AI-Assisted Maintenance and Condition Assessment of Submerged Infrastructure

Purpose and goal:

The project aims to identify the obstacles an facilitate the entry of Machine Learning (ML) for automated inspection of marine structures by developing suitable ML models for specific conditions and environment of such structures, and ultimately autonomous condition assessment of marine structures. Such an inspection system is expected to reduce the inspection and condition assessment costs by at least 50% and increase the speed of the operations by at least 50%.

Expected results and effects:

A hybrid neural network model for image processing was developed. The model can detect cracks in concrete structures in rather complex backgrounds present in marine environments. The expected effects include: - To dramatically reduce the maintenance costs for underwater infrastructure owners,

- To dramatically reduce the risk of human injuries due to underwater inspections by promoting and facilitating autonomy, - To increase the relevant knowledge and competence in Sweden. Approach and implementation:

- Identify the underwater structural distresses: We identified the most common damage types in concrete structures in marine environments,

- Investigate different AI tools and select the most suitable one: Different ML models were studied and a hybrid convolution neural network model was established,

- Validation and evaluation of the results: The model was examined for a set of evaluation images from a concrete pier at the Port of Gothenburg and a promising average accuracy of 93% in detection of cracks was obtained.