(362r) Modeling Dynamics of Material Flows in Coupled Industrial Processes Using Data Driven System Identification

Material flows within an industrial network are typically coupled through interdependence of the industries providing and consuming resources or products. Dynamical analysis of this coupled industrial networks can provide critical information about production, emissions, and energy use of the overall macroscopic system along with the key mechanisms that drive these changes. Currently dynamic models for process plants exist as highly accurate first-principle relationships. However, their integration is computationally intensive, they provide no simplified understanding of underlying driving mechanisms, and their complexity and size makes them difficult to couple with dynamic models of other interconnected systems for wider network analysis. The Sparse Identification of Nonlinear Dynamics (SINDy) algorithm, introduced by Brunton et al. (2016) and proposed by Farlessyost & Singh (2021) to be an efficient method for recovery reduced order dynamical models for complex dynamics in industrial systems and natural systems, is further applied to several industrial manufacturing processes to establish a set of low-order governing equations that model the coupled dynamics of the industrial network. We demonstrate the approach on a small industrial network with three nodes that consist of soybean growth (natural system), soybean to soy-oil and soy-oil to biodiesel (both process systems). We leverage a pre-existing dynamical model of soybean growth in ordinary differential equation (ODE) form and couple this with ODEs derived from our proposed hybrid system identification method for soybean to soybean-oil and soybean-oil to soybean-diesel. This work advances modelling for soybeans to soybean-oil by training SINDy on synthetic data from a Dyssol simulation and a soybean-oil to soybean-diesel model, for which a model was previously recovered and made available in Farlessyost & Singh (2021). In this previous work, authors tested the soybean-diesel dynamic model against two-hundred simulated hours of plant time series data and found that the recovered equations can predict internal molar flowrates and the output rate of soybean diesel with accuracy. Finally, we demonstrate the strengths and limitations of this approach for model dynamics of material flows, resource consumption and emissions for the overall network to provide insights relevant for sustainability assessment of the production network. Our approach makes use of several previous techniques implemented in Farlessyost & Singh (2021), including further optimization of the SINDy model coefficients.

References:

Brunton, S. L., Proctor, J. L., & Kutz, J. N. (2016). Discovering governing equations from data by sparse identification of nonlinear dynamical systems. Proceedings of the national academy of sciences, 113(15), 3932-3937.

Farlessyost, W., & Singh, S. (2021). Reduced Order Dynamical Models for Complex Dynamics in Manufacturing and Natural Systems Using Machine Learning. arXiv preprint arXiv:2110.08313.

Taherzadeh, O., & Caro, D. (2019). Drivers of water and land use embodied in international soybean trade. Journal of Cleaner Production, 223, 83-93.