Forecast Surface Quality of Abrasive Water Jet Cutting Based on Neural Network and Verified by Experiments

In this study, firstly, the YL12 aluminum alloy is used as experimental materials, then in the following experiments it is cut in JJ-I-type water jet machines, and 1,000 group data are gotten by measurement. In each group data, pressure, material thickness, surface roughness, abrasive flow and traversing speed are included. Next, BP artificial neural network is established. In this network, there are four inputs and one output. The inputs are pressure, material thickness, surface roughness and abrasive flow rate; the output is traverse speed. And then the BP artificial neural network is programmed by one toolbox of Matlab. Using the former 1,000 group data, the BP artificial neural network is trained, and its forecast function is obtained. Finally, the BP neural network is tested to verify through using different thickness of aluminum alloy verifies its forecast function. According to given pressure, material thickness, roughness and abrasive flow, traverse speed is predicted. The YL12 aluminum alloy is cut by the predicted traversing speed. The maximum error between the prediction values of surface roughness and the actual values of the surface roughness is 6.5 %. Copyright © 2013 IFSA.