Prediction Of The Inuence Of Cutting Conditions On Surface Morphology After Its Thermal Spraying With Stellite 6 Alloy

The presented paper is devoted to the basic inuences of the cutting conditions on the basic parameters of the roughness of the surface machined by longitudinal turning; the investigated surface was before the machining modied by the technology of thermal spraying. High-speed Oxy-Fuel (HVOF) thermal sprays, in particular, Stellite 6 sprays based on Cobalt and Chromium alloy, are widely used in the industry thanks to their excellent mechanical properties, wear resistance, corrosion resistance and good temperature stability in comparison with steel for example. From experimental results in the specied range of cutting speeds, feeds and cutting depths, new predictive models of roughness parameters were compiled using mathematical and statistical methods, followed by application of neural networks in order to conrm the obtained conclusions. Research results have proved that feed turning technology is a suitable technology for machining coatings generated by Stellite 6 alloy thermal sprays in such user cases where additional surface treatment is required for specic reasons.

An increase in the mechanical strength of the primary structure of the cobalt base alloy depends primarily on its material composition, on the form of carbide in the cobalt matrix, and on the grain boundaries. The alloys Co-Cr-Mo (F75), are usually produced by casting, the alloys Co-Cr-W-Ni (F90), Co-Ni-Cr-Mo (F562), Co-Ni-Cr-Mo-W-Fe (F563) are produced by forming and forging [12]. No less important contribution to the increase in the mechanical strength of cobalt-based alloys is their coating with the use of High-Velocity Oxy-Fuel (HVOF) technology.
At present, new solutions for surface treatments and hard protective coatings are being investigated in order to improve the properties of materials used in the most demanding applications [13]. An example of the demanding applications of surface treatments is the work of the authors [14] who have reported the use of erosion-resistant coatings, such as particularly CrN, TiN, TiAlN and TiSiN in the engine components in the aviation industry. Hard chrome coatings for wear-resistant applications in the automotive industry are also known [15]. Increase in the number and importance of applications of thermal spraying processes makes it possible to replace the traditional applications of hard chromium by thermal spraying, e.g. for electrodes [16]. The WC-Co cermets are one of the main hard materials used for the production of coatings; the cermets are speci cally used for thermal spraying [17]. The main properties of WC-Co coatings include requirements for hardness and wear resistance [18]. At present some other cobalt-based Stellite alloys are also available that offer excellent wear resistance, high heat resistance and high corrosion resistance [19][20].
In most applications, the use of coatings requires a guarantee of an adequate quality even a long time after the manufacture. As a result, conventional machining processes are introduced for an effective nish of the coatings. Machining of such coatings is extremely di cult precisely because these alloys retain their strength and hardness even at elevated temperatures generated by the machining process.
Low thermal conductivity, high hardness at elevated temperature and high wear resistance are the cause of the very di cult processability of cobalt, chromium and molybdenum alloys.
High attention is paid to conventional operations, such as turning, compared to conventional used grinding technology, in order to reduce machining times. Hard turning is considered to be an adequate method of substitution of grinding when machining high hardness materials [21]. In machining processes, however, the high hardness of the coatings often results in high cutting forces and cutting temperatures, and it consequently decreases the tool service life [22]. The relationship between hardness and wear of the tool was experimentally analysed by the authors [23], and it was found that increasing the hardness of the steel from 35 to 45 HRC increases wear of the cutting edge. Increased attention should also be paid to the process of production of the coating as such. For example, coatings formed by application of thermal spraying are anisotropic, which affects several properties: hardness, modulus of elasticity and fracture toughness [24]. According to the authors [25], the machining of coatings, especially after thermal spraying, has to address two main problems: the adhesion of the coating to the substrate and the wear of the tool.
