Deep learning-based belt conveyor roller fault detection research

Aiming to address the low efficiency and accuracy issues of traditional methods for diagnosing faults in belt conveyor rollers under noisy conditions, we propose a fault detection model named DSCNN-BiLSTM. This model integrates a multichannel Deep Separable Convolutional Neural Network (DSCNN) with a Bidirectional Long Short-Term Memory (BiLSTM) network, enhanced by an attention mechanism. The multi-channel Deep Separable Convolutional Network in this model extracts diverse local features by applying different convolutional kernels to each channel, which reduces the number of parameters and computational requirements, thereby significantly improving computational efficiency. The BiLSTM network manages time-series dependencies, allowing the model to leverage both local features and global sequence information effectively. The incorporation of a channel attention mechanism enables the model to adaptively select channels containing fault features, enhancing both accuracy and noise resistance. Experimental results demonstrate that this model performs effectively in fault detection, offering certain reference value and practical significance.