Multichannel fault diagnosis of planetary gearboxes using hybrid feature selection integrating mRMR and binary human evolutionary optimization

Selecting an optimal feature subset with high discriminative power from high-dimensional data is crucial for enhancing the efficiency and accuracy of multichannel fault diagnosis in planetary gearboxes. Hence, we propose a hybrid feature selection framework integrating maximum relevance and minimum redundancy (mRMR) with a novel binary variant of the human evolutionary optimization algorithm (binary-HEO). Initially, mRMR is employed to eliminate irrelevant features and construct a refined candidate set. Subsequently, binary-HEO serves as an efficient wrapper to identify the optimal feature subset from the reduced search space. In binary-HEO, a novel fitness function based on K -nearest neighbors clustering and cross-entropy is designed to enable a more discriminative evaluation of feature subsets while enhancing computational efficiency over conventional evaluation criteria. Finally, support vector machines are utilized for fault diagnosis using the selected features. Experimental results demonstrate that the proposed approach consistently achieves superior multichannel fault diagnosis performance for planetary gearboxes.