Prediction of the Transition-Temperature Shift Using Machine Learning Algorithms and the Plotter Database

The long-term operating strategy of nuclear plants must ensure the integrity vessel, which is subjected to neutron irradiation, causing its embrittlement over time. Embrittlement trend curves used predict dependence Charpy transition-temperature shift, ΔT41J, with fluence, such as one adopted in ASTM E900-15, are empirical or semi-empirical formulas based on parameters that characterize irradiation conditions (neutron flux and temperature), chemical composition steel (copper, nickel, phosphorus manganese), product type (plates, forgings, welds, so-called standard reference materials (SRMs)). (American Society for Testing Materials) E900-15 curve was obtained a combination physical phenomenological models free fitted using available surveillance data from power plants. These data, collected support ASTM’s E900 effort, open way an alternative, purely data-driven approach machine learning algorithms. In this study, PLOTTER database inform fit has been employed train validate number regression (multilinear, k-nearest neighbors, decision trees, vector machines, random forest, AdaBoost, gradient boosting, XGB, multi-layer perceptron). Optimal results were provided value R2 = 0.91 root mean squared error ≈10.5 °C test dataset. outperform prediction ability existing curves, including reducing uncertainty by ≈20%. addition, impurity-based permutation-based feature importance algorithms identify variables most influence ΔT41J nickel temperature, order), individual conditional expectation interaction plots estimate specific each features.