Fault Diagnosis of Blast Furnace Iron-making Process with A Novel Deep Stationary Kernel Support Vector Machine Approach

In blast furnace iron-making process (BFIP), there is a significant push to maintain a stable iron-making process and ensure process at maximum efficiency. While some control systems can compensate for multiple types of disturbances when faults occur, some significant process faults often require precise human intervention to avoid safety hazards. Therefore, it is crucial to develop an efficient and stable diagnostic system to efficiently identify these faults so that operators can deal with them quickly. This paper focuses on a novel approach called deep stationary kernel support vector machine (DSKSVM) for nonstationary BFIP fault diagnosis. To eliminate the impact of nonstationary property on modeling, stationary subspace analysis (SSA) is adopted to estimate consistent underlying features. Then, design a multi-layer stacked deep kernel network to explore deep nonlinear information further. A support vector machine-based classifier and corresponding two-tier model optimization algorithm are constructed to isolate data from different types to achieve fault diagnosis task. At last, an actual case study based on BFIP presents the effectiveness of DSKSVM. The proposed method has outstanding results in fault diagnosis, and is verified that the performances of stationary construction and online computation times are superior to other methods.