Computer vision is one of the most commonly used methods of artificial intelligence in engineering applications. It is often used to detect defects in various products. As the applications are often complex, several different algorithms are often combined to perform the tasks successfully. In this master’s thesis, an algorithm was designed to detect the location of insulation damage on the stator of an electric motor. Methods such as contrast enhancement were used for image preprocessing and algorithms like convolutional neural network and K-means clustering were used for image classification and further processing. The result was a functional algorithm that identified the damaged pole with 95% success rate and identified the location of the damage by stator height, thickness, and width with 87%, 41%, and 94% success rates.
|
