Optimization of hot extrusion using single objective neuro stochastic search technique

This paper presents a new single-objective neuro-stochastic search technique (SONSST) for the economic load estimation problem in hot extrusion which is often used to produce long straight metal products of constant cross-sections such as bars, solid and hollow sections, tubes, wires and strips from materials that cannot be formed by cold extrusion. The shape of the dies and the temperature developed during extrusion and the velocity of the dies significantly influence forging force at which the process is to be carried out. In order to understand the complex relationship between the material and process variables, a few finite element models are developed and simulated in the FORGE2 environment. These finite element simulations are used to train a neural network (NN) model. Later the same model is incorporated along with a genetic algorithm (GA) and simulated annealing (SA) to form SONSST. It incorporates a genetic crossover operator BLX-α and a problem specific mutation operator incorporating a local search heuristic: to provide it a better search capability. Extensive simulations have been carried out considering various aspects and the results are validated with those of the existing finite element method in the literature. These results indicate that the new SONSST heuristic converges to better solutions rapidly. SONSST is a truly single-objective technique as it provides the values of various process parameters for optimizing single objective (extrusion load), in a single run and thus assists in achieving energy and material saving, quality improvement and in the development of sound extruded parts.