(384f) Faut Detection in a Closed-Loop Nonlinear Distillation

With the increasing demand for systems safety and reliability, the design of fault detection (FD) schemes received great attention in the last decade. This paper deals with an early fault detection method in order to improve the safety and continuity of production. It consists to the design of a residual generator, which is based on a Multi-Layer Perceptron Artificial Neural Network (MLP-ANN), to reveal any drift from the normal behaviour of the process. An analysis of the inputs choice, time delay, and hidden neurons is carried out. Three statistical criteria are used for the validation of the experimental data. Also, this study shows another technique for reduction of neural models into account the physical knowledge of the process. After describing the neural modelling, a realistic and complex chemical application as a distillation column is presented in order to illustrate the reliability of the prediction and model reduction. Satisfactory agreement between identified and experimental data is found and results show that the neural model successfully predicts the process behaviour. Then, the proposed FD method, which is based on the standardized residuals, is developed and tested on a real incident. The study shows that it detects the change presence, and pinpoints the moment it occurred. The experimental results demonstrate the robustness of the FD method.

Keywords: Reliability; anomaly detection; normal behaviour; modelling; neural network; distillation