Physics-Informed Data-Driven Prediction and Design in Advanced Manufacturing Processes

Mozaffar, Mojtaba

doi:https://doi.org/10.21985/n2-nzjh-5w36

Work

Physics-Informed Data-Driven Prediction and Design in Advanced Manufacturing Processes

Public

Download PDF

Download All Files (.zip)

Manufacturing processes are known for their intricacies in changing material shapes and properties. New generations of manufacturing technologies, known as flexible manufacturing, are moving toward design freedom, which allows producing parts with optimized geometries and high customizations at an affordable cost even for low-volume productions. Two prominent flexible manufacturing processes that are of interest in this dissertation are additive manufacturing and incremental sheet forming. An important limiting factor in advancing current capabilities in such processes is the difficulty to reliably understand and control them due to the complex multi-physics and multi-scale nature of the processes. As the result, current practices are overly conservative, significantly limiting the vast potential of producing parts with customized material properties.At the same time, we observe a surge in the digitalization of manufacturing processes. Today, manufacturing facilities are more connected to data centers than ever, and various measurement methods are becoming standard components of modern manufacturing pipelines from controlling and monitoring the progress while manufacturing to testing and analysis after the part is built. Therefore, this dissertation is dedicated to developing computational methods to advance modeling and design capabilities with a focus on approaches to optimally use manufacturing data—an underutilized asset of manufacturing systems. My contributions in process characterization and design are organized into five research tasks and briefly discussed below. Predicting the spatiotemporal behavior of manufacturing processes is challenging due to the long history-dependent correlations and complex unstructured geometric features common in manufacturing. Motivated by this challenge, my first research contribution introduces a data-driven methodology to learn material behaviors on unseen geometries over long simulation periods. My method efficiently combines a recurrent neural network to capture material evolution over time and a graph representation to flexibly extract geometric features. This methodology is demonstrated on thermal prediction of additive manufacturing processes and shows great generalizability across industrial-grade parts. Plasticity is one of the important pillars of computational mechanics. Conventional plasticity methods heavily rely on restrictive assumptions to reduce the dimensionality of the problem into so-called “effective” parameters. In a first-of-a-kind research, my second contribution proposed a data-driven approach to material constitutive modeling, where the material behavior under complex elasto-plastic loading conditions can be learned from data. My work not only shows that data-driven constitutive modeling is accurate, but also it is computationally efficient and performs well across multiple material systems including composites and metal alloys. The large design spaces of flexible manufacturing such as the additive manufacturing process present a daunting optimization task, which limits the capabilities of producing highly customized parts. In the third contribution of my dissertation, I proposed a reward-driven solution to the toolpath design problem, where an agent is trained to explore the environment and develop strategies to collect maximum rewards. Four methods (three model-free and one model-based) varying in their exploration and decision-making formulations are developed and tested to design toolpaths for over 400 sections and the results show the effectiveness of this methodology especially in the presence of a dense reward structure. In my fourth research contribution, I developed a differentiable manufacturing simulator that enables a seamless integration between physics-based and data-driven methods. I demonstrate that the gradients of a physics-based thermal simulation of the additive manufacturing process can be computed using automatic differentiation. Furthermore, this differentiable simulation is combined with neural networks to effectively optimize time-series process parameters and reach ideal thermal responses or melt pool behavior over hundreds of simulation time steps. The computational expense of physics-based manufacturing models is a limiting factor in the size of the problem that can be reasonably solved especially applications such as iterative design, model predictive control, and uncertainty quantification. In my fifth contribution, I investigated modern heterogeneous computational hardware to accelerate the simulation of additive manufacturing processes. Using the proposed matrix assembly and flux calculation strategies on graphical processing units, a speedup of 100-150X is achieved compared to an optimized CPU implementation.

Creator