part of the central composite design.All of the process factors are confirmed to have an influence on the two targeted factors under the significance level of 0.05. The hardness average is especially affected
by three process factors even under the significance level of 0.025 and the influence of the liquid fraction is observed to be most dominant. The priority of influence of the remaining two factors is in the order of the compression holding time and the die temperature. However, their influence is determined to be relatively lower than that of the liquid fraction. As for the standard deviation of hardness data, the influence of the die temperature and the compression holding time is shown to be dominant and that of the liquid fraction is resolved to have a smaller
influence on the target factors.
5 Nonlinear regression analysis
5.1 Process optimization using the NN decision maker and the genetic algorithm
After the multi-layer NN decision maker is established by the back error propagation method, with the decision making pattern, it is used for evaluating the raw fitness values of the genetic algorithm.The result of the optimization is revealed in Table 4. It is observed that the optimal value of the liquid fraction exists within the range of the experimented condition, and the load dwell time is suggested to have values as large as possible to improve the quality of the final products. This result means that the load dwell time has a positive and a negative correlation to the hardness average and the standard deviation of the hardness, respectively.
5.2 Verification experiments and microstructural aspects
Verification experiments are performed to confirm the results of variation analysis, the surface response models and the obtained optimal condition of the process parameters. As the result of verification experiments in the optimal condition show only small discrepancy from
that of the surface response model, the variation analysis and surface response analysis performed are found to be valid. Figures 8 and 9 show the microscopic views of the piston product obtained from both the former experiments and the verification experiment. The microstructure in Fig. 8 has been taken from the inner portion of the piston formed at a die temperature of 200°C, a liquid fraction of polynomial regression analysis was performed while setting three process factors as the independent variables and two target factors as the dependent variables. To visualize the surface response equations, one of the factors is fixed and the surface of the equation is demonstrated in three dimensions in Figs. 6 and 7. Graphical demonstration shows that low liquid fraction
and long compression holding time yields high hardness average and high standard deviation, while die temperature does not have a clear correlation with hardness average.
6 Conclusion
In this study, experimental analysis of the thixoforging process for manufacturing automotive pistons is performed by design of experiment and surface response analysis techniques. In addition, optimization of process parameters is achieved using the NN decision maker and genetic
algorithm. Conclusions reached with this study can be summarized as follows.
1. From variation analysis based on the data of central composite design, all process parameters are found to affect every target factor such as hardness average and standard deviation of hardness under the significance level of 0.05.
2. Surface response analysis of the experimental data provides the valid mathematical models and variation analysis of regression affirmed validity of the surface response model.
3. Due to the NN decision maker, the supervisor’s decision is effectively reflected on the dual-objective optimization problem arising from the study and its determination of optimality becomes a bit more obvious. Genetic algorithms are successfully coupled with the NN decision maker because of their flexibility in the form of objective functions. 铝合金汽车活塞制造英文文献和中文翻译(5):http://www.youerw.com/fanyi/lunwen_2047.html