Jominy hardness test is the most representative method for evaluating the hardenability of steel, and is a very important parameter for predicting the microstructure, strength and thermal distortion of steel. However, in order to perform the Jominy test, it is possible to prepare the specimen prepared and the evaluation facilities manufactured according to the ASTM A255 standard. In order to overcome the temporal, cost and equipment limitations in conducting this evaluation, accurate prediction is required to replace the actual test.
Global standards such as ASTM A255 and SEP1664 also provide prediction equations, but the prediction accuracy is low, which makes a big difference from the actual data. In addition, the prediction accuracy of previous stuides is not high because the use of theoretical method that do not use the actual Jominy data or the use of only a few of the alloying elements that affect the Jominy hardness.
In this study to overcome these difficulties, we utilized the all 13 alloying elements of C, Si, Mn, Cu, Ni, Cr, Mo, V, B, Al, Ti, N, Nb and total of 1,197 actual data of 8 steel grades were used. In addition, we tried th implement a model for predicting the Jominy hardness prediction with high reliability by using neural network model which can learn best from the data and generate the best approximation fuction to derive the best approximation.
As a result of this study, we found that the mean square error of the neural network model is 0.9963, which is very high reliability. Also this neural network model has the error within 1HRC compare to actual data.
In the actual industrial field, the range of Jominy hardness narrowing management is 3HRC level, and our Jominy hardness prediction model constructed through the study can obtain the reliability the can be used in the actual industrial field.