Hardness Prediction System for Multi-pass Weld Metal of Low-Alloy Steel Using Neural Network

Multi-pass welds are subjected to various thermal cycles and have a complicated microstructure distribution; therefore, defects can easily occur. To prevent the defects, it is important to select the appropriate welding conditions before welding. Hardness is the most convenient criterion for the safety evaluation. A hardness prediction system for the heat-affected zone (HAZ) of a multi-pass weld has been proposed by the authors. However, in actual welding, defects can occur not only in the HAZ but also in the weld metal (WM) of a multi-pass weld. Therefore, in this study, hardness prediction system for the WM of a multi-pass weld was developed based on a neural network and database of experimental measurements, a new hardness prediction system of WM has been developed by using neural network. In addition, a hardness prediction system for an entire multi-pass weld, including both the WM and HAZ, was developed by combining the new system with the previously proposed hardness prediction system for the HAZ. Therefore, the hardness values for both the WM and HAZ could be predicted based on the simulated thermal cycles of multi-pass welding, and the calculated values were found to be in good agreement with the measured results. This indicated that the newly developed hardness prediction system for multi-pass welds, including both the WM and HAZ, was effective for selecting appropriate welding conditions prior to actual multi-pass welding.