Predicting and analyzing of tribo-performance of NAO friction composites using Taguchi method


 The present study is aimed to investigate the dry sliding behavior of phenolic friction brake pad materials for industrial applications. Low metallic phenolic friction composites with addition copper-graphite (Cu-C) particles produced by traditional powder metallurgy methods. The friction test is carried out by pin-on-disc configuration on universal tribometer MMW-1 with hardened steel as a counterface material. The plan of experiments conducted by Taguchi’s L27 orthogonal array on MINITAB 19.1.1 software using 3-level design model. Analysis of variance (ANOVA) was performed for predicting and analyzing the effect of design parameters like contact pressure (1.9, 5.75 and 9,6 MPa), sliding velocity (0.64, 1.57 and 2.5 m/s) and filler content (5, 10 and 15%) to tribological properties. Results of modeling and optimization of composites has showed that contact pressure has the greatest impact on the friction process following sliding distance and filler content. On the other hand, the most influential factor for the wear process was the sliding velocity, following contact pressure and finally filler content. It has also been determined that 5–10 Wt.% Cu-C filler content has an effective impact on tribological properties. The friction surface examination of the composites using a scanning electron microscope (SEM) revealed that Cu-C content has a significant effect on improving heat resistance properties.


Introduction
One of the most important parts of brake mechanisms is the brake pad or lining materials. The brake pad function is to either slow down or completely stop the movement by creating a frictional resistance against the rotating element of the device during the braking process [1]. The reliable operation and productivity of automobiles and industrial applications are determining by work of the brake systems [2].
The main requirement for friction materials in tribotechnical system is to ensure the rapid conversion of energy into heat and to meet complex requirements such as friction and keeping of tribological properties stability during multiple braking processes [3]. Stable friction characteristics, durability, noise and vibration minimization are considered to be the main norms when designing materials included in brake systems [4]. One of the main demands for brake pad materials is high stability of the friction coe cient in wide ranges of working temperature [5]. The high wear resistance during friction contact and keep its initial physical and mechanical properties, as much as possible, at rising temperatures allow these materials to work in heavily loaded tribotechnical systems [6]. The wear of materials used in friction units became inevitable. For this reason and to meet new industrial requirements there was always a need to study new materials.
Composite materials have become a common material for various industries elds due to its uniqe properties [7]. The brake friction materials are the most complicated structured among composite materials, consisting of multiphase constituent elements such as binders, llers, modi ers and bers [8].
Friction composites are mainly consisting of metallic and non-metallic components. Choosing the right composition is an important factor in the design of modern friction materials for improving reliability and durability. To meet the above discussions is highly depends on the selection of suitable ingredients. The manufacturing and synthesis of these materials cannot be performed by methods generally accepted for other traditional materials, as in the case of melting, the mixing and integration of the ingredients must be provided with de ned parameters [9]. Although polymers have not proved themselves in the industry enough as friction material, recent studies by researchers have shown that it is possibility to obtain completely new properties using nano-scale particles and various functional llers [10,11].
One of the most promising materials in terms of economically and environmentally that can meet the requirements of the tribotechnical system is polymer-based friction composite materials [12]. Due to these properties, polymer-based powder composite materials containing metal elements can be considered an ideal candidate to work in extreme operating conditions. Metal elements not only increase the required mechanical properties, but also have the ability to improve the friction properties [13].
The role of copper and graphite as solid lubricants in brake friction materials already studied by different researchers. An increase in graphite content reduces the coe cient of friction but depending on the shape and size, it also increases the wear resistance [14]. Copper increases thermal conductivity, improves sliding properties for different braking regimes [10]. As the nanocrystals on the surface recrystallize at high temperatures, a new brittle friction layer is forming, which acts as a solid lubricant, like graphite.
Although the new friction layer reduces the coe cient of friction, it eventually mixes with other material particles released from the contact area of the disc and brake pad, stabilizing the tribological properties [15]. However, the copper-graphite (Cu-C) effect is not studied as well and it has a particular interest in terms of studying surface characteristics.
The characteristics of the contact surfaces are very important for friction materials. The surface roughness of the contacting materials is not only related to the nature of the mechanical properties of the material, but also to the friction and wear process itself. Along with depending on the type of rubbing pair materials, as a result of interaction effects such as adhesion, surface fatigue, abrasive wear and tribochemical reactions in the contact surface also in uence friction and wear mechanisms of tribological materials [16]. It is known that the process of friction of metallic materials is often characterized by plastic deformation and oxidation wear [17]. Besides, at elevated temperatures friction layer formation, degradation and decomposition of non-asbestos organic (NAO) also in uences tribological properties [18,19]. However, the simultaneous occurrence of several wear mechanisms makes it di cult to assess the friction process and requires a systematic approach.
In order to study these con icting properties design of experiments (DOE) can be helpful. Taguchi approach is a widely used powerful statistical technique for multifactoral experiments (Taguchi and Konishi, 1987). Unlike traditional DOE methods, Taguchi technique uses "smaller-is-better", "larger-thebetter" or "nominal-the-better" criteria for achieving the best quality performance and allows evaluate the robustness of the process with numerous approaches [20]. The technique may be more valuable when selected control parameters are discrete and substantiated [21].
In the present study, produced friction composites with Cu-C particles subjected to the analyzing impact of the control parameter and the interactions on friction coe cient and wear rate behaviours. The design parameters selected for experiments were determined based on the commercial materials used in the industry.

