The Graduation project is a part of the TISCALI (Technology Innovation for Sewer Condition Assessment – Long-distance Information-system) project at the University of Twente. Maintenance of underground pipelines and sewage systems is a major challenge due to the mere fact that they are underground. If the sewer pipe has defects or voids, it can lead to soil particles and groundwater entering the voids, thus creating an underground cavity that might lead to a collapse of the soil over the pipe. This damages the infrastructure and can also hurt personnel. The most common way of management of these pipes is to dig the ground around the sewer pipes to access them. This work demands highly skilled operators and might lead to mistakes that lead to further destruction of infrastructure.
The TISCALI project aims to create a noninvasive method of inspecting the sewer lines that will thereby decrease any possibility of damage to the infrastructure. "The project aims at utilizing, integrating, and further development of relatively low-cost, off-the-shelf, techniques to arrive at an objective detection and quantification of defects in sewers and to determine the constructive strength and stability of sewers" [1]. The underground voids are hence detected using noninvasive methods by employing a ground-penetrating radar (GPR). GPR is a non-destructive method that uses electromagnetic radiation and detects the reflected signals from the structures underground. The generated radargram hence has information on the underground structures based on their dielectric properties. The radargram images are obtained with an In-pipe GPR from within the sewer walls. This leads to utilising high-frequency signals to obtain a better resolution of the radargram images since the distance between the voids and the radar is decreased.
The graduation assignment will hence focus on the classification of the objects in the radargram using Machine learning techniques.
The problem questions are:
* What pre-processing steps can be taken to remove noise and for signal processing?
* What image segmentation technique can be applied to find the location of the voids?
* What Machine learning algorithm will suit best for pattern recognition? How can the radargram output be used to create a differentiation between a void filled with water and air?
The final goal will be to have an algorithm that will detect only voids in a sewer pipe.
[1] https://www.ram.eemcs.utwente.nl/research/projects/tiscali