Launch Commit Criteria Monitoring Agent Glenn S. Semmel Ladislau Bölöni Steven R. Davis, Kurt W. Leucht University of Central Florida Dan A. Rowe, Andrew 0. Kelly Dept. of Electrical and Computer Engineering National Aeronautics and Space Administration [email protected] Spaceport Processing Systems Branch [email protected] Abstract Sh,rnk Dab Sbam The Spaceport Processing Systems Branch at NASA Kennedy Space Center has developed and deployed a soft- Hrdw Con,b,ablS La,,h ConoI a,ai Mon.tor ware agent to monitor the Space Shuttle 's ground process- Sppo,t Itf Eqaipnbne Mothila Mcen,na Sv,t,n, - ing telemetry stream. The application, the Launch Commit Criteria Monitoring Agent, increases situational aware- Figure 1. Ground Control and Monitoring at ness for systemn and hardware engineers during Shuttle NASA KSC launch countdown. The agent provides autonomous mon- itoring of the telemetry stream, automnatically alerts sys- tem engineers when p redefined criteria have been met, identfles limit warnings and violations of launch coin- The Launch Processing System (LPS) at KSC provides fa- mit criteria, aids Shuttle engineers through troubleshooting cilities for NASA Shuttle system engineers, contractors, procedures, and provides additional insight to verify ap- and test conductors to command, control, and moni- propriate troubleshooting of problems by contractors. tor space vehicle systems from the start of Shuttle interface The agent has successfully detected launch commit crite- testing through various phases including terminal count- ria warnings and violatons on a simulated playback data down, launch, abort, safing, and scrub turnaround. stream. Efficiency anii safety are improved through in- LPS continually monitors the Shuttle and its ground creased automation. equipment including environmental controls and hardware that loads propellants. Consoles with vehicle responsibili- ties communicate information directly to and from the Shut- tle computer systems. Consoles with ground support equip- 1. Introduction ment responsibility communicate information to and from the hardware interface modules which are connected to the This paper describes a software agent that is used for numerous ground support systems. See Figure 1. Each mod- processing Space Shuttle telemetry data and notifying sys- ule is capable of interfacing to approximately 240 sensors tem engineers of warnings and violations. After describing or controls. Overall, some 50,000 temperatures, pressures, the problem and objectives, the environment, interfaces, ap- flow rates, liquid levels, turbine speeds, voltages, currents, plication description, and extension of the agent for future valve positions, switch positions, and many other parame- uses will be presented. ters must be controlled and monitored. For over 25 years, engineers have used LPS to verify 1.1. Background Space Shuttle flight readiness and to control launch count- down. LPS has performed superbly well. Recently, much NASA Kennedy Space Center (KSC) is responsi- of the LPS hardware was upgraded assuring its continu- ble for pre-launch ground checkout of the Space Shuttle. ance for many more years. However, the system architec- ture was not changed and software remains basically the I A demonstration of this system will be available to be shown at the same. As a result, the level of situational awareness has not conference. increased proportionally to what would otherwise be possi- ble with more modem software technologies. Given the possible complexity of limits, the large num- After the Shuttle Columbia disaster on February 1,2003, ber of limits, and the need to get supporting data quickly the Columbia Accident Investigation Board [14] proposed Shuttle engineers needed an advisory tool to provide more recommendations to improve safety from both an organi- insight and situational awareness during launch countdown. zational and technical perspective. The Board indicated the In the latter half of 2003, a software tool was proposed to need to "[adopt] and maintain a Shuttle flight schedule that provide additional insight during Shuttle launch countdown is consistent with available resources." Also, both manage- and increase the level of situational awareness. The