中文摘要：

针对基于不同展开方式的多向主元分析（MPCA）方法在线应用时各自存在的缺陷,提出一种改进的基于变量展开的MPCA方法,实现间歇过程的在线监控与故障诊断。该方法采用随时间更新的主元协方差代替固定的主元协方差进行T2统计量的计算,充分考虑了主元得分向量的动态特性;同时引入主元显著相关变量残差统计量,避免SPE统计量的保守性,且该统计量能提供更详细的过程变化信息,对正常工况改变或过程故障引起的T2监控图变化有一定的识别能力;最后提出一种随时间变化的贡献图计算方法用于在线故障诊断。该方法和MPCA方法的监控性能在一个青霉素发酵仿真系统上进行了比较。仿真结果表明：该方法具有较好的监控性能,能及时检测出过程存在的故障,且具有一定的故障识别和诊断能力。

英文摘要：

Batch processes are very important in most industries and are used to produce high-value-added products, which causes their monitoring and control to emerge as essential techniques. Several multivariate statistical analyses, including multi-way principal component analysis （MPCA）, have been developed for the monitoring and fault detection of batch processes. In this paper, an improved statistical batch monitoring and fault diagnosing approach based on variable wise unfolding was proposed to overcome the drawbacks of traditional MPCA and the AT method proposed by Aguado. The proposed method did not require prediction of the future values while the dynamic relations of data were preserved by using timevarying score covariance, and principal-component-related variable residual statistics was introduced to replace SPE-statistics, thus avoiding the conservation of SPE statistical test and providing more explicit information about the process conditions. As a result, the root cause that violated the Hotelling T2 test but still satisfied the SPE test could be unambiguously identified, which was impossible in the MPCA. In addition, time-varying contribution charts were proposed to diagnose anomalous batch process. The proposed method was applied to detecting and identifying faults in the simulation benchmark of fed-batch penicillin production. The simulation results clearly demonstrated the power and advantages of the proposed method in comparison to the MPCA and AT method.