Design And Preparation Of High Performance Tread Materials:Machine Learning And Experimental Study

The rapid development of the automobile industry puts forward higher requirements for tread rubber materials,that is,a balanced "magic triangle" and good mechanical properties.Each performance is interrelated and restricts each other,and it is difficult to be improved at the same time.How to achieve the ideal balance between each performance is the primary problem faced by designers.The design of tread rubber material involves many factors such as matrix,filling system,anti-aging system,vulcanization system and processing technology,which is a very complicated process.If the traditional univariate analysis method is used,a large number of experiments need to be set up to realize the optimization of the design,which is time-consuming and costly.Besides,it is easy to fall into a local optimal solution and difficult to fully realize the potential of the material.As an emerging artificial intelligence method,machine learning methods have been successfully used in different disciplines such as bioengineering,food science,energy systems,and materials science.Compared with traditional experiment-driven material development methods,machine learning,as a data-driven method,allows multiple variables to be studied at the same time and provides a global optimal solution,thereby greatly accelerating the material design process.On the other hand,the relative importance of the respective variables and the possible interaction between the variables can also be obtained from the machine learning model,helping designers to deepen their understanding of materials.This paper attempts to combine machine learning methods with experimental methods to accelerate the design and preparation of high-performance tread materials,and focuses on the relatively complex wear resistance,which mainly includes the following four parts:The first part(Chapter 2)of this paper studies the effect of filler content on the properties of NR composites in the carbon nanotube,carbon black and white carbon black ternary filling system.With the help of machine learning methods,fast prediction of rolling resistance,wet-skid resistance,tensile strength and hardness of tread rubber materials was achieved,and "low roll resistance scheme" and "high wet-skid resistance scheme" were given.Twentyseven different NR composites were obtained as sample points by orthogonal design with three factors and three levels.The most suitable modeling algorithm was determined by "trial and error" method,which was multivariate linear regression(MLR)algorithm.At this time,the prediction errors of all target performance were less than 5%.The established machine learning model was used to analyze the relationship between the amount of each filler and the target performance.It is found that the total amount of filler is the most important factor affecting the wet-skid resistance,rolling resistance and tensile strength of NR tread rubber.The increase of silica dosage is beneficial to balance the dynamic performance of the rubber,and the increase of CNTs dosage can significantly improve the hardness and tensile strength of the NR composites.Finally,according to the model,a low rolling resistance formula "1.25 phr CNTs,31 phr silica,43 phr total dosage equivalent" and a high wet-skid resistance formula“1.2 phr CNTs,31 phr silica,48 phr total dosage equivalent" were designed and verified by experiments.On the basis of the first part,the second part(Chapter III)of the paper expands the independent variables to matrix,anti-aging system and filling system,and adds the most complex abrasion resistance in the "Magic Triangle"to the target performance.Rapid prediction and effective balance of the wear resistance,rolling resistance,wet-skid resistance and mechanical properties of NR composites were achieved by machine learning method.The representative sample points were obtained by uniform design and the most appropriate algorithm for each target performance was selected for modeling.Visualizing the prediction model,it is found that the interaction between type of NR and type of CB has the greatest influence on Akron abrasion and tensile strength,which reflects the affinity difference between different types of NR and carbon black.For wet-skid resistance and rolling resistance,the most important influencing factors are the interaction between type of NR and antioxidant 4020 dosage(mainly affecting cross-linking density)and type of CB.A series of optimal design schemes that take into account "magic triangle" and tensile strength were derived according to the machine learning model.Taking two of these schemes,OD-1(synthetic NR,2 phr antioxidant 4020,N330)and OD-2(RSS-1#,2.25 phr antioxidant 4020,N330)for experimental verification,the results showed that both OD-1 and OD-2 could effectively balance each target performance,among which OD-1 exhibited the better abrasion resistance and OD-2 possessed the lower 60? tan ?.The third part of the paper(Chapter 4)combines data mining methods and machine learning methods to explore the impact of mechanical properties on the abrasion resistance of tread rubber materials.The sample set was obtained by sorting out the previous experimental results and collecting literature data,and using the gradient boosting decision tree(GBDT)algorithm to gradually eliminate unimportant independent variables.It was found that when stress at 300%,tensile strength,tear strength and elongation at break were used are used as independent variables,the model has the smallest prediction error.The machine learning analysis results showed that for abrasion resistance,stress at 300%was the most important independent variable,and Akron abrasion decreased with stress at 300%and first decreased and then increased with tensile strength.The effect of elongation at break and tear strength on abrasion resistance is relatively small.Decomposition by principal component analysis resulted in a set of new independent variables which was independent of each other and Bayesian optimization was used to optimize the designs.It was found that all tread rubber with high abrasion resistance had high stress at 300%(>12.5 MPa),low tensile strength(<26 MPa)and elongation at break(<465%).We infer that this is because high tensile stress helps reduce the friction of the rubber,low tensile strength and elongation at break mean that the size of the ridge in the wear pattern is small,thus the NR composites have a low crack growth rate.The fourth part(Chapter 5)of the paper simulates the high load and high speed on the tire during the actual driving process through the homemade rubber abrasion device,and explores the effect of load,speed and carbon black(CB)dosage on the abrasion resistance of NR composites.The results show that the abrasion rate of rubber increases with the increase of load,but the relative abrasion rate between samples with different CB dosage would reverse with the change of load.The effect of rotation speed on the abrasion rate is less than that of load.When the rotation speed increases from 600 rpm to 800 rpm the abrasion rate increases.However,the abrasion rate does not change significantly when the rotation speed is further increased.The abrasion rate of samples with 40 or 45 phr CB is similar,but when 50 phr CB is adopted,the wear resistance of sample is significantly improved.Observation on the abraded surface and the wear debris revealed that a sticky degradation layer appeared on the rubber surface,and micrometer sized fine-grained wear debris and largesized crimped wear debris were simultaneously included in the wear debris,indicating that the abrasion resistance of NR composites mainly depends on the dynamic cycle two processes of degradation of surface layer and stripping of degradation layer.Particle wear mainly occurs when the former is dominant,while roll-up wear becomes the dominant wear mechanism when the latter is dominant.The effect of load,rotation speed on abrasion resistance is essentially achieved by affecting these two processes.