(372ac) Using Artificial Neural Networks to Predict Chemical System Stability for Design

Though artificial intelligence is now new to chemical engineering, there has been a recent surge in exploring artificial intelligence applications in chemical engineering. This includes using artificial neural networks (ANNs) to provide estimates for values which may be difficult to calculate directly. The presented work uses feedforward ANNs to predict gain margins and sensitivity numbers for a Haber-Bosch reactor in an ammonia plant which runs entirely on renewable energy. Stability parameters are calculated in Python 3.9 using Python’s control library.

An ANN is subsequently treated as a surrogate model for an optimization problem to identify the maximum stability of the system. The OMLT Python package is used to take an ANN which predicts a single output value to determine optimal input parameters by modelling the ANN in PYOMO (Python’s optimization suite). The results show that ANNs are capable of predicting gain margin and sensitivity number for a chemical system with root mean squared error less than 5% of the minimum value used in the training set. It is further demonstrated that using Python’s OMLT package, the ANN as a surrogate system model optimization problem can identify optimal design and operating variables to maximize system stability. This will be expanded into optimizing for other parameters of interest, with the goal of minimizing total annualized cost or maximizing energy efficiency. This will also be expanded into considering multi-objective optimization using the OMLT package to generate pareto curves from the ANN-surrogate optimization problem.