基于MCUSUM-ForeCA的微小故障检测

Small Fault Detection Based on MCUSUM-ForeCA

针对传统多元统计方法对过程的微小变化不敏感且无法反映过程时序特性的缺点,把多元累积和(MCUSUM)与可预测元分析(Fore CA)相结合,提出了MCUSUM-Fore CA方法。该方法利用MCUSUM对数据进行处理,累积数据的历史信息,使用Fore CA计算可预测元矩阵,选取合适的可预测元,构造能够反映系统运行状况的统计量,以此建立了微小故障检测模型。将此模型用于在线监控,并详细分析了MCUSUM的步长对微小故障检测效果的影响。最后,在TE模型上的仿真结果表明了MCUSUMFore CA方法在微小故障检测方面的良好性能。

Considering the traditional multivariate statistics' insensitivity to the gradual small shifts and the incapability in better reflecting dynamic timing characteristics of the process,a new multivariate statistical process monitoring method was proposed which has the multivariate cumulative sum( MCUSUM) combined with the forecastable component analysis( Fore CA) to work out this MCUSUM-Fore CA method,in which,the MCUSUM answers for processing of data and accumulating historical information; and Fore CA calculates the forecastable component matrix and selects appropriate forecastable components to construct the statistic so as to establish a small fault detection model. Applying this model to online monitoring and analyzing the influence of MCUSUM's step on the fault detection results and the simulating the results on the Tennessee Eastman( TE)model prove the MCUSUM-Fore CA method's good performance in the small fault detection.