tool, ment and engineering support staff must maintain an aware- called the Launch Commit Criteria Monitoring Agent (LC- ness of anomalies and those must not be lost "as engineer- CMA), complements LPS and is capable of autonomously ing risk analyses [move] through the process." Given two and continuously monitoring Shuttle telemetry data. LC- tragic losses of a crew and Shuttle, today NASA engineers CMA automatically alerts NASA Shuttle engineers when have an even greater pressure to be more vigilant in identi- predefined criteria (e.g. limit violations, warnings) have fying problems. Anomalies must be detected and reported been met and guides the engineers through troubleshoot- to prevent problems with Shuttle subsystems, countdown, ing probedures. and launch. The aging LPS hardware has limited resources and precludes the level of automation and notification war- 1.3. Objectives ranted by this domain. LCCMA acts as a software agent for the NASA engineer. For this discussion, an agent is defined as rule-based, au- 1.2. Problem Description tonomous software that reacts to its environment and com- municates results to a human, a NASA engineer in this During launch countdown, NASA Shuttle engineers are usage. Agents have been extensively researched [22, 19]. required to monitor shuttle data for violations of the launch Agents standards [9] and frameworks [1, 15] have also been commit criteria (LCC) and to verify that the contractors developed. troubleshoot problems correctly. When a violation is rec- The primary objectives for LCCMA include: ognized by the system engineers it is reported to the NASA Test Director. The problem report, or call, includes a de- • Monitor Space Shuttle telemetry ground data. scription of the problem, the criticality, whether a hold is • Allow a NASA engineer to specify rules to be applied requested, and whether a preplanned troubleshooting pro- to Space Shuttle telemetry ground data. cedure exists. Many systems have a large number of mea- • Display a visual indication of violated LCCs. surements with associated LCC limits and a large num- • Display a text message of the LCC violation call. ber of LCC requirements. Table 1 shows four representa- tive Shuttle subsystems and their corresponding number of • Display troubleshooting steps from preplanned proce- LCCs and measurements. As illustrated, hundreds of mea- dures. surements must be monitored just for this small set of sub- LCCMA does not send any commands and is used for systems. advisory purposes only. A future release of LCCMA will The Shuttle is composed of many subsystems (e.g. Main include an interactive troubleshooting display that reads the Propulsion, Hydraulics). Each of those subsystems has a data stream and accepts user inputs to direct diagnostic trou- team of engineers responsible for troubleshooting problems bleshooting. for that respective system during a launch countdown. Each team has its own tools for identifying LCC violations. Many 2. Environment and Interfaces of these tools use the LPS software and simply change the color of the displayed data and/or present a text message 2.1. Shuttle Data Stream to the user or set off an audible alarm. Troubleshooting may require other displays such as plots and troubleshoot- Data processed by LPS is distributed on a local area net- ing flowcharts. Valuable time is spent locating these pro- work. As shown in Figure 1, the distributed data is known as cedures and locating the data that supports them. Table 2 the Shuttle Data Stream (SDS) [16] and contains real-time shows some sample limits for the Power Reactant Supply vehicle and ground processing data. Thousands of teleme- and Distribution (PRSD) subsystem. In this case minimum try measurements are published in the SDS and are used and maximum limits are specified for pressure measure- by monitor-only applications such as LCCMA. The SDS ments associated with two hydrogen tanks. Limits can be contains multiple types and subtypes of measurements in- much more complex, involving limits calculated from other cluding discretes (i.e. boolean measurements), analogs (i.e. measurements and limits that apply only to specific times floating point measurements), and digital patterns (i.e. inte- during