Robust and Adaptive Bayesian Network Soft Sensor Development for Multi-Rate and Noisy Data

Abstract

For efficient process control and monitoring, accurate real-time information of quality variables is essential. To predict these quality (or slow-rate) variables at a fast-rate, in the industry, inferential/soft sensors are often used. However, most of the conventional methods for soft sensors do not utilize prior process knowledge even if it is available. The prediction accuracy of these inferential sensors depends mainly on the quality of available data, which can be affected by significant noise, outliers, drift and possible sensor failures. To address these issues, in this work, soft sensors based on Bayesian network (BN) are developed. Compared to the existing soft sensors, the proposed approach will allow users to integrate prior knowledge into the BN structure. Due to the probabilistic nature of BNs, variances of measurement noises and disturbances between hidden states are simultaneously estimated. Moreover, BN based soft sensor can naturally handle multi-rate, missing data, outliers or the problem of drift, which usually arises during online soft sensor implementation stage. Performance of the proposed approach is demonstrated on a benchmark flow-network problem and an industrial process. It is observed that Bayesian network based soft sensors are able to give significantly better and more reliable estimates compared to the conventional approaches.