A novel one-dimensional convolutional neural network with parallel attention for fault diagnosis of rigid guides

Rigid guides are key components of the vertical shaft hoisting system as they guide and stabilize the conveyance during its operation. Excessive clearance and misalignment of the rigid guides can significantly impact conveyance, and serious faults can lead to safety accidents. To accurately diagnose the categories and severity of faults in rigid guides, a new parallel attention one-dimensional convolutional neural network (PA1DCNN) is proposed. The PA1DCNN utilizes multiple small-size convolutions to effectively reduce the dimensionality of the collected eddy current signals. This approach not only reduces the computational cost of the model but also enhances the representation of shallow features in the input data. A parallel channel/spatial attention model is designed to enhance the learning of sensitive channel information and signal segments that are indicative of faults. Experimental results demonstrate the effectiveness of the PA1DCNN. The recognition rates of PA1DCNN for normal rigid guides, clearance and misalignment faults are 100%, 99.21%, and 99.92%, respectively. Moreover, the recognition rate for fault severity reaches an impressive 86.12%, surpassing the performance of the six state-of-the-art networks.