| As one of the three pillar industries, material is the basic needs of human survivaland development. With the rapid development of modern science and technology, thesurface treatment technology of material is also put forward higher and higherrequirements. Therefore, improving the surface properties of the material and thensaving resources is particularly important, no doubt the development of thin-filmtechnology with high value in use. The rigid film material which is an important newmaterial has been widely used in various industries for the rigid film materials possessexcellent properties, such as high hardness, low friction coefficient, high chemicalstability, etc. Therefore, in order to improve the surface properties of the material so asto enhance its life, the rigid film material has become an important research object ofmany researchers. Hard film material has been widely used in the tool and moldindustry, but there are still many deficiencies, which need to be improved by furtherimprovement and investigation. In the process of the preparation of rigid film material,each experimental process parameter directly affects the quality of the fabricated film.Therefore, how to effectively analyze the experimental data and then to select areasonable design of experiment is particularly important.In this thesis, the support vector regression (SVR) modeling theory combining withparticle swarm optimization (PSO) algorithm to optimize the model parameters, wasintroduced to model and analysis the process parameters on the hardness and thebinding force of the TiN film plated by ZL109Multi-arc ion plating, the depositingcondition on the hardness and thickness of TiN/AlN multilayer film deposited viapulsed laser deposition, and depositing parameters on the friction coefficient anddeposition rate of DLC rigid film deposited by magnetron sputtering deposition. Themain contents are as following:(1) SVR-LOOCV strategy was employed to model and analysis the influence of4process parameters (temperature, sputtering time, deposition time and negative bias) onthe on the hardness and the binding force of the TiN film plated by ZL109Multi-arc ionplating. The result revealed that even in few samples, the SVR regression model stillpossesses very high accuracy. And it was found that maximum hardness of the TiN filmmay reach1548HV0.05under the condition of temperature≈268℃, sputtering time≈8.6min, deposition time≈28.4min and negative bias≈240V. (2) Based on the experimental dataset of TiN/AlN multilayer films ablated on themonocrystalline silicon substrate via pulse laser deposition (PLD) technique, the SVRcombined with PSO was proposed to construct models for prediction the thickness ofAlN thin films and hardness of TiN thin films in TiN/AlN multilayer films depositedunder different process parameters. The predicted hardness of AlN films via establishedSVR model is compared with that obtained by immune radial basis function (IRBF)neural network using identical training and test samples. It is demonstrated that the SVRmodel possesses higher prediction accuracy and better generalization capacity thanIRBFNN. The established SVR model for TiN hardness was further employed toanalysis and optimized the PLD deposition process, and was utilized to depict theinteraction influences of multi-factor on the hardness of deposited TiN films. It waspredicted via the constructed SVR model that the available maximum hardness of TiNfilm may be38.6GPa under an optimal process combination.(3) The SVR was also applied to model and analyze the effect of five processparameters on the friction coefficient and the deposition rate of the Zr-DLC filmdeposited by an unbalanced magnetron sputtering system. The result indicated that theestablished SVR model own higher prediction and generalization abilities. It was foundthat the friction coefficient of the Zr-DLC film may be reduced to0.0875under anoptimal combination of target voltage≈-40V, target current≈0.816A, pulse frequency≈89.2kHz, flow rate of methane≈6.6sccm and working distance≈12.6cm, which issmaller than that (0.112) deduced by a grey fuzzy Taguchi approach. In addition, theSVR model was further used to calculate the sensitivities of friction coefficient of thefabricated films on each process parameter, and it was illustrated that the sensitivitybecome more and more weak according to the following order: working distance>pulse frequency> target current> flow rate of methane> target voltage.This investigation demonstrates that using support vector regression to study theexperimental data of preparation rigid film material can effectively explore and analysisof the experimental data, may provide scientific theoretical guidance for the design ofexperiments so as to save manpower, material resources financial resources and time,thus it would play a very important role in the optimization experimental design. |
PDF Full Text Request