Chicken Swarm Optimization and Deep Learning for Manufacturing Processes

Abstract

In this paper we propose an approach that uses the Multilayer Perceptron Classifier (MPC), a type of deep neural network, for classifying the products generated by the manufacturing processes as faulty or not faulty. The number of nodes from each hidden layer of the MPC was determined using an adapted version of the Chicken Swarm optimization (CSo) algorithm. The proposed method was integrated in an experimental prototype and was tested and validated on two representative datasets for manufacturing processes, namely the SECOM and the SETFI datasets. The main contributions of this article are (1) the use of a MPC for classifying the products resulted after the application of the manufacturing processes as faulty or not faulty, (2) the identification of the number of nodes of the hidden layers of the MPC using CSO and (3) the testing and the validation of the proposed approach using an experimental prototype developed in-house.