ML-s-LeAF – AI and Acoustic Monitoring for Quality Assurance in Additive Manufacturing

Additive manufacturing as key technology for lightweight construction

Lightweight construction is central to industries such as aerospace, automotive and mechanical engineering. Additive manufacturing processes, such as LPBF, enable the production of complex, functionally integrated and weight-optimised components that would be difficult to produce using conventional methods. This reduces material usage, weight, and energy consumption, while also creating new design possibilities.

Quality assurance is a major challenge in the LPBF process

Despite its advantages, additive manufacturing places high demands on quality assurance. Defects such as pores, cracks or bonding defects can occur during the manufacturing process, compromising the mechanical properties of the components. Such defects are often not detected until after the manufacturing process is complete, resulting in additional inspection efforts, waste, and increased costs. Therefore, for industrial mass production, methods are needed that can reliably identify defects during the manufacturing process itself.

Acoustic monitoring with artificial intelligence

As part of the ML-S-LeAF research project, Fraunhofer IDMT worked alongside its project partners to develop innovative acoustic process monitoring methods for additive manufacturing. The aim was to create an AI-powered quality assurance system capable of automatically detecting and evaluating manufacturing defects based on acoustic signals. Robust methods for industrial monitoring were developed using both real process data and simulation-based virtual acoustic data.

Structure-borne sound is particularly well-suited to defect detection

The project results demonstrate the great potential of structure-borne sound signals for monitoring LPBF processes. Compared to airborne sound, defects can be identified and classified more reliably. Investigations confirmed that acoustic measurement data provides valuable information about the manufacturing status, which can be used as the basis for automated in-line quality assurance.

AI models for automatic defect classification

This project involved developing, training and validating various machine learning and deep learning methods. These models can automatically detect and classify typical manufacturing defects. Additionally, a comprehensive data analysis and evaluation pipeline was developed for processing acoustic signals. These findings lay the groundwork for future intelligent monitoring systems in additive manufacturing and beyond.

• Development of AI-based methods for automated acoustic quality monitoring in additive manufacturing
• Investigation of the use of virtual audio data to generate robust ML models
• Exploration of approaches to improve the explainability of developed AI models and their testing for project-specific use cases