The influence of technological parameters on surface roughness during turning and roughness prediction using artificial neutral networks

The purpose of this investigation was to determine whether and to what extent the technological parameters of turning (feed, cutting speed) affect selected surface roughness parameters of aluminum alloy EN-AW 7075 (AlZn5.5MgCu). The principal findings indicate a significant impact of feed and show on the surface roughness and simultaneously show that cutting speed has no effect on the value of surface roughness parameters under investigation. An artificial neural network was employed to evaluate the prediction of surface roughness parameter Rz in turning.

Light alloys (aluminum and magnesium) are widely used in various industrial areas as a modern construction material (due to its favorable strength properties and low density). The machinability of aluminum alloys can be influenced by such physical properties as: high expansion and thermal conductivity as well as a small modulus of elasticity. On the other hand, surface roughness can be influenced by: tool material, workmanship quality and tool nose geometry, material properties (e.g. Rm, HB) and technological parameters.
During turning of the AZ31 alloy, the roughness decreases with the increase of the cutting speed (up to vc approx. 160 m/min), and increases with increasing depth and feed. The change in the Ra parameter is in the range of about 1 μm (Ra = 1.1÷2.2 μm) [9]. Similar relationships were observed when processing the AZ61 alloy (vc to 200 m/min). The Ra parameter in experimental studies was Ra ≈ 0.1÷2.8 μm [1].
When rolling aluminum alloys, apart from the right geometry (γ to 30°, α to 10°), it is important to control the breaking or rolling of the chip. As stated in [12], Ceratizit introduced a compressed air cooling system integrated in the tool (the ability to process dry aluminum alloys).
Often when roughing light alloys with carbide tools (up to vc approx. 400 m/min), machining surface roughness Ra <10 μm. Also the change in the angle of attack (increase in γ) may positively affect the improvement of the quality of the final surface [16].
AlMg1Si alloy is difficult to cut (malleable, hard, reinforced by forging). The use of a tool with a PCD blade, with a laser shaped chip winder and a chamfer with the so-called the negative rake angle γ can increase the efficiency of the process (chip breaking at small intervals and easy discharge from the cutting zone) [11].
A similar relationship can occur when treating the surface of pistons made of foundry eutectic or non-toxic aluminum alloys. In this case, optimum cutting data values for vc = 610 m/min and ap ≤ 1 mm [14] were obtained.
Positive influence on machinability indexes and cutting properties of tools, and even on a significant increase in durability of tools, has rounding of the cutting edge (better adhesion of the coating to the substrate and reduced tendency to peeling on the edges of the blades) [2]. It can significantly improve the operational properties of tools, for example in the abrasive machining process. This favorable phenomenon is explained, among others obtaining significantly lower roughness of the blade's working surfaces and thus less friction on the tool contact surfaces with the workpiece and limiting abrasive wear and adhesive blade [2,6]. The radius of the cutting edge curvature can also be used to determine the minimum thickness of the cutting layer, where hmin is the product of the friction coefficient between the cutting edge and the material being machined and the radius of the cutting blade's curvature [10]. It is logical to say that the role of rε grows during the finishing of the finishing work. For the βn (50÷70°) angle, the radius of the cutting edge rounding rn ≈ 6÷11 μm is obtained -expected for the new hard sintered edge. In practice, the values are greater -around 15 μm. They grow along with the progressive wear of the tool tip [13]. On the roughness of the surface after turning, the conditions of mapping the cutting edge on the machined surface can also influence [3].
As reported in [5,12,16] and in the recommendations of tool manufacturers [17], popular aluminum alloys can be processed with vc ≥ 2500 m/min. However, for reasons related to the construction of technological machines, these parameters may be limited under certain conditions. It seems therefore important and purposeful to carry out machinability tests in the area of increased machining parameters in order to increase the efficiency and effectiveness of light alloys processing.
However, the increase in production efficiency may contribute to the increase of requirements for cutting tools. One of the basic criteria for assessing the suitability of a machining tool can be assured to ensure an adequate surface quality [7,15,16]. It has also been proven many times that there are close relationships between surface topography and functional properties of machine parts. So there are practical possibilities to improve the technological and usable quality of parts, including in finishing operations [4].

