Researchers at the University of Toronto have unveiled a machine learning-based framework designed to accelerate and refine the optimization of laser-based metal additive manufacturing. Named AIDED (Accurate Inverse process optimization framework in laser Directed Energy Deposition), the system allows users to predict and inversely determine optimal printing parameters for Laser Directed Energy Deposition (L-DED) processes, significantly reducing the time and experimentation required for high-quality results.
The study, published in Additive Manufacturing, demonstrates how AIDED uses two trained machine learning models and a genetic algorithm to recommend the best process parameters based on user-defined performance goals such as high print speed and precise melt pool geometry. The framework achieved an R² score of 0.995 for predicting single-track melt pool areas and 0.969 for the tilt angle of multi-track melt pools, indicating extremely high accuracy in replicating experimental results.
AM process outcomes
By reconstructing multi-layer prints from ML predictions, AIDED enables users to explore the effects of laser power, scan speed, powder feed rate, and hatch spacing with high accuracy and without extensive physical testing. When printing cubic test parts, the system achieved dimensional accuracy within 1.75% for width and 12.04% for height. The larger deviation in height is partly due to the complexity of predicting vertical build dynamics, such as thermal accumulation across layers. Despite this, all parts maintained densities above 99.9%, indicating minimal porosity and strong structural integrity, which are essential benchmarks for functional metal components.
The inverse optimization capability allows manufacturers to input design goals and receive an optimized parameter set within 1-3 hours. This is a major improvement over traditional trial-and-error approaches, which are often time-consuming and costly.
Transfer learning across material systems
A notable feature of AIDED is its ability to transfer models across different materials with minimal data. The team successfully applied a model trained on 316L stainless steel to predict melt pool geometries for pure nickel using just 56 data points, a 7.4-fold reduction from the original training set size. This efficiency suggests AIDED could be widely adopted in industrial settings where material variety and production agility are key.
Data-driven optimization in additive manufacturing
The use of machine learning in metal additive manufacturing has gained traction as researchers seek ways to streamline production while maintaining quality. Previous studies, such as those by Akbari et al. and Feenstra et al., have used ML to estimate melt pool geometries, but often focus on single-track prints and lack inverse optimization or real-world validation.
AIDED addresses these limitations by integrating direct prediction, full geometry reconstruction, inverse design, and experimental validation in one end-to-end workflow. The framework is currently available as an open-source project on GitHub, inviting further experimentation and industrial adoption.
While the framework has proven effective in controlled lab settings, the authors note that more work is needed to generalize AIDED to complex geometries and additional materials such as titanium and aluminum alloys. They also point to potential improvements via convolutional neural networks and physics-informed simulations for higher accuracy and broader applicability.
As additive manufacturing continues to scale in aerospace, biomedical, and energy sectors, tools like AIDED offer a pathway to more automated, precise, and cost-effective production workflows.
AI-Driven enhancements in additive manufacturing
The integration of artificial intelligence (AI) and machine learning (ML) into additive manufacturing workflows is steadily transforming how 3D printed parts are designed, qualified, and produced. Across the industry, companies are developing AI-driven tools to enhance process control, reduce iteration cycles, and streamline material optimization. One example is 1000 Kelvin’s AMAIZE 2.0, a fully AI-automated build preparation workflow for metal 3D printing. Introduced at Formnext 2024, this system features a Printability Checker and Exposure Strategy Optimization, reportedly reducing redesign cycles by 40%, improving quotation accuracy by 30%, and cutting material costs by up to 20%.
Back in 2022, Authentise advanced its data-driven capabilities by acquiring software developer Elements, expanding its Manufacturing Execution System (AMES) to support real-time analytics and flexible workflows in new industrial verticals. These capabilities enable manufacturers to better analyze and respond to production data as it’s generated, improving decision-making on the shop floor.
In the materials qualification space, Senvol has demonstrated a machine learning method for predicting material allowables, key values required for certifying parts in regulated industries. This innovation holds potential for significantly reducing the cost and time associated with traditional qualification testing. Likewise, research into AI-powered material discovery is helping identify optimal combinations of print parameters and feedstocks, accelerating the development of novel 3D printing materials.
Collectively, these developments point to a growing movement in additive manufacturing toward intelligent, adaptive systems, echoing the goals of the AIDED framework in enabling data-driven optimization and predictive control over complex metal 3D printing processes.
Read the full article in Additive Manufacturing.
Subscribe to the 3D Printing Industry newsletter to keep up with the latest 3D printing news.
You can also follow us on LinkedIn and subscribe to the 3D Printing Industry YouTube channel to access more exclusive content.
At 3DPI, our mission is to deliver high-quality journalism, technical insight, and industry intelligence to professionals across the AM ecosystem.
Help us shape the future of 3D printing industry news with our 2025 reader survey.
Featured image shows cross-section comparisons of real and predicted multi-layer L-DED prints. Image via Shang et al. / Additive Manufacturing (2025).
