Improving industrial safety and maintenance: Cenosco and EDIH Adria test AI for corrosion detection

In demanding industrial environments such as oil and gas installations, chemical industry and mining, corrosion poses a constant threat to equipment integrity, worker safety and environmental protection. Timely detection and accurate corrosion assessment are essential to prevent costly failures, unplanned downtime and potential environmental incidents. Cenosco, a leading company in the development of Asset Integrity Management (AMM) software, has recognised the need for innovative solutions that would build on existing inspection and monitoring methods. Traditional approaches, although reliable, can be time-consuming and subjective, while the amount of visual data generated during inspections is constantly increasing.

Solution through Test-Before-Invest (TBI) program

To explore the potential of the latest technologies, Cenosco teamed up with the EDIH Adria consortium through the “Test Before Invest” (TBI) programme. The aim of this collaborative project was to develop and test a prototype of an artificial intelligence (AI) model capable of automatic identification, classification and assessment of corrosion grade based on photographs of pipelines and other components of industrial plants. The focus was on the application of advanced computer vision and deep learning techniques to automate and objectify the process of visual inspection data analysis.

As part of the TBI activities, experts from the Juraj Dobrila University of Pula, a partner in the EDIH Adria consortium, worked with a comprehensive set of data provided by Cenosco. Key steps included:

• Detailed analysis and data preparation: The initial meeting consisted of high-resolution photographs and related annotations. Data was cleaned and validated to ensure consistency.
• Developing a high-resolution image processing strategy: Due to the loss of detail in standard image scaling for AI models, an innovative technique of dividing original images into smaller, overlapping segments has been implemented. This approach has proven to be crucial for preserving the fine details necessary for accurate corrosion detection.
• Training and evaluation of AI models: YOLOv11 (You Only Look Once) architecture was used, known for its efficiency in object detection tasks. The “nano” and “small” versions of the model were tested, with the “small” model showing superior performance after 38 hours of training.

Results and benefits for Cenosco

TBI project has successfully demonstrated the technical feasibility and significant potential of AI application for automated corrosion detection:

• Satisfactory performance achieved: The YOLOv11 small model achieved a mAP50 value of 0.756 and a mAP50-95 value of 0.589 on the validation set, which are promising results for a complex industrial task. The model showed the ability of precise classification (precision 0.857) and good recall (recall 0.662). A more detailed performance analysis by defect class shows that the model achieves the best results for corrosion detection on coatings, while more challenging tasks are corrosion detection on medium grade piping equipment and insulation problems. The model achieves a recall of 0.79 at a confidence threshold of 0.000, which confirms its ability to reliably identify different types of defects in industrial conditions.

• Confirmation of the effectiveness of the ‘tiling’ strategy: Sharing images into smaller segments allowed the model to learn efficiently from the details, which is crucial for detecting subtle signs of corrosion.
• Insight into the potential for automation: The AI model showed the ability to quickly analyze a large number of images, identifying areas affected by corrosion and classifying their type and degree. This paves the way for a significant reduction in the time required for manual examination and an increase in the consistency of assessments.
• The basis for future development: Through the project, Cenosco received a detailed insight into the applicability of AI technology for corrosion detection, including a demonstration of the functionality of the developed prototype approach, an evaluation of its performance and a comprehensive report with recommendations. These insights serve as a solid basis for Cenosco to make decisions about further improvement and potential integration of advanced AI functionalities into its Integrity Management Suite (IMS). The project has successfully demonstrated technological feasibility at TRL 3-4 level.

Lessons learned and future steps

This TBI project provided Cenosca with valuable insights:

• Importance of quality data: The success of AI models inevitably depends on a large, diverse and precisely annotated dataset.
• Adapting to specific challenges: Standard approaches often require innovative adaptations, such as image-sharing strategies, to address the specificities of industrial data.
• Iterative development: AI solutions require continuous improvement, experimentation with new architectures and techniques, and feedback-based adaptation.

Cenosco plans to continue researching and developing AI solutions, focusing on further improving model accuracy, expanding the data set. The aim is to provide its clients with even more advanced tools to proactively manage asset integrity, reduce risks and optimise maintenance costs.

Conclusion

The collaboration between Cenosco and EDIH Adria through the TBI program has successfully demonstrated how the application of artificial intelligence can transform inspection and maintenance processes in demanding industrial sectors. This project not only confirmed the technical feasibility of AI corrosion detection, but also laid the foundations for future innovations that will contribute to greater safety, efficiency and sustainability of industrial facilities. EDIH Adria continues to support companies like Cenosco in their digital transformation, allowing them to test and implement advanced technologies that bring real business value.