NASA Technical Memorandum 88263 An Engineering Approach to the Use of Expert Systems Technology in Avionics Applications Eugene L. Duke, Victoria A. Regenie, Marylouise Brazee, and Randal W. Brumbaugh May 1986 National Aeronautics and ‘Space AdministrationNASA Technical Memorandum 88263 An Engineering Approach to the Use of Expert Systems Technology in Avionics Applications Eugene L. Duke, Victoria A. Regenie, and Marylouise Brazee ‘Ames Research Center, Dryden Flight Research Facility, Edwards, California Randal W. Brumbaugh PRC Kentron, Edwards, California NASA National Aeronautics and Space Administration ‘Ames Research Center Dryden Flight Research Facility Edwards, California 93523AN ENGINEERING APPROACH TO THE USE OF EXPERT sySTEMS TECHNOLOGY IN AVIONICS APPLICATIONS Eugene L. Duke, Victoria A. Regenie, and Marylouise Brazee NASR Anes "Research Center Dryden Flight Research Facility Edwards, California Randal W. Brunbaugh PRC Kenton Edwards, California Abstract This paper presents the concept of using a knoviedge' compiler to transform the knowledge base id inference mechanism of an expert systen into a conventional program. The motivation for this discussion ts the need to accomodate real-time systens requirenents_in applications. such as embedded avionics. The paper presents. an over- view of expert systems and a brief conpartson of expert systems and conventional prograns. Avi ‘onics applications of expert systens are briefly Siscussed before the detailed discussions of applying the proposed concept to example sys- tens using forward- and backward-chaining. Introduction Expert, systens technology offers trenendous potential for the next generation of aviontes Systens. The power of this technology Ives both in its Separation of domain speci ic. knowledge from program control mechanisms and in the devel- ‘opsent methodology and development environnent « However, as expert systems technology moves. out af the research laboratory and into severely ‘demanding applications such as avionics, new problens arise. While the power of expert. sys tens is evident in the laboratory environment, many of the most attractive features of this, technology becone burdensome in applications. Te is argued in this paper that, by treating expert systens a5 development téols for more conventional prograns, many of the prablens emerging in applications may be solved. It s the thesis of this paper that an engineering approach to the use of expert systens technol ogy in avionics application may minimize the need for special purpose computers. The concept of using an expert system as 4 developnent: tool for a conventional program arose from the applications research at. the Dryden Flight Research Facility of the NASA Anes Research Center (AneS-Dryden) and was Suggested by the research at the University of California, Los Angeles (UCLA), under the direction of Professor Jacques Vidal. This research in real-time mechanization: of expert, systems has influenced some of the key fd Presented in this paper. In particular, the doctoral thesis of John Helly [1] has provided 2 unique and original view of the transforma tion of a knowledge base into an equivalent logic representation. At Anes Dryden, expert systems technology is being applied to avionics systems on hight performance aircraft in two projects: the expert system flight status monitor for the X-29 forwsrd- swept-wing atreraft [21,3] and in the joint NASA DARPA automated wingnan’pragram. Both NASA proj- ects focus on the use of expert systens technology in real-tine applications. The application areas, in which this research is being conducted are such that extreme computational denands.are placed on the host computer system. The complexity of nigh- perfornance aircraft place severe demands on expert systens technology. The following sec~ tions present an overview of expert systems. and @ comparison of expert systems and convent tonal Prograns as background before proceeding with ‘the main argument of the paper: expert. systens can be converted automatically into convent onal Programs and the process from development. to deployment retains the best features of both types of systens. It is the thesis of this paper that the knowledge base and inference mechanism of an expert system can be converted into a convent onal rogram using a knowledge compiter. The concept of a knowledge compiler is explained using example forward- and backward-chaining expert. systens. The main benefit of converting the expert system into a conventional program 1s increased execution speed and, hence, reduced processor requirenents. The knowledge compilation method described in this