Early failure monitoring of complex multi-variable dynamic processes based on empirical likelihood

Abstract

Dynamic processes are prevalent in actual industrial processes, and existing dynamic process research methods are lacking and ineffective in detecting faults. In this paper, an empirical likelihood-based early fault detection method for complex multivariable dynamic processes is designed. For the dynamic process with autocorrelation of process variables, a dynamic model is established and combined with the Kalman filter method, the estimated values and residuals of the dynamic process are monitored and analyzed respectively based on the empirical likelihood to achieve fast and accurate detection of early process faults. The application results of the dynamic process of blast furnace smelting show that the method has better detection effect and sensitivity than the traditional dynamic PCA, and meets the real-time demand of blast furnace production fault detection.