Predictive Control Based on Reduced Order Model for temperature homogeneity in a resin transfer molding tool for thermoset materials

TL;DR

This paper presents a novel predictive control approach using a reduced order model for achieving temperature homogeneity in RTM molds, addressing control complexity and variability challenges.

Contribution

It introduces an optimized MPC architecture with a simplified ROM derived from FEM data, incorporating adaptive parameter estimation and enhanced observers for improved temperature control.

Findings

Successfully implemented in an RTM tool

Achieved tight temperature control in the mold

Handled large actuator and sensor coupling

Abstract

Resin Transfer Molding (RTM), which has attracted much attention in the last years for lightweight manufacturing, represents an important challenge in terms of control technology. During the process, a resin fills the cavity where a reinforcement fabric has previously been layered. This resin undergoes a chemical reaction which is thermically activated. Therefore, assuring a proper reaction requires a precise control of temperature in the entire mold cavity. Three factors make this control problem especially hard: the coupling among the large number of actuators and sensors, the variability of the test conditions and the power limitations of the electric actuators which do not offer cooling capability. The present work describes an optimized Model Predictive Control (MPC) architecture capable of handling these difficulties and also achieving the tight control requirements needed in the application. The thermal distribution inside the mold cavity is included into the controller by a simplified Reduced Order Model (ROM). This representation is obtained by data from an experimentally validated Finite Element Model (FEM), using AutoRegressive model with eXogenous terms (ARX) identification. In order to maintain the simplicity of this linear representation, the time-varying model parameters are estimated by using a perturbation observer. Additionally, the performance of the basic algorithm is improved: firstly, an augmented observer to estimate the temperature distribution of an extended spatial resolution; and secondly, a symmetry condition in the calculation of the control commands. The developed architectures have successfully been implemented in a RTM tool with the fulfillment of the control requirements.