Enhancing Fault Diagnosis in Process Industries with Internal Variables of Model Predictive Control Article Swipe

View

Related Concepts

Principal component analysis Fault detection and isolation Internal model Fault (geology) Model predictive control Kernel principal component analysis Galvanization Autoencoder Process (computing) Engineering Component (thermodynamics) Computer science Reliability engineering Artificial intelligence Artificial neural network Control (management) Support vector machine Kernel method Materials science Operating system Composite material Thermodynamics Geology Seismology Layer (electronics) Actuator Physics

A Diallo , Lazhar Homri , Jean‐Yves Dantan , Frédéric Bonnet , Thomas Boeuf ·

YOU? · · 2024 · Open Access · · DOI: https://doi.org/10.1016/j.ifacol.2024.07.274 · OA: W4401630773

This paper introduces the use of internal variables, estimated through Model Predictive Control (MPC), for fault detection and diagnosis in process industries. To do so, a data-driven methodology is proposed. Three reconstruction techniques - Principal Component Analysis (PCA), Kernel Independent Component Analysis (KICA), and Autoencoder (AE) - are compared using data sets that combine plant measurements with internal variables. The methodology was tested on a hot-dip galvanizing line dedicated to the production of automotive steel and compared to the use of only plant measurements for the development of the reconstruction methods. The results showed that the incorporation of internal variables significantly enhances the overall fault detection rate. Finally, contribution plots were used to identify thefaulty sensor.