• Tidak ada hasil yang ditemukan

The objectives of the present thesis are to study the cutting performance, modelling of the performance parameters and optimization of cutting conditions in an environment-friendly air-cooled and dry turning processes. During dry and air- cooled turning, cutting fluid is not used, thus reducing pollution on the shop floor.

To study the effect of air-cooling on cutting performance, several experiments were conducted with different tool-workpiece combinations in dry and air-cooled condition. The comparative study was carried out mainly on the results of surface finish of machined workpiece, tool wear, machining forces and vibrations. It is observed that air-cooling has a great influence in reducing the tool wear, but does not have much effect in improving the surface finish and reducing the cutting forces.

Neural networks have been used successfully for prediction of surface finish, tool life and machining forces. A novel neural network modelling procedure has been developed in this thesis for the prediction of surface finish and tool life considering limited and noisy experimental data. The method utilizes both MLP and RBF networks. However, neural networks could not be employed successfully for predicting the tool wear based on the machining force measurement. A simple method has been developed for the indirect estimation of tool wear in a probabilistic manner using the rate of change of wear with respect to cutting/feed force. In the present thesis, a few strategies have been developed for efficient utilization of cutting tools in finish turning operation `in both dry and air-cooled condition.

Finally, a simple heuristic method for the optimization of cutting conditions that does not require a priori information of tool life has been developed for finish turning operation.

The conclusions of the thesis can be summarized as follows:

• From the experiment, it has been observed that HSS tool is not suitable for finish turning (corresponding to N7 surface roughness) of grey cast iron. However, non-coated carbide and ceramic tool can be employed for finish turning operation of grey cast iron. It is observed that ceramic tool is capable of providing a much lower surface finish compared to carbide tool.

• It is observed that during dry turning of grey cast iron with a ceramic tool at a cutting speed of more than 480 m/min, rapid flank wear occurred and the surface roughness of the machined workpiece increased. At a combination of high speed and high feed, the crater wear was observed along with the flank wear. The crater wear was not seen when the cutting speed was less than 400 m/min.

• It has been observed that the air-cooling has a great influence in reducing the tool wear and increasing the tool life compared to dry turning for various tool-workpiece combinations studied in this thesis. However, the air-cooling does not help in improving the surface finish. In some cases, the air-cooled turning produced slightly higher surface roughness compared to dry turning. It is also observed that air-cooling does not help in reducing the cutting and feed forces.

• It is observed that air-cooling is highly useful in hard turning (workpiece hardness more than 45 HRC) of H13 steel with CBN tool. Air-cooling reduces the tool flank wear, crater wear and the built-up edge during hard turning.

• During the turning of grey cast iron with ceramic tool, it was observed that acceleration of vibrations did not provide any correlation with surface roughness or flank wear.

• It has been observed that the neural network modelling helps in the prediction of surface finish, tool life and cutting forces during turning. The MLP network used for the prediction of surface finish

and tool life in the machining of mild steel with coated carbide tool gives a good prediction with RMS error within 15%. For the prediction of the cutting force in turning of grey cast iron with ceramic tool, an RBF network was used. The prediction accuracy of the model was found within an RMS error of 15%. However, in the case of the prediction of surface roughness and tool life in turning of grey cast iron with ceramic tool, the RMS prediction error was found to be around 27% due to limited and noisy data. Therefore, a novel neural network modelling was developed to suppress the noisy data generated during machining. This improved the accuracy of modelling to an RMS error less than 15%.

• The different neural network models that have been developed in the thesis require lesser number of learning data compared to traditional neural network models. This is because factorial design and effect of factors have been used to produce the learning data in a systematic way. For eliminating the spurious data, a novel methodology has been developed which is found effective in improving the prediction accuracy. Also, a test of hypothesis using Student’s t-test was carried for the analysis of statistical variations of experimental data to asses the results of fitted neural network models.

• It has been observed that neural networks could not be used successfully for the prediction of the tool flank wear based on the measurement of machining forces. However, a simplified procedure that has been proposed for the indirect estimation of tool wear, using the rate of change of wear with respect to cutting/feed forces, can estimate the tool wear in a probabilistic manner.

• In the thesis, some simple strategies have been proposed for the efficient utilization of a cutting tool. It is observed that tool life based on surface roughness in dry turning is more than that in air- cooled turning. Thus, the tool failed in air-cooled turning may be utilized in dry turning. Also, the failed tool can be further utilized by

reducing the feed. The strategies presented in this thesis enhance the effective tool life of a cutting tool.

In the present thesis, a heuristic method has been developed for the optimization of cutting conditions in a finish turning operation. The advantage of this method is that it does not require a priori information of tool life, which is mostly required in conventional optimization strategies.