This paper proposes a moving window recursive sparse principal component analysis (MWRSPCA)-based online fault monitoring scheme, aim at providing an online fault monitoring solution for large-scale complex industrial processes (e.g., chemical industry processes) with time-varying and dynamically changing characteristics. It establishes a sparse principal component analysis (SPCA) model based on the sliding window block matrixes to perform process monitoring and incorporates normal process monitoring data set simultaneously to the model training set to update the monitoring model online, so that the process monitoring model has strong adaptability to time-varying processes. A recursive computing procedure of the corresponding sparse loading matrixes is derived based on a modified rank-one matrix approximation algorithm, so that the computational complexity of the process monitoring model is greatly decreased and the real-time monitoring capability can be guaranteed. The effectiveness of the proposed method is verified by the benchmark Tennessee-Eastman process. Compared with traditional fault monitoring methods, the proposed method can effectively improve the fault detection accuracies with lower false alarm rates, which is suitable for the fault monitoring of time-varying, long-term and continuous complex industrial processes.

本文提出了一种基于移动窗口递归稀疏主成分分析（MWRSPCA）的在线故障监测方案，旨在为时变且动态变化的大型复杂工业过程（例如化工过程）提供在线故障监测解决方案。特征。它基于滑动窗口块矩阵建立稀疏主成分分析（SPCA）模型以执行过程监控，并将正常过程监控数据集同时纳入模型训练集以在线更新监视模型，从而使过程监控模型具有强大的功能。对时变过程的适应性。基于改进的秩一矩阵近似算法，推导了相应的稀疏加载矩阵的递归计算过程，从而大大降低了过程监控模型的计算复杂度，可以保证实时监控能力。基准田纳西-伊士曼过程验证了该方法的有效性。与传统的故障监测方法相比，该方法可以有效提高故障检测的准确性，降低误报率，适用于时变，长期，连续的复杂工业过程的故障监测。