(60db) Data-Driven Techniques for Process Fault Classification

Data-driven techniques are crucial in the detection and isolation of faults in multivariate systems that may exhibit complex or nonlinear dynamics. The availability of accurate mathematical model is not always guaranteed, and, thus, the ability to extract useful features and patterns from big data is invaluable. As a result, applications involving the identification or classification of faults benefit from the use of these robust data-driven techniques, thereby reducing maintenance costs over time and improving the overall operating safety of the process. The use of multiple fault detection techniques to build a system of binary classifiers, capable of multiclass fault classification has shown promising preliminary results, based on a case study carried out on the benchmark Tennessee Eastman process, where it was capable of outperforming deep learning techniques, such as artificial neural networks, applied to the same process.