Fabrication of composites
The friction brake pad composites were fabricated using traditional powder metallurgy methods. Three specimens containing eleven ingredients were produced by ball milling, mechanical mixing in dry conditions (for 16 hours with a speed of 60 cycles/min on a horizontal drum mixer), pre-forming at 10MPa and hot pressing for 20 min. at 160°C under 25.5MPa pressure. The glycerin used as a plasticizer for improving powder processing. The manufacturing process is nished with the post-curing procedure in a mu e furnace for 5 hours in 140°C. The information about produced specimens formulation is given in Table 1.  [22]. The base composition (67% Wt.%) was kept constant and Cu-C content was used in three proportions (5, 10 and 15 Wt.%). The remaining composition was compensated with barite ller. Specimens were fabricated in cylindrical pin form in size of Ø 4.7 x 12.8 mm (Fig. 1a). Brake pad composites codes called CG-5, 10 and 15 represent specimens containing 5, 10 and 15 Wt.% Cu-C particles respectively.

Plan of experiments
In this study, MINITAB 19.1.1 software used to analysis and correlate responses between different variances. L 27 (3 3 ) orthogonal array was chosen to design the planning matrix according to the Taguchi technique. Analysis of variance (ANOVA) was performed with 95% con dence level. Three control factors were used to study the friction and wear performance of the composite materials: contact pressure (A), sliding velocity (B), and ller content (C). Each control level combination was repeated three times and the data evaluating was based on obtained average test results. Information on factors and selected levels is presented in Table 2. Friction coeffcient S/N ratio was calculated with setting goal of the experiment to "larger-the-better" and wear rate S/N ratio was calculated using "smaller-is-better" criteria. S/N ratios friction and wear was determined using logarithmic loss function (1) and (2) respectively.
where y is the response for the given factor level, n is the number of experiments.

Friction and wear test
The dry sliding friction tests were performed on a vertical MMW-1 tribometer with "pin-on-disc" con guration at room temperature (25°C). The principle of friction tests is based on the simultaneous clockwise rotating of three cylindrical pin samples on a stationary steel disc (Fig. 1). The quenched steel (hardness: 44-46HRC, surface roughness: ~0.2µm) with an outer diameter of 31.7 mm, an inner diameter of 16 mm and a thickness of 10 mm was used as the counter-face material. Load applied to the disc by the movement of the rotor which is installed at the bottom part of the machine (Fig. 2.).
Prior to each test, the surface of the disc and specimens was cleaned with ethyl alcohol to prevent lubrication. The surface of the specimens was grounded on a polishing machine with 100, 1000 and 2000-grid SiC papers. Worn samples were weighed on an electronic balance (Mettler Toledo) with an accuracy of ± 0.1 mg. The sliding distance was 1.5996 km for all tests.
The following formula was used to calculate the wear rates using weight loss measurements: where m 1 and m 2 are the mass before and after wear respectively (gram), τ is the duration of the experiment (hours), and J is the wear rate (g/hour).

