Rocket Force University of Engineering
The National Natural Science Foundation of China (61673387，61833016，61903375)
在工业生产中，对系统进行故障检测具有十分重要的作用.改进的偏最小二乘(Modified partial least squares，MPLS)是在PLS基础上提出的一种扩展算法，在质量相关故障检测中具有良好的检测效果.但当测试数据中含有质量无关故障时，MPLS算法漏报率较高；另外，MPLS算法的阈值为固定值会导致其误报率增加，这些问题会对工业过程监控产生较大影响.为此，本文提出一种基于局部信息增量与MPLS的质量相关故障检测方法(Local information increment-MPLS, LII-MPLS)，在MPLS基础上，通过使用局部信息增量技术对测试数据进行实时更新检测后，质量相关故障的漏报率明显降低.同时过程复杂化导致静态控制限不能满足故障检测的需求，现存的动态控制限适用范围具有一定局限性，因此本文改进静态控制限将其推广为局部动态阈值.最后，通过田纳西伊士曼过程(Tennessee Eastman process,TEP)仿真实验验证了所提算法的有效性.
Fault detection of the system has a very important role in industrial production. Modified partial least squares (MPLS) is an extended algorithm based on PLS, which has a good detection effect in quality-related fault detection. However, when the test data contains quality-unrelated faults, the MPLS algorithm has a high fault missed alarm rates; in addition, the fault false alarm rates of MPLS will increase because of its static threshold, and these problems have a great influence on industrial process monitoring. To this end, this paper proposes a quality-related fault detection method based on local information increment and MPLS (LII-MPLS). On the basis of MPLS, the fault missed alarm rates of qualityrelated fault is significantly reduced by using local information incremental technology to update and detect the test data in real time. Meanwhile, the complexity of the process results in static control limits that cannot meet the needs of fault detection and existing dynamic control limits have certain limitations. Therefore, this paper improves the static control limit and generalizes it as a local dynamic threshold. Finally, the effectiveness of the proposed approach is verified on an industrial benchmark of Tennessee Eastman process.