Moreover, the authors [26] state that the e ciency of the machining process depends on the structure and properties of the materials in contact. The expected and desirable results, therefore, depend to a large extent on selected tool materials and selected machining parameters [27]. An adequate selection of machining parameters can help to achieve acceptable surface quality and tool wear. Typically, cutting rates of approx. 30 -40 m·min−1 and relatively low values of cutting depth and feed rate are used [25]. Although machining is a complex process that is in uenced by many factors, the current studies focus on selecting the machining parameters for turning operations. An empirical and mechanical approach to selecting machining parameters is identi ed by the authors [28] for surface roughness analysis, although both approaches could be used for assessment of other outputs. The considerable complexity of the mechanical approach is simpli ed by the empirical approach, that's why usually several experimental tests with different machining parameters are usually performed to predict the nal effect on the results.
The empirical methodology is particularly suitable for machining processes, which represent a strong background for stabilising the initial values of the machining parameters. However, this background is not large enough for machining of coatings, cutting of which is di cult. Although the choice of machining parameters plays an important role in the process results, it should be emphasized that other factors can also have a signi cant impact. For example, the authors [29] identi ed the tool vibrations, the appearance of ruptures, adhesion, material deposits, elastic and plastic deformations and tool wear as basic factors for achieving surface quality. The in uence of machining parameters is more accepted in the scienti c literature than it is monitored by technical practice; in this sense, more results of the machining process are affected. In the case of hard turning, the authors [30] con rmed the in uence of cutting rate, cutting depth, feed rate and machining times, surface roughness and wear of the tool. The effects of machining parameters on the surface quality are given by the authors [31], who evaluated the turning of sintered WC-25Co with the use of cutting tools by the tests, which identi ed an unequivocal relationship between cutting rate and surface roughness. At cutting rates of 15 and 40 m·min−1 the surface roughness was limited to less than 0.2 µm for all machining times tested. However, at a cutting rate of 100 m·min−1, the surface roughness achieved signi cantly higher values and reduced the machining time.
On the other hand, the authors [32] found that the choice of feed rate (from 0.03 to 0.3 mm·rev−1) played an important role in the development of surface roughness. When analyzing the effect of cutting depth, it was found that the use of sharp tools caused an increase in cutting depth and lower values of surface roughness. The authors [33] evaluated the in uence of cutting rate and feed rate on the surface roughness by turning Stellite 6 using tungsten carbide and ceramic-reinforced ceramic tools. They found a great in uence of the feed rate on the surface roughness value, but the effect of cutting rate was lower. Test tested feed rates ranged from 0.1 to 0.2 mm·rev−1 for tungsten carbide tools and from 0.25 to 0.35 mm·rev−1 for bre-reinforced ceramic; the cutting rate was set from 30 to 50 m·min−1 (for tungsten carbides) and from 30 to 90 m·min−1 (for ceramics). The optimal surface roughness was achieved by using low feed rates and high cutting rates.
It means that the hardness of Stellite coatings is the main factor that makes the machining process more di cult. Based on published ndings, the authors of this paper ask the key question as to how e cient

