Soft Computing Applications on SR-30 Turbojet Engine

AIAA 2004-644415, the result from the second attempt is shown. Asteady state was achieved in about 5.0 seconds with anear no-steady error. Although the settling time waslonger than the one observed in the simulation using thePID controller, the performance of the fuzzy controllerfor this case showed an improvement over the first case.Figure 15 shows a good agreement with the simulationresult shown in Figure 12.to Pearl’s treatment of the subject [4]. The specificBBN approach to be employed in this effort is shown inFigure 16 and its basic procedure is beyond the scope ofthis paper. The basic steps are listed below.λ X (u)Uπ X (u)Thrust [lbs] and Vavle Position [deg]5.04.54.03.53.02.52.01.51.0desired thrustthrustvalve position161 163 165 167 169 171 173 175Time [sec]Figure 15: Step Response from SR-30 Engine Case 2V. BAYESIAN BELIEF NETWORKSCONTROLLER DESIGN [2]For the engine start phase, the primary softcomputing technology to be utilized is Bayesian beliefnetworks (BBN). The sole intent of the BBN is toqualify each of the states during engine start-up prior toreaching main-stage. This will further assure certaintyin the health of the engine and proceeding into mainstagein addition to providing added assurance intopreventing any premature engine shutdowns.BBN’s have been proven to be good predictive anddiagnostic mechanisms for reasoning about the state ofevents in environments where uncertainty is universal.Suppressing the details, the genealogy of this SCT isstrongly rooted in classic statistical Bayesian inferencetheory where a subjectivist viewpoint is taken. In shortBayesian inference uses a different interpretation ofprobability where one’s degree of belief in some eventis part of the reasoning. BBN’s are computationalarchitectures that permit declarative (prior conditionalprobabilistic values) and subjective opinions (posteriorprobabilistic values) about world (factual) knowledge tobe part of the reasoning and assessment through avisual network representation and a unique syntacticmessage-passing feature. Furthermore, the visualnetwork makes it easier to view the top-down cause andeffect (or condition to consequence) relationships. For amore detailed account of this SCT, the reader is referred8American Institute of Aeronautics and AstronauticsV X WYλ Y (x)π Y (x)π Z (x)λ Z (x)Figure 16 Bayesian Belief Network1. Belief Updating – When node X is activated toupdate its parameters for belief updating, it firstinspects all messages transmitted to it by its parent (π)and its children nodes (λ). Then using all input, itupdates its belief.2. Bottom-up Propagation – Using messagestransmitted by Y&Z, compute message to transmit toparent node U.3. Top-down Propagation – Node X then computesnew messages to be sent to its children nodes Y&Z.A first design of the Bayesian Belief Networks forthe SR-30 Engine is presented in the Figure 17a and17b. Table 4 gives the initial design for the BBNalgorithm.LegendVirtual NodeUnidirectional BeliefPropagationBidirectional BeliefPropagationContinuationBoundary BetweenSequence &Belief NetworkAirSwitchFuelSwitchEngineVibrationOil/FuelLeakageSwitchStatusIgnitionSwitch3 Sec.TimerTempLimitsMasterKeySwitchEngineStartInitiatedTerminateEngineStart Seq.Figure 17a Bayesian Belief Networks forSR-30 Turbo Jet Engine Start-UpThe design required analysis of the engine-start dataand careful choice of the parameters for the design. ForZOilPressureFuelSwitchElectricMasterSwitchEngineStartInitiated