A Hybrid Intelligent Fault Diagnosis Framework for Rolling Bearings Using CNN‐Based Feature Extraction and MFOA‐Optimized KELM

This study presents a hybrid intelligent diagnostic framework that integrates convolutional neural networks (CNNs) with kernel extreme learning machines (KELM), aiming to overcome key difficulties in extracting discriminative fault representations and achieving precise classification of bearing fault conditions. First, the pooling layer of the CNN is innovatively designed to use stochastic pooling following a uniform distribution, thereby enhancing the retention of spatial characteristics and promoting diversity within extracted features. Such a design enables dynamic feature learning, compresses feature dimensionality, and suppresses irrelevant signal components present in rolling bearing vibration data. Then, the moth‐flame optimization algorithm (MFOA) is applied to dynamically optimize the penalty coefficient and kernel width parameters of KELM, thereby constructing a fault classification model based on MFOA‐KELM. Finally, simulations are conducted using the rolling bearing dataset from Paderborn University, and the proposed method is compared with traditional approaches. The experimental results demonstrate that the proposed model achieves high diagnostic accuracy and excellent generalization performance, with fault state recognition rates exceeding 99%, thereby validating the effectiveness of the proposed method.