Detection with k-nearest neighbour algorithm for cracked concrete

Mahadzir, Muhammad Zharfan (2022) Detection with k-nearest neighbour algorithm for cracked concrete. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

	Text (24 Pages) Detection with k-nearest neighbour algorithm for cracked concrete.pdf - Submitted Version Download (262kB)
	Text (Full text) Detection with k-nearest neighbour algorithm for cracked concrete.pdf - Submitted Version Restricted to Repository staff only Download (3MB)

Abstract

Nowadays, the construction sector plays a vital role in country development. Many high and unique buildings were made to show how the economic status and progress of the country developed. Thus, solid and high efficiency materials such as concrete are needed to construct modern and robust buildings. Even though concrete has high material durability compared to other materials, that specialty cannot deny that the concrete has one main problem if not well maintained. The problem is the cracked wall surface structure. Generally, three types of cracks can be found at the concrete surface, there is a minor crack, moderate crack, and severe crack. This will make the building safety low. The current method nowadays to perform structural health monitoring (SHM) for the concrete wall is by using manual inspection that is visual inspection. The reason why SHM is performed is to detect concrete surfaces and monitor the current structure condition. Several problems have been carried out in this study when the manual inspection is applied to inspect the presence of cracks toward the concrete surface. The first problem is that manual inspection consumes more time to inspect the presence of cracks at the surface since the inspection is done visually and can be done by an experienced inspector only. Next is, hazardous environment to perform inspection increase the safety risk toward the inspector since the presence of crack indicates that the building is not in good condition. Lastly, a limited number of experienced inspectors has become one of the problems for current crack building condition inspection. Thus, the main objective of this study is to produce a technique that can detect the present crack at the concrete surface by using the K-Nearest Neighbor (KNN) algorithm, which can reduce the crack inspection time, reduce the number of inspectors at the hazardous area and lastly required fewer experiences manpower for crack inspection. This technique can detect the presence of crack by using crack concrete surface images. The highest crack classification accuracy obtained in this work was 94.40% correct class classification. The suitable number of greyscale intensity level that had been used is 0.4 for features extraction and the K value is 3. The best number of training datasets that had been used to archive 94.40% accuracy level is 800 crack datasets and 100 non-crack datasets that use 90.00% training and 10.00% testing from the datasets. In conclusion, the crack detection method was able to detect the crack dataset with high correct classification accuracy.

Actions (login required)

View Item