Research on the digital twin for robotic carbon fiber reinforced plastics machining processes

Abstract

Carbon fiber-reinforced plastics (CFRP) find many applications given their superior properties. These materials are usually formed using a near-net-shape method that requires secondary machining, such as drilling and trimming, after molding. Industrial robots are becoming increasingly popular machining tools in industries exhibiting high demand for CFRP. However, it remains challenging to achieve high dimensional accuracy when using such robots and dynamic performance is poor. Serial robots exhibit nonlinear changes in dynamic properties depending on their posture. Therefore, research is necessary to predict dynamic properties based on posture during robotic CFRP machining processes and to optimize the robot's posture to improve the machining performance. Additionally, even with optimized processes, CFRP materials, being composed of fiber and resin, show anisotropic characteristics, necessitating a monitoring system to detect defects occurring abnormally during the machining process. Therefore, research is needed on a digital platform that can integrate and manage these necessary elements for robotic CFRP machining processes. The aim of this study is to develop a digital twin platform designed to improve the machining performance in robotic CFRP machining processes. AI technology is an efficient approach to solving these nonlinear problems, hence a system based on AI models has been developed. A model was developed to predict dynamic properties according to the posture using a twin model synchronized with the actual robot before the process. Based on these results, the robot's posture was optimized to improve the machining performance. Additionally, a model for predicting CFRP defects using the multi-sensor data generated during the machining process was developed, proposing a monitoring system for the robotic machining system. Ultimately, the developed predictive models were integrated into a single platform, leading to the development of a digital twin platform for the robotic CFRP machining system. First, the dynamic properties of the robot, which vary nonlinearly based on its posture, were analyzed, and their correlation with defects during CFRP drilling and trimming processes was investigated. Based on these results, a model was developed to predict dynamic properties according to the robot's posture for improving dynamic performance. Multi-Layer Perceptron (MLP) models were developed to accurately predict the dominant natural frequency and dynamic stiffness of each tool tip axis according to the robot's posture. Based on the developed models, posture optimization was performed, and an improvement in machinability during the robotic machining was confirmed. Second, existing research on CFRP defect monitoring has primarily focused on machine tools, with a lack of monitoring systems tailored to robot-based machining system. Investigations into parameters capable of monitoring defects during the robotic CFRP machining process led to the development of a monitoring system based on these findings. A model was developed to predict defects using multi-sensor data, based on a multimodal 1D CNN model, with each sensor data forming its own modality. The model's precision was enhanced by integrating the varying machining characteristics captured from each sensor. Additionally, a model was developed for visualizing the machined surface in the monitoring system, using only force sensor data from the multimodal 1D CNN. Finally, a digital twin platform for robotic CFRP machining processes was developed, integrating and managing all the preceding research results. This platform, synchronized with the actual robot, uses a twin model to optimize the robot's posture based on predicted dynamic properties before the process, and operates the process based on this information. Furthermore, a system was developed to monitor CFRP defects based on the multi-sensor data generated during the machining process, with the monitored data being stored in a database for future use as historical data. In conclusion, the developed models and platform have been proven to improve the machinability and performance in robotic CFRP machining processes.