Application of crested porcupine optimized-back propagation neural network under feature dimensionality reduction for rolling bearing fault diagnosis

To address the demands for precision and generalization in fault diagnosis of rolling bearings within resource-limited industrial settings, an intelligent diagnostic model utilizing Mean Impact Value (MIV), Crested porcupine optimizer (CPO) algorithm, and Backpropagation Neural Network (BPNN) (MCB-Net) is proposed. First, MIV ranks and filters features based on feature significance, thereby diminishing input dimensionality and enhancing model interpretability. Second, the CPO technique is implemented to improve BPNN parameters, thereby improving global search capabilities and expediting convergence, and addressing the conventional BPNN’s propensity to become trapped in local optima. Finally, MCB-Net was assessed utilizing rolling bearing fault datasets from Case Western Reserve University and Southeast University. Experimental results indicate that MCB-Net surpasses 97 % classification accuracy on three distinct datasets, exhibiting minimum performance variability compared to other approaches, confirming the model's efficacy and practicality.