Sensor Integration For Inflight Icing Characterization .
c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.MMSMM MM A01-16414AIAA 2001-0542SENSOR INTEGRATION FORINFLIGHT ICINGCHARACTERIZATION USINGNEURAL NETWORKSJames W. Melody, Devesh Pokhariyal, Jason Merret,Tamer Basar, William R. Perkins, Michael B. Bragg,University of Illinois,Urbana, IL39th AIAA Aerospace Sciences Meeting and ExhibitJanuary 8-11, 2001Reno, NevadaFor permission to copy or republish, contact the American Institute of Aeronautics and Astronautics1801 Alexander Bell Drive, Suite 500, Reston, VA 20191-4344
c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.AIAA 2001-0542SENSOR INTEGRATION FOR INFLIGHTICING CHARACTERIZATION USINGNEURAL NETWORKSJames W. Melody* Devesh PokhariyalJ Jason MerretfTamer Ba§arf William R. Perkinsf Michael B. BraggfUniversity of Illinois,Urbana, ILThis work advances a neural network that characterizes aircraft ice accretion inorder to improve flight performance and safety. Neural networks have been developedpreviously for use within an ice management system that monitors inflight aircraft icingand its effects upon performance, stability, and control. The previous work has appliedthese networks to stability and control derivative estimates provided by an H parameteridentification algorithm during a longitudinal maneuver. This paper extends those resultsby addressing ice characterization in the absence of pilot input when poor excitation ofthe flight dynamics limits the accuracy of parameter estimates. To compensate for thisshortcoming inherent to steady-level flight scenarios, the neural network presented in thispaper integrates steady-state characterization and hinge moment sensing with parameterestimates. The neural network provides icing characterization in terms of an estimateof the previously developed icing severity factor, rj. Extensive simulation results arepresented that indicate the accuracy of neural network characterization during steadylevel flight in the presence of sensor noise and turbulence over a broad range of flight trimconditions and turbulence levels. Furthermore, the relative utility of each informationsource is investigated via consideration of network accuracy of networks trained only onthat information source.INTRODUCTIONCurrent aviation research and development has begun to focus more upon creating aircraft that are safeand reliable during severe weather conditions. Aircrafticing is a large area of concern due to the detrimentaleffects of accumulated ice upon aerodynamic performance. Small amounts of ice can have an extremeimpact upon aircraft dynamics and consequently, icing has been one of the most visible causes of severeaccidents. Icing was determined to be a factor in803 aircraft accidents that occurred between 1975 and1988.l Nearly half of these accidents resulted in fatalities. Commercial accidents such as the AmericanEagle ATR-72 crash near Roselawn, Indiana, whichkilled 68 people in October 1994, have also led to national recognition of icing problems.2 In response tothe abundance of aircraft icing accidents, NASA and* Graduate Research Assistant, Department of Electrical andComputer Engineering and the Coordinated Science Laboratory,Member AIAA.t Graduate Research Assistant, Department of Aeronauticaland Astronautical Engineering, Member AIAA.* Professor, Department of Electrical and Computer Engineering and the Coordinated Science Laboratory.§ Professor and Head, Department of Aeronautical and Astronautical Engineering, Associate Fellow, AIAA.Copyright 2001 by Coordinated Science Laboratory, University of Illinois. Published by the American Institute of Aeronauticsand Astronautics, Inc. with permission.the President's Commission on Aviation Safety and Security have developed an aviation safety research planthat places a high national priority upon icing protection and prevention.3Most icing-related accidents occur because ice accretion affects the performance and stability of anaircraft by altering the shape of its aerodynamic surfaces. Other icing incidents include engine failure andpropeller ice, but this work will focus only on the effects of airframe icing. Currently, there are two mainapproaches that deal with the dangers of ice accretion.First, pilots are given complete weather informationbefore and during flights in order to avoid potentialicing conditions. Second, aircraft are thoroughly deiced before take-off and then operate an ice protectionsystem (IPS) to accomplish in-flight ice removal.An IPS functions in either an advisory or primarycapacity. Advisory systems rely upon the flight crewto activate ice protection devices based upon data received from icing and environmental sensors. On mostcommuter aircraft, icing sensors are not available andpilots determine the level of ice accretion by visualinspection of the wings and control surfaces. Thistype of visual ice detection is inadequate because pilots usually cannot see all of the wing or any of the tail.Systems that function in a primary capacity utilize information from icing sensors to automatically activate1 OF 9AMERICAN INSTITUTE OF AERONAUTICS AND ASTRONAUTICS PAPER 2001-0542
