Enabling Data-Driven Decision Making in Additive Manufacturing

The challenge

Laser powder bed fusion AM workflows, and other AM processes, are frequently forced to make quality decisions late in the process chain. This creates cost, delay, and risk, especially when a part is only identified as scrap after significant downstream effort is already completed. Post-processing and inspection alone can account for up to 40% of the total part cost, so late scrappage can waste substantial time and resources.

To avoid unnecessary and costly post processing, operators may need to manually gather and interpret very large datasets across multiple systems, including in-process monitoring signals, machine process variables, and other build information after a build completes. Even where monitoring and analytics tools already exist, they are often siloed, making it difficult to connect data sources and generate actionable insight. Furthermore, shopfloors with diverse manufacturing equipment, like the one at the MTC, are faces with an exponentially increasing challenge to pull useful data from their equipment, due to the broad range of connectivity protocols, data formats, and lack of semantic clarity. This is a challenge faced across industry and limits progress towards more advanced approaches such as digital verification and certification. 

MTC's solution

MTC developed and demonstrated an automated AM part quality decision making workflow. To achieve this, first a series of controlled AM builds were conducted, followed by inspection to generate a high-quality dataset. The build strategy allows us tto reduce the variability and create a clear training dataset for AI models.  

• Process: Laser Powder Bed Fusion (LBPF)
• Machine: EOS M290
• Material: Aluminium CP1

Our EOS M290 platform provided us with two key in-process monitoring methods:

• Melt pool monitoring (MPM) - producing intensity signals that indicate melt pool characteristics
• Optical thermography (OT) - producing intensity signals that indicating the long exposure heat distribution across each layer

A staged build plan was executed:

• Build 1 created the material-specific calibration needed for MPM on Aluminium CP1.
• Build 2 used a simplified parameter strategy and a planned variation of process parameters to generate a wide range of part quality outcomes.
• Build 3 included additional sample sizes to assess size effects in monitoring signals.
• Build 4 used standard parameter strategies and more component-relevant geometries to test the approach in a more representative scenario.

Once the builds were completed, Archimedes density measurements were collected for all parts to provide a porosity target, and these were validated using X-ray computed tomography (XCT) and Ultrasonic testing (UT). The inspection quality measurements were then linked to in process monitoring signals statistics, such as average emissions per layer. Exploratory data analysis confirmed a strong and repeatable relationship between in-process monitoring part statistics (MPM and OT signal summaries) and bulk porosity, providing a basis for predictive modelling.

A single source of truth for connected AM datasets was then established to address data silos and handle the diverse collected datasets. Granta MI was utilised as a centralised data platform to store and link:

• Build records, materials, geometries and part identifiers
• In-process monitoring summaries
• Test and inspection outcomes

This created a repeatable structure where each part could be traced from build conditions through to measured outcomes, and where other analytics could access consistent, connected datasets without relying on bespoke spreadsheets. Finally, a predictive model with a simplified decision tool user interface was created and deployed through a web-based interface for practical use:

• Users select a completed build and then select a part from that build
• The model instantly returns a predicted porosity value for the selected part
• Prediction quality indicators are presented with clear, human readable explanations to support confidence and consistent decision-making 

The outcome

The project delivered a working toolset that integrates build data, in-process monitoring statistical analysis, and inspection results into a single decision workflow. It demonstrates the feasibility of predicting part porosity using data available at build completion, supporting earlier go or no-go decisions and reducing reliance on late and manual data inspection.

The work also established a scalable foundation for future development, with plans to expand datasets across additional builds, materials, and data sources to consistently improve performance and move closer to industrial deployment readiness. The same data platform approach also enables future extension into broader defect classifications (beyond porosity), defect localisations to inform targeted inspection, automated process window development, and supporting activities aligned to digital part and process qualifications. 

Benefits to industry

• 40% Cost Saving - Earlier and faster AM part quality insights, reducing cost and delay associated with late decisions
• 3 Hour Time Saving per Part – Accelerated data collection and analysis, presented in one location with reduced errors
• 100% Consistency – Fully consistent data driven decision making supported by connected monitoring and historical inspection evidence
• Transparency - A traceable data record linking build conditions, monitoring signals, and test outcomes
• Scalability - A scalable digital foundation that can be expanded to support future digital verification activities