Can J Chem Eng 2026 Qin A data‐augmented meta‐learning framework for few‐shot fault diagnosis under varying operating

Industrial process monitoring and fault diagnosis are essential for ensuring high-quality production and safe operation in manufacturing systems. However, insufficient labelled data, imbalanced fault samples, and frequent changes in operating conditions limit the diagnostic performance and generalization capability of existing deep learning-based methods. Moreover, conventional models often struggle to adapt rapidly to newly encountered operating conditions when only limited labelled samples are available. To address these challenges, this paper proposes a data-augmented meta-learning (DAML) framework for fault diagnosis under varying operating conditions. The framework aims to extract transferable meta-knowledge from different diagnostic tasks, thereby enabling rapid model adaptation to new operating conditions with limited labelled data. First, to alleviate fault class imbalance in practical industrial processes, a generative adversarial network is employed to generate minority-class fault samples and improve the balance among different fault classes. Multiple fault classification tasks are then constructed through random sampling within the meta-learning framework. Second, to address few-shot diagnosis under new operating conditions, a meta-learning model integrating a long short-term memory network and a fully convolutional network (Meta-LSTM-FCN) is developed. Meta-training is performed over a task domain comprising multiple related fault diagnosis tasks.