paper is completely compatible with the usual environment available for expert systens develop- ment and allows the system developer to exploit the desirable features of expert systems while developing conventional programs. Overview of Expert Systens Artificial intelligence (AI) ts described fn Ref. a: ‘the part of computer science concerned with designing intelligent computer systems,that ts, systens that exhibit the characteris tics we associate with intelligence tn hunan Dehaviors ss [AL research focuses on understanding the basic Processes of intelligence as well as on computer- based methodologies for solving difficult. problens that would otherwise require human intelligence. The field of AI 1s concerned with a wide range of problem classes that are associated with Intel - Vigence in humans: problem solving, reasoning, understanding language, learning, robotics, auto= matic programaing, and’ vision. The research in problem solving and reasoning has Ted to the developnent of the sudtield of applied AI known as expert systens in which ‘general-purpose reasoning engines (inference mmechanisns) are utilized to reason about domain Specific knowledge in a target. application area. An expert systen is described in Ref. "hn expert system 1% one that has expert rules and avoids Blind search, performs well, Feasons by manipulating symbols, grasps fund3- imental doaain principles, and has conplete weaker reasoning methods ‘to. fall back on when expert rules fail and to use in producing explanations. It deals with gifficuit problems tn a complex domain, can take a problem description tn lay terns and convert it to an internal represent tion appropriate for processing with its expert rules, and it. can reason about. its own knowledge (or lack thereof), especially to reconstruct inference paths rationally for explanation and self-justification." This description of an expert systen is more of a future goal than a current reality. The Sncorporation of fundamental donain principles ‘into what are known as "deep" expert systens 1s at best a research topic; most expert systens have an extrenely limited and_ shallow knowledge base. This lack of knowledge of fundamental principles contributes to the inability of current. generation expert systens to reason about their knowledge 4n general and about the Imttations of their know- edge in particular. However, many of the fea- tures listed in this description do exist in what inight be termed the state of the art in applica tion systems. Figure 1 shows a common (although highly sinplified) representation of the basic expert aystens architecture, In this representation, the main structural features of an expert. system are illustrated: knowledge acquisition facility, Knowledge base, inference mechanism, and input- output system." The knowledge acquisition faciTity $s the main Interface between the expert system and the expert. This facility provides a mech- anism for developing a knowledge base. The know! edge base consists of domain specific knowledge, ‘generally in the form of conditional (1f-then) Files, The inference mechanism reasons about Specific facts using this knowledge base. In this simplified representation of an expert system, the inference sechanisa would also pro- Vide explanations of conclusions reached and rules Used to reach those conclusions. The knowledge acquisition facility, inference mechanism, and ‘input-output system are part of the expert system program. After the specific facts that are to be Feasoned about are provided, the knowledge base 1s treated as a data source that 1s used to dicect the Inference process. Perhaps the most. significant feature of cur- rent expert systems 1s the use of knowledge in the form of expert rules that represent. domain principles. Encoding the knowledge and problen Solving techniques of a domain expert into rules Provides the real power of expert. systems. These Fules are usea by an inference mechanism oth to. reason about specific facts in a given situation land to provide explanations of the deduct tons of the expert system to its user. Figure 2 shows what might be typical rules used in avionics applications, These rules might be used in {ight control systems, guidance systens, and leven more powerful integrated systens such as. those being developed for the pilot's associate Progran. "Figure 3 shows how these sane rules, would be coded in a higher order Tanguage (in Enis case, FORTRAN). As can be seen by comparing Figs. 2 and 3, the representation of knowledge 1s mich ‘clearer and nore eastly verified in the rules (Fig. 2) than in the FORTRAN code (Fig. 3). Expert. Systems and Conventional