the countdown. ger measurements). Subsystem Number of LCCs Number of Measurements APU/HYD 50 252 ECLSS 29 136 PRSD 15 113 OMS 18 434 Table 1. Number of Measurements for Various Shuttle Subsystems LMeasurement Id Description Mm Max Units V45P21 1OA Tank 1 Heater Control Pressure 196 298 psia V45P2100A Tank 1 Pressure 192 294 psia Table 2. Example LCC Requirements for PRSD H2 Tank 1 2.2. LCCMA Context Diagram Jess' predicate logic lends itself to capturing and spec- ifying the heuristics and engineering rules of this space- Figure 2 shows the context diagram for LCCMA. The port domain. The declarative paradigm of this rule-based agent process, represented in the middle circle, commu- agent application also makes it highly modular and scal- nicates with various sources and data stores. A measure- able to span multiple subsystems of the Shuttle. Jess also ment database is used to decode the SDS into usable mea- includes a fourth generation scripting language and interac- surements. The SDS source broadcasts measurements as tive command line which are very conducive for prototyp- data packets over local area networks. LCCMA monitors ing and testing. this stream for measurement violations and warnings spec- Jess is written entirely in Java and has access to the ified by the Shuttle engineers. The Troubleshooting Proce- full Java application programming interface from the script- dures source represents html or pdf files containing the trou- ing language. It provides standard control flow constructs bleshooting steps, often in flowchart format. LCCMA sends and supports variables, strings, objects, and function calls. limit violations to the NASA Engineer via the Status Board Jess automatically converts between its own types and Java Display. The Rules data store represents the Jess scripts and types insulating the developer from manually performing knowledge base that defines the rules for the limit viola- the conversions. Its use as a Java library made Jess' selec- tions. tion more appealing since Java supports multiple platforms with its "write once, run anywhere" paradigm. Beyond that, the need for NESTA to support web enabled clients also 3. Application Description made Java a natural fit given its origins and strong support for developing Internet based applications. 3.1. Languages and A! Tools Used in Application The Java Expert System Shell (Jess) [121 was selected as 3.2. Design the agent's rule engine. Jess was developed and supported by another government agency, Sandia National Labs. As Java classes were developed to parse and decode the data such, our development team and customer have full usage stream and represent measurements as facts in Jess' work- of the tool via government licensing without any fees. This ing memory. To interface Jess' rule engine with the SDS, includes access to all the Jess source code. each data measurement is modeled and implemented as a Jess' forward chaining reasoning system was modeled Java bean [21]. Java beans provide a component architec- after production systems such as OPS5 [3] amd CLIPS [23]. ture to enable easier integration of applications. A property It contains highly efficient and sophisticated pattern match- change notification mechanism is supported that allows one ing based on the Rete algorithm [ 11]. This enables its in- object to become a registered listener of another object. The ference engine to process many rules and data rapidly. The listener object will then automatically receive changes from engine repeatedly processes through a match-select-act cy- the source object. This is also known as a publish-subscribe cle. As a production system, its consequents can be actions. or observer pattern [13]. Within Jess, each Java bean cone- A conflict resolution strategy determines the precedence of sponds to what is known as a shadow fact. A Jess shadow rule firings. fact is a mirror image of a Java bean, such as a pressure LiN Rule: e Data Format Rule Status Shuttle Measurement / 0.0 ) _L_i_m_i_t _v_i_o_la_t_io__n_ Data I LCCMA Enabled Stream Client DM1 Sequence NASA Engineer Message leshootin/ ure j _ Log Figure 2. LCCMA Context Diagram measurement, within Jess' working memory. All shadow in the text box. The user reads the text