(101e) Machine Learning Aided Process Design and Intensification Using Sparse Experimental Data: An Ammonia Production Example

With the burgeoning growth of computing power and algorithmic advances, machine learning (ML) has become a powerful tool to identify multivariate relationships and find statistical patterns behind large amounts of data. ML has also been extensively leveraged to optimize chemical process design and operations by generating reduced order surrogate models from well-established mechanistic models [1-2] and/or enhancing the solution efficiency of sophisticated optimization algorithms [3]. A research question lies in how ML techniques can assist the development of new process designs at the early experiment stage, during which prior process knowledge and experimental data are both limited [4]. The lack of training data presents formidable challenges to generalize the analytics to new test data inputs due to overfitting.

In this work, we present a ML-aided process design strategy with application to a novel microwave-assisted ammonia production process [5]. Current commercial ammonia production routes via thermal Haber-Bosch process feature very high energy intensity at high temperature and pressure. Unconventional microwave-assisted catalytic ammonia synthesis offers the unique advantage to achieve high reaction activity under moderate operating conditions via selective heating [6-7]. A set of experiment data is collected totaling 46 data points to comprise the training and test data sets. Four input features are included: pressure, temperature, hydrogen to nitrogen ratio, and feed flow rate. The goal of process design is to maximize the resulting ammonia concentration over a given amount of CsRu/CeO2. We investigate and compare three classes of methodologies: (â…°) Statistical analysis using Response Surface Methodology (RSM), in which a polynomial input-output function is correlated for use in optimization [8], (â…±) Artificial Neural Network (ANN), in which a regression model is developed using the limited training data set [9], (â…²) Deep neural network, in which synthetic minority oversampling technique is further utilized to generate new data samples following the original data distribution to improve prediction accuracy [10]. Optimal process design conditions suggested by the aforementioned computational methods are compared with the designs optimized via experimental analysis. Extensions of the methodologies will also be discussed to unravel the characteristics of microwave-assisted catalytic processes.

References

[1] Cozad, A., Sahinidis, N. V., & Miller, D. C. (2014). Learning surrogate models for simulation‐based optimization. AIChE Journal, 60(6), 2211-2227.

[2] Kim, S. H., & Boukouvala, F. (2020). Machine learning-based surrogate modeling for data-driven optimization: a comparison of subset selection for regression techniques. Optimization Letters, 14(4), 989-1010.

[3] Wilson, Z. T., & Sahinidis, N. V. (2017). The ALAMO approach to machine learning. Computers & Chemical Engineering, 106, 785-795.

[4] Cao, B., Adutwum, L. A., Oliynyk, A. O., Luber, E. J., Olsen, B. C., Mar, A., & Buriak, J. M. (2018). How to optimize materials and devices via design of experiments and machine learning: Demonstration using organic photovoltaics. ACS nano, 12(8), 7434-7444.

[5] Wang, Y., Khan, T. S., Wildfire, C., Shekhawat, D., & Hu, J. (2021). Microwave-enhanced catalytic ammonia synthesis under moderate pressure and temperature. Catalysis Communications, 159, 106344.

[6] Muley, P. D., Wang, Y., Hu, J., & Shekhawat, D. (2021). Microwave-assisted heterogeneous catalysis.

[7] Chen, W., Malhotra, A., Yu, K., Zheng, W., Plaza-Gonzalez, P. J., Catala-Civera, J. M., & Vlachos, D. G. (2021). Intensified microwave-assisted heterogeneous catalytic reactors for sustainable chemical manufacturing. Chemical Engineering Journal, 420, 130476.

[8] Evans, M. D., Ring, J., Schoen, A., Bell, A., Edwards, P., Berthelot, D., & Baldino, C. M. (2003). The accelerated development of an optimized synthesis of 1, 2, 4-oxadiazoles: application of microwave irradiation and statistical design of experiments. Tetrahedron letters, 44(52), 9337-9341.

[9] Golkarnarenji, G., Naebe, M., Badii, K., Milani, A. S., Jazar, R. N., & Khayyam, H. (2019). A machine learning case study with limited data for prediction of carbon fiber mechanical properties. Computers in Industry, 105, 123-132.

[10] Liu, X., Xu, Y., Li, J., Ong, X., Ibrahim, S. A., Buonassisi, T., & Wang, X. (2021). A robust low data solution: dimension prediction of semiconductor nanorods. Computers & Chemical Engineering, 150, 107315.