Fault Diagnosis for Cascaded Multilevel Inverters: An In-Memory Feature Extraction Approach With Convolutional Neural Networks

This is the authors' abstract. We don't add key points for this paper.

The extensive convolution computations, repetitive data transfers, and storage in convolutional neural networks (CNNs) result in prolonged fault diagnosis time. To address this issue, an in-memory convolutional neural network (IMCNN) is proposed for fast open-circuit fault (OCF) diagnosis of cascaded H-bridge multilevel inverters (CHMLIs). To be specific, an extended matrixization encoding (EME) method is first designed to convert the original 1-D diagnostic voltage signal. Then, convolutional calculations are physically performed based on an in-memory feature extraction (IM-FE) using memristor crossbar arrays, with pooling operations used to extract fault features. The extracted features are then input into a fully connected layer, and the Adam algorithm is used to train the model parameters to counter resistance drift. The trained model is then used to identify fault types. Experimental results on the three-phase five-level grid-connected CHMLI platform validate the performance of the proposed approach.