Methods
The Co-Cr-W Stellite 6 alloy, which is most widely used in industrial applications, was used for the presented experiments. Its exceptional wear resistance is mainly due to the unique properties of the hard carbide phases dispersed in the CoCr matrix. Various coating technologies, such as plasma transfer arc (PTA), inert tungsten gas (TIG) welding, thermal spraying or laser coating, can be used to deposit the Stellite layer on the surface. The chemical composition of the Stellite 6 spray obtained from the scanning electron microscopy/energy dispersive X-ray spectroscopy (SEM/EDX, Carl Zeiss s.r.o., Prague, Czech Republic) measurements [34], with the density of the test material being 8.44 g cm −3 .
[ Table 1 about here.] Table 1 Nominal composition (mass %) and basic physical properties of the Stellite 6 coating Thermal spraying is an advanced technology for depositing thick coatings (50-500 µm), its principle is illustrated in Figure 1. The coating material is usually delivered as small particles to a spray device where it is heated and accelerated towards the surface of the substrate. The molten particles gradually touch the surface of the substrate, where they deform into scales and lamellae in the form of a disc and then quickly solidify. Repeated impact of particles creates a coating with a typical lamellar microstructure and with speci c anisotropic properties [34].
[ Thermal spraying is an advanced technology for depositing thick coatings (50-500 µm), coating material is usually delivered as small particles to a spray device where it is heated and accelerated towards the surface of the substrate. The molten particles gradually touch the surface of the substrate, where they deform into scales and lamellae in the form of a disc and then quickly solidify. Repeated impact of particles creates a coating with a typical lamellar microstructure and with speci c anisotropic properties [34]. Carrier gas Nitrogen, 6.5 l×min -1 Offset 6 Number of passes 7 In addition to deformed particles, the microstructure of the coating applied by thermal spraying also contains numerous defects, such as oxides, pores and impurities. Oxides originate in particular from the oxidation of the surface of the particles formed during the process of spraying. Porosity and unprocessed particles are formed as a result of non-optimal heating and acceleration of ying particles and their improper propagation after impact. The presence of defects naturally reduces the mechanical properties of the coating, such as hardness, strength in cohesion, toughness and consequently also functional properties, such as wear resistance [35]. The nature of the coating defects determines its quality [34].
Although different techniques of thermal spraying were developed (e.g. plasma spraying, ame spraying, cold spraying, etc.) used for different types of coating materials in connection with various applications, the HVOF technology for spraying metal and hard metal is the most suitable. High speed combined with medium temperatures reduces the amount of oxidation of particles' surface during the spraying process and allows the formation of coatings with low porosity and small amounts of internal oxides [34,35].
The HP/HVOF JP500 spraying technology was used for the realisation of the presented experiments at the Research and Testing Institute in Pilsen, Ltd. The parameters of spraying are summarised in Table 3.
It is evident from the SEM microstructure that the amount of porosity is low without the occurrence of any cracks or deformations. At larger magni cation, the gap between the splask and the inner dendritic microstructure can be seen. Using the HVOF technology, the Stellite 6 coating microstructure, consisting of two cohesive Co-based solids, the main cover centre (fcc) and the less dominant hexagonal tightly closed (hcp), was applied by spraying. At a mechanical load, the fcc Co tends to convert to hcp-Co. The pressure-induced martensitic transformation that is responsible for hardening is one of the main sources of low wear and high erosion resistance of cobalt-based alloys. Transformation mechanisms of pure cobalt and its alloys were described in various studies [35]. Unlike cast Stellite 6 or in the form of laser-applied microscopic paint, HVOF does not contain carbide particles due to the high speed of solidi cation at spraying. The microstructure of the test coating is analysed in more detail in the works [35]. Several experiments were carried out as part of experimental veri cation. For the rst experiment, cylindrical specimens of 54.7 mm in diameter from the C45 base material were used, on which a Stellite 6 was applied by thermal spraying with an average value of 0.55 mm. The thickness of the machined layer was set to 88 mm. The tool TUNGALOY RNGN 120400 LX11 43 was used, which was clamped in the tool holder MRGNR2525M12. The experiments were veri ed on the MASTURN 50/C80 engine centre lathe according to the conditions outlined (Table 4). A view of the experiment is shown in Figure 1.
