Page - 184 - in Adaptive and Intelligent Temperature Control of Microwave Heating Systems with Multiple Sources

6. SummaryandConclusion Finally, all aforementioned control methods are tested in practical ex- periments,anddifferentpropertieshavebeenobservedfor individual controlmethods. For the adaptive control scheme, both MPC and the NNC are able to letmultipletemperaturesconvergeatthetargettemperature. Thefinal temperaturedistributionandthetemperaturewindowinbothcontrol methods are significantly improved compared with the conventional PID controller. The control models used in the MPC method imply that they are well suited to materials and setups with lower ther- mal conductivities or multiple independent workpieces. The NNC method can be used in more general situations without special re- quirementsonthepropertiesof theheatedmaterial. For the adaptive control scheme, the intelligentQ(λ) based reinforce- ment learning controller also shows great control performance, re- flectedbyitseffectivecontrolof thetemperaturewindowbetweenthe maximum and the minimum temperatures. It is suitable for applica- tions where the entire temperature profile can be monitored in real- time. Compared with conventional control schemes, it is more pow- erful because the temperature distribution can be directly controlled andimproved. Themaincontributionsof thisdissertationare: • Thenonlinearstate-spaceandtheneuralnetworkmicrowaveheat- ing models developed in this dissertation provide powerful al- ternatives to the traditional linear microwave heating models (as in [HPE97] and [RCVI99]). The dynamics of the distributed mi- crowave heating systems are accurately described and estimated bythesemultiple-inputmultiple-output(MIMO)modelsthathave beenconstructed in thisdissertation. • Based on the MIMO models, different MIMO control systems are designed and applied to control the distributed microwave heat- ing system - HEPHAISTOS. The temperature homogeneity (re- flected by the final temperature window ∆T) obtained using the the advanced MIMO control system (∆T = 4◦C - 6◦C at 100◦C forbothMPCandNNC)ismuchbetter thantheconventionalPID controller (∆T ≥ 13◦C at 100◦C). The experimental results pre- sented in this dissertation provide evidence that the heating per- 184