Signal-to-noise (S/N) calculation
Taguchi generated design with three factors namely contact pressure (A), sliding velocity (B), and ller content (C) at three levels is used for conducting experiments. Table 3 exhibits the experimental results of tribotechnical tests based on the selected planning matrix and calculated S/N ratio values in accordance with the obtained data. Based on the S/N result, it is possible to determine which is the most in uential factor in increasing the coe cient of friction. The average S/N values for each factor level are given in Tables 4 and 5. As can be seen from the results, the contact pressure is the most signi cant factor in uencing the change in the coe cient of friction followed by sliding velocity and ller content, respectively. In contrast to the coe cient of friction, the most signi cant factor in uencing the wear rate was the sliding velocity followed by contact pressure and ller content, respectively (Table 5).   Figure 3a and b show the control factor dependence graphs of S/N ratios values for the friction coe cient and wear rate, respectively. Delta (∆) is the average difference between the maximum and minimum response values for each factor. Taguchi suggests that response with a large S/N ratio indicates its importance on the process. As can be seen from Figure 3, graphs whose starting and ending points are farther from the horizontal line show that they have a greater effect. The graphs demonstrate that the most important factor in uencing the increase in the coe cient of friction is the contact pressure. The most important factor in uencing the wear rate is the sliding velocity. The obtained curves re ect the given values in Tables 4 and 5 and demonstrate the most signi cant factor effects. De ned impact ranges for both response values helped to determine the optimal regime parameters.