c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.Ice ryInformation!- .Pilot InputEnvelopeProtectionIce Protection Ice EffectsSystemIce ol AdaptationPrimary IPS OperationFig.1Aircraft icing encounter model.anti-ice and de-icing devices. The flight crew is givenupdates concerning IPS status and may manually override the system. Current ice protection equipmentconsists mainly of devices that bleed hot engine exhaust onto the wings to prevent icing or inflatableboots that break off accumulated ice.Recent icing accidents have shown that the IPS approach does not always adequately provide safe andreliable flight during icing conditions. In fact, theATR-72 accident resulted from ice that accreted aftof the wing de-icing system.4 In response to the deficiencies of the current IPS, a new approach has beenintroduced.2 This approach adds a new Ice Management System (IMS) that works in cooperation withthe existing IPS. The purpose of the IMS is to continually monitor ice accretion and its effects, automaticallyoperate the existing IPS and provide the flight crewwith an assessment of the aerodynamic performance.It may also adapt flight controls to allow safe and reliable flight through icing conditions.Ice Management SystemThe IMS approach is being developed by the University of Illinois Icing Center and is a cooperativeeffort among researchers from several disciplines including control systems, aerodynamics, flight dynamics, and human factors.2 The objective of the IMS isto provide an additional layer of defense that guardsagainst aircraft icing accidents. This objective is accomplished by monitoring ice accretion and its effectsupon aircraft flight dynamics. The IMS works in cooperation with existing ice protection systems and theflight crew in order to effectively use all available data.A block diagram depicting the operation of the IMSduring an icing encounter is shown in Figure 1. Thesolid lines portray the current state of the art and thedashed lines show the additional capability providedby the IMS. The IMS adds complexity but also createsan aircraft that is much more robust to the effects ofice accretion.The purpose of the IMS is to allow safe and controllable flight when hazardous icing conditions cannot be avoided. The IMS accomplishes this task byperforming three main functions. First, ice must bedetected and then classified in order to determine thedetrimental effects upon aircraft stability and control.Second, the IPS must be automatically activated andoperated while the IMS provides the pilot with continual updates concerning aircraft performance. Third,the flight envelope and aircraft control laws may beadjusted when severe ice creates potentially uncontrollable conditions. This allows the IMS to protect theaircraft against traditional icing handling events suchas roll upset and tailplane stall. Following envelopemodification, the flight crew will be notified and canthen successfully navigate using limited maneuvers until the icing conditions are eliminated.Effective IMS performance depends heavily uponaccurate detection and classification of icing events.Since icing is a concern precisely to the extent thatit affects the flight dynamics, parameter identification of the flight dynamics is a critical element ofthe IMS.5'6 Exhaustive simulation results have demonstrated the superior robustness and convergence properties of H identification techniques in the presenceof disturbances and measurement noise; hence in thiswork an H parameter identification technique hasbeen adopted wherein stability and control derivativescritically related to icing are estimated. Along with2 OF 9AMERICAN INSTITUTE OF AERONAUTICS AND ASTRONAUTICS PAPER 2001-0542
c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization.these parameter estimates, a rich array of measurements are available that contain information on theicing degradation, including steady-state characterization, hinge moment sensing, and environmental, aerodynamic, and icing sensor information. The IMS alsoincorporates this information in the "sensor fusion"function referred to in.2 Based on this information, theIMS provides (i) an initial indication of the presence ofice accretion that relies heavily on ice probe measurements, if they are available, followed by (ii) a characterization of the type and severity of the degradationof the flight dynamics, where by "type" we mean todiscriminate between, for example, effects leading totailplane stall and those leading to roll upset. Basedupon this characterization, the flight envelope will beadjusted and control laws may be reconfigured.ICING CHARACTERIZATIONIn,7 we have advocated a neural network that characterizes degradation of the aircraft flight dynamicsbased upon sensor data and parameter estimates. Thischaracterization may be accomplished through a variety of other methods, but neural networks are usedbecause of their ability to extract information simultaneously from multiple data sources that depend on thedesired information in a complex manner. It is alreadyknown8 that a feed-forward neural network with atleast one hidden layer is able to approximate any continuous function to an arbitrary level of accuracy onany bounded set given ideal training. Neural networksalso have inherent parallel properties that provide a robust and fault-tolerant structure. Networks are practical for aircraft applications because, following initialtraining, they process information very rapidly. Rapidcomputation can be achieved because the majorityof mathematical operations involve addition, subtraction, or multiplication.Previous work by the authors, reported in,7 appliedneural networks to icing detection and classificationduring a normal operational maneuver modeled as anelevator doublet. In that case, the neural networksincorporated only longitudinal