Applications of Artificial Neural Network Simulation for Prediction of Wear Rate and Coefficient of Friction Titanium Matrix Composites

Abstract The Artificial Neural Network (ANN) techniques were utilized to predict wear rate and CoF of the Ti-5Al-2.5Sn matrix reinforced with B4C particle manufactured by the powder metallurgy. TMCs and wear test samples were characterized by the Scanning Electron Microscope (SEM). Dry sliding wear narrative of the composites was estimated on a pin-on-disc machine at various loads of 20-60N, sliding velocity of 2-6m/s and sliding distance from 1000m-3000m. The wear rate of the composite was reduced by augmentation in weight fraction of boron carbide from 3-9%. The benefits of interfacial TMCs with B4C are: increase in strength, wear-resistance, and volume fraction. ANN was planned and utilizes a Levenburg-Marquardt program algorithm to reduce the mean squared error using a back-propagation technique. The input parameters are considered to include load, sliding velocity, and sliding distance. The experimental results of an ANN model and regression model are compared. ANN replicas have been urbanized to foreshow experimental rate of wear and CoF of TMCs and examined that ANN predictions have exceptional concord with deliberated values. Accordingly, the prediction of wear rate and CoF of TMCs using ANN in earlier actual manufacture will significantly save the manufacturing time, exertion, and expenditure.