Programs nen comparing an expert system with a con- ventional program, differences in knowledge rep- resentation, control structure, and operational processes are most obvious. Expert systems use Symbolic representations of knowledge and symbolic Inference; conventional programs use nuneric and logical representations.” An expert system might be said to execute a compilation of knowledge whtle a conventional program might be character ized'as a compilation of pracedires. The struct ture Of an expert system in which the knowledge base and inference mechanism are separate differs from a conventional progran in which ‘the knowledge base and inference mechenisn are essentially combined in the program code. The sain operational differences arise from the ad{1~ ‘ity of an expert systen to offer explanations of its inference process and to have that inference process modified by the addition, deletion, or modification of rules. In a conventional pro- gram, explanation features are lacking and the modification of the reasoning process. Involves Fewriting the code. Obvigusly, the features that characterize an expert system are intertwined and cannot be separated. In addition to the representational, struc- tural, and operational differences, expert sys- tems 3150 differ from conventional programs 1a their developnent. Ina conventional program, the possible lines of ‘reasoning mist be mapped Ut ahead of time and then rigidly encoded into tthe program structure. In an expert systen,because the lines of reasoning are enbodied in tthe knowledge base, the systea can be developed incrementally. The environments tn which these two types of prograns are developed also differ radically. Conventional programs are most often developed in the target programming language with few support tools beyond an editor and, perhaps, 2 debugger. The environnent for expert systens development 1s fundamentally far richer than that for conventional programs. Figure & shows the components of an expert system and illus- trates the benefit of using an expert systens shell in the development process. Expert sys- tens shells are available with inference mech- anisas, 2 knowledge acquisition facility, and explanation capabilities already developed and in place. The knowledge engineer need only con- pile the domain specific knowledge of an expert land enter it into the shell. By using an expert. Systens shell, an expert system can de developed rapidly. Entering only a few key rules into a Shell allows. a. systen to be prototyped quickly and allows early feasibility denonstrat ion. It should be noted at this point that, when discussing expert systens, sone, dist igctian ie. aceastonaly made between development and deliv- ery systems. “The distinction is not entirely clear. A-delivery system must at least have a Knowledge base, an inference mechanism, and an explanation facility. “A delivery system is probably not one with the capability for knowl~ edge base modification, while the capability for knowledge dase modification 1s an essential feature of a development system. This distine- tion between development and delivery systens WITT be avoided in this discussion; it will be assuned that all expert systens are capable of Knowledge base sodi fication. Avionics Applications of Expert Systems The main interest in using expert systens technology in avionics arises. from the complex- ity of the missions to be perforned and the demanding environments in which these missions faust be performed. Consideration of these fac- Tors have forced fequirenents for increasingly ‘nore conplex systems. The energing generat ion Of tactical vehicles (fighters and attack heli- Copters) are near the Timit of what can be accom plished using a single pilot. ana conventional technology. Additionally, these systens require the pilot to perform at a'level that 1s near (1f fot beyond) the Timits of human performance. When all systens are operational, the basic tasks facing the pilot are data interpretation and sub- systeas integration. When problens arise in the avionics subsystens of these energing systess, the pilot may be unable to cope with the situation Because of the complexity of the sudsystens and their interactions within the vehicle system as a whole, Expert systems technology offers the pos sibility of solving these problens by providing the pilot with useful kxcutedge and asstetance. The main interest. in using expert systems tech- nology in avionics applications is that it may be the only means of providing the next level of sys- tens integration. Many avionics applications of expert sys- tems technology are possible. These applications include