Bearing fault diagnosis via parameter optimization feature mode decomposition and improved 1DCNN

Owing to the harsh operating environment, bearing vibration signals tend to exhibit non-stationary characteristics and are highly susceptible to heavy noise interference. As a result, the fault features are not prominent, which is a challenging for intelligent fault diagnosis. To address the above issue, a rotating machinery fault diagnosis method based on parameteroptimized feature mode decomposition (FMD) and an improved one-dimensional convolutional neural network (I1DCNN) is proposed. First, to eliminate the influence of manually set FMD parameters, the Subtractive Average-Based Optimization (SABO) algorithm is employed to automatically optimize the key parameters of the FMD algorithm. Second, the optimized FMD is used to filter and denoise the bearing fault signals to obtain the desired mode components and remove the interference information. Finally, by introducing sparse channel attention mechanism into the conventional one-dimensional convolutional network, we developed an improved model, termed I1DCNN, which is designed to learns discriminative features intelligently from the decomposed feature mode and realized bearing fault diagnosis. Experimental results indicate that this method markedly enhancing the accuracy and robustness of bearing fault identification under heavy noise condition, and demonstrates clear advantages over other representative methods.