Accuracy enhancement in mobile phone recycling process using machine learning technique and MEPH process

Abstract The technological development, market expansion and increased population will lead the increased use of electronic equipment and production of e-waste worldwide. Disposal of electronic equipment is a challenging problem across the globe. Improper way of electronic waste disposal leads human health risk and environmental pollution. The report shows that 50,000 million tons of e-waste generated across the globe. Electronic waste includes CRT (Cathode Ray Tube), PCB (Printed Circuit Board), unused Television (TV), computers and mobile phones. In this paper we focus on recycling of unused mobile phones. The main objective of this research is introducing automation in metal purification measurement and improvement process using machine learning. The mobile phones contain different toxic metals such as cadmium, beryllium, lead and arsenic. Improper ways of unused mobile phone disposal contaminate water, air and soil. This research work has two parts. First part uses MEPH (Magnetic separation, Eddy current, Pyrometallurgical and Hydrometallurgical) process for metal separation, metal extraction and purification process. In the second part the purified metal is captured through camera and the captured image is subject to noise removal and given as input to the Convolutional Neural Network (CNN) classifier. The classification process is done in two ways; first one is taking input and classing the output. Second one is find the percentage of similarity to the particular class. We used the later one for finding the percentage of similarity between recycled metal and the pure metal. Suppose similarity is less than 90%, the purification process will be improved to enhance purity. In Machine learning different methods are available for feature extraction and classification among which we used the CNN. It easily find the spatial and temporal dependencies of input image by applying proper filters and also extracts high level features, dominant features using convolution and max pooling operation. This operation also reduces the computational power needed to process the input. Activation function plays important role in the feature extraction and classification process. In our research we used the ReLU (Rectified Linear Unit) function for validating the features learned from the input image. The most important advantage of using CNN is that it discovers the significant features without the human command. Also we used the image augmentation to increase the input image data set. The accuracy of metal classification measured using confusion matrix. From this research we got the purified metal and it is directly used for other product manufacturing.