Introduction
Titanium and its alloys have amazing physical, chemical, and mechanical characteristics such as low density, high corrosion resistance, and comparatively low elastic modulus making them more suitable for industrial, aerospace, marine, power plant, chemical, sport, and biomedical applications [1][2][3][4][5][6] .TMCs possess significant mechanical characteristics, such as high specific strength, thermal stability, erosion, corrosion, and creep resistance 7,8 .
Ti-5Al-2.5Sngives an impressive performance and is considered as the workhorse in the titanium and alloys product manufacturing engineering since its inception in the early 1950s 9,10 , the industry has evolved.Ti-5Al-2.5Snalloys have exceptional weldability and are usually utilized in airframe applications, particularly on weldability of the aerospace structural constituents close to engines where they can withstand super high temperatures 11,12 .
Ti-5Al-2.5Sn also acquires superior ductility and fracture toughness at tremendously low down temperatures whereas the prominent Ti-6Al-4V alloy It loses its ductility 9,13,14 .Ti-5Al-2.5Snalloy is low in cost, economics and cheaper than the well Ti-6Al-4V is an alloy.Ti-5Al-2.5Snalloy, which formed near to and under alpha, is preferred for use in high-temperature cryogenic applications.Due to its low heat conductivity and chemical reactivity, the efficiency of machining titanium alloys is negatively influenced by early tool wear when using standard machining techniques.Rapid tool wear is a problem that needs to be solved when titanium alloys cutting 13 .
A metal matrix composite has two categories based on the form of reinforcement.The first form of reinforcement is squashy reinforcement as of graphite and the next form is of tough particulates like SiC, Al 2 O 3 , TiC, TiO 2 , and others.The use of hard particulates minimizes loss of wear, mainly contrasting to the matrix main alloys 15,16 .The TMCs produced by the addition of B 4 C showed a greater tensile elongation.The tensile elongation of TMCs attained with B 4 C improved as the coarse reinforcements manufactured by B 4 C could be more easily cracked at the fracture surface 17,18 .
The hardness of the composites increased as compared to the base material (Ti-6Al-4V) and TMCs/10 B 4 C which is accredited to the incidence of the rigid ceramic stage.ANOVA was utilized to verify the legality of the urbanized model.The optimal constraints of WR and CoF were recognized and so was abridged wear rate 19,20 .Powder metallurgy is commonly used for the production of processed MMC materials with sufficient potential that can be used in many applications instead of obtaining a few admirable properties over other competing methods 21 .
An ANN-based replica is very critical and complex than another pragmatic modelling approach.It also necessitates significant acquaintance and knowledge to pertain efficiently and has been functioning predominantly for examining the link among input and output constraints of materials dispensation method [22][23][24][25] .ANN is an excellent pertinent technique due to its significant merits of superfluous precision, low cost, and shorter time.Recognizing every type of complex nonlinear interaction between input parameters and prediction of the output parameter is probable with ANN [26][27][28][29] .The complexity of ANN-based modeling is often higher than that of other empirical modeling techniques.In order to use it properly, it also requires a great deal of knowledge and expertise, and it has mostly been used to examine the relationship between the input and output parameters of materials processing processes 27 .
The ANN and regression analysis is used in optimization of the manufacturing process and proved that ANN showed improved results [30][31][32][33][34] .An ANN technique was pertained earlier for predicting the process parameter routine in several applications 26,27,29,31,[33][34][35] .Mondal et al. 30 used Flower Pollination Algorithm (FPA)-based optimization of drilling burr using regression analysis and ANN model and concluded that ANN produced better results.Ivan I Argatova (2019) investigated the usage of the neural network as an aspect of a necessary type nonlinear regression model guided by the spatial understanding of the experimental collection of data.It was for a pedestal on a scrupulous group of investigational data for the composites showed, the efficiency of the ANN-supported Archard -Kragelsky regression model 36 .Palanivel R (2019) examined and found the link among friction welding titanium joints by the relevance of an ANN technique.The joints were modelled by the relevance of the RSM 37 .Ashan et al. 32 studied that an ANN model was more precise as compared to RSM.Hassan et al. 31 observed the fatigue performance of composite shafts by ANN and discussed that it forecasted predictions of superior quality.Wang et al. 34 utilized an ANN model to depict the bending force deeds in a hot strip rolling process.Chen et al. 33 found an ANN model for the prediction of the surface roughness of titanium alloys while making an allowance for cutting forces and tool vibrations as input parameters.IlkerKucukoglu researched artificial neural network technique, an artificial intelligence tool commonly used for identification and it was tested and evaluated.A capable ANN with a two-layer feed-forward structure has been identified in this analysis.
The ANN rating precision for the study results were 88.9%and 95.0%for all calculation details.Details of the experiments carried out for the neural network have shown that ANN is a useful and appropriate tool for identifying faulty assembly operations by the considered wearable devices 38 .
The current research work executes the artificial neural network technique to they predict the wear rate of TMCs under different load conditions.Rahmath Ulla Baig (2020) developed and tested ANN predictions that were precise with an overall regression of 0.9993.The addition of reinforcement has augmented the wear resistance of aluminium MMC 39 .Sosimi et al. 40 , analyzed the ANN-LM model and the results showed the prediction precisely and the model provided an enhanced routine.Mutuk et al. 41 observed an ANN model for predicting the wear rate values of the composite that was created.This entails that the wear rate can be predicted by the structure of ANN is predicated with good accuracy by the experimental data.
Kannaiyan and Raghuvaran 42 analyzed the evaluation of wears of ANN modelling methods which were tested, and the performance of the model was found satisfactory.The ANNbased process was found to be the most suitable one in the calculation of performance response.Dinaharan et al. 43 used an ANN to foreshow the wear rate composites formed by friction stir processing (FSP).The DoE approach was utilized to direct the investigation.Pati 44 studied the ANN approach to predict used nonlinear regressive predictive equation showing the wear rate of the composites and showed improved superior wear resistance of the composite.Megahed et al. 45 examined that on augmentation of both the load and sliding distance it leads to an augment of the wear rate.The augment in the weight fraction of Al 2 O 3 diminishes the wear rate appreciably.The ANN approach is a successful model for predicting the wear rate of Al-Si/Al 2 O 3 composite.
Sweety Mahanta et al. 46 performed a wear test and considered three input parameters L, D, V; to investigat WR and CoF creation of the output parameters.The investigational resultant is examined by RSM at the same time as the wear performance extrapolative model is urbanized by using ANN.Pramod et al. 47 measured wear analysis of damage to parts of the system, and, to investigate this specific phenomenon, research was conducted by using the L27 orthogonal array method.For predicting the tribological behaviour of the Al7075-Al 2 O 3 composites, an ANN was developed using the Levenberg -Marquardt optimization model.Gampala Satyanarayana et al. 48reported a mathematical regression model and ANN models were urbanized to foreshow the conjectural rate of wear rate of the Al/red mud nanocomposites and found that ANN calculation showed exceptional conformity with premeditated values than the other models.Hanief and Wani 49 formed a model for the wear analysis.ANN's findings are significantly higher than the standard model.It is manifested from the analysis that together the developed and ANN models could be used to determine where wear rate could run through with precise accuracy.
The literature survey concludes the TMCs were utilized to enhance physical, chemical, thermal, tribological and mechanical properties like that hardness, density, thermal conductivity, wear, corrosion, abrasion, erosion, tensile, compression, impact, and flexure.In generally preparations of the composites were used stir casting, squeeze casting, die casting and powder metallurgy.The powder metallurgy is the one of the best technique to develop metal matrix composites and the ANN tool is the best predict the wear resistance.
The objective of the current investigation is to generate and utilize the ANN technique that could be associated in measuring the tribological behaviour of the Ti 5Al-2.5Sn/B 4 C composites under different load, velocity, and distance conditions.An influential ANN technique is described to foresee the wear and CoF deeds of TMCs using neural network software MATLAB R2018b for the assessment and ANN outcomes be presently thought about.

