PROBLEM TO BE SOLVED: To heat-treat a rail having a stable hardness distribution. SOLUTION: The hardness of a rail after forced cooling of a rail having an austenite temperature or higher by a cooling facility 7 is predicted. Cooling conditions using a model that uses a cooling condition data set that has at least the surface temperature of the rail before the start of cooling and the operating conditions of the cooling equipment 7 as input data and calculates the internal hardness of the rail after forced cooling as output data. Acquire a plurality of sets of training data consisting of a data set and output data of hardness. By machine learning using a plurality of sets of acquired learning data, a hardness prediction model is generated in advance using a cooling condition data set as at least input data and information on the hardness inside the rail after forced cooling as output data. A hardness prediction model is used to predict the hardness of the rail from the hardness inside the rail for a set of cooling condition datasets set as cooling conditions for forced cooling. [Selection diagram] Fig. 1
展开▼