A fuzzy logic controller(FLC) and a fuzzy logic filter(FLF), which have a special type of fuzzifier, inference engine, and defuzzifier, are applied to the water level control of a nuclear steam generator (S/G). It is shown that arbitrary two-input, single-output linear state feedback controllers can be adequately expressed by this FLC. A procedure to construct stability-guaranteed FLC rules is proposed. It contains the following steps: (1) The stable sector of linear feedback gains is obtained from the suboptimal concept based on LQR theory and the Lyapunov's stability criteria; (2) The stable sector of linear gains is mapped into two linear rule tables that are used as limits for the FLC rules; (3) The construction of an FLC rule table is done by choosing certain rules that lie between these limits. This type of FLC guarantees asymptotic stability of the control system. The FLF generates a feedforward signal of S/G feedwater from the steam flow measurement using a fuzzy concept. Through computer simulation, it is found that the FLC with the FLF works better than well-tuned PID controller with variable gains to reduce swell/shrink phenomena especially for the water level deviation and abrupt steam flow disturbances that are typical in the existing power plants.
A neurofuzzy logic controller (NFLC), that is implemented by using multi-layered neural network to have the same function as the FLC discussed above, is designed. The automatic generation of NFLC rule table is accomplished by using back-error-propagation (BEP) algorithm. There are two separated paths at the error back-propagation in the S/G. One is to consider the level dynamics depending on the tank capacity, and the other is to take into account the reverse dynamics of S/G. The amounts of error back-propagated through these paths show opposite effects to the BEP algorithm each other at the swell/shrink phenomena. Through the computer simulation, it is found that the BEP algorithm adequately generates NFLC rule tables according to given learning parameters.
Although the FLC introduced above can guarantee the stability of the control system, the FLC have a structural handicap that the number of rules to be determined goes numerous by adding new input variables. Thus the number of input variables are normally limited by two, or separated rules are developed for the other major variables. This means that the FLC cannot use all of the available information necessary to make the control system robust. In order to overcome these difficulties, a robust fuzzy logic gain scheduler (robust FLGS) is designed based on the synthesis of fuzzy logic inference and $H_∞$ technique which is one of the most commonly used one in the field of robust control system design. The robust FLGS has a set of robust control gains in a gain pool. The fuzzy inference engine chooses one of the most desirable control gain considering the S/G operatin condition. The robust gains in the gain pool are determined by the time domain $H_∞$ control technique such that both of the weighted $H_∞$ norms of the two transfer functions are minimized. One of the transfer functions is from the steam flow disturbance to the mass capacity portion of S/G level, and the other is from the steam flow disturbance to the reverse dynamics of S/G. The robustness of FLGS in the face of S/G parameter variation is checked. The result shows that the robust FLGS with $H_∞$ control gain is stable as long as the sign of S/G parameter is not changed due to variation. The FLGS rules are automatically generated using a genetic algorithm (GA). The genetic algorithm (GA) gives a set the gain scheduling rules that reflect given performance design specification well. The computer simulation confirms that the proposed controller shows both of guaranteed stability and good performance in terms of small overshoot and fast settling time in spite of S/G parameter variations and steam flow disturbances.