Materials
Titanium alloy (Ti 5Al-2.5Sn)matrix material with powder mean size of 35−40 μm was considered for this experiment.Ti 5Al-2.5Sn is a non-heat immoderation able alloy that can attain superior weldability with stability.It also contains good high-temperature steadiness, high strength, superior corrosion opposes and fine creep opposes.Creep mentions the occurrence of plastic strain for longer periods of time, which ensues at high temperatures.Ti 5Al-2.5Sn is generally utilized in aircraft, airframe, and cryogenic applications 50 .B 4 C with an indicate size of 20−40 μm was used as reinforcement in this work for requirement for dispersal strengthening.The yield strength of TMCs with 3, 6, and 9% of the addition of B 4 C is a further representation of the believed collective effects of next phase strengthening, grain refinement, and solution strengthening.Specification of the powder details used in this work for the powder compaction process is provided in Table 1.

Preparation of composite
The powder metallurgy method was used for embryonic TMCs reinforced with B 4 C made during the chemical vapor authentication process.Predominantly, the powders were weighed up in conformity with diverse compositions as Ti 5Al-2.5Snwith 3%, 6%, and 9% B 4 C.The weighed proportion was then intermingled to make use of a planetary ball mill for a total of 7 hours through an interlude of 45 minutes for each and every 1 hour for the progress of dispersal of B 4 C into the matrix.Scanning's of SEM of B 4 C subsequent to intermingling amalgamation of powders are shown in Figure 1.These micrographs carry an obvious vision for the dispersal of B 4 C in TMCs.The composite combination was then dense at 750 MPa using a hydraulic press to attain cylindrical specimens with 10 X 30 mm in diameter and height respectively.Subsequent to compacting, composite specimen as per standard measurements was sintered in the incidence of air at 950 °C for 1 hour 51 .

Microstructure of the composite
The titanium matrix composites specimens remained erudite for removing rubbish present on the surface.Particle distribution remains estimated at the support of optical microscopes.The powder metallurgy process remained inspected under the optical microscope to determine the reinforcement pattern of the powder metallurgy method.They remained grained using 100 grit silicon carbide paper tracked by 220 to 1,000 grades of emery paper ahead of optical surveillance the specimen were automatically polished and engraved by Keller's reagent to attain an improved disparity.The samples remained pictured on diverse magnifications to display the occurrence of reinforcements and its allocation of the metal matrix diverse elements/compounds which were presented in the graphite and boron carbide are difficult to distinguish by optical micrographs.
The microstructure of the Titanium matrix alloy formed by the powder metallurgy method is shown in Figure 2 The matrix particle distribution was uniform, according to microstructure analysis.The inclusion of 3%, 6%, and 9% boron carbide in this situation provides a benefit by minimizing the corrosion caused by the interaction with the titanium matrix.It was obvious that the 9% graphite interacted with the titanium matrix, causing corrosion, and that its higher volume in respect to the substrates might enable powder to escape from the matrix and onto the reinforcement.This resulted in the reinforcement fracturing and degrading, and the strength qualities deteriorating as a result.It might be because of the creation of boron carbide as a result of the interaction between carbon and boron, which results in an increase in strength and hardness.

Wear test experiment
Dry sliding wear tests for a different sample were carried out by using a pin-on-disc machine supplied by DUCOM as shown in Figure 4.In pin on disc wear test equipment, the disc material is grade H11 tool steel and followed ASTM G99 standard.The disc has a 500 HB hardness rating.H11 is a low carbon, high chromium, hot work tool steel with good toughness.By air quenching, this alloy has the capacity to undergo profound hardening.This alloy is suitable for severely stressed structural components in the aerospace sector because to its low carbon content and toughness.The Figure 5 shows the samples of the wear test specimen.The sample was whipped beside the counteract face of a revolving disc with a wear pathway diameter of 60 mm.The sample was loaded beside the disc during a deadweight loading system.The wear test for different specimens was carried below the normal loads of 20N, 40N, 60N, and a sliding velocity of 2, 4, and 6m/s.Wear tests were conducted for a total sliding distance of 3000 m beneath parallel surroundings as conversed below.The wear test parameters details are shown in Table 2.

