Neural network analysis of experimental data for air/water spray cooling

Abstract To obtain appropriate strength properties, nickel-based superalloy or titanium materials used in the aerospace industry are heat treated by cooling from high temperatures. Unacceptably high residual stresses may result, if the rate of cooling is too high. After extensive investigation, it was found that air-assisted atomised water sprays offer an excellent capability of controlling the rate of cooling, and they are a viable alternative to the widely used techniques of quenching in oil or water. The heat transfer data were obtained for a wide range of pressure ratios and hence water flows for surface temperatures of up to 850°C. This paper provides a neural methodology for heat transfer analyses of data obtained experimentally during the investigation of the use of air-assisted atomised water spray systems for the controlled cooling of high temperature forgings. The model created to train the neural network relates the spray input variables to the corresponding heat transfer data for the range of conditions observed experimentally. For comparison purposes and accurate evaluation of the predictions, part of the data is used to train the neural network and the remainder to test the model. It is described in detail how a neural network can be trained to successfully predict the resulting heat flux for specific input spray parameters. This particular knowledge can then be used to optimise the process, i.e. to establish the spray conditions that would yield the cooling rate required to attain the pre-specified mechanical properties, and to minimise the residual stresses.