and, if there is an as- facts are registered listeners of their Java bean counterparts. sociated troubleshooting file, clicks on the file button next to Thus, whenever a measurement changes in the data stream, the text. This brings up a Troubleshooting Display for that a property change event is automatically generated for the particular LCC and limit. The LCC text remains bold un- given measurement and its sibling shadow fact is updated in til the Acknowledge button is pushed. Message text can be Jess' working memory. Figure 3 illustrates this path. displayed with one of three icons representing a violation, After a shadow fact is updated, the Jess pattern matcher warning, or informational cue. will determine if the premises of any rules match the new The text messages can be read over the Operational Inter- or modified facts. Rules are compared to working mem- communication System as LCC calls during the countdown. ory to identify premises that are matched by the data in Calls will change based on what limit is violated (e.g. warn- working memory. For LCCMA, this data represents mea- ing, LCC, high/low limit), the time criticality of the call, and surements from the SDS and rules represent data monitor- LCC effectivity. The application aids the NASA engineer in ing criteria submitted by NASA Shuttle system engineers. making a Go/No-Go decision. Rules with matching premises are activated and placed onto an agenda. Next, the agenda is ordered according to Jess' default conflict resolution strategy. The highest priority rule is then fired and executed. This match-select-act cycle re- peats until no more rules are available to fire. An action han- 3.4. Execution dler class was developed and is used to build and send the notification message to the Shuttle engineer whenever a rule fires. At startup, LCCMA connects to a single data stream based on user input and reads a rules file containing LCC violation and warning limits. Table 3 shows the conditions 3.3. Graphical User Interface and actions associated with an LCC warning and violation for the hydrogen (H2) tank I from Table 2. For example, if A graphical user interface currently exists for LCCMA either of the H2 tank I pressures are above the upper limit, called the Status Board Display. It is being upgraded and the agent should notify the NASA engineer by displaying Figure 4 shows a storyboard representative of that future in- the violation in red font and direct the engineer to the corre- terface. The Status Board Display shows the health of the sponding troubleshooting file (i.e. PRSDO6Hi.pdf) for that network connection, data stream status, countdown time, violation. The troubleshooting file shows the steps neces- and other relevant information. sary to be taken by the engineer when the specified limits of When LCC limits are violated, the LCC call is displayed a given subsystem are violated. FD Measurement (JavaBean) Shuttle Data Stream Shuttle Data Stream Reader [mtvcharraeSu000c1 Jess Shadow Fact J getNextPacket() set VaJueO tirePropertyChange() propertyChange() Figure 3. Sequence Diagram Illustrating Update to Jess Working Memory from Shuttle Data Stream LCCMA Status Board Display Stream ID; 7 Limit File: Fuel CeU Activation Stream Status: GiWI':328:0055154 User Inlsjbiled: I TCID: SAAII3B CDT:-fOOdJOlO/23 Masked Values/Rules: 2/I LCCMAVer:OJc Liantt tictaits 0 I - 322:0052/12.359 GMT (+00.0009/07 CDI) L PRSD-02 02 Manifold Isolation Valve Indicates Closed V45X214IE1 {PRSD 02 MANF 3 TSLN VLV-OPEN} is OFF V45)c214d51 {PF2D 02 MANF 4 ISU1 VLV-OFENI is OFF. 0 328:0051/12.359 GMT (+00:0008107 CDT) PRSD-04 02 Manifold Isolation Valve Indicates Closed V45X4141E1 (PRSD 02 MANF 7 ISLN \YLV-OFEN) is OFF 328:0049/12. 