Fault Diagnosis of Quadruped Robot Joint Modules Based on TDIC and Random Forest

As robotics technology advances, quadruped robots are increasingly instrumental in diverse applications, including rescue operations and healthcare. However, the intricate operating conditions expose the components within the joint modules of these robots to alternating and impact loads. Consequently, research into signal processing and fault diagnosis techniques for planetary gearboxes in joint modules holds significant engineering importance. Given the challenges associated with fault signal extraction from planetary gearboxes, this study proposes a novel fault feature extraction method based on Time-Dependent Intrinsic Correlation (TDIC), integrated with a random forest classifier for fault classification. This approach leverages signal frequency analysis to adaptively extract segments characterized by high correlation among different physical field signals as fault features. The random forest model is utilized for fault classification and its performance is compared against CNN, KNN, and XGBoost models. Experimental results validate the superior fault classification efficacy of the proposed method.