Prediction of Wear Resistance of In-situ Zinc Matrix Composites Reinforced by Silicon Phase Based on Neural Network

Silicon phase reinforced zinc-aluminium alloy composites were prepared by in-situ method. The effect of silicon content and external load on the wear resistance of in-situ zinc--aluminium composites was studied by BP artificial neural network. The test results show that the silicon phase, as a hard material, plays the role of supporting load and improves the wear resistance of the alloy. With the increase of Si content, the wear resistance of in-situ zinc--aluminium composites increased first and then decreased. When the amount of silicon added exceeds 2.5-3%, the wear resistance will be reduced, which is related to the silicon phase aggregation and the easy separation of large silicon phases from the matrix. With the increase of external load, the relative wear continues to increase, which is related to the increase of friction. The neural network platform can successfully predict the general trend of wear resistance with the change of silicon content and load.


Introduction
Zinc-aluminium alloy composite material has higher wear resistance and can replace some copper alloys to produce wear-resistant parts such as bearing shell.However, in the process of dry friction, zinc alloy creep occurs due to surface temperature rise, which greatly limits the use of zinc-aluminium alloy [1].In order to improve the wear resistance of zinc-aluminium alloy under dry friction conditions, inorganic materials reinforced zinc-base composites have been studied [2][3][4].For example, after materials such as SiC, Al 2 O 3 or graphite are added to zinc-aluminium alloy, the wear resistance of the composite materials formed is better than that of zinc-aluminium alloy.However, these materials also have their own shortcomings: such as SiC and Al 2 O 3 , although they help to improve the wear resistance of zinc-aluminium alloy, but because of their very high hardness, it is easy to scratch the parts in use.Zinc-aluminium-graphite composite is self-lubricating, but because of the low strength of graphite, this material is easy to wear.Silicon phase reinforced zinc matrix composites were prepared by in-situ method to avoid these shortcomings [5].In this paper, the influence of silicon content and positive load during wear on the wear resistance of zinc-matrix composites reinforced by silicon phase is predicted by using neural network platform.

Experimental Method
The main components of the test material are Al 23-27%, Si 1.5-4.5%,trace Cu and the rest Zn.The test raw materials are zinc ingot with 99.99% mass fraction of Zn, aluminium ingot with 99.99% mass fraction of Al, copper plate with 99.99% mass fraction of Cu and Al-Si master alloy ingot with 20% Si.
The resistance furnace crucible was used to melt the alloy.The melting temperature was controlled at 740-780℃.The molten alloy was degassed and kept warm for 5-10 minutes, then poured into a metal mold preheated to about 200℃.Wear test was carried out on ML-10 abrasive wear testing machine with disk rotation of 60rpm.The friction relative material is 1200 # sandpaper, and the positive pressure is in the range of 0.5-3kg.In the wear test, a new piece of sandpaper is used for each sample to ensure that the wear test conditions of each sample are consistent and comparable.The test is based on the weight loss during wear.The relative wear amount is used to reflect the wear performance of the material.The relative wear amount is the difference between the mass before and after wear divided by the friction surface area of the specimen.Before and after wear, the sample should be cleaned, dried, and weighed with 1/10,000 optical precision balance.Each sample was weighed three times and averaged.
After the experimental test results are obtained, the neural network can be used to make predictions.The structure of neural network model consists of input layer, hidden layer and output layer.Wear resistance test data can be used for training in predictive experiments.By adjusting the controllable parameters in the neural network platform, the error between the predicted value and the training value is less than the preset error.In this paper, the mean square error of system training is set at 10 -8 .After many trials, the neural network model platform needed for work is formed.