Statistical analysis (ANOVA)
Processing of experimental test results by ANOVA helped to identify the factors affecting the coe cient of friction (Table 6). As can be seen from Table 6, the P-values of the friction coe cient is zero for the main factors, which shows their statistical signi cance, but the α-value of each of the interaction factor conditions was higher than 0.05. DF-degrees of freedom, Seq SS-sequential sum of squares, Adj SS-adjusted sum of squares, Ppercentage of contribution Signi cance levels below α = 0.05 indicate their signi cance is low. Thus, the higher the value of F, the higher the effect on the process, and according to this principle, the impact percentage is determined in the following order. Contact pressure is the most signi cant parameter with the highest percentage contribution (55.75%), following sliding velocity (28.48%), and nally ller content (2.92%). The most important of the interaction factors was the combination of ller content -ller content (C*C), which was 4.44%.
The percentage of factors in uencing the wear rate is determined in the following order ( Table 7). The most dominant factor is sliding velocity (52.24%), following contact pressure (24.53%) and ller content (16.96%). The most important of the interaction factor conditions was the combination of contact pressure -sliding velocity (A*B) at 2.63%.
The Pareto chart can be useful to better understand the effect of interactions. Based on the results obtained, a 3D interpretation of the dependence of the friction coe cient and wear rate on the regime parameters is given in Figs. 5 and 6, respectively. As can be seen from Pareto charts, factor C ( ller content) is the least important among the main factors. Therefore, in order to understand better the effect of process parameters on changes of response values, material factor (C) is not taken into account in 3D plots. As can be seen from Fig. 5, the increase in sliding velocity reduces the frictional characteristics of the material. This is due to the increased intensity of wear rate at high sliding velocity (Fig. 6). Changes in the contact surface as a result of wear lead to a decrease in the coe cient of friction. This dependence can be better seen in Figs. 9 and 10.
During sliding, material removal or gain occurs due to interactions between two solid surfaces. In the initial phase of sliding distance meeting of contact surfaces with each other, mainly resulting with plastic deformation. The occurrence of plastic deformation leading to the transformation of mechanical energy into frictional heat. Under frictional force in the surface of heterogeneous structured brake pads materials primary and secondary contact plateaus is forming [23]. In closer a look, if we approach it on a nanoscale, it is known that friction surface asperities are very rough. During the rst stage of friction under big contact stress, harder asperities of the disc are breaks off softer brake pad asperities and form new real contact areas. Although the increase in the real contact areas has a positive effect on the friction characteristics, the structure of the contact surfaces changes again as a result of wear [24,31]. For this reason, the highest coe cient of friction was at a contact pressure of 9.6 MPa and a sliding speed of 0.64 m/s (0.445). The coe cient of friction began to decrease more intensively after ~ 1.5 m/s. Kragelski, I. V. (1962) attributed the decrease in the coe cient of friction with the increase in the sliding velocity to the weakening of the formation of a strong bond between the two friction surface due to generated oxide layer. This is a phenomenon related to the mechanical properties of materials and may have different effects to friction depending on the properties of the individual components. Simultaneously, increasing the sliding velocity also increases the frictional temperature at pairing surfaces. Generated heat during friction process increases interface temperature, which is resulting in decrease in friction coe cient [25]. As the temperature rises, changes in the tribological behaviour of contact materials occur. A great amount of dissipated energy ow is absorbing by the brake pad and affecting material properties. The high temperature generated on the contact surface is distributed from the interface layer into the material. The distribution of temperature within a material depends on the thermo-physical characteristics of that material [26]. At the end of the friction tests, two models of thermal effects were identi ed in the samples, especially in high contact and sliding velocity conditions. A graphical description of observed heat effects based on the cross-section of the specimens is given in Fig. 7a.
The rst model (Fig. 7a) describes the completely burned contact surface of the sample with low ller content . This can be understood as the low thermal conductivity of the CG-5 sample, due low content of copper-graphite. In samples CG-10 and 15, no signi cant difference was found in the contact surface and the other parts. This effect shows that heat is rapidly distributed and transferred within these materials, which also indicates that possibly increase of thermal conductivity by Cu-C particles [27]. The difference between CG-10 and 15 was only noticeable in images taken with scanning electron microscope (SEM) (Fig. 8). Figure 8 shows the SEM observations of worn surfaces of CG-5, 10 and 15.
These observations cover 9.6MPa pressure and 2.5 m/s sliding velocity regimes (except Fig. 8c, it shows worn surface in 9.6MPa pressure and 0.64 m/s sliding velocity combination). Unlike the CG-15 sample, the number of cracks on the surface of the CG-5 was larger and deeper (Fig. 8b). Due to the low mechanical properties of CG-15, the wear was high and its complex wear mechanism re ected in the surface morphology. However, in the CG-15 sample, an increase in the sliding velocity was observed with microcracks ( Fig. 8d). It is also possible to see the deepening of micro-voids, which indicates the intensi cation of the participation of metal particles such as aluminum oxide and silicon dioxide in the abrasive wear process with the degradation of organic components. Signs of plastic deformation are also visible in various areas of the friction surface which indicates that high Cu-C content affected the surface structure ( Fig. 8c and d). Oxidation of Cu as a result of local temperature increase can in uence wear mechanism by forming tribo-layer [28]. Oxidation in uences were not observed in CG-15 as much as CG-5 and CG-10 specimens due to low copper content. Generated heat during friction affects the oxidation, thermal strength and plasticity of the material. This leads to changes in surface following complicated physical-chemical reactions (diffusion and adsorption of atoms or molecules on the friction surface) [29]. The change in the interaction between the surfaces as a result of structural changes is re ected in the dry frictional behaviour of brake pad composites. Graphs show that, the coe cient of friction does not decrease in high contact pressure regimes, on the contrary, an increase is observed ( Figs. 5 and 9). This difference is especially noticeable in setting with low sliding velocity and high contact pressure settings. The reduction of friction coe cient can be explained decrease in the contact time at the asperities due increase of speed [30]. Due reduce of the contact time possibility of forming enough junctions for friction by adhesion is reducing. However, this process may be different depending on the properties of the brake sample surface and friction layer characteristics. When the pressure is high, the heating of the contact surface can cause to formation of a friction layer of solid lubricants, which leads to the stabilization of the coe cient of friction [31]. Depending on termo-mechanical properties of material sliding velocity is leading generation of non-uniform pressures with high local contant stresses in relative motion [32]. These effects are impacting surface morphology and cause non-stability of friction coe cient.
Mentioned effects on friction surface can be seen clearly in Fig. 8a and c. Since the wear of the friction layer is not high at low sliding velocity, this prevents the friction coe cient from decreasing over a period of time.
The lowest friction coe cient was observed at a contact pressure of 1.9 MPa and a sliding velocity of 2.50 m/s (0.367) (Fig. 9). Wt.% 15 of ller content in this combination is one of the reasons that reduces friction. This can be attributed to the formation of a copper-graphite lubricating lm at the sliding surface.
The graphite layer signi cantly reduces the contact of the specimen with the rotating disc surface. The formed lm thickens due to the worn particles of the sample, further reducing the coe cient of friction.
An increase in wear rates was observed in a typical sequence -the increase in contact pressure and sliding velocity resulted in an increase in wear rate. The highest wear rate responses were observed at 9.6 MPa contact pressure and 2.50 m/s sliding velocity conditions (0.068 g/hour). The lowest was observed at a contact pressure of 1.9 MPa and a sliding speed of 0.64 m/s (0.003 g/hour) (Figs. 6 and 10). It should also be noted that, contour plots implies that ller content up to 10Wt.% is improving the wear resistance of samples. But 15Wt.% ller is not appropriate in terms of tribological properties. In CG-15 specimens wear rate values were high in comparison. When the contact pressure is high the graphite particles are squeezed out from the contact area and this leads to an increase in the wear rate [33]. This phenomenon causes a change in the structure of the friction surface. Surface roughness parameters are one main factor which is determining the contact behavior of rubbing pair materials. As discussed above frictional heat formed a surface roughness by changing the surface morphology in the CG-15 brake pad sample (Fig. 8c). During dynamic asperity -asperity contacting, depending on initial surface roughness elastic or a plastic deformation may occur [34]. These tribo-mechanical in uences are impacting friction stability and wear of mating brake materials. Thus, surface roughness can also be considered as one of the main reasons for high wear rates in CG-15 specimens.

