Fault Detection and Diagnosis in a Chemical Process using Long Short-Term Memory Recurrent Neural Network

Long short-term memory recurrent neural networks have been proved to be especially useful for learning sequences comprising longer-term patterns of unknown length as they are able to preserve long-term memory. The learning of higher level temporal features could be achieved by stacking recurrent hidden layers for faster learning with sparser representations. In this work, we propose a novel approach to data driven fault detection and diagnosis of a chemical process. The method employs a state-of-the-art deep-learning technique, viz. the long short-term memory recurrent neural network. An application of the proposed approach is performed with realistic simulated data from a chemical process literature benchmark. Results point out an excellent performance when compared to already published linear and nonlinear fault detection and diagnosis methods.