The following parameters of surface roughness were selected as the basic indicators of the morphology of the machined surfaces as part of the experimental veri cation: arithmetical centre of absolute deviations of the ltered roughness pro le from the centre line within the basic measuring length l r -Ra; medium depth of roughness, i.e., the average value calculated from 5 values of Rz and 6 values of basic lengths l r .
For these plates, a clamp C5-SRDCN00060-12AHPA was selected, which was xed using the CAPTO C5 onto the dynamometer (Spike by Promicron), and the dynamometer was then clamped to the milling spindle. A plan of the experiment was drawn up, according to which the selected plates were tested at a cutting rate of 320 and 80 mm·min-1and at a feed rate of 0.3 mm·rev -1 . In both cases, the cutting blades were quickly destroyed due to the considerable positive geometry of the tool (γ = 20 ° From the viewpoint of the chip machining of the coating after it's spraying with the alloy Stellite 6, two speci c problems arose that caused worse machinability. From the viewpoint of machinability, it was mainly the boundaries of the splat, which did not concern the grain boundaries but the boundaries of the deformed particles. Cohesiveness between splats is lower, particularly for Stellite 6. The splat boundaries can be formed by the oxides formed during the ame particle ight during ame annealing. The second problem was the alloy material as such. The Stellite within the splat undergoes a deformation induced martensite transformation from the fcc to the hcp grid. During machining, it hardens and becomes brittle, similarly as some types of steel. The behaviour of the investigated objects can generally be described using models that can explain the logical relationships of objects, but sometimes they must do with nding a mathematical function that determines the dependence between factors and responses for the objects in question, although it is impossible to answer unequivocally why is it so. The monitored responses in the presented experiment can be searched and veri ed using a mathematical model using statistical tools. This method makes it possible to describe how the given response depends on individual factors. Due to the fact that, in the case of surface roughness pro le parameters the attention was paid mainly to the basic parameters of roughness Ra, Rz, two models were sought for these two responses. For the Ra parameter, the model is formulated in (1) as a relation with variables v c cutting rate, f rev feed rate and a P depth of cut.
The analysis shows that the model (1)  Furthermore, the adequacy of the selected model was tested using analysis of variance (Table 5) by testing the zero-statistic hypothesis H 0 , which resulted from the nature of the test and provided information that none of the effects used in the model affected the signi cant change of the investigated variable. It followed from the subject test that the achieved level of signi cance (Prob> F) was less than the chosen signi cance level α = 0.05 and that it could be concluded that there was not enough evidence for accepting the H 0 hypothesis, i.e. that it could be stated that the model (1) was signi cant. It also follows that part of the total variability of the experimentally obtained values, which corresponds to random errors, is signi cantly smaller than the variability of the measured values in accordance with the model.
[ On the basis of testing of the adequacy of the used statistical model lzr it is possible to determine the regression coe cients of statistical dependence de ned in (1) using the smallest square method (Table  6).
[ An analysis of the in uence of the individual investigated input variables on the change of the Ra value is shown in Figures 2 and 3. It is obvious that with an increase in the cutting speed the value of the mean arithmetic deviation of the roughness pro le also increases. Figure 2 illustrates an experimentally obtained model describing the given relationship for constant parameters: constant feed rate, constant cutting depth, and constant machining length, due to the in uence of the given input factors on the change of the Ra value. The in uence of the cutting rate on the change of observed response Ra is 15.75%. Given the validity of the predictive model (1) at the selected range of the cutting rate from 150 to 250 m·min -1 , it can be stated that within the given interval an increase of the cutting rate by 10 m·min -1 will cause an increase of the Ra value by 0.027 μm, with ful lment of other conditions ensuing from the nature of mathematical and statistical modelling. The validity of the model (1) is limited to the interval of the used cutting conditions, and the validity beyond of those intervals has to be always veri ed experimentally.

