AI-Anomaly Project: Condition Assessment of Urban Water Assets through the Detection of Cracks based on Artificial Intelligence Methods

The current paper aims at presenting the project AI-Anomaly which contributes to innovate, extend and automate the existing condition assessment approach of urban water assets by identifying and classifying anomalies in inspectable components. A dataset of photographic recordings collected from in situ inspections of water storage tanks and pumping stations is used and processed, including drawing and labeling of cracks, image binarisation and division into patches with a specific resolution. An existing convolutional neural network pre-trained with images of concrete surfaces with and without cracking is applied to these urban water assets. Different explicability artificial intelligence methods are applied to ensure the transparency of the used artificial intelligence methods and the identification of bias in the results.