Project title: Automatic crack classification in large scale tunnel structure using deep learning
Researcher: Darragh O'Brien
This research aims at the integration of technology into the system of inspection of large-scale underground infrastructure environments. This incorporation is operated through the creation of two distinctive convolution neural network (CNN) algorithms: one to detect cracks in the tunnel lining and the second to classify the subsequently detected cracks. The inevitability of deterioration of any structure over time is unavoidable. Cracks are the initial signal of deterioration in a structure and are a very common phenomenon that can be instigated by a singular or a culmination of numerous forces on the structure be they internal or external. The corrosion of reinforcement, chemical weakening of the concrete tunnel lining, and application of adverse loading to the structure are the most common forces on the structure causing shortcomings in the lining in the form of water leakage, spalling, cracking, and deformation. To ensure tunnel serviceability, periodic tunnel inspections must be carried out. Traditionally these inspections are undertaken by trained professionals where they examine the surface of miles of tunnels in search of defects. This conventional manual method is costly, time-consuming, and inevitably subjective method of inspection leading to variation in results.
This research project will use Transfer Learning (TL) with a pre-trained Deep Convolution Neural Network (DCNN) for automatic crack detection and classification. The CNN will be trained/validated and tested on images acquired from the European Centre for Nuclear Research (CERN) while CERN facilities are shut down from 2019 to 2021.
The proposed research in the area of Geohazards is a fundamental step towards an inspection method that gives an unbiased factual view on the structural health and the behaviour of the tunnel lining. This development in tunnel inspection will highlight deterioration in the tunnel lining earlier, giving an insight into the behaviour of the tunnels thus prolonging the lifespan of the tunnel while decreasing inspection time enabling less inconvenience to the serviceability to the tunnel users.