A Novel Approach to Jominy Profile Prediction Based on 1D Convolutional Neural Networks and Autoencoders that Supports Transfer Learning

This paper introduces a novel method for the estimation the Jominy profile of steel based on its composition, by combining autoencoders and 1-D Convolutional Neural Networks. The approach has two goals: firstly, to enhance the accuracy of hardenability prediction by exploiting the capability of the 1-D CNN to learn how the chemical composition of steel affects the shape of the Jominy profile; secondly, to use transfer learning to apply the knowledge gained from training on a specific dataset to new types of production with less available data or data with different characteristics as it often occurs in the industrial context. The proposed approach was tested on two industrial datasets aiming to assess the effectiveness of the methods on the two goals achieving satisfactory results.