Methodology and purpose of research
The main stages of the research included machining at high cutting speeds (vc up to 1700 m/min), 2D surface roughness parameters analysis and prediction of the selected surface roughness parameter (Rz) using artificial neural networks.
The treatment was carried out on a DMG MORI CTX450 turning center with the Sinumerik 840D control system. The tool used was a folding knife with the symbol SDJCL 25x25M11 CORO Turn 107. For turning, a DCGT 11T304-FN-ALUH210T carbide insert (Ceratizit) with a corner radius of rε = 4 mm and a radius of rounding of the cutting edge of approx. 11 μm were selected. A constant cutting depth ap = 1 mm and a variable range of other technological parameters were assumed: f = 0.05÷0.15 mm/rev, vc = 1000÷1700 m/min.
The EN-AW 7075 (AlZn5.5MgCu) alloy was used in the T6 state (supersaturated, artificially aged). A machining emulsion was used during machining. The length of the turning path was L = 20 mm. The Hommel Tester T1000 profilographometer was used for surface roughness measurements. Roughness measurements were repeated five times on each surface. The value of the elementary segment was lr = 0.8 mm, and the measurement section ln = 4.8 mm.
The results of experimental tests (average values of roughness parameters) were the input element in simulations of the Rz roughness parameter using artificial neural networks, which were carried out in the Statistica software. The simulations used MLP networks (multi-layered perceptron) with activation functions: linear, exponential, logistic, tanh and sine. The indicators of the correct selection of the network were: the quality of teaching, the quality of validation, as well as the learning error determined by the least squares method.
To make the network construction as simple as possible, they have one layer hidden with the number of neurons in the 1÷10 range (experimentally selected), the input layer with two neurons (cutting speed vc, feed f) and the output layer with one neuron (Rz parameter roughness). BFGS gradient method (Broyden-Fletcher-Goldfarb-Shanno) was used to teach the network. Statistica Automatic Neuron Networks enables the use of two fast learning algorithms: the conjugate gradient algorithm and the BFGS algorithm. The BFGS algorithm provided better results when simulating this process. When teaching the network, 75% of the measurement results were a learning group and 25% were validated. Due to the small number of data sets, the test group was dropped [15,18].

Results of tests and simulations and their analysis
After taking into account the corner radius rε and the feed value f, one can calculate the theoretical surface roughness -Rzt parameter. For the analyzed range of Rzt parameters it was respectively: 4.5 μm -in case of change in vc, 0.8÷7.0 μm -in case of change f.
The graphs ( fig. 1 and fig. 2) show the change of 2D surface roughness parameters (Ra, Rz, Rv, Rp) depending on the change in the cutting speed vc and the feed rate f. Additionally, the numeric values of the Rsm parameter are given. It can therefore be seen that the cutting speed vc does not significantly affect the surface roughness. It follows an important conclusion that machining with high vc values is possible without deterioration of the treated surface quality. This is an important statement in the context of efficiency and efficiency of machining processes. For the entire feed range f, the numeric value of the Rsm parameter was within 85÷147 μm. It can therefore be seen that the feed affects the value of surface roughness parameters quite significantly.
As a result of experimental research, it was possible to predict the selected roughness parameter -Rz -using artificial neural networks. During the simulation of each parameter, 100 networks were designated, of which the best three were selected based on the above mentioned indicators. Their characteristics are presented in the table.
Based on the analysis of the obtained neural networks, it can be concluded that the best results have been achieved for MLP 2-3-1 networks. The simulation results for this network are shown in fig. 3.  . 4). The error in determining the parameter values did not exceed 8%. Thus, it can be concluded that the results of the simulation are characterized by an acceptable level of error and can be a tool to determine the processing parameters in order to obtain the assumed roughness.

Conclusions
The analysis shows the importance of the correct selection of technological cutting parameters. Measurements and analysis of results allow to formulate more important general conclusions: • cutting speed does not significantly affect the surface roughness parameters after turning (Ra value was approx. 1 μm, Rz -approx. 4.5 μm), • a significant effect on the surface roughness has a feed rate, this effect is approximately proportional to the feed rate; this is due to m.in. increase in cross-section of the cutting layer, • it is possible to use high cutting speeds (up to vc 1700 m/min) without worrying about the roughness of the treated surface (increase in the efficiency of the process), • it is possible to model the roughness parameters depending on the cutting speed and feed using artificial neural networks (the error between the actual values and obtained from the model for the Rz parameter does not exceed 8%), • models obtained as a result of the simulation enable determination of a set of machining parameters allowing to obtain the assumed surface roughness after machining.
Further analysis of 2D and 3D roughness parameters can be an important prerequisite for engineers designing turning technology for aluminum alloys.