Research On Vermicularity Of Ductile Iron Based On Machine Learning

Since the mid-20th century, ductile iron has been widely concerned since it emerged. As a industrial production, it is widely used for it's well structural material properties. Due to the cast iron is mainly influenced by the percent of vermiculation of the spheroidal graphite cast iron in mechanics performance, it has became an important research field finding a fast and efficient method to test the percent of vermiculation's grade. In the past, artificial determination and image pattern recognition are the chief methods, which need to cut, polish and amplify the photograph of the spheroidal graphite cast iron when it cools down. This method costs a large number of manpower and financial capacity, and it is difficult to realize the large-scale production. According to the thermal analysis method proposed by the material science, this paper applies machine learning to learn the attribute information of the spheroidal graphite cast iron to predeterminate the grade of vermiculation of the spheroidal graphite cast iron.Firstly, this paper puts forward a self-adaption filter algorithm based on the minimum mean square error and high frequency signal, which is in the light of the multi-resolution analysis theory and the traditional self-adaption filter ideology. The measured temperature data has been adaptive denoising from the temperature logger by this method.Secondly, this paper builts a classifier for the data after filtering by support vector machine(SVM) whose input information is the attribute information of the temperature of the the spheroidal graphite cast iron and the output is the grade of the the percent of vermiculation of the spheroidal graphite cast iron. The basic input data sequence is the corresponding feature information abstracted from the physical feature.The target data sequence is the level of creep rate of iron. When building a classifier, due to it is difficult to determine the penalty factor and deformability factor in SVM, this paper applies the immune memory clonal selection algorithm which belongs to the evolutionary algorithms as the parameters'optimization algorithm of SVM. The algorithm comes from the simulation of the immune evolution principle and clonal selection theory in the biological immunology, it not only has the ability of global optimization and remains the local search function, but also has immune memory ability and can convergent to the global optimal solution.This experimental data is applied by professor Jinhai Liu who is the leader of the team researching on the spheroidal graphite cast iron in the material science and engineer school of Hebei university of technology. The experimental results demonstrate the self-adaption filter algorithm based on the minimum mean square error and high frequency signal has a better self-adaption ability, and the support vector machine which is optimized by the immune memory clonal selection algorithm shows the superiority on predicting the classification accuracy.