59 GMT (±0 .0006/07 CDI) [OUTDATED MEAGEJ PRSD-02 02 Manifold Isolation Valve Indicates Closed V45X21 4JEI IPRSD 02 MANF3 ISLN FL V-OPEN/is QPFi V45X21 46E! [PRSD 02 M4NP 4 JSLN VI V-OPEN/is OFF 328:0048/12.359 GMT (+00:0005107 CDT) PRSD-01 02 Manifold Isolation Valve Indicates Closed L V45X1 141E1 IPRSD 02 MANF 1 ISLN VLV-OFE11) is OFF. V45X1 146E1 /PRD 02 MANF 2 ISLN VLV-OPEN} is OFF. V Act on selected Select All Acknowledge Pause Rules Print... j Select Noise Remove . :i Help Exit Figure 4. LCCMA Status Board Display Condition Description Message Action (V45P21 bA > 270 H2 Tank 1 Pressure Heater Control Pressure Reading Display [Description], OR V45P2100A > 270) AND Warning High [V45P21 bA], [Message] in Yellow V45P21 bOA <= 298 AND Tank Pressure Reading V45P2100A <= 294 _______________ [V45P2100A] _________________ (V45P2 11 OA > 298 OR H2 Tank I Pressure Heater Control Pressure Reading Display [Description], V45P2100A > 294) Violation High [V45P21I0A], [Message] in Red. Tank Pressure Reading Open file ________________________ ________________ [V45P2 IOOA] PRSDO6Hi.pdf Table 3. Example LCC Conditions and Actions for PRSD H 2 Tank 1 3.5. Deployment of them is exceeded, the rule will fire indicating an anomaly in the cabin oxygen pressure. Once fired, the right hand LCCMA was delivered to the customer and has suc- side of the rule executes. The not ifyActionHandler cessfully detected LCC warnings and violations on an SDS call has three arguments. The first one contains two trou- recorded playback. It has not been used during an actual bleshooting web page links that are made available to the launch countdown yet since NASA has not returned to flight NASA engineer. The second argument specifies the color subsequent to the Columbia disaster. However, LCCMA's of the message, a violation in this case. Finally, the third ar- potential was already recognized by other projects at NASA gument species the three measurements that may be plotted KSC and it is in the process of being integrated into a larger to investigate the anomaly further. monitoring application. For that one, hundreds of customers will use LCCMA to enter not just LCC monitoring criteria, but many types of simple and complex measurement con- 4. Future Exploration Agents straints. As indicated in the national Vision for Space Exploration 3.6. Knowledge Representation [18], an increased human and robotic presence will be cul- tivated in space, on lunar and Martian surfaces, and other This is an actual LCCMA rule written in the Jess script- destinations. Spaceports will now span from the Earth to ing language: the moon and beyond. A new set of challenges is presented by this Exploration Vision. In particular, the need for auton- (defruie orbiter-cabin-o2-prassure-anonaly-rule ECL-060rsrgsncy Condition Yellow Orbiter Cabin 02 Pressure Anocraly omy significantly increases as people and payloads are sent ?aotivatton-faot u- (activate-orbi ter-cabin-o2-p_-sssure-anona ly-rule) greater distances from Earth. )AnalogFd (fdNace V61?2511A1) (value ?V6i?2SLIAS_val) (AnaiogFa (fdnare V6iP2513Ai( )vaive'V61P251301_val() Agents for these future applications will demand much (AnaiogYd (fdNve V6lP2Si5A1( (valve ?V61P2515A1_vui) (teat higher degrees of autonomy than today's Shuttle agents. Few or no human experts will reside at remote lunar or Mar- (0 (abs (- ?V61P2511A val ?V6102513A1_val 0.15 )v )abs(- ?V61?251301 vol ?V61P251051_val () 0.15 tian sites to correct problems in a timely manner. More au- )v (abe)- 'V61?251101_val ?V61?255A1_val )) 0.15 (u (abc)- ?V61025l1h1_val 3.1)) 0.3) tomation will be required along with advanced diagnostics (uu (aba)- ?VS1?2513A1_voi 3.1)) 0.3) ( (aba)- ?V6i025iShLvai 3.1) 0.3) and prognostics. This requires higher levels of reasoning. Today on Earth, system and hardware engineers and technicians leverage multiple skills when monitoring, di- (retract ?activatlnn-faot) agnosing, and prognosticating problems in Shuttle ground assert (orbiter-cabn-e2-pressure-anona1y-ru1 c-reactivation-activate)) )nntifybctinntandlcr support equipment. For the Exploration Vision, the need for (creates http://xb?0.ksr.nasa.gnv/EcL/Ec:Hove/Launrh.htnl extending these skills to support other vehicles at remote http //rbO0.ksc.nasc.gnv/0CL/CCL_Hcve/Cab±n_Leai.htcl locations from the Earth to Mars becomes essential. These (get-venber L000aColnreessageviolatlov) (createS V610240101 V61P240551 V61125525l( skills include being rational, collaborative, goal driven, and the ability to reason over time and uncertainty, The agent discussed earlier in the paper, LCCMA, is capable of shal- For this rule, three analog measurements, V6lP25llAl, lowing reasoning of short inference chains within the Shut- V6IP25I3A1, and V61P2515A1, are monitored. The