examples of most of the class of expert systems problens described in Ref. §: interpre- tation, prediction, diagnosis, planning, moni toring, debugging, ‘repair, and control.” Most of ‘these applications wil) be endedded in other systems or will serve to Integrate subsystens. Sensor fusion is an excellent example of inter Pretation in which multiple sensors provide diverse pieces of data that must be integrated into a coherent picture of the world outside the aircraft. the prediction prodlen is exemplified by tactics prediction in which the future tac. tics of an opponent would be predicted from past perforsance. Diagnosis could be applied to the Hsolation of a failure within a flight control, system. Planning systems have obvious and immediate applications tn route planning and target allocation problens, Monitoring systens could be used to assess the health and status of any of the various subsystems within the aircraft or of the effectiveness of the aircraft as a Weapons systen. Expert systens that. diagnose, debug, or repair could be employed to provide the expertise needed in a reconfigurable control sys~ ten, In this application, problems in a flight contro} systen could be corrected by recontig- uration (repair) after specific malfunctions determined from sensor and aircraft behavior (aiagnosis) had been analyzed. and corrective easures had been identified (dedugging)- Finally, control expert systens could be used to integrate the subfunctions of an avionics. system at the mission level. Obviously, this Vist of applications isnot exhaustive, and the problem areas are not disjoint. However, this, Mist of applications provides sone insights into the problems associated with the use of expert systems in avionics applicattons.. AVL avionics software must execute in a real- tine environment in which the tinefrane 1s deter mined by the application; the slowest of these applications are those that interface with the pilot or involve Tong-term planning, and the Fastest are those that involve flight control systems or weapon sensors. The need for real- tine operation makes speed of execution a criti- cal consideration in any examination of the tradeoffs between two competing pieces of soft- ware. Even with developnent of newer and more powerful avionics computers, software efficiency MITT continue to de an issue. Nore capable con- puters will merely result in greater denands; nore Computational power will simply result in greater expectations. “Avionics computers will continue to be utilized to the maximum that can be obtained from then. The probleas of real-time progranming WITT continue to be a concern, and it 1s here that expert systems becone burdensose. One of the first rules of real-time progran- ming 15 to develop an effictent code. AS {lustrated in the Transforming Expert Systens_Into Conventional Prograns- sectors, expert systems ~ ire and the tine required to Converge to an answer varies with the situatton being analyzed. while expert systens are more efficiently developed than conventional prograns, expert systems are necessarily more inefficient Snvexecution. Further, this inefficiency has nothing to do with the speed at which a computer executes LISP, FORTRAN, or any other computer language. The inefficiency of expert systens is innerent in the separation of the knowledge base ana the control mechanism and in the iterative nature of the control process. Because the inef- Ficiency of an expert system increases rapidly as the size of the knowledge base increases, sone consideration has been given to dividing knowledge bases into smaller partitions. However, this Vatter approach does not eliminate the problen of inefficiency; it merely alleviates it. Aithough sone of the previously discussed applications require a direct interface with the Pilot, not all do. In fact, many of these appl ations such as sensor fusion and flight control system reconfiguration will be inbedded in other Systems and isolated from the pilot. The value of the high-level user interface characteristic Of expert systens is thus unnecessary in many avionics applications. However, even if it ts assumed that an explanation feature 1s required Of the man-machine interface, the capability of knowledge base or program control modification is almost certainly not a requirenent of an avionics system, whether it {san expert systen or a conventional system, In fact, if any pro- gram control modifications are to be permitted in-avionics systens, these will almost certainly be known, well-defined options that are buflt into the’systea. The Teast consideration in avionics systems should be abandoning standard systens