Prediction Of Aging Hardness And Crystal Properties Of Aluminum Alloy Based On Machine Learning

In the past,most of the discovery and design of new materials were based on the trial and error of experiments,which resulted in a lot of pointless consumption and could not explain the relationship between structure,composition and properties of materials.But in the discovery of materials,it is very important to explore the relationship among structure,composition and properties.At present,the calculation of properties in the field of materials is mainly based on establishing professional mathematical model.Although calculation results are relatively accurate,it needs a lot of time and source.In recent years,with the rapid development of artificial intelligence,the combination of material database and machine learning has driven the progress of material informatics.Machine learning based on data driven are widely used in the field of materials.For example,in the field of prediction of material properties,many scholars have built machine learning models for property prediction,and then applied these models for material screening and new material mining.In this thesis,the properties of aluminum alloy are predicted.Choosing aluminum alloy as the research object has certain research significance,because aluminum alloy has the characteristics of low density,good mechanical properties,excellent conductivity,heat transfer and corrosion resistance,which is widely used in marine industry,chemical industry,aerospace,metal packaging,transportation and other fields.The main contents of this thesis include the following contents:(1)This part uses data of Al-Cu-Mg-X(X: Zn,Zr,etc.)alloys,including composition,aging conditions(time and temperature)and hardness to predict mechanical properties of aluminum alloys.In order to solve the problem of how to build an effective ensemble learning model,this part applies the automatic machine learning framework auto＿ml to the construction of ensemble learning.The experiment firstly selects the model based on automatic machine learning and uses extreme random tree,random forest,gradient boosting regression model and gradient boosting decision tree to predict hardness.The experiment also uses random search and grid search to optimize hyper parameters of models to obtain models with higher prediction accuracy.On the basis of these models,this part uses the combination strategy based on average method and learners to fuse the models.It is worth noting that the secondary learners in the learning strategy are screened by auto＿ml.This part analyzes and finds out the deficiency of auto＿ml and introduces attention mechanism to optimize the secondary learner.The experimental results verify the feasibility of building an ensemble learning model using auto＿ml and attention mechanisms,and show that machine learning can accurately predict aging hardness of Al-Cu-Mg-X.(2)This part predicts crystal properties of aluminum alloy,including average atomic volume,average atomic energy and atomic formation energy.In order to optimize the building process of the deep learning models,this part uses automatic machine learning model.This algorithm can automatically adjust the structure and hyper parameters of the neural network and finally provide an end-to-end solution.This part builds parallel neural network,deep neural network,residual network and attention-based network based on the model structure and hyper parameters provided by automatic machine learning.Neural network degradation occurs in the experiment and this part proposes Dense Based Net to solve "degradation phenomenon".Since the data amount of average atomic volume and average atomic energy is the same,a multi-attribute model is built to predict two attributes simultaneously.The experimental results verify the feasibility of optimal models based on automatic machine learning and Dense Based Net,and show that machine learning can accurately predict the average atomic volume,average atomic energy and atomic formation energy of aluminum alloy crystals.(3)This part uses best algorithms for predicting aging hardness and crystal properties of aluminum alloys to develop GUI,using python and Pyqt5 framework.When using GUI,if you want to predict the aging hardness of aluminum alloy,you need to input the composition,aging temperature and aging time.If you want to predict the crystal properties of aluminum alloys,you need input elements composition and pressure.