about here.]
The feed rate f rev as the second investigated input factor has a 53.79 % in uence on the change in the value of the observed response of the surface roughness Ra. It may be stated that with an increase in the feed value the value of the mean arithmetic deviation of the roughness pro le increases as well. The in uence of feed rate on the roughness of the machined surface results from the kinematics of the machining technology. From the machining theory, it is obvious that with the increase in feed rate the corrugation of the machined surface increases. Research performed by many authors shows that, at higher feed rates, the cutting rates have a smaller in uence on the quantitative values describing the micro-geometry of the machined surface. These conclusions are also documented by the results of this experiment ( Figure 3).

about here.]
More profound analysis of the statistically most signi cant factor -the feed rate frev on the change of the observed response Ra, shows that the conditional value Ra (Figure 3) also increases within the range of the experimentally used values of feed rate with its increase. Since the feed rate plays a major role in changing the Ra value, its in uence can be further analysed by the Kruskal-Wallis nonparametric test ( It is necessary to realise in this context that the Kruskal-Wallis test does not work with the original values but with the sequential numbers that were assigned. From the values of the sum of the order (Figure 8), it can be seen that the lowest values of the observed response Ra were achieved at the feed rate of 0.4 mm/hr -1 and with the continuous increase of the feed rate value the Ra values also increased. The results are consistent with the previous conclusions. When comparing the achieved levels of signi cance p (Table 8), it can be seen that the statistically signi cant difference in the achieved Ra value is only between the feed rate of 0.4 mm·rev -1 and 1.0 mm·rev -1 . The difference of the observed response Ra between the mm·rev -1 and 0.6 mm·rev -1 , as well as between 0.6 mm·rev -1 and 1.0 mm·rev -1 was not proven.
[ The third independently examined variable factor was the cutting depth a P . It is affected by 16.32 % in the case of a change in the investigated Ra response. The in uence of the cutting depth on the surface roughness results from changes in the deformation processes in the cutting zone when changing the depth of cut. The literature sources document a negligible in uence of the cutting depth on the roughness of the surface. By increasing the cutting depth, the roughness will decrease slightly if the cutting rate and feed rate are constant.

