Fault Detection in Industrial Wastewater Treatment Processes Using Manifold Learning and Support Vector Data Description
Tian Chang, Tianlong Liu, Xiaobo Ma, Qiyue Wu, Xinyuan Wang, Jinlan Cheng, Wenguang Wei, Fengshan Zhang, Hongbin Liu
Abstract
The treatment of industrial wastewater is becoming increasingly important due to growing environmental concerns. Untreated wastewater carries hazardous substances that can severely damage water resources and lead to further negative impacts on the environment. To ensure treated wastewater meets the discharge standards, real-time monitoring for prompt fault detection, adjustments, and system stability is required. This study introduces a novel fault detection approach employing uniform manifold approximation and projection (UMAP) coupled with support vector data description (SVDD). This innovative approach tackles the challenges posed by high-dimensional, non-Gaussian, and nonlinear process data by projecting it into a more manageable lower-dimensional feature space. The UMAP retains the data’s global structure and reduces dimensionality, yielding significant intragroup and intergroup distances in low-dimensional mapping. After dimensionality reduction, the SVDD algorithm is adeptly employed to refine the fault detection process further. Experiments using benchmark simulation data demonstrate the high sensitivity of the UMAP-SVDD model to fault information and its high generalization ability for modeling in different scenarios. This model significantly outperforms conventional linear fault detection models in terms of detection rates and versatility and offers a promising new approach for wastewater treatment fault detection, ensuring rapid adaptation and system integrity restoration.