Proceedings of SerbiaTrib '13

Comparative analysis of results 1n the figures 17and 18 indicates that the MQL technique providesbetter results in aspect of tool wear and tool life, butslightly worse results in aspect of the surfacequality as consequence of thermodynamicprocesses in the cutting zone.4. MODELING OF RESULTSExamination of the experimental results wasperformed by multiple regression analysis. (see Fig.19). The output values from the regression modelshowed a significant correlation with theexperimentally measured values. The averagerelative error of the regression models does notexceed 5%. The models presented in the form ofregression equations can be used with highaccuracy of prediction. The models of main cuttingforce (F c ) have errors less than 2%, while the mainsquare errors for models of forces (F f ) and (F p ) arehigher. This corresponds to the theoreticalassumptions of cutting parameters behaviour in thecase of machining of steel C45E.the main cutting force and resultant cutting force.Cutting speed and feed have a great influence onthe force (F f ) and (F p ), and the resultant forces(F f,p ), especially during machining with MQLtechnique. It can be concluded that the increase offeed and depth of cut increases the value of thecomponents of the cutting force. Increasing cuttingspeed reduces the values of cutting force becausethere is no negative phenomenon, such as the burrson the tool edge.Table 7. Models of cutting forces with coefficients ofmachiningMaterial: C45ETool: SNMG 1204 08 NXMCollerationcoefficientRelative error (%)F c = 2485 · a 0.878 · f 0.844 · v -0.047 · K 1 0.99 1.9F f = 1154 · a 0.838 · f 0.391 · v -0.157 · K 2 0.93 3.9F p = 822 · a 0.589 · f 0.644 · v -0.052 · K 3 0.93 3.7Techniques K 1 K 2 K ‚3Conventional 1 1 1MQL 0.99 0.96 1.02Figure 19. Comparison of outputs regression model withexperimental resultsThe biggest errors is expected in the predictivemodels of forces (F f ) and (F p ) in machining withMQL technique; 4.93% and 4.44% respectively. Inaddition to modelling the of cutting forcecomponents, also a resultant force F f,p , which is aresultant of (F p ) and (F f ) was modelled. Theseresultant forces have a higher value than the maincutting force. The resultant force (F f,p ) is a mainindicator of tool wear, and with its growth usuallyintensive tool wear occurs.Analyzing the models, and their correspondingexponents (Figure 19 and Table 7), it can beconcluded that the depth of cut has the highest,while the cutting speed has the least influence onModelling of cutting force for differenttechniques of CLF dosing using regression analysiswas done. The developed models are presented inTable 7, where the influential factors representedby the corresponding coefficient Ki. Developedregression models have error less than 4%, whichindicates the high accuracy of the model. Theapplying MQL technique reduces the energyconsumption compared to the conventionallubrication technique. MQL technique should befavoured in highly productive processes.Often, multiple regression analysis is notsuitable for the modelling of complex processes,which depends on a many number of factors. Formodelling of such processes a large number ofexperimental data are needed. In our study ANNtechnique was applied. For this technique a specialmodule Neural Toolbox in the software packageMATLAB, is used. Modelling with ANN wasconducted using a model of two-layer neuralnetwork with forward propagation.298 13 th International Conference on Tribology – Serbiatrib’13