Artificial Neural Network (ANN) structure
The predictions play a vital role in modern science and technology and in this ANN play an amazing role.ANN is a physically stirred computer program designed to replicate the system in which the human natural brain processes information.ANN provides a range of powerful new techniques for solving problems in materials science and engineering applications 52 .
The predictive capabilities of ANN-based on the training on experimental data and then validated by an assortment of proper neural network structure are important.The expansion and implementation of the network model were approved by means of neural network utensils from MATLAB R2018b software 53 .ANN architecture is one among the back-propagation method which has been working as per the flow chart shown in Figure 6.
In this investigation, ANN was shown to be working to predict the wear rate and CoF of TMCs reinforced with B 4 C.The ANN network is defined to have three layers such as input, hidden, and output layer.The input layer utilized 3 parameters (L, V, D), the ten hidden layers of neurons and two outputs (WR and CoF), were speckled according to the optimum output requirement.The function of the ANN model is alienated into two significant steps such as forward computing and rearward erudition.The input parameters, pertaining to the neurons of the first layer, are meager stimuli to the network in forwarding computing step 54 .The hidden neurons are associated to each other in the output layer so that change in the value of one node has an effect on the value of another 55 .Figure 7 shows urbanized ANN architecture among single based buried layer.In this urbanized architecture, ANN 3 stands for the input parameter layer (L, V, D), 10 means of neurons (hidden layer) and 2 is the output parameter, b and w mentioned are for the biases and weights respectively.In Figure 7, the transfer function associated with the buried layer is Tansig and is signified by a curvature.The transfer function pure lin is used in the output layer and is shown by the appearance of a straight line.The design of experiment (L27 array) for wear of the TMCs is presented in Table 3.

Analysis of wear debris and worm surfaces
The investigation analysis of the wear debris and the worm surface was carried out with the help of SEM.The morphological examination of the wear wreckage confirmed the results.SEM micrographs of titanium alloy, after dry sliding wear conditions are shown in Figures 8(a), 8(b) and 8(c).Figure 8(a) shows the wear surface clearly exhibiting the presence of delaminated layers, and deep grooves.The worn surface of the TMCs clearly reveals the incidence of wear tracks, delaminated layers, and deep grooves.The layer has stimulated the sliding surface clearly, as shown in Figure 8(a).The surfaces also appear to be smooth because of the boron reinforcement content and the incidence of deep grooves, which may have augmented the wear loss.The surface of the TMCs (Figure 8(b)) apparently reveals the incidence of delaminated layers and debris, observation shows the occurrence of debris and wrecked particles.The surface of the TMCs (Figure 8(c)) visibly exposes the prevalence of cracks, delamination, and deep grooves.Imprecise grooves and fine grazes were created on the worn surface.The wear mechanisms are disposition by the arrangement of the undulation, which is fashioned by the humanizing exploit of rigid abruptness on the counter disc and toughened worn debris.Augmentation in B 4 C would consequently diminish in wear.At immense velocities, the temperature over the sliding surface increases consequence of oxidization of material and thus material relocates to transpire between the pin and contradict face leading to the creation of layer.These layers smooth the progress in accomplishing greater tribological possessions more than immense velocities.

Best Validation performance of Network training for foreseeing of wear rate of TMCs
Assured and successfully prediction of the wear rate of TMCs and assorted intends of the ANN architecture are   endeavoured for this kind of examination, investigating different parameters, perceptibly, is imperative in promising the most excellent and consistent ANN structure which is associated to take care of the model and its associations to the full extent for sequence developed.In this research, the single based hidden layer of 3-10-2 (input-hidden-output) ANN structure was finally decided.
The ANN structure is designed with 3 credentials suggesting the number of input parameters in that order and the complementary '10' suggests the number of neurons (hidden layer), used as an ingredient of the buried layer, at the same time and the last '2' specifies the measure of a parameter in the output.The ANN was programmed to use a Levenburg-Marquardt algorithm.MATLAB R2018b software was wont to instruct the network for innumerable periods nearly allowing 9000 epochs to accomplish an adequate recital 27 .In this investigational process, the network was instructed for the best validation performance of WR and CoF for 2087 epochs and 3253 epochs.The recital of the ANN is finalized by MSE.Biases and weights are attuned iteratively throughout training by taking into consideration the MSE 56-58 .All experimental data is reduced by MSE. Figure 9 shows MSE is abridged for the purpose of the number of epochs for best validation.The most excellent recital is acquired for WR and CoF at 2087 epochs and 3253 epochs and displaying an MSE value of 2.3939e-06 and 1.1865e-08 respectively.This minimum value shows accuracy of good prediction.

