Retrofitting of a Cyber-Physical System to a reactive transfer molding machine for thermoset resins

This thesis presents the retrofit of an Cyber Physical System autonomously control the resin transfer molding (RTM) machine for the encapsulation of Printed Circuit Board Assemblies (PCB-As). In order to produce high-quality parts with an RTM process, typically substantial domain expertise and continuous production improvement is crucial. As both these factors are cost and time intensive, this master thesis tackles these challenges by implementing a CPS to an existing RTM machine of the industrial partner Ottronic E-Systems GmbH. The CPS is designed to optimize the machine's production parameters without expert intervention, and ultimately reduce operational costs while retaining highest quality. The proposed methodologies is: First, analyzing historical RTM data to identify correlations between machine parameters, control (set) parameters, and part quality evaluations. Based on the data audit, the correlation of specific control parameters was investigated and interlinked with a physical model, concerning the RTM-specific viscosity. Second, based on the gained correlation parameters and the physical model which serves as a basis for the CPS, an optimization algorithm for this process was developed. Thereby, the CPS was developed, based on two different optimization approaches in order to evaluate the efficiency and precision of both methods.: In detail, a model based on the response surface methodology (RSM), a gradient-based technique, and the other utilizing Bayesian optimization, which employs a probabilistic surrogate model, were implemented and validated. The results yielded that both CPS approaches successfully identified optimal set-parameters that produced perfect processed parts. However, the RSM-based CPS demonstrated a continuous need for manual adjustments of control variables during operation, which made it unsuitable for fully automated optimization. In contrast, the Bayesian optimization CPS operated autonomously without such adjustments. Additionally, the Bayesian optimization CPS explored a significantly larger parameter space, reducing the likelihood of missing global maxima. This broader exploration also generated a more comprehensive dataset, forming an excellent foundation for the future development of a digital twin. The digital twin would serve as a basis for further enhancing the optimization process. Consequently, the Bayesian optimization CPS was selected as the superior approach. The developed CPS successfully achieves the objective of creating a self-optimizing machine parameter controller for the RTM process. This in turn enables the replacement of expert supervision with operator-based control, thereby reducing development costs. Moreover, the Bayesian Optimization-based CPS demonstrated a user-friendly operation at high performance, requiring minimal training and thus offering an improved time-to production at lower costs. Finally, by the implementation of a novel user interface minor process adjustments during production are significantly simplified as historical process data linked with quality results are constantly presented to the user. Thus, trend in quality degradation can be easily identified.