Deep Learning Diagnosis of Membrane Fouling Based on the Gramian Angular Field Method

Because of the complex formation mechanism of MBR (Membrane Bioreactor) membrane fouling, the existing membrane fouling diagnostic model lacks the performance of spatiotemporal multi-feature fusion, making it difficult to judge the membrane fouling diagnosis under complex working conditions accurately. Therefore, this paper proposes a diagnostic model based on ResNet learning architecture and Gramian Angular Field aiming to improve the accuracy and reliability of membrane fouling diagnosis. First, the contaminated data are transformed using the Gramian Angular Field (GAF) method, and the effective features of the temporal series in them are analyzed and preserved. Then, the processed images are fed into ResNet incorporating the spatiotemporal attention mechanism for feature extraction. Finally, the fault codes are output using the SoftMax classifier and the performance of the model is compared with other networks. The experimental results show that the ResNet model incorporating the Gramian Angular Field method has high comprehensive performance and provides a new solution for membrane fouling failure analysis.