SBIR-STTR Award

Global Engineering and Materials, Inc. (GEM) and its team members, University of Illinois Urbana-Champaign (UIUC) and Professor Richard Sisson from the Center for Heat Treating Excellence at Worcester Polytechnic Institute (WPI), propose developing a coupled tool based on integrated computational materials engineering (ICME) and a machine learning (ML) approach to optimize quenching processes with coupled thermal, metallurgical, and mechanical approaches. The ML-driven ICME toolkit will utilize physics-informed neural networks (PINNs) that integrate fundamental physical principles and data from high-fidelity simulations and experiments, to generate the inter-relationship between residual stress/distortion and quenching process parameters. A high-fidelity thermal multi-phase fluid-structure interaction (FSI) model will be enhanced to simulate fluid dynamics with phase transitions and temperature fields in the quenchant tank. An immersogeometric approach will be used for efficient FSI model generation directly based on 3D workpiece CAD models with various dipping orientations. Given the temperature and phase profiles predicted from the thermal multi-phase FSI module, a residual stress and distortion prediction module will be developed by including fields mapping, temperature and phase dependent property evolution, and a user-defined material model for Abaqus. The high-fidelity IMCE and its ML-based tools will be validated using water quenching data of aluminum coupons collected from WPI.

Benefit:
The research will help us greatly to improve the understanding of heat transfer between the immersed metallic workpiece and surrounding quenchant in the presence of film boiling, nucleate boiling, convection, fluid circulation, and phase transformation in the workpiece. The research will also result in a high-fidelity and computationally efficient toolkit for prediction of residual stress and distortion. Using physics-informed neural networks (PINNs), an inter-relationship among the fluid, thermal, phase transformation, and the stress/strain can be established to control residual stress and reduced distortion in large aerospace aluminum forging parts. At present, the lack of physical understanding of the quenching process and the use of a trial and error approach have resulted in a high scrap rate of large airframe aluminum forging parts. Development of quicker and reliable qualification and certification procedure is commercially viable given stringent constraints on cost and schedule. Methods and techniques developed can be included for broad use in the aerospace industry, auto industry, and heavy-equipment industry for reduction of scrap rate of heat treated metallic parts with a controllable level of residual stress and distortion.

Keywords:
Distortion, Distortion, Residual Stress, Phase transformation, quench, Immersogeometric Analysis, machine learning., fluid-structure interaction, physics-informed neural network

Global Engineering and Materials, Inc. (GEM) along with its team members, University of Illinois Urbana-Champaign (UIUC), Weber Metals, and Spirit AeroSystems, proposes to develop and demonstrate a coupled framework based on the integrated computational materials engineering (ICME) and machine learning (ML) approach to optimize the quenching process with coupled thermal, metallurgical, and mechanical analysis. The key features of our novel approach include: 1) a coupled thermodynamics and turbulent multiphase flow based FSI for quenching process simulation; 2) an immersometric FSI with adaptive mesh/quadrature refinement/coarsening for multi-part dipping of large scale structures; 3) a domain decomposition simulation module to capture the system dependent agitation mechanisms; 4) a mechanistic constitutive modeling approach with coupling of microstructure and macro-level constitutive behavior; 5) a digital property evolution module and its coupling with a zoning approach; 6) a multi-level physics-informed ML framework for quenching process modeling and uncertainty quantification; and 7) a performance informed process optimization via a surrogate modeling approach. The resulting QLAB toolkit will be firstly validated using data collected from Web Metals at coupon and component levels. The high-fidelity toolkit along with its multi-level PINNs will be demonstrated using the data collected from the V-22 and CH-53K forging parts doing and after the quenching. To further enhance the feasibility of the technology transition, we will explore the process tailoring to reduce the distortion of large scale structures.

Benefit:
The research will help us greatly to improve the understanding of heat transfer between the immersed metallic workpiece and surrounding quenchant in the presence of film boiling, nucleate boiling, convection, fluid circulation, and phase transformation in the workpiece. The research will also result in a high-fidelity and computationally efficient toolkit for prediction of residual stress and distortion. Using physics-informed neural networks (PINNs), an inter-relationship among the fluid, thermal, phase transformation, and the stress/strain can be established to control residual stress and reduced distortion in large aerospace aluminum forging parts. At present, the lack of physical understanding of the quenching process and the use of a trial and error approach have resulted in a high scrap rate of large airframe aluminum forging parts. Development of quicker and reliable qualification and certification procedure is commercially viable given stringent constraints on cost and schedule. Methods and techniques developed can be included for broad use in the aerospace industry, auto industry, and heavy-equipment industry for reduction of scrap rate of heat treated metallic parts with a controllable level of residual stress and distortion.

Keywords:
fluid-structure interaction, Residual Stress, Immersogeometric Analysis, physics-informed neural network, Distortion, quench, thermodynamics, Phase transformation