Multiplier case: AI assistance system in X-ray imaging<\/h2><\/div>Preliminary study on the reliable identification of defects in image sections<\/h4><\/div><\/div><\/div><\/div><\/div>Summary<\/h3><\/div><\/div><\/div>The project is investigating the feasibility of an AI system for inspecting X-ray images that detects and localises defects. The key challenge lies in the reliable detection of very small defects (1\u20132 mm) when their size relative to the overall image increases the likelihood of false positives. The aim is to use real-world data to demonstrate that the system achieves a reliability of \u2265 99 per cent in identifying defect-free components.<\/p>\n<\/div><\/div><\/div>
![](\"https:\/\/humaine.info\/wp-content\/uploads\/2025\/06\/doncasters_sentin_mulitplikatorfall.bmp\" "\\"doncasters_sentin_mulitplikatorfall\\"")
<\/span><\/div>Photo:<\/em> DONCASTERS Precision Castings-Bochum GmbH<\/em><\/p>\n<\/div><\/div><\/div>Description of the Approach<\/h3><\/div>To develop the AI system for inspecting X-ray images, the images were divided into smaller sections (patches) to enable the precise detection of defects. Various approaches were tested, including image segmentation and classification. Given the characteristics of the data, a classification approach was chosen in which the patches were categorised as \u2018OK\u2019 (fault-free) or \u2018not right\u2019 (faulty). To prepare the AI models, around 10,000 patches were created from annotated images, and a neural network was trained and validated to ensure robust and reliable detection.<\/p>\n<\/div><\/div><\/div>
Goals & Results<\/h3><\/div>The project aims to reduce the workload on skilled staff by providing objective decision support and to improve the efficiency of defect detection. The AI system achieves a 40% increase in evaluation efficiency by reducing the number of image sections to be inspected. It enables reliable classification and localisation of defects, whilst ensuring that no defective image sections are mistakenly classified as fault-free.<\/p>\n<\/div><\/div><\/div><\/div><\/div>
Partner<\/h1><\/div><\/div><\/div>Practice Partners<\/h3><\/div>
Preliminary study on the reliable identification of defects in image sections<\/h4><\/div><\/div><\/div><\/div><\/div>Summary<\/h3><\/div><\/div><\/div>The project is investigating the feasibility of an AI system for inspecting X-ray images that detects and localises defects. The key challenge lies in the reliable detection of very small defects (1\u20132 mm) when their size relative to the overall image increases the likelihood of false positives. The aim is to use real-world data to demonstrate that the system achieves a reliability of \u2265 99 per cent in identifying defect-free components.<\/p>\n<\/div><\/div><\/div>
<\/span><\/div>Photo:<\/em> DONCASTERS Precision Castings-Bochum GmbH<\/em><\/p>\n<\/div><\/div><\/div>Description of the Approach<\/h3><\/div>To develop the AI system for inspecting X-ray images, the images were divided into smaller sections (patches) to enable the precise detection of defects. Various approaches were tested, including image segmentation and classification. Given the characteristics of the data, a classification approach was chosen in which the patches were categorised as \u2018OK\u2019 (fault-free) or \u2018not right\u2019 (faulty). To prepare the AI models, around 10,000 patches were created from annotated images, and a neural network was trained and validated to ensure robust and reliable detection.<\/p>\n<\/div><\/div><\/div>
Goals & Results<\/h3><\/div>The project aims to reduce the workload on skilled staff by providing objective decision support and to improve the efficiency of defect detection. The AI system achieves a 40% increase in evaluation efficiency by reducing the number of image sections to be inspected. It enables reliable classification and localisation of defects, whilst ensuring that no defective image sections are mistakenly classified as fault-free.<\/p>\n<\/div><\/div><\/div><\/div><\/div>
Partner<\/h1><\/div><\/div><\/div>Practice Partners<\/h3><\/div>
Summary<\/h3><\/div><\/div><\/div>The project is investigating the feasibility of an AI system for inspecting X-ray images that detects and localises defects. The key challenge lies in the reliable detection of very small defects (1\u20132 mm) when their size relative to the overall image increases the likelihood of false positives. The aim is to use real-world data to demonstrate that the system achieves a reliability of \u2265 99 per cent in identifying defect-free components.<\/p>\n<\/div><\/div><\/div>
<\/span><\/div>Photo:<\/em> DONCASTERS Precision Castings-Bochum GmbH<\/em><\/p>\n<\/div><\/div><\/div>Description of the Approach<\/h3><\/div>To develop the AI system for inspecting X-ray images, the images were divided into smaller sections (patches) to enable the precise detection of defects. Various approaches were tested, including image segmentation and classification. Given the characteristics of the data, a classification approach was chosen in which the patches were categorised as \u2018OK\u2019 (fault-free) or \u2018not right\u2019 (faulty). To prepare the AI models, around 10,000 patches were created from annotated images, and a neural network was trained and validated to ensure robust and reliable detection.<\/p>\n<\/div><\/div><\/div>
Goals & Results<\/h3><\/div>The project aims to reduce the workload on skilled staff by providing objective decision support and to improve the efficiency of defect detection. The AI system achieves a 40% increase in evaluation efficiency by reducing the number of image sections to be inspected. It enables reliable classification and localisation of defects, whilst ensuring that no defective image sections are mistakenly classified as fault-free.<\/p>\n<\/div><\/div><\/div><\/div><\/div>
Partner<\/h1><\/div><\/div><\/div>Practice Partners<\/h3><\/div>
The project is investigating the feasibility of an AI system for inspecting X-ray images that detects and localises defects. The key challenge lies in the reliable detection of very small defects (1\u20132 mm) when their size relative to the overall image increases the likelihood of false positives. The aim is to use real-world data to demonstrate that the system achieves a reliability of \u2265 99 per cent in identifying defect-free components.<\/p>\n<\/div><\/div><\/div>
<\/span><\/div>Photo:<\/em> DONCASTERS Precision Castings-Bochum GmbH<\/em><\/p>\n<\/div><\/div><\/div> To develop the AI system for inspecting X-ray images, the images were divided into smaller sections (patches) to enable the precise detection of defects. Various approaches were tested, including image segmentation and classification. Given the characteristics of the data, a classification approach was chosen in which the patches were categorised as \u2018OK\u2019 (fault-free) or \u2018not right\u2019 (faulty). To prepare the AI models, around 10,000 patches were created from annotated images, and a neural network was trained and validated to ensure robust and reliable detection.<\/p>\n<\/div><\/div><\/div> The project aims to reduce the workload on skilled staff by providing objective decision support and to improve the efficiency of defect detection. The AI system achieves a 40% increase in evaluation efficiency by reducing the number of image sections to be inspected. It enables reliable classification and localisation of defects, whilst ensuring that no defective image sections are mistakenly classified as fault-free.<\/p>\n<\/div><\/div><\/div><\/div><\/div>Description of the Approach<\/h3><\/div>
Goals & Results<\/h3><\/div>
Partner<\/h1><\/div><\/div><\/div>
Practice Partners<\/h3><\/div>
![](\"https:\/\/humaine.info\/wp-content\/uploads\/2024\/02\/doncasters-logo.png\"){kind=logo}
<\/a><\/span><\/div>![](\"https:\/\/humaine.info\/wp-content\/uploads\/2021\/05\/Logo_sentin-GmbH.png\"){kind=logo}
<\/a><\/span><\/div>