Digital Twin-based Machine Control for Laser Powder Bed Fusion of Metallic Aerospace Components

The main objective of this Phase II project is to develop a Statistical Parametrized Physics-Based Machine Learning Digital Twin (S-PPB-ML-DT) monitoring with a feedback control approach to autonomously tailor local microstructure, heal defects, and minimize residual stresses and surface roughness in near real-time for metallic aerospace components manufactured using Laser Powder Bed Fusion (LPBF). The major deliverable will be an adaptive S-PPB-ML-DT  LPBF process that assures repeatable, reliable, and optimal mechanical performance. The project goal is accomplished by constructing a mechanistic data-driven machine learning digital twin (ML-DT) framework that integrates real-time experimental data collection and process control with machine learning-based process modeling and optimization.  The framework starts with training the adaptive ML-DT based on the statistical physics-based models and experimentally collected data. Within this digital twin, both forward and inverse models are encapsulated. The forward model serves as a diagnostic tool for in-situ monitoring of defects and anomalies. When combined with the reference processing conditions, the diagnostic tool is employed for forward process control. The inverse model, on the other hand, considers the real-time sensed information and predicts both the formation of defect/anomaly and corrective strategy. Both models feed information to the process controller and implement different strategies based on the nature of the defects/anomalies.Three major strategies will be developed for defect control:1)    The first strategy is to set process parameters based on ML-DT predictions to prevent defects proactively. This approach relies on swift predictions generated by ML-DT and is purely prognostic or feedforward.2)    In the second and third strategies, predictions of ML-DT are combined with real-time diagnostic/sensing information. ML-DT provides predictions (prognostic data) and devises specific process steps to either mitigate or repair defects. These steps are feedback control and often include rescanning or remelting of specific process zones.The project objectives will be achieved with a team consisting of HIDENN-AI, LLC, Northwestern University (NU), DMG-MORI and Product Innovation and Engineering LLC (PINE).   Five major tasks are proposed:Task 1: Statistical parameterized Physics-based model development (Phase II base and option for HIDENN-AI, NU, PINE)Task 2: Data Fusion and Analytics to Build Machine Learning the Forward and Inverse Digital Twin Models (Phase II base and option for HIDENN-AI)Task 3: Sensor technology and machine integration (Phase II base and option for NU and DMG-MORI)Task 4: Demonstration of machine control (Phase II base and option for NU and DMG-MORI)Task 5: Data Management Plan (Phase II base and option for HIDENN-AI and NU)