qualifica- tion procedures for the Ture of exotic technology Chat allows @ pilot to modify a system in real tine. The requicenent for an explanation in an aviontes application is almost. certainly weaker than the requirement for-an explanation in a de\ opment system. At most, it might be expected that a'firstclevel explanation would be required. A Tirstelevel explanation 1s an explanation of the last rule used to reach a conclusion. This type of explanation is in contrast to the sort of detailed backward justification available in Sone current expert systeas. Transforming Expert Systens Tnto Convent ional Prograns This section describes a mechanism for con- verting expert systems into conventional programs. The requirements outlined in the previous. section are taken as the requirements for the conversion process described here. While the examples that Follow are only for simple forward and backward- chaining faference mechenisns, these inference ‘mechanisms can be and are being utilized for a wide vartety of applications in aviontes. To iVtustrate how expert systems can be mapped into convent tonal. programs, tuo exanples Of production rule systems are givens. one with a formard-chaining inference mechanism end one with a backward-chaining inference mechanism. These exanples, witle extremely sinple, will facilitate an understanding of the proposed approach. Further, these examples will serve to ilusteate the tradeoffs involved in convert- ing an expert systen into a conventional program. The first two subsections provide a brief intro- duction to the inference sechanism, present 2 Sample set of rules, and then show the equivalent FORTRAN representation, In the final subsection, expert systens and their transformations into con- ventional prograns are discussed, In the following exasples, FORTRAN 1s used to AMustrate the use of a higher order language. FORTRAN was chosen for two reasonss because. of the authors" familiarity with the language and Because FORTRAN" 1s the primary computer, fanguage used in conventional. sctentific and engineering applications,” This language 1s sT30" supported: on most machines by fast, efficient optinizing cone Pilers.. The perception that FORTRAN ts probably Unsuitable for Al applications also influenced the decision. Although atleast one expert. systens, Shell has been implemented in FORTRAN (the TIAA expert. systens generation tool by General. Research Corporation), FORTRAN 1s generally considered the antithesis of a suitable Al language. All examples ‘Of FORTRAN code presented in this paper could as easily have been’ shown in LISP, Ada, Cy oF any ‘other higher order Tangua: Forard-chaining Example Forvard-chaining 45 often referred to as aatacdrivens inference because the rules are applied to the established facts to reach wnatever conclusions. are consistent with the given fects and the rules.” Forvardechaining inference stops sinen‘a pass. {iteration or eyeley thraugh the roles Yields fo’ now facts and the inference: process ts, Zonplece. "A'set of example rules. inwricn clauses have Been replaced by symbols 15 a5 follows: Ife on f then d If d then e If b and a then ¢ If a then b Figure 5 shows the results of applying these rules with a forward-chaining inference mech: anism in a situation ia wnich only the fact a is established initially. In this simple example, Tive inference cycles are required before the stopping rule 1s satisfied. Figure 6 shows three FORTRAN representat ons of a logically equivalent reasoning process. The three examples of FORTRAN Correspone’ to steps Inthe conversion process from the expert systen to Conventional code. To aenteve the direct repre- Sentation,. all clauses would be assigned a vart= ble nane’and the rules would be translated directly into code. In tre second step of the Conversion process, al? clauses. that are only lised as antecedents are taken as the baste 1a~ pot syabots (prinitives), and then each rule is expanded until it is expressed entirely in terms Of these prinitives. " (Here, to simplify the ‘iscussion, ehese primitives are assuned to rep Pesent the Input data to the expert system.) The Final simplified FORTRAN code 1s established by applying standard methods for reducing Boolean expressions, such as. the Quine-NeClustey minint~ sation method [i]. Auset of rules to be used by a forvara-chaining inference mechanism 1s thus Transformed into'a single-pass, conventional Set of migner oeder language expressions. Sackward-Chaining Example Backward-chatning inference ts often referred to as "goal driven inference because the infer= nce mechanism begins with an ordered 11st. of goals’ (hypotheses) and uses the