A systematic approach to model and optimize wear behaviour of castings produced by squeeze casting process

The present work is an attempt to produce squeeze cast component with excellent wear resistance property. The material wear rate in squeeze casting depends on appropriate selection of pressure duration, squeeze pressure, die temperature and pouring temperature. Experiments are conducted and data is collected as per central composite and box-behnken design approaches. The input-output relationship developed by utilizing central composite design is found to be statistically adequate and yielded better prediction accuracy. Recurrent and back propagation neural networks are trained by using data generated from best response model. The huge training data in batch mode helps to capture fully the dynamics of squeeze casting process. The recurrent neural network outperformed both, the back propagation neural network and central composite design. Genetic algorithm, desirability function approach, and particle swarm optimization are used to determine best set of squeeze casting conditions that locate the extreme values and will result in minimum wear rate. Particle swarm optimization and genetic algorithm outperformed desirability function approach, as the former carried out search in many directions at multi dimensional space, simultaneously. The results of non-linear regression, neural network based models, the performance of different optimization techniques are compared and some concluding remarks are made.