A Data-Driven Framework for Direct Local Tensile Property Prediction of Laser Powder Bed Fusion Parts

Area of Science:

• Materials Science and Engineering
• Additive Manufacturing
• Data Science

Background:

• Traditional qualification of additively manufactured parts relies heavily on statistical process control and feedstock characterization.
• In situ monitoring offers rich data but integrating it for part qualification remains a challenge.
• Existing methods often lack the granularity to predict localized material properties accurately.

Purpose of the Study:

• To develop a generalizable, data-driven framework for qualifying laser powder bed fusion (LPBF) parts using part-specific in situ data.
• To predict localized tensile properties of stainless steel parts by fusing multi-modal sensor data with machine learning.
• To reduce reliance on statistical process control and feedstock characterization for part qualification.

Main Methods:

• A cyberphysical infrastructure was used for robust spatial tracking of 6299 tensile specimens.
• Multi-modal in situ sensor data (powder bed imaging, machine health, laser scan paths) were fused with a priori information.
• A sequence of machine learning models, combining deep learning, machine learning, and feature engineering, was trained to predict tensile properties.
• Computer vision techniques registered ground truth tensile measurements to the co-registered 230 GB in situ dataset.

Main Results:

• The trained machine learning models achieved a 61% error reduction in ultimate tensile strength predictions compared to models without in situ data.
• The framework demonstrated the capability to locally predict tensile properties based on fused multi-modal in situ sensor data.
• A publicly released, co-registered dataset of 230 GB was created to support further research.

Conclusions:

• The proposed data-driven framework enables robust part qualification in LPBF by leveraging in situ monitoring and machine learning.
• Accurate prediction of localized tensile properties is achievable, leading to improved quality assurance in additive manufacturing.
• Further improvements in sensor technology and mechanical testing procedures can further enhance the qualification framework.