ab- tle domain. However, this existing agent can be endowed solute value of the difference among pairs of these analog with higher levels of rationality enabling a deeper reason- measurements must not exceed a given quantity. If anyone ing. We are investigating how to mature LCCMA into a Spaceport Exploration Agent (SEA) in support of the Ex- 4.2. Types of Questions to be Answered ploration Vision. SEAs will need to communicate and collaborate along SEAs will be able to answer the following types of ques- multiple and lengthy logistics chains. This does not simply tions: include agents monitoring pre-flight checkout of vehicles at • Is there one or more faults in the support equipment? a terrestrial spaceport (e.g. LCCMA monitoring Shuttle ac- tivities). Rather, SEAs will reside in multiple locations at • Where is the fault most likely located? great distances. Logistics, scheduling, and planning are just • What combination of data and events lead up to a fault some of the activities that these agents will manage. (i.e. explanation generation)? Within this virtual collaborative management chain, • If a fault remains, what are its effects both locally and SEAs will be inundated with massive amounts of data that systemically, now and in the future. must be sorted and processed. It becomes necessary for them to revise their sets of beliefs as new data arrives. It is • What other systems and personnel need to be notified simply not enough to revise singular data points within an of the fault including both internal and external clients agent's working memory and to have an agent blindly re- (e.g. command/control system, hardware subsystems, act to those changes. Rather, an agent must possess the abil- integrated health management, logistics, business sys- ity to revise previously concluded assertions based on what tems, etc.)? may be now stale data. This activity is called truth main- • What actions need to be taken by the system to address tenance [4, 10], also known as belief revision, and is the fault from both a hardware and personnel perspec- particularly important when deep reasoning of long in- tive? ferences is necessary. An assumption based truth main- • Did the fault stress other previously healthy equipment tenance system (ATMS) can reason over many contexts that now needs to be repaired or replaced? simultaneously. By capturing, maintaining, and deploy- ing spaceport expertise within ATMS-enabled SEAs, the costs and manpower required to meet the Exploration Vi- 4.3. Extending LCCMA with Truth Maintenance sion are reduced while safety, reliability, and availability are increased. As indicated earlier, LCCMA uses Jess as its inference engine. Jess implements a lightweight version of truth main- tenance that is much simpler than a full blown ATMS. Jess 4.1. Benefits of Endowing Spaceport Exploration uses a logical keyword that keeps track of the "here and Agents with Belief Revision now" for specified premises. Other rule based systems, such as Clips and Lisa [24], implement a similar level of truth SEAs enabled with belief revision will provide the fol- maintenance. lowing: Premises on the left hand side of a rule can be tied to • SEAs will continuously monitor spaceport telemetry assertions of facts on the right hand side via the logical streams for expected and anomalous conditions during keyword. A dependency is created between the facts of the operations and launch countdowns. SEAs wil analyze premise and the fact of the conclusion. After the rule fires data from networks of sensors and draw inferences and the consequent's fact(s) is asserted, if the premise ever over time to deduce further action. Results are pro- becomes false, the consequent's facts will be automatically vided to humans, agents, and other subsystems which retracted. In contrast to Jess' version of truth maintenance, may compose an integrated health management func- an ATMS dependency network offers a full history of de- tion. pendencies using an efficient labeling algorithm. It offers a • SEAs provide an automated explanation generation fa- history of everything that has happened as opposed to just cility and diagnostic capabilities. The inferences and the "here and now" as provided by Jess' textitlogical key- facts that lead to a conclusion will be available to the word. Dependency tracking and proof histories have been human expert and other agents for further processing. researched [4, 10, 8] and implemented in other rule based expert system shells such as MIKE [7]. • SEAs provide prognostics to predict where and when failures may occur in support equipment and what if scenarios to assess chains of events. 