Development of the new soft sensor method and the application to process control

Abstract

Soft sensors are widely used in chemical plants to estimate a process variable which is difficult to measure online. An inferential model is constructed between variables which are easy to measure online and one which is difficult to measure online, and an objective variable is estimated by the model. However, soft sensors have some practical difficulties. One of the crucial difficulties is that predictive accuracy drops due to changes of state of chemical plants, sensor and process drifting and so on. If a problem of degradation of soft sensors is not solved, it is difficult to identify reasons of abnormal situations. There is no effective method to solve these difficulties under the circumstances. In this study, we have developed the new soft sensor method that combines independent component analysis (ICA) and PLS. ICA is a method that is used in many fields such as signal processing. We can comprehend the state of plant by an ICA model and estimate an objective variable by a PLS model, updating PLS model appropriately. We showed the superiority of this method over a traditional one with real industrial data.