Design of hybrid deep learning using TSA with ANN for cost evaluation in the plastic injection industry

Abstract

In the plastic injection industry, plastic injection molding is one of the most extensively used mass production technologies and has been continuously increasing in recent years. Cost evaluation is essential in corporate operations to increase the market share and lead in plastic part pricing. The complexity of the plastic parts and manufacturing data resulted in a long data waiting time and inaccurate cost evaluation. Therefore, the aim of this research is to apply a cost evaluation approach that combines hybrid deep learning of a tunicate swarm algorithm (TSA) with an artificial neural network (ANN) for the cost evaluation of complicated surface products in the plastic injection industry to achieve a faster convergence rate for optimal solutions and higher accuracy. The methodology entails the ANN, which applies feature-based extraction of 3D-model complicated surface products to develop a cost evaluation model. The TSA is used to construct the initial weight into the learning model of the ANN, which can generate faster-to-convergent optimal solutions and higher accuracy. The result shows that the new hybrid deep learning TSA combined with the ANN provides more accurate cost evaluation than the ANN. The prediction accuracy of cost evaluation is approximately 96.66% for part cost and 93.75% for mold cost. The contribution of this research is the development of a new hybrid deep learning model combining the TSA with the ANN that includes the calculation of the number of hidden layers specifically for complicated surface products, which are unavailable in the literature. The cost evaluation approach can be practically applied and is accurate for complicated surface products in the plastic injection industry.