4.4. ATMS Background • If a human expert leaves the program or moves onto Using de Kleer's model [4, 5, 6], an ATMS is com- other opportunities, SEAs remain and can virtually posed of a set of nodes, n1, n2, .. . , , where each node mentor the human replacement leveraging its knowl- is a propositional variable. A proposition represents either a edge base. premise, contradiction, or assumption. A premise is a node Justification Node" I I "Node I Antecedent Consequent rinformant I Figure 6. ATMS Node Class Diagram Figure 5. Justification Class Diagram that is always true. A contradiction is a node that is al- ways false. An assumption is a node whose values may be changed by the inference engine during rule firings. The in- ference engine incrementally transmits these propositions (i.e. nodes) to the ATMS. When the inference engine fires a rule that results in a new or modified fact, a justification is transmitted to the ATMS. As shown in Figure 5, a justification is a tuple con- sisting of the rule's antecedent and consequent forming the inference. Suggesting an "if-then" type of implication, a justification may only contain positive literals and be rep- Figure 7. Datum Class Diagram resented as a horn clause. From Figure 6, an ATMS node has a datum, justifica- tion, and label associated with it. The datum represents a pling, the match step is modified. This requires changes to rule or fact within the inference engine as indicated in Fig- the Rete algorithm of the inference engine. ure 7. The justification is composed of an antecedent, con- To extend LCCMA with full dependency tracking via sequent, and informant. The antecedent represents facts on an ATh'IS, Jess offers sophisticated event handling that will the left hand side of the rule that caused the rule's premises readily enable communication between the Jess inference to be true and resulted in an activation and firing. The con- engine and the ATMS. Event handlers are supplied and in- sequent represents facts that were asserted on the right hand voked when, for example, a fact is asserted, retracted, or side of the rule upon firing. The informant describes the type modified. In conjunction with an ATMS facility, these han- of deduction and is never used in any ATMS computations. dlers could build and maintain a complete history and de- It may be supplied to the inference engine to provide tex- pendency network. tual cues for explanation generation. Inspired from the Lisp interface definition of Forbus and de Kleer [10], Figure 8 shows the Atms class with anal- 4.4.1. Interfacing an ATMS to the Jess Rule Engine In- ogous Java method signatures for an Atms interface def- terfacing an ATMS to a production rule system has been inition supporting the Jess inference engine. The Atma previously investigated by Morgue and Chehire [17]. In class realizes the Atms Interface and thus implements their study, two levels of coupling were described with re- the interface's methods. The InferenceEngine class spect to the match-select-act cycle of an inference engine. depends upon the Atms Interface. The createNode When an ATMS is loosely coupled with an inference en- and justifyNode methods of the Atms class are not gine, the select and act steps are modified to enable inte- publicly accessible as indicated by the leading minus signs gration. This is a simple form of interaction between the by the method names. They are called after the Atms ob- ATMS and inference engine and is more prone to becom- ject receives a message from the inference engine indicat- ing intractable than a tight coupling approach. In tight cou- ing that a new fact was created or a fact was modified by a and their applicability for constructing probability distribu- tions from an ATMS [20]. Brachman and Levesque [2) propose description logics to implement a production system, act as the working mem- ory, or provide some other service to such a system. In this paper, the sub sumptive power of description logics might be __________ leveraged by the label update algorithms of the truth main- tenance system. Further, both agent applications and Jess it- 4 self are implemented in Java, an object oriented language. Description logic taxonomies might be constructed to natu- rally mirror the object oriented models of the agents. References [1] L. Boloni and D. C. Marinescu. An Object-Oriented Framework for Building Collaborative Network Agents. In H. Teodorescu, D. Mlynek, A. Kandel, and H.