A Regularized DBN Based on Fault Diagnosis Model for Inductively Coupled Plasma System

The data-driven approach is an ideal method for modern system fault diagnosis. However, due to the complexity and comprehensiveness of inductively coupled plasma (ICP) generating systems, it is necessary to monitor the parameters of each sensor and the movement of each position, then, how to extract features effectively from massive monitoring data is the key to fault diagnosis of the system. This paper proposes a fault diagnosis model for Deep Belief Networks (DBN) based on regularized constraint. The regularized constraint is imposed on the Restricted Boltzmann Machine (RBM) to preserve the feature manifold structure in the data while ensuring the sparsity of the learning model. Finally, the model is trained for fault classification of inductively coupled plasma generating systems. The research results show that our method has a more accurate fault diagnosis rate than the existing fault diagnosis methods.