Blast Furnace Ironmaking Process Fault Detection Using Canonical Variate Analysis and Support Vector Data Description

Anomalies detection plays a vital role in ensuring the stability and smoothness of the blast furnace ironmaking process. Due to its complexity, data-driven fault detection has become a hot topic for the blast furnace ironmaking process. To monitor the process of the blast furnace, this paper proposes a fault detection method based on Canonical Variate Analysis and Support Vector Data Description (CVA-SVDD). CVA is a state space-based monitoring tool that is effective in dealing with process dynamics. SVDD is a data domain description method, also known as a one-classification method, which can be used for outlier detection and has the advantage of solving the problem of describing nonlinear and non-Gaussian distributed data. Specifically, a CVA model is first constructed for the process variables to generate a series of feature vectors, and then the features are further monitored using SVDD, which is used to fit the data into a hypersphere so that its radius can be used as a health indicator in fault detection. A real blast furnace ironmaking process is employed to verify the effectiveness of the proposed method.