基于正交信号修正与高效偏最小二乘的质量相关故障检测方法

Quality-related fault detection method based on orthogonal signal correction and efficient PLS

偏最小二乘(PLS)是一种广泛应用于多变量统计过程监控中的有效算法,高效偏最小二乘(EPLS)是近年提出的一种PLS改进算法,在质量相关故障检测中具有良好的检测效果,但当测试数据含有质量无关故障时,EPLS算法的误报率偏高,可能导致误报警,对工业过程中的故障检测有较大影响.为降低检测质量无关故障的误报率,将EPLS结合4种正交信号修正(OSC)方法提出4种OSC-EPLS算法.用质量无关故障样本建立OSC模型对在线监测数据进行预处理,将处理后的信息用EPLS算法进行故障检测,误报率明显降低.最后结合田纳西-伊斯曼工业过程,应用OSC-EPLS、PLS、EPLS算法进行故障检测,分别比较误报率和有效报警率的大小,体现所提出算法在故障检测中的优势.

The partial least squares algorithm(PLS)is an effective method,which has been widely used in multivariate statistical processes.Efficient projection to latent structures(EPLS)is an improved PLS algorithm proposed in recent years and has a good detection effect in quality-related fault detection.However,when the test data has a quality-independent failure,the false alarm rate of the EPLS algorithm is high,which may lead to false alarms and has a great influence on fault detection in industrial processes.In order to reduce the false alarm rate of detection-independent faults,four OSC-EPLS algorithms are proposed in combination with four orthogonal signal corrections(OSC)and the EPLS.In this paper,we establish an OSC model with quality-independent fault samples to preprocess on-line inspection data,and use the EPLS algorithm for fault detection to reduce the false alarm rate.Finally,combined with Tennessee Eastman industrial process,four kinds of the OSC-EPLS,PLS,and EPLS algorithms are used for fault detection.The false alarm rate is compared with the effective alarm rate to reflect the advantages of the proposed algorithm in fault detection.