Knowledge base to Httanpt to find. a set of rules. thet allows these hypotheses tobe concludes. If the First hypothe= S15 cannot be satisfied, using the knowledge base Ang established facts, the second hypotnests 1s, Sttenpted. This process continues untiT a hypath- esis can be asserted or until the Tist of hypoth. ses fs exhausteg. A 11st of rules and an ordered Set of hypotheses are given beTow. tule 1 Waadbadethnd 4 2 Ifeandf thena > 3 fg orn then b a 4 tt add then e 5 theanda then 6 Ite and t then g To ‘Mustrate sackyard-chaining, the facts « and fare established initially. Because the hypoth ser. are ordered, the backward-enalniag mechantsa first attenpes to conclude d. This ts. done By Finding a rule with d-as the consequent — in this Caney Tule I above, The backwardvchatning mech nian then compares the established facts and attenpts to satisfy the rule antecedents. tron those facts. If ay by and ¢ are not established facts (and, {n this ease, they are not), the Dackuardechaining sechanisa mist repeat’ the process of finding rules with each of the sub- hypotheses as consequentss In doing so, it must test those rule antecedents against the estab Vythed’ facts and continue until either the Tist of rules has been exhausted or all antecedents for sone hypothests can be sat tated. Attempt ng to assert d results in the seareh-tree, shown in Figs 7s In tnis example, the hypothesis d would be Sbandonea and the baciward-cheining wechantss: would test to see f the next hypotnests © could be asserted, Given the rules and establishes facts. 1n this exanple, a can de asserted using fale 2." The backward-chaining nechanisn would assert a, after trying to assert @ and b in turn, and stop: Figure @ shows how the evles and nypatheses iscussed above could be represented 28° FORTRAN code. As in the forward-chaining example, the Conversion from cules. {0 FORTRAN code I 4 three= Step process. To achieve the direct represen- tation, all clauses would be assigned vartable hanes" and the hypotheses would be trans tated Girectly into the code. The secona step of the ransformation process once again requires that the clauses that are used only as antecedents be gent ified at prinitives (and that these prini= tives are assumed to represent input data). The Fepresentations of the ordered hypotheses are, expressed in terns of the priaitives.. The Tisal Step of the transformation ageln requires the Spplication of a method for reducing Sooiean expressions and results In the FORTHAN cose pre Sentes as the "representation after substitution and reduction" tn Figs 8. The aifference Detween EMs single-pass code’ and the backvard-chaining example ts thet all hypotheses that can be satts- Fea'wit! be satisfied. To provide an equivalent rnechanism to the backward-chaining example, @ test fd return must. be Inserted after each represen tation of a hypothesis (Fig. 8) and a vartable UhULATP) created to tnatcate wen no hypothesis could be satisfied. Conpartson of Exanple Expert systems and ‘Their Conventional Code Representations In the exanples of forward and backward chaining, the conversion of ules into a conven Honat eee results ina tagicalty equivatent Fepresentation of the knowledge base and inference rechantsn. However, the flexiaility of the expert Systens has been converted into the cigidiey of ¢ Conventional cade.” Yet. if this conversion is automatic, nothing has been Tost. Another step inthe development, process has been Tnsersed (the conversion process). However, this 1s a sinor nconventence shen considered” tn the context. of "program enat would exceste faster and could tasty be hosted on any of a number of numeric processors. “Obviously, the conventional code is Yess readable and self-docunent ing than the rules would be for an expert system. Nevertheless, the Feadabiity and code documentation are critical nly If the code must be maintained and nodified by humans. In the processes described above, mod- SEication and’ maintenance of the knowledge bese ould occur within the context of the expert. sys~ ens development environment When the knowledge base {8 modified, a new piece of conventional Source cose could be gonerated and the old code Could be éiscarded. This. process would then be SHatlar to the use of a standara compiler. In fact, 4 knovtedge conptter is exactly what 15, being proposed tn this paper Te 4s tmportant to understand what vould be lost In the transforsation of expert systens toconventional code in the examples given. While the conventional code 1s logically equivalent to ‘the knowledge base of the example expert systens, fo provision is mide in these exanples of & con- ventional code for an