Obtaining linear regression equation & cheking adequacy of model
The linear regression equation was applied to establish a correlation between the important factor conditions obtained with based to the ANOVA results, and the following equations (4) and (5)  Information about the coe cients related to the models can be found in Tables 8 and 9.  (3) it is observed that contact pressure plays a major role on the friction coe cient. A number of factor combinations (B*B, C*C, A*B and B*C) were negatively related to the coe cient of friction. This is due to the processes occurring on the friction surface which is discussed above. Eq. (4) showed that wear rate highly in uenced by sliding velocity and contact pressure. And thus, these factors increase the wear rate of brake pad specimens. Almost all of the selected factors and their combinations contribute to the increase of wear rates. Obtained regression equations can be used to predict tribological behaviours of the brake pad composites.
Residuals vs. Fitted Value graphs showing the adequacy of the model obtained by linear regression equations (4) and (5) are given in Fig. 11. The graph is re ecting the distribution of negative residual values below the zero line, and the positive results above that zero line in the vertical direction along the ordinate axis. Residual values are calculated by the following formula [35]: where is residual, is experimental response value and is predicted response value.
According to the accepted distribution patterns, obtained results prove the adequacy of the model. Thus, in adequate models, residual values are usually characterized by random distribution in different directions above and below the zero line (presented in Minitab). For the friction coe cient, the largest deviations in the graph are the results for rows 15, 18 and 27in the planning matrix given in Table 3 ( Figure 11a). Each of the above-stated rows is containing a sample where the ller content is 15 Wt.%. In specimens from 5 to 10 Wt.% ller content, an error of row 15 occurred due to increased friction and wear.
The plots showing the distribution of the residual between the experimental results and the predicted values for the wear rate also show that this model is adequate for the process (Fig. 11b). Despite this, the unequal distribution of residuals can be seen in Fig. 11b. Red arrows show that residuals are increases with tted response values magnitude. All large residuals are referring to the results for the rst starting point of the third level of sliding velocity (2.5 m/s). The residual was higher in these rows because the effects at level 3 of the sliding velocity were different depending on the ller content.
As can be seen from basic information on both models in Tables 8 and 9

