Performance Analysis of Identification Strategies Based on the Partial Least Square Method

ZURITA NOELIA; SANCHEZ MABEL; RODRIGUEZ LEANDRO; CEDEÑO MARCO

Bahía Blanca

Congreso; IX congreso Argentino de Ingeniería Química; 2017

PLAPIQUI - CONICET

Process monitoring is turning into data driven strategies due to its relatively easyto-apply methods. These approaches have several advantages and drawbacks when theinformation of the process have to be clear in order to solve a diverse root cause of anabnormal behavior. Due to the diverse process measurements the selection of anappropriately monitoring strategy is critical. Also the fault diagnosis and the prediction ofthe product quality plays an important role in the statistical monitoring of the process in order to ensure the safe and profitable scenario. In this sense several approaches have beendeveloped to accurately signal the measurements that better explain the withdrawal from thenormal operation. The selected method to monitor the benchmark process is the extendedpartial least square strategy due to its ability to handle a predictor and a predicted matrix.Also the performance of some of the lately developed fault identification strategies arecompared under a benchmark that is characterized to holding continuous processmeasurements and quality variables making a large amount of process measurements. Forthis, the use of some performances indices is needed so that a good identification of a faultcan be characterized different from an incorrect one.