inputs data, with several similar statistical models designed under this hypothesis. Key parameters are obtained for each statistical model via the maximum likelihood ratio method, and from this, a fault diagnosis index based on the log-likelihood ratio is designed. The proposed methodology is validated using public datasets and a scaled-down harbor test bench. The research results indicate that the proposed signal feature extraction method is effective in extracting key signal features, and the proposed fault diagnosis index can accurately identify early weak signal faults.