explanation feature, At the user interface, no provision 1s made for rule modi icat fon or even for the display of rules. As discussed in the Avionics Applications of Expert Systems. section, the proviston for knovtedge base ‘maintenance in the application system 1s neither required nor desirable, but sone form of explana- tion feature ts, at tines, desirable. A Fiestalevel explanation can easily be gen- erated using a subroutine (procedure) in addition to the subroutine that performs the logical Infer~ fence. This subroutine would, in essence, contain formatted representations of the knowledge base rules. Returning to the example rules in Figs. 2 and 3and assuming that all rules would be used io Situations requiring explanation, the subroutine shown in Fig. 9 could be generated from the rules to provide the needed explanation. In the exanple shown in Fig. 9, the explanation in the format. statement would’be displayed to the user (pilot) whenever the logical variable (Cl, C2, C3, C4y or £5) corresponding to the rule consequent of the formatted rule was true, That 1s, if the equiva ent of the first rule in Fig, 2 could nave been used to conclude that the “longitudinal rate damping mode 1s. inoperative" and Cl in Fig. 3 would have been true, then the rule encoded in ormat statesent 101'in Fig. 9 would be displayed as an explanation. Concluding Remarks A brief introduction to expert systems 18 p sented in this paper. Expert. systens are conpared fo convent ona progeans ang then aisclssed within the context of avionics applications. After des- cribing the need for expert systens in avionics applications, the problems posed by this technology are discussed. Two example inference mechanisms (forward= and backward-chaining) are described and used to exenplify the proposed technique of con verting expert systems knowledge bases Into con- ventional programs. This paper presents the concept of using # xnoutedge conpiter to convert the Knowledge base developed for an expert. aysten Into a convent ional Brogran.. This concept. allows the most. desirable Features of expert systens to be retained and also provides a neans for producing fest, efficient ode capable of exacutton on any processor. WniTe af seussed within the context of avionics. appl¥e Hons, Enis concept nas uelTiey in oener applic tions where execution speed 1s not the prinary consideration but where the options available for Larget machines are Timited. Wnile the expert systens examples presented in this paper are extrenely simplified, they are representative of two powerful inference nech= anisms that are applicable to a wide range of problens. By denonstrating that the knowledge bases and Inference mechanisas used in these. fexanple expert systens can be converted into conventional programs, it has been shown how Sone of the problens of using expert systens in avionics applications say be minini zed, In fact, the proposed approach need not be Limited to aviontes applications. For any sys- tem that can de converted using this technique, the advantages are significant. while the use of symbolic. processors in research and develop- ment. Taboratortes has many benefits, the costs assoctated with these single-user, spectal= purpose systens may make then unsuttable for Earget applications. The technique described in this paper provides a means for converting expert systens from symbolic processors to rumeric processors. References (11 2.9. Helly, dre, "A distetbuted @ {or’space shuttle ftgnt contro, Thesis, Unty. of California, Los Angeles, 1984. [2] V.A. Regenie and E.L. Duke, “Design of an expect-systen flight status monitor." NASA 4W=86739, Aug. 1985, [31 EsL. Duke and VAs Regenie, "Description of an experimental expert system flight. statue monitor." NASA TH-86791, Oct. 1985. [4] A. Bare and €.A. Feigenbaum, The Handbook of Artifical Intelitgence, vor. T. Los ATtos, TA: item kautnann, Ines, 1901. [5] Rid. Brachman, S. Anarel, C. Engtenan, RIS. Engelmore, E.A. Feigenbaum and DLE, witkins, *wnat are expert systems?” in Gutlding Expert systems, F. Hayes-Rath, Deke Wateraan-and Dobe Canis Ed. ending, MA: Addison-Wesley, 1983, pp. 31°57. [6] F. Hayes-Roth, D.A. Waterman and 0.8. Lenat, "hn overview of expert systems," in Sui ldin Expert. Systens, Fu Hayes-Roth, OuA. Materaane Gnd U.B tense, Ed. Reading, HA: Adcison~ Westey, 1983, pp. 3-29. [71 Fad. WII and 6.8. Peterson, Introduct ion to Settening Theory and Logical Design, tnd ed- ew York! on WITey Sons, Tees’ 1978,— t Taamnigicaai] [Romeo nam ey ] ‘Specitic a [Knowledge base }-—[ inferencemechanism Lone ean asic export systens architecture. igs Le then ” then then Fig. 2 the primary pitch