Generalized Out-of-distribution Fault Diagnosis (GOOFD) via Internal Contrastive Learning

TL;DR

A unified fault diagnosis method based on internal contrastive learning and Mahalanobis distance is put forward to underpin the proposed generalized framework and can be applied to multiple fault diagnosis tasks and achieve better performance than the existing single-task methods.

Abstract

Fault diagnosis is crucial in monitoring machines within industrial processes. With the increasing complexity of working conditions and demand for safety during production, diverse diagnosis methods are required, and an integrated fault diagnosis system capable of handling multiple tasks is highly desired. However, the diagnosis subtasks are often studied separately, and the current methods still need improvement for such a generalized system. To address this issue, we propose the Generalized Out-of-distribution Fault Diagnosis (GOOFD) framework to integrate diagnosis subtasks. Additionally, a unified fault diagnosis method based on internal contrastive learning and Mahalanobis distance is put forward to underpin the proposed generalized framework. The method involves feature extraction through internal contrastive learning and outlier recognition based on the Mahalanobis distance. Our proposed method can be applied to multiple faults diagnosis tasks and achieve better performance than the existing single-task methods. Experiments are conducted on benchmark and practical process datasets, indicating the effectiveness of the proposed framework.

Figures (5)

• Figure 1: (a) The difference between the GOOFD framework and existing single-task methods. A single-task approach can only handle a specific task. However, the GOOFD framework is dedicated to integrating and solving various fault diagnosis tasks through a unified method. (b) The basic process of our approach addresses multi-task issues.
• Figure 2: Flow chart of the TE processLAWRENCERICKER1996205
• Figure 3: (a) The confusion matrix for the OSFD task in the TE process with Fault 6 as the unknown class. (b) we visualize the distribution of prediction scores generated by each method. The images in the second row are the enlarged display of the yellow region in the first-row images, providing a better view of details. The prediction scores of the Softmax, OpenMax, and CenterLoss methods are softmax probabilities. Scores below the threshold are classified as unknown. The prediction scores of the EOW-Softmax method constitute the (K+1)th-dimensional probability of its output, estimating open-world uncertainty. Scores surpassing the threshold are classified as unknown. The GEN method can be applied to any pre-trained softmax-based classifier to generate the entropy-based score, and scores below a threshold are determined to be unknown. The prediction score of the PULSE method is the output of the discriminator. Scores below the threshold are labeled as unknown classes. The prediction score of ICL+ and our method are the opposites of the output score. Data with prediction scores below the threshold are classified as the unknown class. The smaller the overlap between the prediction scores of known and unknown categories, the better the method's performance.
• Figure 4: The histogram of prediction scores for unknown classes in ICL+ and our methods. The prediction scores of the ICL+ and our method shown in Fig. \ref{['te_cm']} (b) are overly dense, making it difficult to discern specific distributional features. Therefore, we present a clearer display of the distribution of the prediction scores for unknown classes.
• Figure 5: The average F1-score of OSFD tasks in TE process when applying various threshold ${\theta}_{u}$.