Weighted multi-source information fusion network-based framework for mechanical fault diagnosis

To alleviate the loss of potentially valuable and discriminative features as the number of networks increases in deep learning (DL). A weighted multi-source information fusion network-based framework is proposed for mechanical fault diagnosis. Firstly, a feature improved Deep Belief Network (FIDBNs) and a feature improved Convolutional Neural Network (FICNNNs) are designed to repetitively extract valuable information from original data. Meanwhile, an attention mechanism is adopted to further extract important fault characteristics Then, a weighted feature fusion module is used to integrate better for more diagnostic performance. Finally, an experimental dataset based on bearing failures is used to prove the effectiveness and superiority of the proposed framework compared with other methods.