Abstract In general, batch processes cover two-directional dynamics, in which the batch-wise dynamics are related to different operation modes, while the time-wise variations correspond to different phases within each batch. The problem of unevenness is common as a result of various factors, particularly in multimode batch processes. In order to address these issues, this paper proposes a two-directional concurrent strategy of mode identification and sequential phase division for multimode and multiphase batch process monitoring with the uneven problem. Firstly, pseudo time-slices are constructed in order to describe the process characteristics regarding the sample concerned, which can preserve the local neighborhood information within a constrained searching range and effectively prevent the synchronizing problem caused by uneven lengths. Secondly, mode identification is conducted along the batch direction and the phase affiliation is sequentially determined along time direction by determining the changes in variable correlations. The two-directional steps are implemented alternatively in order to identify the mode and phase information, which can also guarantee the time sequence within each mode. Thirdly, for online monitoring, the mode information and phase affiliation are simultaneously judged in real time for each new sample, from which the fault status is distinguished from the phase shift. The division results can indicate the critical-to-mode phases from which a certain mode begins to be separated into different sub-modes. In order to illustrate the feasibility and effectiveness of the proposed algorithm, it is applied to a multimode and multiphase batch process (namely an injection molding process) with the uneven problem.

中文翻译：

---

不等长多模多相批量过程监控的模式识别和顺序分相的双向并发策略

摘要 一般来说，批处理过程涵盖了双向动态，其中批处理动态与不同的操作模式有关，而时间变化对应于每个批处理中的不同阶段。由于各种因素，特别是在多模式批处理中，不均匀问题很常见。针对这些问题，本文提出了一种模式识别和顺序分相的双向并发策略，用于多模式和多阶段批处理监控不均匀问题。首先构造伪时间片来描述有关样本的过程特征，可以在有限的搜索范围内保留局部邻域信息，有效防止长度不均匀引起的同步问题。其次，通过确定变量相关性的变化，沿批处理方向进行模式识别，并沿时间方向依次确定相位关联。双向步骤交替执行，以识别模式和相位信息，这也可以保证每个模式内的时序。第三，对于在线监测，同时实时判断每个新样本的模式信息和相隶属关系，从中区分故障状态和相移。划分结果可以指示某个模式开始分离成不同子模式的临界模式阶段。为了说明所提算法的可行性和有效性，