Speaker
Description
INTRODUCTION Zero-Defect Manufacturing (ZDM) is an emerging European industry paradigm focused on reducing waste and costs while maximizing product quality. Non-Destructive Inspection (NDI) systems integrated into production lines enable proactive quality assessment. This work presents openZDM project results, focusing on InfraRed Thermography (IRT) across three industrial use cases. The research demonstrates how thermographic data, integrated into an open platform based on RAMI4.0 and Asset Administration Shell standards, drives real-time decision-making, enabling efficient ZDM implementation.
USE CASES AND METHODOLOGIES Advanced IRT solutions were deployed in three pilot lines:
1) VDL Weweler, production of trailing arms for heavy vehicles by hot-forming of steel bars: two thermographic NDIs were installed to measure the temperature profile of the product and residual oxidation (scale) after the induction furnace and the descaling unit. Calibrated for high temperatures (about 1200°C), these sensors ensure the steel bars meet specific thermal requirements before hot-forming.
2) APTIV, EV battery production: for the assembly of electric vehicle battery trays, a thermography-based system was implemented to inspect the welding process. The system utilizes AI-driven models to analyze thermal cooling patterns and spatial attributes, identifying welding defects such as missing welds or irregularities.
3) VIDRALA, glass container manufacturing: in the hot-end forming stage, where bottles reach temperatures of approximately 700°C, a dual-camera setup was deployed. This system captures infrared emissions to indirectly estimate the wall thickness of the bottle and predict defects 45 to 90 minutes before the cold-end quality control.
RESULTS AND DISCUSSION The implementation achieved significant successes while also highlighting practical limitations.
At VDL Weweler, the thermal cameras successfully provided real-time data to the digital twin, enabling process parameter recommendations. However, the hostile environment due to extremely high temperatures requires continuous diagnostic strategies in the system. Acquisitions are made during a thermal transient that must be considered and corrected.
In the APTIV use case, the integration of IRT with data-driven quality modules achieved a defect detection rate of nearly 99%. The system has been operational for over a year, though it has required iterative hardware improvements regarding refrigeration and camera positioning to withstand the industrial environment.
For VIDRALA, the IRT models demonstrated high correlation for thickness prediction at specific control points. Nevertheless, the challenges encountered are related to the ability to generalize the model to all mold cavities and to ensure the representativeness of the data for complex cases and to have a sufficiently large dataset in the whole range of thickness values for training the neural model.
CONCLUSION The openZDM project demonstrates that IRT is a mature enabler for ZDM when supported by robust data platforms. Different thermographic systems have been designed and developed for various industrial contexts to improve efficiency and effectiveness, successfully addressing the challenges encountered in different scenarios. While AI models fed by thermal data proved highly effective for defect prediction, successful deployment requires addressing on-the-edge challenges, including network latency, hardware durability, and the scalability of calibration models across variable production conditions.
