Seite - 168 - in Adaptive and Intelligent Temperature Control of Microwave Heating Systems with Multiple Sources

5. ExperimentalResults 5.25b). Theproblemiscausedbytheerrorfromeitherthesystemiden- tification or the GA controller. The high control input dimension of HEPHAISTOS is a large computation burden for both of them, espe- cially for the GA algorithm. The overall searching space in the GA algorithm increases exponentially with the input dimension. When all other parameters (such as the number of generations used and the numberoftheindividuals ineachgeneration)arefixed,ahigherinput dimensionleadstoamuchlargersearchingspaceintheGAalgorithm, which correspondingly lowers the probability that a good control so- lution can be obtained. In practical experiments, the most simple and effective way to reduce the temperature overshoot and enhance the entirecontrolperformance is touse lessheatingsources. For example, in the experiment shown by figure 5.28, all heating sources below the metal table are not used during the heating pro- cess (in each module only sources No. 2, No. 4, No. 10 and No. 11 are switched on, see figure 2.13). Overall 8 heating sources are used for the controlling. On the one hand, the computation burdens for both the system estimator and the controller are largely reduced. On the other hand, the total 8 sourcesstill guarantee a huge control space and control diversity. With fewer heating sources, the accuracies of thesystemidentificationandtheGAcontrollerbothcanbeimproved. Hence, from the results in figure 5.28, it is clear that both the temper- ature overshoot and the final temperature window are significantly reducedusing lesssources. As aforementioned in the system modeling part (chapter 2) and the beginning of this section, thermal conduction plays an important rule in the model derivation as well as controlling process. In order to fur- ther testhowdifferentconductioneffects influence thecontrolperfor- mance, a different setup was implemented in our experiments, such as in figure 5.29. This setup has the same aluminum plate, vacuum bagging and other components, except one big difference. Instead of one big silicone rubber workpiece, in the second setup 5.29 there are five independent small workpieces. Compared with the first setup 5.21,apparentlythesecondsetuphasamuchlowerinfluencefromthe thermal conduction and convection parts. In other words, all cooling effects includingthethermalconvection,radiationandconductionfor the five workpieces are almost the same. Therefore in such cases, the propertiesofeachcontrolmethodcanbemoregreatlyreflected. 168