(394d) Integrating Process Knowledge with Data-Driven Approaches for Industrial Soft Sensor Development

In industrial production, the in-situ measurement of critical quality variables is usually limited by current technical advances and high equipment costs^[1,2]. Data-driven soft sensor technology has become a promising solution to this challenge^[3]. Traditional methods mainly rely on data-driven modeling strategies emerged decades ago have attracted a great attention and achieved a wide application in process industry^[4]. With the surge of all kinds of new Artificial Intelligent, AI, methods, data-driven methods are getting even more popular^[5]. On the other hand, the curriculum of chemical engineering has been established for over a century, including systematic theory and methodology, which have been the starting point for process design and operation, and shouldn’t be ignored when AI dominates the application of soft sensor development. A key assumption for data-driven method is that data collected from facilities are redundant, and shall contain all information of the studied process, but it is not always the case due to the availability of measurements. In such a situation, how to incorporate the well-established domain knowledge to data-driven method would be an interesting topic in near future.

This paper proposes a novel soft sensor modeling strategy that synergistically integrates process knowledge with data-driven methods. By constructing secondary variables to represent the inherent correlation between variables, and merging them with original measured data as input variables, this fusion method not only retains the advantages of data-driven techniques in dealing with complex nonlinearities, but also leverages the prior information embedded in process knowledge, thus ensuring the calculation efficiency and significantly improving the prediction accuracy and generalization ability. Experimental results on the Dow dataset demonstrate the superior performance and reduced computational costs of the proposed method, which provides a promising way for soft sensor development in process industry.

References:

[1] Kadlec, P., Grbić, R., & Gabrys, B. (2011). Review of adaptation mechanisms for data-driven soft sensors. Computers & chemical engineering, 35(1), 1-24.

[2] Jiang, Y., Yin, S., Dong, J., & Kaynak, O. (2020). A review on soft sensors for monitoring, control, and optimization of industrial processes. IEEE Sensors Journal, 21(11), 12868-12881.

[3] Sun, Q., & Ge, Z. (2021). A survey on deep learning for data-driven soft sensors. IEEE Transactions on Industrial Informatics, 17(9), 5853-5866.

[4] Kadlec, P., Gabrys, B., & Strandt, S. (2009). Data-driven soft sensors in the process industry. Computers & chemical engineering, 33(4), 795-814.

[5] Perera, Y. S., Ratnaweera, D. A. A. C., Dasanayaka, C. H., & Abeykoon, C. (2023). The role of artificial intelligence-driven soft sensors in advanced sustainable process industries: A critical review. Engineering Applications of Artificial Intelligence, 121, 105988.