Fault diagnosis of planetary gearbox using multi-criteria feature selection and heterogeneous ensemble learning classification

Fault feature selection and early fault diagnosis for planetary gearbox are important tasks and have been widely investigate. This paper proposes a novel fault diagnosis scheme for planetary gearbox using multi-criteria fault feature selection and heterogeneous ensemble learning classification. Vibration signals collected by the acceleration sensors are imported for fault diagnosis of planetary gearbox. High dimension fault features are extracted by analyzing the vibration signals in time domain, frequency domain and time-frequency domain. The criteria for selecting lower dimension optimal fault features of planetary gearbox are designed, and the mathematic model for fault feature selection with multi-criteria is established. After that, a new feature selection method using Multi-Objective Evolutionary Algorithm based on Decomposition (MOEA/D) is applied to obtain diverse lower dimension quasi optimal fault feature subsets. Then each quasi optimal fault feature subset is transferred to a base classifier for primary fault diagnosis. Those base classifications are performed by support vector machine and sparse Bayesian extreme learning machine respectively. Dezert-Smarandache rules are applied for classifier-level fusion to achieve and evaluate overall accuracy of the fault diagnosis for planetary gearbox. Experimental results state that the proposed method constantly gets diverse lower dimension quasi optimal fault features smoothly, and significantly improves the accuracy and robustness of fault diagnosis.

Automated summary

A novel fault diagnosis scheme for planetary gearbox using multi-criteria fault feature selection and heterogeneous ensemble learning classification is proposed in this paper, significantly improving the accuracy and robustness of fault diagnosis. Techniques such as vibration signal analysis, MOEA/D for feature selection, and Dezert-Smarandache rules for classifier-level fusion are utilized to achieve diverse lower dimension quasi optimal fault features and enhance overall accuracy of the fault diagnosis.