A Lightweight Meta-Learning for Few-Shot Fault Diagnosis of Control Valve

Due to the distribution of control valves across different control loops within industrial systems, posing a significant challenge in collecting adequate fault data that covers various working conditions. Furthermore, industrial equipment often faces resource constraints when deploying data-driven models. To this end, a lightweight meta-learning fault diagnosis method is proposed to address the problem of few-shot fault diagnosis under varying operating conditions. First, to effectively address the unsteady time-varying fault signals, the two-dimensional (2-D) cloud model (CM) is employed to reconstruct the fault signals, thereby characterizing the states and features of the fault signals more comprehensively. Subsequently, a lightweight meta-learning model is constructed, the initial network parameters are randomly generated, and the parameters are updated by gradient descent. Finally, the parameters of the meta-learner are improved on the new task, and a few-shot data can be used to achieve rapid adaptation to new tasks. The experimental results show that the proposed method has significant advantages compared with other traditional meta-learning methods.