Thermal treatment of rails having a stable hardness distribution is enabled. The hardness of a rail, obtained by forced-cooling of a rail in the austenite region temperature or higher in a cooling facility (7) is predicted. A plurality of sets of learning data made up of a cooling conditions dataset and output data of hardness was acquired, using a model performing computation with the cooling conditions dataset having at least surface temperature of the rail before starting cooling and the operating conditions of the cooling facility (7) as input data, and the internal hardness of the rail after forced cooling as output data. A hardness prediction model is generated in advance by machine learning using the obtained plurality of sets of learning data, in which the cooling conditions dataset is input data at least, and information relating to internal hardness of the rail after forced cooling is output data. Hardness of the rail is predicted from the internal hardness of the rail as to one set of cooling conditions dataset set as cooling conditions of forced cooling, using the hardness prediction model.
展开▼