Production is an activity that has its roots in human existence. The initial production activities were carried out only for the man himself and his family. In other words, there is no customer to whom goods or services are offered. Therefore, the concept of customer satisfaction has not emerged yet. After this satisfaction has emerged, the need for quality control of the products arose. Along with the developing systems, quality control systems have been evolved through image processing. Thanks to the image processing-based control systems, the capacity and efficiency of the production facilities can be increased and perfect products can be delivered to the end-user by making precise measurements.
This paper studies to detect the undetectable defects in casting such as air holes, pinholes, burrs, tensile defects, mold material defects, metal casting defects, metallurgical defects, and etc. through images via deep learning methods. An automatic reading review for submersible pump impeller is proposed and a deep learning model is developed. The data set of the images include defective and smooth submersible pump impeller has been utilized to prove the performance of the designed network architecture. According to the obtained results, maximum accuracy of 89% for the classifier has been achieved in the training stage and it reached a maximum accuracy of 93% in the testing stage.
Anahtar Kelimeler: Convolutional neural network, Image processing, Faulty product
|
