A machine learning approach to dynamic simulation of electromagnetic heating
Topic(s) : Manufacturing
Co-authors :
Tim KOENIS (NETHERLANDS), Albert DE WIT (NETHERLANDS), Robert MAAS (NETHERLANDS), Niels VAN HOORN (NETHERLANDS)Abstract :
In the Netherlands R&D mobility program “Thermoplastics for a sustainable aviation” industry is working on developing technology to make lighter, more circular and more cost-efficient transportation, e.g. aircraft. Thermoplastic composite is a material that can be important in helping to achieve this and offers the versatility to make structures that are lightweight, more recyclable and easier to maintain than was previously possible. For such structures, development of advanced joining technologies and more effective use of materials are both necessary to enable economically and environmentally competitive component assemblies.
One example of such advanced joining techniques is induction welding (Ahmed, et al. 2006). Thermoplastic CFRP can be re-melted allowing them to be joined via welding. For a successful and reproduceable weld that is acceptable for certification authorities extensive knowledge and control over the welding process is necessary. At present, the inductive heating process requires substantial data logging, control of welding parameters, design of the weld, and process experience. This is to create a successful and repeatable process with similar mechanical quality. To obtain control over the inductive welding process a Digital Twin (Jones, et al. 2020) of the welding process could assist to predict and adjust the necessary welding parameters. As fluctuations occur during welding between the measured weld parameters and those computed by the digital twin, adjustments could be suggested to the equipment operator.
For the digital twin, a prediction model that can run in real time during welding is necessary. This model should predict the outcome of a welding process based on input parameters given to the model. In the present work the authors focus on developing a combined physics-based and machine learning data-based modelling procedure for the prediction model. In previous work (de Wit, et al. 2023) an Abaqus (Simulia 2023) electromagnetic-thermal simulation model was developed.
The temperature distribution across a weld is generally considered to be a good indicator for weld line success. For static inductive heating the Abaqus model predicts the temperature distribution of a Uni-Directional (UD) CFRP plate sufficiently. Unfortunately, this Abaqus implementation is unable to model coil movement. Therefore, in this work the authors share an implementation that uses Abaqus static electromagnetic simulation for different coil positions (Figure 1) to arrive at a moving coil simulation. Since physics-based simulations require a significant amount of time to run (on the order of hours), a neural network is trained. Using simulation results the temperature distribution in the UD CFRP is predicted while the coil is moving with respect to the workpiece (Figure 2). The results of this work can be used to predict the success of heating a CFRP UD laminate to melt temperature while moving the coil and predict the necessary weld parameters.
One example of such advanced joining techniques is induction welding (Ahmed, et al. 2006). Thermoplastic CFRP can be re-melted allowing them to be joined via welding. For a successful and reproduceable weld that is acceptable for certification authorities extensive knowledge and control over the welding process is necessary. At present, the inductive heating process requires substantial data logging, control of welding parameters, design of the weld, and process experience. This is to create a successful and repeatable process with similar mechanical quality. To obtain control over the inductive welding process a Digital Twin (Jones, et al. 2020) of the welding process could assist to predict and adjust the necessary welding parameters. As fluctuations occur during welding between the measured weld parameters and those computed by the digital twin, adjustments could be suggested to the equipment operator.
For the digital twin, a prediction model that can run in real time during welding is necessary. This model should predict the outcome of a welding process based on input parameters given to the model. In the present work the authors focus on developing a combined physics-based and machine learning data-based modelling procedure for the prediction model. In previous work (de Wit, et al. 2023) an Abaqus (Simulia 2023) electromagnetic-thermal simulation model was developed.
The temperature distribution across a weld is generally considered to be a good indicator for weld line success. For static inductive heating the Abaqus model predicts the temperature distribution of a Uni-Directional (UD) CFRP plate sufficiently. Unfortunately, this Abaqus implementation is unable to model coil movement. Therefore, in this work the authors share an implementation that uses Abaqus static electromagnetic simulation for different coil positions (Figure 1) to arrive at a moving coil simulation. Since physics-based simulations require a significant amount of time to run (on the order of hours), a neural network is trained. Using simulation results the temperature distribution in the UD CFRP is predicted while the coil is moving with respect to the workpiece (Figure 2). The results of this work can be used to predict the success of heating a CFRP UD laminate to melt temperature while moving the coil and predict the necessary weld parameters.