(363c) Process Operability Mapping Using Neural Networks

Process operability concepts have emerged in the past two decades as an efficient tool to take into account control and intensification targets of a production system [1]. Currently, the process operability mapping concepts rely on Non-Linear Programming (NLP)-based approaches to compute feasible operating regions for process intensification employing first-principles models of the process in focus [2]. However, such NLP-based approaches are not tractable for a high-dimensional, non-linear process system. The computational cost associated with solving the optimization problem may be prohibitive and thus solving this problem in a tractable manner would require parallel computing systems [3], which may not be easily available in user programming platforms. To address the aforementioned challenge, this work proposes the use of Machine Learning (ML) algorithms employing Neural Networks (NN) as surrogate models to represent the high dimensional non-linear systems. Such surrogate models can help in reducing the computational cost associated with the inverse mapping to compute the desired input set (DIS) from the desired output set (DOS).

In particular, the NN model is trained and tested against the data generated from the first-principles model. The produced surrogate models can be used to predict the outputs of the system under consideration. These outputs are then used in the NLP-based approach to compute the feasible DIS (DIS*) from the DOS. The proposed approach can replace the current existing operability framework by employing ML-based surrogates in the current tools.

To give an illustration of the proposed framework, the ML NN-based operability method is applied to process systems of non-linear nature, such as a membrane reactor for the Direct Methane Aromatization system. Results of the proposed approach will be discussed regarding the obtained error margins and the reduction of computational time that could be achieved with the proposed method when compared to the traditional approaches. This work thus has the potential to significantly change the way process intensification and design problems are addressed with the help of the Machine Learning-based process operability framework.

References

[1] Vitor Gazzaneo and Fernando V. Lima. Multilayer operability framework for process design, intensification, and modularization of nonlinear energy systems. Industrial & Engineering Chemistry Research, 58(15):6069–6079, 2019.

[2] Vitor Gazzaneo, Juan C. Carrasco, David R. Vinson, and Fernando V. Lima. Process operability algorithms: Past, present, and future developments. Industrial & Engineering Chemistry Research, 59(6):2457–2470, 2020.

[3] Juan C. Carrasco and Fernando V. Lima. Bilevel and parallel programing-based operability approaches for process intensification and modularity. AIChE Journal, 64(8):3042–3054.