Contrastive Conditional Adversarial Autoencoder With Class-Specific Forces for Imbalanced Open-Set Fault Detection

In real-world industrial scenarios, fault detection faces the widely recognized challenge of data imbalance, which not only refers to the scarcity of fault data but also includes the imbalance in healthy data. This article is concerned with imbalanced open-set fault detection (IOSFD), a practical yet challenging scenario in industrial applications where multiple healthy operating conditions and multiple fault types are imbalanced. In this article, we propose a new contrastive conditional adversarial autoencoder for IOSFD. It constructs an end-to-end unified model based on multi-class known healthy and faulty data to address the reliance of traditional methods on fault samples, while optimizing with class-specific weighted forces to ensure equal attention to imbalanced known classes. Input and feature reconstruction conditioned on operating modes are utilized to learn a compact decision plane and achieve both unknown fault detection and known data classification. Significantly, we formulate the optimization objective of conditional reconstruction based on contrastive learning and introduce adversarial training to further enhance the model’s performance. The effectiveness of the proposed method is validated through real-world pipeline leak detection and Tennessee-Eastman multi-fault detection. Note to Practitioners—Existing data-driven fault detection methods often assume a balanced training dataset, where the quantities of data for various fault types and healthy conditions are roughly equal. However, in real-world industrial settings, fault data is frequently scarce, and the amount of data for different healthy operating conditions can vary significantly. This imbalance can lead to biased models that do not perform well in practice. To address this challenge and leverage the available data effectively, this paper develops a contrastive conditional adversarial autoencoder. The proposed method directly models multiple classes of known fault and healthy data to identify unknown faults, eliminating the need for unknown fault samples in model training. Additionally, the model employs class-specific weighted forces to ensure that all categories, regardless of their frequency in the training set, are given equal attention during learning. This helps to mitigate the effects of data imbalance, ensuring that the model learns from both the abundant healthy data and the sparse known fault data. The proposed approach has been validated through real-world applications, including pipeline leak detection and the Tennessee-Eastman multi-fault detection task. These case studies demonstrate the effectiveness of the method in practical settings. This technique offers a robust solution for industries where fault data is scarce and operating conditions are varied, making it a valuable tool for practitioners working with imbalanced datasets.