A Hybrid PSO-MLP-based Inverse Modeling of Manufacturing Process: Case Study with ADI Process

Manufacturing processes in most of the industries involve complex interactions between its input processing parameters, such as, input materials and temperature, pressure, humidity, etc., in which their interdependence is highly complex and nonlinear. The physical properties of the output product, such as, hardness, tensile strength, etc., depend on these interactions. However, there may not have any exact analytical solution to determine the input process parameters for a given physical property of the output. To fulfill one of the objectives of sustainable manufacturing, it is necessary to predict an optimum set of input process parameters for a given physical property of the output product, in order to reduce the use of precious natural resources and energy, and to reduce wastage. In this paper, a novel inverse modeling of the manufacturing process is proposed that is based on a hybrid scheme of particle swarm optimization (PSO) and multilayer perceptron (MLP) algorithms. By taking a case study of manufacturing of austempered ductile iron (ADI), with extensive simulation results, it is shown that the proposed technique is able to predict an optimum set of input process parameters which can save materials and reduce the cost of production. It is shown that the model predicted processing parameters can save US$40-$85 per tonne of ADI produced (i.e., about 13-25% in cost saving).