Optimization process parameters
The results of the optimization for the friction coe cient are given in Fig. 12. In the presented optimization graphs blue lines show optimal settings and black lines show the predicted response values at different regimes. As described in Fig. 9, the optimal friction condition is a combination with a contact e = y − ˆ y y ˆ y pressure of 9.6 MPa, a sliding velocity of 0.64 m/s and a content ller of 10 Wt.%. The predicted value of the friction coe cient in this regime is 0.451898.
The results of the optimization for the wear rate are given in Fig. 13. As shown in the gure, the optimal condition for the wear is at the lowest level of each factor: the contact pressure is 1.9 MPa, the sliding velocity is 0.64 m/s and the content ller is 10 Wt.%. The predicted value of wear rate in this condition is 0.0020093 g/hour.

Con rmation test
Checking the accuracy of the results is the last step in the planning methods of experiments. Table 10 exhibits the conditons selected for both response parameters. According to these parameters, the obtained results are given in Tables 11 and 12. showed that the error was less than ~ 3% for both con guration, which is an acceptable result.

Conclusion
Using the Taguchi tecnique following conclusions are drawn based on statistical processing of obtained data from experimental friction and wear tests: P-values of selected design factors were higher than 0.05, which is proves their importance to tribological properties.
ANOVA results showed that contact pressure (55.75%) and sliding velocity (33.16%) is a most signi cant factor on the friction coe cient. Filler content impact was only 2.92%.
The highest friction coe cient (0.445) and lowest wear rate was obtained in brake pad composition containing 10 Wt.% copper-graphite particles.
The lowest friction coe cient (0.367) and highest wear rate was found in brake pad composition containing 15 Wt.% copper-graphite particles.
Friction coe cient increasing with the increase of contact pressure with lowest sliding velocity. but decreases with the increase of sliding velocity.
Surface characterization with SEM and graphical scheme for frictional heat effect indicates that copper-graphite particles are increasing of heat conductivity of the composite materials. Brake pad specimen containing 10 Wt.% copper-graphite particles have better friction surface.
From generated regression equations it is observed that contact pressure is positively related to process and plays a major role on friction coe cient. Also, the contact pressure and sliding velocity are positively related to the output parameter and serve to increase the wear rate.
Optimal combination for friction coe cient was determined to be contact pressure 9.6 MPa, sliding velocity 0.64 m/san and for wear rate typically at the lowest level of parameters.
Declarations Figure 1 a:left) Cylindrical specimens and b:right) disc used in friction tests  Pareto chart of the standardized effects for a:left) friction coe cient and b:right) wear rate Surface plot for dependence between friction coe cient and regime parameters Figure 6 Surface plot for dependence between wear rate and regime parameters  Contour plot for dependence between friction coe cient and regime parameters Figure 10 Contour plot for dependence between wear rate and regime parameters Figure 11 Residual plots of linear regression model for a:left) friction coe cient and b:right) wear rate Figure 12 Setting and optimal solution for friction coe cient Figure 13 Setting and optimal solution for wear rate