基于过渡时段分析的多阶段建模及故障诊断

Multiphase process modeling and fault detection based on transition phases analysis

针对间歇过程多阶段硬化分和误分类导致监控效果不理想的问题,提出了1种基于过渡时段分析的多阶段MPCA监控策略。该方法按照过程动态特性的变化,依据模糊C-均值算法(Fuzzy c-Mean,FCM)将过程数据划分为多个阶段,根据隶属度函数处理过渡阶段数据,建立阶段之间的联系;之后采用动态时间规整算法(Dynamic Time Warping,DTW)将分段后数据非线性化规整对齐,较好地解决了过渡阶段的监控问题,并通过在青霉素发酵过程的应用验证了该方法的有效性。

According to the multiple phases with hard-partition and misclassification problems, an improved multi-stage MPCA method for on-line monitoring based on transition phases analysis was proposed. The proposed method firstly divided the process data into multiple stages with FCM algorithm, dealed with the transition phase data with fuzzy membership grade, then used DTW algorithm to synchronizing batches and building multi-stage model. The simulation result of penicillin fermentation process platform shows that monitoring performance of this method is more reliable.