Fault diagnosis of industrial processes using dynamic global–local preserving projection and genetic algorithm‐based feature selection

In the realm of industrial production, where the scale is continuously expanding, chemical process variables often exhibit complex characteristics such as nonlinearity, multi‐modality, and dynamic behaviour. Traditional fault diagnosis methods based on multivariate statistics, like principal component analysis (PCA), generally operate under the assumption that current values are independent of historical statistical values. Additionally, most of these fault diagnosis algorithms focus on feature extraction, which, despite reducing the number of features, often results in a loss of the original data's characteristics. To address this issue, the fault diagnosis and monitoring algorithm introduced in this study integrates genetic algorithm (GA)‐based feature selection with dynamic global–local preserving projection (DGLPP). This approach not only accounts for the dynamic nature of multivariate data but also reduces dimensions while retaining the original features of the data. The effectiveness of this methodology is demonstrated through comparative experiments using the Tennessee Eastman process dataset. This paper compares the proposed model with four existing models: dynamic principal component analysis (DPCA), global–local preserving projection (GLPP), DGLPP, and GA‐DPCA and establishes a significant enhancement in performance with the proposed method.