Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Modelling and Closed-loop Control of Selective Laser Melting
Selective laser melting (SLM) is an additive manufacturing (AM) process in which a layer of fine metal powder is selectively melted and solidified to form a three-dimensional object in a layer-by-layer fashion. Due to its flexibility and capability to produce near-net shape products, SLM has been one of the most used AM processes in the industry. Due to the complex physics involved in the process, there are numerous sources of uncertainty and disturbance in the process. For this reason, understanding the relationship between various physics, process parameters, and material properties is an active research field that is ever-growing. Due to these sources of error, ensuring the mechanical properties of an SLM printed part is often an important challenge that requires numerous empirical studies in practice. One promising way to address these challenges is incorporating closed-loop control by utilizing accurate and efficient models of the thermal dynamics in the process. While high-fidelity numerical models for SLM exist in the literature, such models are often computationally demanding and not suitable for closed-loop control applications.
Keywords: Additive Manufacturing, Repetitive Processes, Thermal Modelling, Reduced Order Modelling, Iterative Learning Control, Model Predictive Control, Learning-based Control
In this project, we are interested in developing control-oriented thermal dynamics models and learning-based controllers for a multi-layer SLM process. More specifically, we will build on our previous research to integrate melt pool dynamics of the process within a pre-existing implementation framework for temperature modeling. In the existing model, we utilize a network of nodes to approximate the thermal distribution in the spatial domain of each layer of the process, given the laser power input, path, and other necessary process properties. However, the existing model does not capture the melting dynamics, which is a highly nonlinear process. The thermal dynamics of the process, and the melting behavior, are inherently PDEs with spatial and temporal components which are computationally prohibitive. Our goal in this work is to develop reduced order, low fidelity approximations of the melt pool dynamics, and integrate them with the existing model to demonstrate closed-loop control applications using the developed model at various fidelity.
To this end, the project is expected to investigate several approaches for the melt pool dynamics approximation models and control approaches.
Possible directions for the project include:
- Investigation of learning-based control methods for layer-to-layer additive manufacturing applications.
- Modelling of the melt pool at low and high fidelity to study the accuracy and computation trade-offs.
- Development of robust and efficient learning-based controllers to work with the computationally intensive ODE/PDE-based high-fidelity models.
- Development of model order reduction techniques to reduce the spatiotemporal process models into more computationally efficient ones.
- Work on the experimental test setup at the Technopark in collaboration with groups from D-MAVT.
In this project, we are interested in developing control-oriented thermal dynamics models and learning-based controllers for a multi-layer SLM process. More specifically, we will build on our previous research to integrate melt pool dynamics of the process within a pre-existing implementation framework for temperature modeling. In the existing model, we utilize a network of nodes to approximate the thermal distribution in the spatial domain of each layer of the process, given the laser power input, path, and other necessary process properties. However, the existing model does not capture the melting dynamics, which is a highly nonlinear process. The thermal dynamics of the process, and the melting behavior, are inherently PDEs with spatial and temporal components which are computationally prohibitive. Our goal in this work is to develop reduced order, low fidelity approximations of the melt pool dynamics, and integrate them with the existing model to demonstrate closed-loop control applications using the developed model at various fidelity.
To this end, the project is expected to investigate several approaches for the melt pool dynamics approximation models and control approaches. Possible directions for the project include: - Investigation of learning-based control methods for layer-to-layer additive manufacturing applications. - Modelling of the melt pool at low and high fidelity to study the accuracy and computation trade-offs. - Development of robust and efficient learning-based controllers to work with the computationally intensive ODE/PDE-based high-fidelity models. - Development of model order reduction techniques to reduce the spatiotemporal process models into more computationally efficient ones. - Work on the experimental test setup at the Technopark in collaboration with groups from D-MAVT.
- Understand the thermal modeling problem and possible control architectures;
- Development and test of the approximate melt pool dynamics models with the existing modeling and simulation framework at varying modeling fidelity;
- Testing closed-loop controllers with the developed models in simulation to demonstrate process improvement;
- Developing an understanding of what closed-loop control approaches and what modeling approximations work best for the given problem
- Understand the thermal modeling problem and possible control architectures; - Development and test of the approximate melt pool dynamics models with the existing modeling and simulation framework at varying modeling fidelity; - Testing closed-loop controllers with the developed models in simulation to demonstrate process improvement; - Developing an understanding of what closed-loop control approaches and what modeling approximations work best for the given problem
Efe Balta (ebalta@ethz.ch)
Alisa Rupenyan (rupenyan@inspire.ethz.ch)
Efe Balta (ebalta@ethz.ch) Alisa Rupenyan (rupenyan@inspire.ethz.ch)