Online learning‐based predictive control of crystallization processes under batch‐to‐batch parametric drift

Abstract

This work considers a seeded fesoterodine fumarate (FF) cooling crystallization and presents the methodology and implementation of a real‐time machine learning modeling‐based predictive controller to handle batch‐to‐batch (B2B) parametric drift. Specifically, an autoencoder recurrent neural network‐based model predictive controller (AERNN‐MPC) is developed to optimize product yield, crystal size, and energy consumption while accounting for the physical constraints on cooling jacket temperature. Deviations in the kinetic parameters are considered in the closed‐loop simulations to account for the B2B parametric drift, and two error‐triggered online update mechanisms are proposed to address issues pertaining to the availability of real‐time crystal property measurements and are incorporated into the AERNN‐MPC to improve the model prediction accuracy. Closed‐loop simulation results demonstrate that the proposed AERNN‐MPC with online update, irrespective of the accessibility to real‐time crystal property data, achieves a desired closed‐loop performance in terms of maximizing product yield and minimizing energy consumption.

Implementable Stability Guaranteed Lyapunov-Based Data-Driven Model Predictive Control with Evolving Gaussian Process

“…To build the initial Gaussian process model, we use the plant (36) to generate N td 0 = 10 data as t 0 by Latin hypercube sampling (LHS) from a local subset around the steady-state x s = [0, 0] T . Next, we can get the initial predictive model of single state according to (15) and then the dynamic expressions m td 0 (•) and v td 0 (•) of the whole system according to (16).…”

“…The MPC applies the first element of this optimal control sequence to the plant. Many MPC algorithms have been designed for different systems. − Recently, to design MPC involving machine learning-based or data-driven models has been a topic of interest. ,− Considering that there exists an unknown mismatch between the estimated states and the real states when updating the model online, how to keep the MPC feasible and the closed-loop system stable with enhanced performance is one of the problems in machine learning-based MPC design.…”

“…Reference uses an online updating RNN to achieve a more accurate prediction in MPC for nonlinear dynamic characteristics of processes in the presence of uncertainty. Then, ref gives an LMPC of crystal process based on Autoencoder RNN which is updated according to an event-trigger mechanism.…”

Physics-Informed Online Machine Learning and Predictive Control of Nonlinear Processes with Parameter Uncertainty

“…Previous works have utilized purely data-driven online model update strategies for RNNbased MPCs subject to time-varying disturbances, where realtime sensor data are collected and used to update the RNN model to improve its prediction accuracy, yielding improved control performance. 21 In our previous work, a PIRNN model has been developed for the nominal system without model uncertainties; 22 however, at this stage, PI-informed ML modeling of nonlinear systems with model uncertainties has not been studied. Therefore, in this work, we consider nonlinear systems subject to parameter uncertainty and further develop an inverse PIRNN modeling method to estimate unknown process parameters in order to improve the prediction accuracy of PIRNN models.…”

Data-Driven Distributed Model Predictive Control of Continuous Nonlinear Systems with Gaussian Process

“…al. designed the Lyapunov-based MPC (LMPC) and adaptive LMPC based on recurrent neural network (RNN) for nonlinear processes, ,, the MPC for systems with scheduled mode transitions, and the MPC using statistical machine learning model; Gaussian process (GP)-based MPC is proposed in refs , , ; and a neural network to approximate the MPC controller by reinforcement learning method, etc. is used in refs and .…”