| Weathering test occupies a pivotal position in the aging test of automotive coatings. It can be the most close to simulation the actual use state of product. It’s authenticity、reliability is another artificial accelerated tests can not be compared, it’s the final resort of test the car paint anti-aging properties. However, the test period is too long has been a drawback for weathering test, along with the high development speed of automobile coatings products, automotive coatings products R&D- production- elimination cycle is very short, too long weathering test cycle is not suitable for product development and design. In order to accelerate the cycle of weathering test, people used a lot of accelerated weathering test device. Such as: the exposure rack which can accelerate corrosion by spraying brine, the sun tracking exposure rack which can enhance solar radiation and the sample surface temperature, accelerated exposure device of enhanced condenser by 10 flat mirrors. While these devices can be varying degrees of speed up the natural aging test cycles, but often still can not meet the rapid product development process.In this paper, we try to use the property that neural network can quickly and accurately fit the characteristics of nonlinear function, establish model by using a large number of different types of car coating weathering test data. It’s significance lies in the fact that it can be used to predict the late changes in the later period of the test, and to provide an important guidance for the development and design of automotive coatings. In order to achieve this goal, the main work of this paper is as follows: 1.how to determine the acceptability of the prediction results; 2.how to determine the kind of neural network to model; 3.how to determine the neural network input layer and the output layer.First of all, the paper introduces the color theory of the automobile coating, which leads to the concept of the critical region of the human eye. According to the standard of the color of the automobile coating, it’s acceptable that the prediction error of chromatic aberration is ±1.0.Secondly, the typical artificial neural network is used to predict is the BP neural network, RBF neural network, GRNN neural network, in order to select the appropriate neural network, this paper chose two complex functions, to compare the function approximation ability of these three kinds of neural networks. The results show that the approximation ability of GRNN neural network to two kinds of function is better than BP neural network and RBF neural network, So we chose to use GRNN neural network to predict chromatic aberration of automotive coatings weathering test.Again, in the process of GRNN modeling, the 4 sets of experiments were designed. The chromatic aberration data of 8 months, 12 months, 16 months and 20 months were the network input, the data of 24 months were the network output. The accuracy of the 4 groups was 70%, 70%, 80% and 90%. In the research and selection of automotive materials, it can be acceptable that the prediction accuracy of material’s aging properties can reach 70%,and the results are positive for the selection of product development.Although the accuracy of data prediction with the 20 months is the highest, but it is Pointless that to predict the test data of 24 months.Therefore, this paper will choose the 8 months which is shortest test period for the network input node.Finally, using GRNN to build model, choose 2 years chromatic aberration data of 520 different types of coating as the network learning samples, the samples are the current mainstream automotive coatings products. After learning and training, the network has been proved to be qualified, then predict 30 groups sample which did not participate in the network learning, the accuracy of the results reached 80%, and it’s proved that the use of GRNN modeling can be used to predict the chromatic aberration data of automotive coatings of weathering test. |
PDF Full Text Request