Machine learning is currently a popular concept in the technology world, and when combined with simulation, it is a valuable tool for product development. SIMULIA Roles Portfolio Engineering Specialist Yangzhan Yang will be discussing the combination of machine learning and physics-based simulation for product development at the NAFEMS online seminar on April 28^th. The theme of this year’s seminar is, “AI, Data Driven Models and Machine Learning: How Will Advanced Technologies Shape Future Simulation Processes?”

As Yang explains in her presentation, physics-based simulation and machine learning are both paradigms that build predictive models for engineering problems. Machine learning is a data-driven approach that discovers underlying patterns in data, while Finite Element Analysis, as a physics-based approach, describes a physical system with differential equations and solves them numerically.

Data-driven models use large amounts of data but do not discover new laws of physics, while physics-based models look at understanding and explaining processes through laws of physics without utilizing big data. The combination of the two has numerous new applications in a variety of industries including transportation, aerospace, energy, high tech, life sciences, and, as Yang will be discussing, product development.

Yang’s presentation will include several detailed use case examples, including a Composites Airplane Wing Design Evaluation using machine learning. A composite airplane wing model consists of a quarter million design variations, an overwhelming number that can be addressed by using machine learning. A small subset of variations can be evaluated by engineers, and machine learning can then quickly evaluate the rest of the design space.

Yang developed a program with a user interface using the Process Studio app on the 3DEXPERIENCE platform. With a small amount of user input, the program can generate a given number of design variations. The simulation output of stresses can be collected to train a machine learning model that can evaluate new design variations with 91% accuracy for stress.

Another example Yang includes in her presentation involves the use of machine learning to predict melt pool physics for metal additive manufacturing. The DS team used the technology to predict the shape of the melt pool and the temperature near the laser center during selective laser melting (SLM) of a block of titanium alloy.

Yang will also be discussing Living Human Heart Modeling with machine learning. Over the course of several years, SIMULIA has built a multi-physics model of a healthy adult male human heart that has been useful in predicting the mechanical and electrical response during heartbeat cycles. Representing a disease state, however, is challenging due to the complexity and cost of the living heart model. The DS team hopes to use machine learning to address these challenges, gain deeper insights of the heart parameters, and simulate different population groups.

In addition, Yang talks about accelerating FEA crash simulations with machine learning. During the vehicle design process, a vehicle’s chassis must be evaluated for multi-disciplinary aspects such as crash, stiffness, NVH, ride and handling performance. Finite element simulations are used to minimize time and cost for multi-disciplinary evaluation and design improvement, but the DS team aims to accelerate concept crash simulations by correcting mesh discretization error.

The final use case Yang will present involves Constitutive Materials Modeling using machine learning. The DS team proposes a novel artificial neural networks-based approach to address the problems of stability and extrapolation in hyperelastic and plastic material modeling.

Machine learning and simulation are two technologies that can enhance one another in various fields including and beyond those Yang discusses: product development, composites design, additive manufacturing, life sciences, crash simulation and material modeling. To see the entire presentation, you can register for the NAFEMS event here.