不等长批次过程的有序时段划分、建模及故障检测

对具有不等长时段的多时段批次过程进行监测是十分重要而且具有一定难度的. 时段在批次间的错位现象导致时间方向的不同过程特性混合在一起, 这给时段分析以及在线应用带来了一系列的问题. 为了解决不等长所带来的问题, 本文提出一种基于不等长时段有序识别及建模的故障检测方法. 该方法的主要贡献包括以下方面：1) 该方法通过步进地衡量过程的变量相关性对模型精度以及监测性能的影响, 自动有序地识别出每个不等长时段; 2) 在每个时段内,通过对不规则的过程数据进行整合建立了时段模型以捕捉不规则的时段特性; 3) 本文提供了一种简单而有效的在线判断新样本隶属时段和监测其运行状态的方法. 最后, 本文通过一个实例–具有不等长批次长度的注塑过程阐述了本方法的有效性.

Sequential unequal-length phase identification and modelling method for fault detection of varying-duration batch processes

The work of monitoring multiphase batch processes with unequal phase durations is of great importance but difficult. Due to misaligned phases, process characteristics are mixed along time direction which causes problems in phase analysis and modeling as well as online application. In order to solve the uneven-length problem, this paper proposes a sequential unequal-length phase identification and modeling-based fault detection method. The main contribution of the proposed method includes: 1) multiple unequal-length phases are sequentially identified by evaluating the changes of process variable correlations step-wise regarding their influences on model accuracy and monitoring performance; 2) irregular phase characteristics are captured by irregular data-arranging-based modeling strategy; 3) the proposed method provides an easy but effective way to judge the phase affiliation and check the operation statuses of new samples in real time. Its online monitoring performance is illustrated by an injection molding process with varying durations.