Results and discussion for parameter Rz
Similarly, as for the investigated parameter Ra, it is also possible for the investigated parameter Rz to analyse the assumed dependences and to express the model (3) An Analysis of the investigated parameter Rz shows that the model (3)  [ Table 9 about here.] It follows from Table 9 that a statistically signi cant factor in uencing the change of the investigated response Rz is primarily the feed rate, with a 58.305 % in uence on its change, as well as the cutting depth that contributes to a change in Rz variability of 26.252 %. For the parameter Rz, at the level of signi cance of α = 5 %, no signi cant in uence of the cutting rate and an absolute element, or of the model constant (3) was proven. Precisely the statistical insigni cance of the constant indicates that within the realised experiment only important input factors were used. By substituting the coe cients from Table 9 to the model (3), we obtain the nal form of the statistical prediction function of the dependence of the value of the investigated response Rz on the cutting conditions for the feed turning of the coating Stellite 6 with the use of the tool RNGN43 LX11-TUNGALOY (4). These conclusions are also con rmed by simulation of experimentally obtained results using the neural network. At the nodes of a neural network, analogously called neurons, each input variable xi at the input of the jth neuron is multiplied by the weighting factor wji. The sum z j = w 0j + Σw ji ·x i is in a neuron transformed by the activation function. The activation function expresses the intensity of the neuron response to the given input. The most commonly used activation functions include the logistic function, σ j (z)=1/(1+e-z), which resembles the biological sensoric response function. The weights wji represent the intensity of the linkage between the variable and the neuron, or, in the case of multilayer networks, between the neurons in the layers. These linkages are sometimes called synapses. The output variables are predicted as weighted linear combinations of outputs from the last hidden neuron layer. Thus, the neural network is formally a special case of multiple nonlinear regression; the neural network can be virtually considered to be a nonparametric regression. If the neural network contains no hidden layer of neurons, only the input and output variables, it would be a linear regression model. The neural network is optimised using the smallest square criterion. This means that the network is set in a way that the sum of squares of the predicted and measured value of the output variable is minimal. This setting is the objective of an iterative optimisation procedure, called learning or training of a neural network. The optimisation procedure uses adaptive Gauss-Newton algorithms. For the application of the neural network, the experimentally obtained response values Rz were divided by a ratio of 85 % to 15 %, the rst group being used as data for learning and the second group as data for testing. The basic used neural network with three layers (the rst containing 3 neurons and the other two hidden layers contained 2 neurons each) reached a maximum error for learning data value of 0.004346201313 and a mean error for learning data value of 0.0009920144294. The maximum error for the tested data is 0.003247050035 and the mean error is 0.001291113035. These values, along with the graphical representation of the learning process (Figure 4), demonstrate good quality of the model and they express both the improvement of the model and the improvement of the prediction of valid data. The model in question therefore predicts well from unknown data, and at the same time, it predicts both curves, i.e. both for learning data and for testing data.
[ Fig. 4  When observing the change in the Rz value in dependence on the change in feed rate with simultaneous change in the cutting rate (Figure 7), it can be observed that by increasing the cutting rate, the Rz value increases with the increased feed rate, while the change in the Rz value caused by the cutting rate is small. It varies from 2.55 % to 3.75 % with an increase of the cutting rate from 150 m·min -1 to 250 m·min -1 in increments of 25 m·min -1 . The cutting depth has a more signi cant in uence on the change of the Rz value. An increase of the cutting depth from 0.1 mm to 0.15 mm reduces the Rz value by approx. 59 %. However, this reduction in the given cutting depth interval must be attributed to the machined material as such. When increasing the cutting depth by more than 0.15 mm, it is possible to expect on the contrary a deterioration of the roughness Rz. Each mechanical treatment of the surface leads to the introduction of stress into the coating, and in any case, it leads to rearrangement of the already existing stress in the coating.
Coatings applied by thermal spraying consist in principle of individual deformed particles of the additive material deposited on each other, in any case with a certain amount of porosity. During mechanical treatment of such a coating composition, it is necessary to take into account that the contact tool-coating is characterised by an intermittent cutting, which results in further adverse consequences consisting of a shock effect on the coating.
As a result of oxidation and other physical-chemical processes, the coatings are mostly composed of very hard and brittle phases, which in themselves bring considerable di culties during their mechanical treatment.
Mechanical treatment of applied coatings results in a loss of one of the signi cant advantages of thermal-spraying, which is the formation of thin coatings at certain parts of the surface of components or parts, without any subsequent mechanical treatment of the coatings, thus saving labour and often also de cient and economically expensive coating material.
For further study of the technological possibilities of turning of hot sprays, attention should be paid, in addition to the in uence of cutting conditions, to the appropriate geometry of the cutting wedge and to the cutting forces, which are formed in the process, as they often play a key role in relation to the mechanical properties of the spray application. Because the structure of the applied coating is composed of deformed particles of the additive material, the surface roughness of the applied coatings is greatly in uenced by the granularity and grain size composition of the additive material. Generally speaking, the larger the particle of the deposited material, the greater is the surface roughness of the applied coating.
The good message is that the roughness after spraying is mostly su cient for the functional use of coatings. This fact has led the producers of the additive materials to the idea of producing additive materials with different grain sizes ranging from coarse to very ne ones. Of all the methods of mechanical treatment of coatings applied by thermal spraying, the grinding became the most used, and in the case of ceramic coatings, the grinding with the use of diamond tool became the most used. The grinding requires in all cases an intensive cooling. Poor cooling, or even no cooling, leads to the formation of a mesh of surface cracks in coatings, especially in ceramic coatings. The efforts to achieve the maximum economic effect of mechanical machining of applied coatings lead to the veri cation of the possibilities of machining of these coatings by turning. At present, turning can be used only for machining of the coatings applied by thermal spraying having relatively good plastic properties and relatively low hardness. This type of turning commonly uses higher cutting rates and small feed rates. However, a big problem exists with the use of turning in the case of an intermittent cutting. The allowances required for mechanical machining of the coating, and thus the required thickness of the coating, are also governed by the surface roughness in the state after spraying, and it can be said that they are dependent also on the technological method of the thermal spraying.

Figure 2
Dependence of the change of the surface roughness Ra on the cutting rate v c .

Figure 3
Page 21/23 Dependence of change of the value of surface roughness Ra on the feed rate f rev .

Figure 4
Diagram of the learning process for the response Rz.

Figure 5
Page 22 /23 Relative in uence of predictors on change of the value Rz.

Figure 6
In uence of the change of the Rz value on the feed rate.

Figure 7
Dependence of the change in the value Rz on the feed rate with simultaneous change of the cutting rate.