(463a) A Machine Learning-Based Approach for Chemical Process Sustainability Analytics

Sustainability has become a focal point of future development manufacturing sectors, as it aims at continuous improvement of economic, environmental and social sustainability performance of manufacturing systems. In the process of developing strategies and solutions for goal achievement, the time-variant sustainability performance needs to be monitored, and the collected system input-output data needs to be analyzed. For a sophisticated manufacturing system, the amount and types of process data and beyond could be huge. The sustainability status of a chemical system is highly dependent on process parameters that need to be optimized for improving process sustainability performance. Data-driven modeling can build a mapping from process parameters to sustainability indicators. This is a very challenging problem because of high dimensionality of the input space (i.e., large number of process parameters) and many uncertain internal and external parameters.

In this paper, we propose a machine learning approach for chemical processes sustainability analytics. LASSO, a machine learning technique, can be used to build a regression model between process parameters and sustainability indicators. Such a technique can perform both parameter selection and regularization in order to enhance the prediction accuracy and interpretability of the statistical model. Using this technique, the most important process parameters with the highest effect on sustainability of the system can be specified. The trained model then can be used to generate the trade-off among sustainability objectives. This type of information is very valuable for process engineers to better understand the effect of process parameters on sustainability performance of the system. A case study on the methanol synthesis process is illustrated to show the effectiveness of the proposed approach.

___________________________________________________________________________

* All correspondence should be addressed to Prof. Yinlun Huang (Phone: 313-577-3771; Fax: 313-577-3810; E-mail: yhuang@wayne.edu).