rete gyro has falled ‘and the Backup pitch rate gyro has taled the longitudinal rate damping mode is inoperative the pilot has selacted the alr-torir gun mode cor the puldance system has selected the al-to-alr gun ‘mode ‘he fight control system Is Ia the fuselage pointing tha mission i intoreaption and the ful le not suffctent for a minimum-time Timerception with maximum trust Lise the minimum-time to eruse energy algorithm yound-based thst data base has been updated and the Hok associated with the new situation Is ‘unacceptably high er ‘aN, A2) ce OR Aa) a OR Aa) ca ‘AND, AB) cs ‘AND. AB) vanes ‘At =the primary plten rate gyro has flied ‘AZ “the backup pitch rate gyro has a as ae Interception with maximum thrust ‘Ay =the ground-based tvest deta base has been updated ‘Ag =the rak associated wi Te unacceptably high the now situa the route should be replanned seanple roles for avionics applications. C1 =the fongitinal rate damping mode is Tnoperative oc ca C4 Suse the minimum-time tow crise energy ‘igor 1204 C5 _=the roule should be replanned Fig. 3. FORTRAN representation of example rules for avionies applications. Doman foecie oa setae tee A Knonteage craton ge war expen ream Tawa el feptantion| “ase | ntronce Tete” |managarent | acne Pig. 4. Componente of an expert systTrforenes ae Facts estbished (a) (abe) (abe) abe {abode} renee Fig. 5. Example of forvard-chaining. Direct representation: . za =A Fig. 6. FORTRAN representation of forvard-chaining example. ule Rule ue ule Pulte Rule Fig. 7. Search-tree associated with example hypothesis.Direct represertation: D =((A-AND. 8).AND. Cc) B =(G.0R H).OR. (CAND. A)) A= (E AND. F) Representation ater substitution: 1D. =((E .AND. F) AND. ((E.OR. 1) OR. H)) AND. (1 AND. J)) 1B =((E AND. 1) .OR H) .OR.((| AND. J) AND. (E AND. F))) A= (AND) Representation after substitution end reduction: 1 =((E AND. F) AND. 1) AND. J) IF RETURN = (E .AND. 1) OR H) IF 8 RETURN A= (E.AND.F) IF A RETURN NULHYP=. TRUE. Pig. @. FORTRAN representation of backward-chaining example. ‘SUBROUTINE EXPLAN COMMON ICFACTS! Ct ,C2 ,C3 ,C4 ,CS IF (.NOT. -EXPLIN) RETURN ) WRITE (*,101) ) WRITE (-/102) ) WRITE (-/103) (C8 ) WRITE (°/104) (C5) WRITE (*/105) RETURN 101 FORMAT( The longitudinal rate damping mode is", : * tnoparative ‘because the primary pitch rate gyro has tailed”, , nd the backup piteh rate gyro hes faled.") 102 "FORMAT(“« The flight control system Is In the fuselage”, . “ pointing mode" J, 108 ‘Gun mode Indicator Ison”, * because either the pllot has selected the”, “ aleto-alrgun mode", 104 “FORMAT(* Une theminmuntime 1 else «Use the minimum time to #eruise energy", * alge boceuse the missions ntercapion the fol isnt sutican or ‘intmm-tie itereeption with maximum tvs”) The route shouldbe roplanned” because the ground-based test deta base has", « boan pda, “ and the ak accited with he now tutions” * fsunasoopeably igh") 105 " FORMAT( Fig. 9. FORTRAN subroutine to provide first explanation. evel2, Goverment en No, NASA TH-88263, [AN ENGINEERING APPRORCH TO THE USE OF EXPERT SYSTEMS May 1986, TECHNOLOGY IN AVIONICS. APPLICATIONS casera Cousin oa 7 autor 1 Fotorming Organization Repo Na Eugene L. Duke, Victoria A. Ragente, and Naryloutse Brazee (NASA |” H-1368 ‘ines-Dryden), and Randal W. Brumbaugh (PRC Kentron) RASA Ames Research Center 16 Wok On We 7 Faery Oeianon Nar ad Aros TOP §05-65-11 Dryden Flight Research Facility TW, Canvas or Gan Wo Oe fox 273 Edwards, CR 93523-5000 TE Spomarng Agency Ram ond Ares Technical Memorandum ational Aeronautics and Space Adninistration TH Seri Ramey Gade Mashington, D.C, 20586 TE Sapemetry Rae Prepared as an invited IEEE pager for presentation at National Aerospace ‘and Electronies Conference (NAECON), Dayton, Ohio, May 19-23, 1986, This paper presents the concept of using a knowledge compiler to transform the knowledge base and inference fnechanism of an expert system {nto a conventional progran. ‘he motivation for this discussion ts the need to aceon rnodate real-time systens requirenents in SS enbedded avionics. The paper presents an over-view of ‘expert systems and a brief Comparison of expert. systems and Conventional programs. Avionics appl ‘cations of expert sys~ ems are briefly discussed before the detatled discussions of applying the proposed concept to example systens using forward and Backward-chaining. ppt ications. such 7 Rey Was Sapaed by Rivera) Ta Oaeibaion Satenant Avionics unclassified — Unlimited Expert. systens Knowledge compiler STAR category 05 1. Searty Gas ft eer a Say Oe oF i pa Fi Ne. ot Pe unclassified unclassified 10 a naz “tor sale by the Rational Teohviaal Information Service, Springfield, Virginia 22161.