Correlation coefficients plots of training, testing, and validation using ANN for Wear prediction
The precision of the planned and ANN model for wear rate can be calculated significantly different from Figure 10.It is obvious that every experimental value is secure in accordance with the predicted values of the planned and ANN models.Figure 10 provides proof of the precision of the ANN evaluation by contrasting curves of best fit for training, testing, validation, and rate of wear against experimental data.The subsequent correlation coefficients were attained for training, testing, validation, and all wear rate: 0.99287, 0.99986, 0.99814, and 0.99417, respectively.

Correlation coefficients plots of training, testing, validation using ANN for COF prediction
Most suitable curves were urbanized for training, testing, and validation of the ANN used.Figure 11 gives a hint of the accurateness of the ANN assessment by contrasting curves most suitable for training, testing, validation, and CoF adjacent to experimental data.Correlation coefficients of 0.99063, 0.9982, 0.99955, and 0.9935 were obtained for the training, testing, validation, and all sets of CoF respectively.

Comparison of experimental, ANN and regression model
The regression and ANN models are compared with experimental results.The percentage of error is determined among the experimental results.It is perceptible that taken together the models are reliable with the experimental results.MAD, MSE, RMSE, and MAPE values imitate the intention of the ANN model which is excellent and precise as contrasted to the regression model and the values are given in Table 4 for wear rate and CoF.The results observed from the investigations are illustrated in Table 5 and its graphical representation of wear and CoF is shown in Figures 12 and 13.
Regression analysis is applied to find the rapport amid sovereign erratic and reliant variables and to predict reliable variables.In this analysis, R is of the value of the correlation coefficient and it is utilized to explain the validity of urbanized regression models.The WR of homogeneous TMCs composites following equation is given, Regression Equation:   These values signify an extremely secure relationship between the experimental data and ANN model.

Conclusions
The conclusions of present investigation are as the follows: • The TMCs/B 4 C composite is prepared by the powder metallurgy route.

•
The pin-on-disc wear test machine has revealed that the wear resistance increase in the midst of augmentation B 4 C up to 3, 6, and 9%.

•
Regression and ANN models have been effectively urbanized and reveal exceptional precision and reliability.In this research both models were compared and it is observed that the accuracy of the ANN model is better than the regression model.

•
From the ANN study, the predicted data was related to investigational data and high degree of accuracy was observed and the R-value was in the assortment of 0.9 to 0.999.The error in prediction was calculated to be less than 4%.The use of ANN for prediction played a vital role.

•
The excellent conformity was realized between experimental and ANN values of the output of the wear parameter.

Figure 3
Figure 3 displays the Ti alloy and TMCs' XRD patterns.The suggested powder metallurgy composites' phase structure was ascertained using X-ray diffraction analysis.It can be seen from Figure 4 that the suggested composite's needed composition is clear.The existence of the matrix is guaranteed by the high intensity of the Ti peak, and the difference in intensity of the B 4 C phases distinguishes between variations in the weight % of composites.All composites shared the XRD phases that were discovered.These patterns demonstrate the incorporation of B 4 C particles into the Ti matrix, demonstrating the suitability of the powder metallurgy process for producing B 4 C reinforced TMCs.

Figure 6 .
Figure 6.Flow chart of ANN architecture.

Figure 7 .
Figure 7. Representation of single concealed layer in ANN architecture.

Figure 8 .
Figure 8. (a), 8(b) and 8(c) show the wear worn surfaces of the TMCs/B 4 C that were studied using scanning electron microscope (SEM) for the optimal parameters at (a) 20N (b) 40N (c) 60N.

Figure 9 .
Figure 9. Network training to (a) predict WR of TMCs (b) predict CoF of TMCs.

Figure 10 .
Figure 10.Line of best fit for a correlation coefficient between actual and predicted values for training, validation, testing and all wear rate data.

Figure 11 .
Figure 11.Correlation coefficients plots of training, testing, validation, and combination with all sets of COF using ANN for TMCs.

Figure 12 .
Figure 12.Graphical representation of Experimental, ANN and Regression model for wear rate and CoF.

Table 4 .
Comparison of Regression and ANN Models.