Open-Set Fault Diagnosis Based on 1D-ResNet With Fusion of Cross-Class and Extreme Information for Out-of-Distribution Detection

This article proposes a unified open-set fault diagnosis (OSFD) strategy based on deep learning classification models and posterior detectors. We utilize a 1-D residual network (1D-ResNet) to build a classification model and introduce an out-of-distribution (OOD) detection method that integrates the fusion of collective and extreme information (FCEI). The Feature Center Distance Matrix (FCDM) is derived from Mahalanobis distances between class feature centroids. For test samples, cosine similarity between their class-wise Mahalanobis vectors and corresponding FCDM rows is computed as the discriminative collective measure, and finally, using the maximal softmax value of the model prediction class as the extreme information. By combining and analyzing cross-class and extreme information, the method can identify OOD samples more efficiently. Experiments on the Tennessee Eastman (TE) benchmark process show that the proposed method can not only correctly classify the known faults but also accurately identify unknown faults.