stability and controlderivative estimates, as the estimates were fairly accurate even in the presence of disturbances and measurement noise due to the excitation of the maneuver.In this limited scenario, neural networks were found toprovide an accurate icing indication along with a lessaccurate but still sufficient classification of the icingseverity.In this study, we address the more common steadylevel flight conditions, where the absence of excitationdue to pilot input limits the effectiveness of parameteridentification. Even so, it has been shown that excitation due to turbulence can be exploited to provideuseful parameter estimates, albeit estimates that converge much more slowly.6 In general, a longer delayin icing indication is more acceptable during steadylevel flight than during a maneuver since precipitation of an icing event is less likely in the absence ofpilot action. Moreover, in this paper we address thesensor fusion function of the IMS by incorporating,in simulation, information not taken advantage of inthe previous work.7 Whereas environmental and iceprobe measurements primarily provide information onthe rate of ice accretion, increased hinge moments andsteady-state effects provide information on icing degradation.9 By steady-state effects, we mean specificallychanges in trim conditions consistent with increaseddrag and decreased lift characteristic of icing eventsduring flight conditions with minimal aircraft accelerations (e.g., steady, level flight). The unpredictability ofice shedding and the highly complex nature of the dependence of flight dynamics degradation on the shape,roughness, and location of ice makes correlation of theice accretion rate and the flight dynamics degradationdue to icing difficult. While future efforts will be madeto incorporate traditional atmospheric and ice probemeasurements, we are focusing now on demonstratingthe feasibility of the more novel aspects of the IMS approach. Hence in this paper we incorporate only thesteady-state characterization and hinge moment sensing along with the parameter estimates into the icingcharacterization. Finally, we again restrict our attention to longitudinal dynamics.Neural Network ApproachAs with the previous neural network results, we takethe icing severity factor rj as a measure of the degradation of the flight dynamics due to ice accretion.However, for the sake of clarity, we normalize rj by thenominal icing condition that corresponds to the NASATailplane Icing Program simulated icing condition.10*Since rj is actually a measure of the cumulative potential atmospheric icing severity for an aircraft, the useof 77 as a measure of the flight dynamics degradationis consistent with the assumption that no ice shedding(due to activation of the ice protection system, for example) occurs. Furthermore, we are largely ignoringthe complex relationship between atmospheric conditions on the one hand and flight dynamics degradationon the other. At present, the icing severity factor isthe best measure of icing degradation available, andhence we adopt it despite its limitations in this capacity. As in,9 we address icing encounters duringperiods of steady, level flight wherein the icing severity increases linearly over ten minutes from an initialclean condition.A block diagram of the IMS is depicted in Figure 2.In the upper left corner of the figure, the flight dynamics are subject to the unknown turbulence andmeasurement noise input. As discussed in5'6 theseunknown exogenous signals fundamentally limit the*In9 this nominal icing condition corresponds to a value of77 0.0675.3 OF 9AMERICAN INSTITUTE OF AERONAUTICS AND ASTRONAUTICS PAPER 2001-0542
c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring terizationNeuralNetworkClean TrimCharacterizationFig. 2IMS neural network block diagram.accuracy of the parameter estimates, and hence mustbe included in any realistic simulation. From the flightdynamics we have measurements of the wing and tailhinge moments (7 and C along with correspondingroot-mean-square hinge moment measurements C%rrn8and Cfrrrns. The hinge moment model is describedin.11 From the flight dynamics, we also have measurements of total velocity V, angle of attack a, pitch rateg, pitch angle 0, and elevator angle SE- From thesemeasurements we estimate the flight dynamics trimvelocity V", trim angle of attack a, and trim elevatorangle 1 by lowpass filtering the corresponding measured signals at 1/30 Hz. This estimated trim is usedby the dynamic parameter identification as well as being provided to the neural network directly, since theH ID algorithm is based on a dimensional derivative flight dynamics model5 and hence depends uponthe flight dynamics trim condition. The trim conditions are also used to calculate the expected clean S/Cderivatives M and M% against which the estimatedS/C derivatives Ma and Mq must be compared inorder to ascertain icing degradation. Furthermore, expected clean-aircraft trim values Vc, ac, and 8CE mustbe calculated to provide a reference for interpreting theestimated trim values. In the following simulations,the initial trim values are used as expected clean trim,consistent with steady, level flight scenarios. Whenmore general scenarios are considered in the future, amore sophisticated calculation of expected clean trimwill have to be included. Since the dynamic identification algorithm is expected to perform better underlarger excitation and the steady-state characterizationis expected to have better accuracy