Experimental Results and Discussion
The microstructure of zinc-aluminium alloy can be obviously changed by adding a certain amount of Si element.Figure 1 shows the morphology of silicon phase in in-situ zinc matrix composite reinforced by silicon phase after friction test.It can be seen that silicon phase is firmly embedded in the matrix of zinc-aluminium alloy in the form of flowering shape or random block.When the amount of Si is relatively low, Si is mainly dissolved in α phase of matrix, and only a small amount of primary crystalline silicon phase appears in the matrix.However, when the addition amount of Si reaches 1.5-2.5%,due to the low solid solubility of Si in the α phase of the matrix structure, Si is precipitated in large bulk primary silicon in the alloy and distributed in the matrix.Because the hardness of the primary crystalline silicon phase is higher, and it can be firmly embedded in the matrix of the material, the hardness and wear resistance of the material are greatly improved.If the specimen is subjected to a heavy load in the process of friction and wear, the soft matrix will be dented, and the hard silicon phase will stand out from the surface of the matrix and bear the main friction load.Because the silicon phase not only has higher hardness, but also is not easy to deform and stick, it plays a good anti-wear and anti-friction role.Obviously, increasing the content of hard silicon phase within a certain range helps to improve the wear resistance of the material.In the network prediction, the silicon content (Si) and the loading force(F)are taken as input parameters, and the relative wear amount (W) measured by the friction and wear test is taken as the characteristic parameter of the wear degree, so as to predict the change rule of the wear resistance of the material with the help of the neural network platform.In the neural network prediction, the accuracy of the output target is approximated by repeated training and self-learning of the input and output parameters.In the neural network platform, there are a variety of training methods and various learning equations, such as trainbfg, traingd, learngdm, learngd and Learning.The training function of trainlm network was adopted.Wear test data can be used for training in predictive experiments.The neural network model with different number of hidden layer nodes (n=3, 4, 5, 6) is retrained for 1200 times, and the maximum determination coefficient R2 and regression coefficient R in 1200 cycles are obtained.They all have values in the range [0, 1].The closer the two values are to 1, the better the performance of the model.By adjusting the controllable parameters in the platform, the error between the predicted value and the training value is less than the preset error.After many trials, the neural network model platform needed for work is formed.
Figure 2 shows the comparison between the predicted relative wear amount and the real relative wear amount of in-situ zinc-matrix composite reinforced by silicon phase when the number of hidden layer nodes is 5.It can be seen directly that the error between the predicted results of the optimized neural network model and the real value is small, and the determination coefficient R2 is 0.9913.Figure 3 shows the fitting diagram of the neural network model, in which the regression coefficient R of the training group, test group, verification group and all data groups is above 0.95.Therefore, the optimized neural network model has a higher fitting accuracy.Figure 4 shows the relationship between the relative wear amount and the load and silicon content of in-situ zinc matrix composites reinforced by silicon phase.In this paper, the Si content in the range of 2.5-3.5% is called medium silicon content, and those below or above this range are called low silicon content or high silicon content respectively.As can be seen from Figure 4 (b), no matter in the medium silicon state, low silicon state or high silicon state, with the increase of load, the position of the relative wear amount curve of zinc-aluminium alloy composite material will increase, that is, the relative wear amount will increase.This is because the load increases, the pressure increases, the friction increases, and so the amount of wear increases.It can also be seen from Figure 4 (b) that the relative wear amount of zinc-aluminium alloy composites decreases with the increase of the Si content at the beginning.When the Si content is in the range of 2.5% to 3.5%, the relative wear amount is the lowest and the wear resistance is improved.When zinc-aluminium alloy composite is in high silicon state, the relative wear amount will increase.Figures 5 and 6 explain this effect of silicon on the relative wear amount.Figure 5 shows the SEM image of the wear surface of the in-situ zinc matrix composite containing medium silicon.The cut marks are long and shallow.When the zinc-aluminium alloy containing medium silicon (1.5-3.5%) is rubbed, the three-body wear particles of the alloy in the wear process are mainly the sand particles peeled off from the sandpaper.In the process of friction and wear, the force of sand on the surface of the friction pair can be divided into normal force and tangential force.The normal force can create an indentation on the specimen surface, the tangential force pushes the sand forward, and the sand being pushed is like a tool to plow the specimen surface.In fact, in most cases, the friction surface is not only subjected to the action of ploughing, but also subjected to shearing, cutting or combined action.Due to the high plasticity of the zinc-aluminum composite matrix, the groove is ploughed out when the sand grain slides, and the alloy material on both sides of the groove is piled up along the groove edge due to plastic deformation, but will not be cut off from the surface.The subsequent friction flattens the part of the metal that has just been pushed away by the plow.Such repeated plastic deformation, accumulation, and flattening eventually cause some metals to break away from the surface of the material and adhere to the sandpaper.At this time, deep regular grooves are formed on the surface of the alloy sample.When the amount of Si is more than 3.5% in the high silicon state, the wear particles of the wear process come not only from the sand particles that are peeled off the sandpaper, but also from the larger hard silicon phase that is peeled off the sample.When the load increases, the hard silicon phase that fall off causes great damage to the friction surface under the action of load pressure, forming large dents and deep grooves on the alloy surface, and the lower part of the dents contains a small amount silicon residue.As can be seen from figure 6, the wear surface of Si-phase reinforced in-situ zinc matrix composites with high silicon content is seriously worn, which is related to the separation of large silicon phases from the metal matrix surface.When the silicon content is too high, the silicon phase is easy to agglomerate to form large silicon phase, which seriously cuts the matrix.At the same time, it will also crack under the action of load, so it is easy to fall off.These large pieces of silicon peeled off from the sample have adverse effects on the strength, toughness and wear resistance of the material.

Conclusion
Through the neural network platform, the nonlinear relationship between input parameters (the silicon content, the load) and output parameters (the relative wear amount) is established.The wear resistance of zinc matrix composites increases with the increase of the Si content.Composites with moderate Si content show good wear resistance under heavy loads.However, in the high silicon state, the silicon phase is easy to agglomerate to form a larger silicon phase, which will seriously destroy the matrix.At the same time, it is easy to crack under the action of load and fall off from the matrix.These have adverse effects on the strength, toughness and wear resistance of the material.Not only can the neural network platform successfully predict the general trend of wear resistance with the change of the silicon content and the load, but also the neural network model can provide a data set that is impossible to obtain by experiment.

Figure 1 .
Figure 1.Microstructure of in-situ zinc matrix composites reinforced by silicon phase.

Figure 2 .
Figure 2. A comparison of the estimated relative wear amount and the actual value.

Figure 3 .
Figure 3.The fitting diagram of the neural network model.

Figure 4 .
Figure 4. Relationship between the relative wear amount and the load and silicon content of in-situ zinc matrix composites reinforced by silicon phase.(a) Three-dimensional surface diagram; (b) Two-dimensional curve diagram.

Figure 5 .
Figure 5. SEM image of wear surface of in-situ zinc matrix composites reinforced by silicon phase containing medium silicon.

Figure 6 .
Figure 6.SEM of silicon phase enhanced wear surface of in-situ zinc matrix composites with high silicon content.