Seite - 12 - in Adaptive and Intelligent Temperature Control of Microwave Heating Systems with Multiple Sources

1. Introduction of thedistributedfeedingsources iscalculatedbasedontheestimated model. Theentire temperaturedistribution iseffectively improvedby driving separate temperatures to the predefined target temperature. This is the most common control strategy used in industrial applica- tions. It is suitable for heating scenarios like the heating of multiple small workpieces or situations where the whole temperature profile oftheworkpieceisnotavailable. Differentsystemmodelingmethods, includingthestate-spacemodel [Dor95]andtheneuralnetwork(NN) based models [Hay98], are constructed. The model predictive control (MPC) [MHL99] and neural network control (NNC) [CK92] methods areapplied, respectively. The second intelligent temperature control approach (figure 1.2b ) is basedonthethermalimagefromtheinfraredcamera. Informationab- stracted from the thermal image, such as the maximum temperature, the minimum temperature and their corresponding positions, is used for the control. Unlike the conventional input-output models used in theadaptivecontrolsystem,aMarkovdecisionprocess(MDP)[Bel57] is constructed to simulate the power-temperature state transient re- lation in the intelligent control system. The final control objective is to limit the temperatures of the entire workpiece within a prede- fined range, and hence, obtain a homogeneous temperature distribu- tion. Conventional control methods are not appropriate for this task, therefore a reinforcement learning based intelligent controller [Bar98] isusedhere. Comparing above two temperature control approaches with temper- ature controllers proposed in other papers such as [SAB98] [SBA00] [RCVI99] [HS07], the novel adaptive temperature concept developed in this dissertation has evident advantages in many aspects. Benefit- ing from the MIMO heating and control systems, it is possible to not only control the heating rate but also improve the temperature homo- geneity inreal time. Theonlinesystemidentificationmethodsusedin this control concept able to estimate influences from different factors numerically, without knowing any specific value of the influencing factor, which makes them more flexible be implemented in practical heatingapplications. 12