THE WELDING of in-service pipelines by the hot tapping technique requires skill and knowledge to avoid potential problems such as burn-through and hydrogen assisted cold cracking (HACC). Predicting the hardness of the heat affected zone (HAZ), and hence trying to control it, is a useful way of minimizing the risk of HACC. HAZ hardness, which is ultimately determined by the cooling rate of the weldment and the chemical composition and structure of the as-received steel, must be restricted by producing microstructures that resistant to HACC. Currently the industry uses several models and software packages to predict safe operating ranges for in-service welding [1]. The establishment of improved procedures was highlighted as a key requirement at a recent international conference [2]. This paper reports on a potential alternative approach using back-propagation neural networks to predict the HAZ hardness of various Australian and USA pipeline and tap fitting steels that have undergone simulated weld thermal profiles using a dilatometer. Predicted and experimental hardness values were well matched and the sensitivity of key input variables agreed with findings in the experimental investigation and in the literature. The neural network model was benchmarked against results from existing numerical models, and found to have a lower error between predicted and experimental HAZ hardness values.
展开▼
