Identifikasi Visual Cacat Produk Menggunakan Neural Network Model Backpropagation (Studi Kasus: PT. Panasonic Gobel Eco Solution)

Abstrak

Product defects are common in the production process. Visual identification of product defects is first carried out when the product is produced. Identification of vague defects in very small shapes with different sizes and positions is difficult to do with ordinary eye sight, so that often results in decisions about the status of the product that is not right. Product defects in visual form can be identified by patterns such as shape, size and position on the product image. In this study, we will apply a neural network with the backpropagation model as a classification of the pattern. Product images will be processed using image processing by converting the RGB pixel value of the image into a numeric value. Data in numerical form will be input for training values in the backpropagation model. Training results are used to identify identified product defects and produce product status decisions. The results show that the backpropagation neural network model is able to recognize product patterns with an accuracy of 99.24% and based on simulation test data with the final weight and bias of training results, able to identify product defects with success up to 91%.