under smaller excitation, we have also calculated measures of excitationin order to aid the neural network in discriminating between dynamic parameter ID and steady-statecharacterization. As excitation measures we use anestimate of the dynamic portion of the power of themeasured signals, denoted as Pa,Pg, and PSB andcalculated according toP« (a - a)2,andThe excitation measures and all other input to theneural network are lowpass filtered by batch averagingeach signal over the past 1/2 second in order to reducethe effect of measurement noise. Finally, the neuralnetwork provides an estimate of the icing severity r\at any given time instant t based only on the lowpassfiltered input at that time t. In summary, the neuralnetwork incorporates the following batch filtered input hinge moment measurements: wing and tail hingemoments CJ[ and C and rms wing and tail hingemoments C%rms and firms'steady-state characterization: estimated trim values V\ a, and SE and expected clean trim valuesFc, ac, and S4 OF 9AMERICAN INSTITUTE OF AERONAUTICS AND ASTRONAUTICS PAPER 2001-0542
c)2001 American Institute of Aeronautics & Astronautics or Published with Permission of Author(s) and/or Author(s)' Sponsoring Organization. dynamic parameter identification: estimated S/Cderivatives Ma and Mq and expected clean S/Cderivatives M and M* excitation measure: Pa, Py, Pg, and P In order to be useful, the neural network mustprovide accurate icing characterizations over a broadrange of trim conditions and turbulence intensity. Furthermore, the icing characterization network must accurately identify clean aircraft in order to avoid falsealarms. The neural networks are applied in 600-ssimulations modeling a rich set of steady, level flightscenarios that correspond to combinations of the following variations final icing severity levels of 0, 0.02, 0.04, 0.06,0.08, or 0.1, initial total velocities of 60, 65, 70, or 80 m/s, tail-only icing or wing and tail icing, turbulence level standard deviations of 0, 0.15, or0.3 /, and two each of turbulence and measurement noisesample paths for all cases.Bach simulation was run under the assumption of constant engine power, with a zero flight path angle (z.e.,level flight), and at an altitude of 2,300 m. Simulations were performed with the aid of the FDC Matlab/Simulink toolbox as described in.9 Furthermore,measurement noise consistent with Twin Otter instrument accuracy specifications, listed in Table 1, wereincorporated as zero-mean, bandlimited white Gaussian noise. The simulations provided all measurementinformation at a 30 Hz sample rate.sets through the neural network and adjusts variousweights and biases until the most accurate output isgiven for all training sets.Several networks were trained and tested in orderto investigate the performance of the proposed IMSas well as the utility of the various sets of measurement information. The f
SENSOR INTEGRATION FOR INFLIGHT ICING CHARACTERIZATION USING NEURAL NETWORKS James W. Melody* Devesh PokhariyalJ Jason Merretf Tamer Ba§arf William R. Perkinsf Michael B. Braggf University of Illinois, Urbana, IL This work advances a neural network that characterizes aircraft ice acc
SAE AS 6332 (Quality Assurance Program for De-Icing/Anti-Icing of Aircraft on the Ground) SAE ARP 6257 (Aircraft Ground De-Icing/Anti-Icing Communication Phraseology for Flight and Ground Crews) SAE ARP 1971D (Aircraft De-Icing Vehicle - Self-Propelled) SAE AMS 1424/1 (De-Icing/Anti-Icing Fluid, Aircraft, SAE Type I)
understanding of how travellers use the inflight Duty Free channel, on a regional and demographic basis, and more specifically, insights into: the inflight shopping behaviour the key barriers to inflight Duty Free browsing and purchasing the influence of other Duty Free purchases on inflight usage
Bruksanvisning för bilstereo . Bruksanvisning for bilstereo . Instrukcja obsługi samochodowego odtwarzacza stereo . Operating Instructions for Car Stereo . 610-104 . SV . Bruksanvisning i original
Divide between the cases using a spoon, scraping it off with a knife. Ask a grown-up helper to put the tray in the oven for twenty minutes. Step Five Mix the butter and icing sugar to make a creamy icing. Add colouring, if you like. Push an icing nozzle into an icing bag, then scoop the icing into the bag. Step Six
D 1 Blonde Icing / Blonde Glam Stylist Guide E 1 Blonde Icing “Open the Door to Luxury” Business Opportunity Guide F 1 Blonde Icing / Blonde Glam Backbar Wall Chart G 1 Blonde Icing / Blonde Glam Easel Card with pocket H 25 Blonde Icing / Blonde Glam Client
RS-422: Hexidecimal 24-byte string (ASCII format) 9600 Baud (1 Start Bit, 8 Data Bits, No Parity, 1 Stop Bit) RS422 Outputs: for Icing and Status No Icing – 0, Icing – 1 Status OK – 0, Status Failure – 1 Icing Signal Period: 60 second activation from start of icing measurement (Discrete or R
the inflight industry and usher in a new era of innovation, to make inflight connectivity fast, reliable, and easy-to-access. This paper begins with a review of the current challenges with inflight connectivity and the passenger experience. Next, the
Textbook of Algae , O. P. Sharma, Jan 1, 1986, Algae, 396 pages. Aimed to meet requirements of undergraduate students of botany. This book covers topics such as: evolution of sex and sexuality in algae; and, pigments in algae with their. An Introduction to Phycology , G. R. South, A. Whittick, Jul 8, 2009, Science, 352 pages. This text presents the subject using a systems approach and is .
