(601c) Multi-Rate Hard and Soft Sensors Fusion for Monitoring Chemical Processes

Product quality variables of chemical processes, such as product compositions and purities, are usually measured off-line using analytical techniques in a laboratory. The sampling frequencies are often too low for process monitoring and control purposes. To obtain more frequent estimations of product quality variables, software sensors (also known as soft sensors or inferential sensors), including partial least squares (PLS) models [1, 2] and other data-driven mathematical models [3, 4], have been widely applied over past three decades [5]. These data-driven soft sensors predict the infrequent measured product quality variables using frequently sampled inputs, such as temperature, pressure, and flow rate. To address the time-varying characteristics of chemical processes, caused by factors like fouling, catalyst aging and operating condition changes, several online adaptive soft sensor algorithms have been proposed, including the Kalman filter based method [6] and the means and variance update approach [7].

The accuracy of individual sensor can be easily impaired by varieties of factors, such as instrument malfunctioning, operator mistakes and the inherent measurement error. To improve the monitoring accuracy and reliability for chemical processes, the sensor fusion methodology has been proposed to combine the estimations from lab analyzer and soft sensor using Kalman filter [8, 9]. The lab measurements serve as references and are used to correct the soft sensor estimation when they are available. With the improving capability of sampling instruments, the frequent online measurements of product quality variables by hardware sensors (or online analyzers) become increasingly available. Yet the sensor fusion approaches using online analyzer as well as lab analyzer and soft sensor have not been well discussed.
Here we present a multi-rate sensor fusion scheme based on maximum-likelihood approach [10]. It was examined in a chemical process at Dow. The product impurity was estimated by fusing the measurements from a lab analyzer, an online analyzer and a PLS soft sensor with different sampling rates. The soft sensor was developed using mean and variance update algorithm to track the time-varying process dynamics. The online analyzer measurements were filtered to reduce their variability. These two sensor measurements were fused together with the lab measurements using maximum-likelihood approach. It has shown that the sensor fusion approach improves the process monitoring reliability, quantified by the rates of correctly identified impurity alarm, comparing to the case of using an individual sensor.

References
1. Wold, S., M. Sjöström, and L. Eriksson, PLS-regression: a basic tool of chemometrics. Chemometrics and intelligent laboratory systems, 2001. 58(2): p. 109-130.
2. Chiang, L.H., et al., Diagnosis of multiple and unknown faults using the causal map and multivariate statistics. Journal of Process Control, 2015. 28: p. 27-39.
3. Bishop, C.M., Neural networks for pattern recognition. 1995: Oxford university press.
4. Jang, J.-S.R., C.-T. Sun, and E. Mizutani, Neuro-fuzzy and soft computing: a computational approach to learning and machine intelligence. 1997.
5. Kadlec, P., B. Gabrys, and S. Strandt, Data-driven Soft Sensors in the process industry. Computers & Chemical Engineering, 2009. 33(4): p. 795-814.
6. Chen, K., et al., Soft Sensor Model Maintenance: A Case Study in Industrial Processes∗∗The authors would like to acknowledge the support from the DOW chemical company and the natural sciences and engineering research council of Canada (NSERC). IFAC-PapersOnLine, 2015. 48(8): p. 427-432.
7. Lu, B. and L. Chiang, Semi-supervised online soft sensor maintenance experiences in the chemical industry. Journal of Process Control, 2017.
8. Xie, L., et al. Kalman filtering approach to multirate information fusion for soft sensor development. in Information Fusion (FUSION), 2012 15th International Conference on. 2012. IEEE.
9. Fatehi, A. and B. Huang, Kalman filtering approach to multi-rate information fusion in the presence of irregular sampling rate and variable measurement delay. Journal of Process Control, 2017. 53: p. 15-25.
10. Ernst, M.O. and M.S. Banks, Humans integrate visual and haptic information in a statistically optimal fashion. Nature, 2002. 415(6870): p. 429-433.