-J. Zimmer- man, editors, Intelligent Systems and Interfaces, Interna- Figure 8. ATMS Class Diagram tional Series in Intelligent Technologies, chapter 3, pages 31-64. Kluwer Publising House, 2000. [2] R. Brachman and H. Levesque. Knowledge representation and reasoning. Morgan Kaufmann, May 2004. rte,.rmenthm I .ee_eFr utaee] [3] L. Brownston, R. Farrell, E. Kant, and N. Martin. Pro- gramming Expert Systems in OPS5: An Introduction to Rule- Based Programming. Addison-Wesley, Reading, MA, 1986. () gNattChenge() [4] J. de Kleer. An assumption-based TMS. Artificial Intelli- 1 A gence, 28(2):127-162, Mar. 1986. b,ouamEmnl(DEFRULE_FlRED) oam,tHpmd(ACTlVATION) [5] J. de Kleer. Extending the ATMS. Artificial Intelligence, Shuttle Data Striate 28(2):163—l96, Mar. 1986. era,toNodeO [6] J. de Kleer. Problem solving with the ATMS. Artificial In- telligence, 28(2):197-224, Mar. 1986. tettyNodaO [7] M. Eisenstadt and M. Brayshaw. Build your own knowledge i,Nodeln(node.Nodo. etvEn*oatnant) engineering toolkit. Technical report, Human Cognition Re- search Laboratory, The Open University, UK, June 1990. [81 R. Filman. Reasoning with worlds and truth maintenance in Figure 9. AIMS Sequence Diagram a knowledge-based programming environment. Cominuni- cations of theACM, 3l(4):382-401, Jan 3-6 1988. [91 FIPA. Foundation for intelligent physical agents abstract ar- rule firing resulting in an ACTIVATION event. See the Se- chitecture specification, Dec. 2002. quence diagram in Figure 9. [10] K. D. Forbus and J. de Kleer. Building Problem Solvers. MIT Press, Cambridge, MA, 1993. 5. Conclusion and Future Work [11] C. L. Forgy. Rete: A fast algorithm for the many pat- tern/many object pattern match problem. In Artificial Intelli- An agent that monitors Space Shuttle ground telemetry gence, volume 19(1), pages 17-37, 1982. data was presented. LCCMA provides an increased insight [12] E.Friedman-Hill. Java Expert System Shell. Manning Pub- lications, Greenwich, CT, 2003. for NASA system and hardware engineers. LCCMA has successfully detected launch commit criteria warnings and [13] E. Gamma, R. Helm, E. Johnson, and J. Vlissides. Design violations on a simulated playback data stream. We are in- Patterns: Elements of Reusable Object-Oriented Software. Addison-Wesley, Greenwich, CT, 1995. vestigating extending this agent with truth maintenance ca- [14] H. Gehman, S. Turcotte, J. Barry, K. Hess, J. Hallock, pabilities to support advanced diagnostics and prognostics. S. Wallace, D. Deal, S. Hubbard, R. Tetrault, S. Widnall, Future work includes incorporating probabilities of oc- D.Osheroff, S. Ride, and J. Logsdon. Columbia Accident currence of faults within support equipment. In terms of the Investigation Board (C'A!B), Volume 1. NASA, Washington ATMS, this translates into the probabilities of a fact being D.C., August2003. derivable and the context within which it would appear. Pre- [15] JADE. Java agent development framework. vious research has shown the utility of Bayesian networks http://jade.tilab.com/, 2004. [16] Lockheed. Pcgoal requirements document. Technical Re- port KSCL-1 100-0804, Lockheed Space Operations Com- pany, Oct. 1991. [17] 0. Morgue and T. Chehire. Efficiency of production systems when coupled with an assumption based truth maintenance system. In Proc. ofAAAI-91, pages 268-274, Anaheim, CA, 1991. [18] NASA. The vision for space exploration. http://www.nasa.gov, Feb 2004. [19]S. Russell and P. Norvig. Artificial Intelligence: A Modern Approach. Prentice HaIl, 2nd edition, 2003. [20] S. Srinivas. A probabilistic atms. Technical Report KSL 94-13, Knowledge Systems Laboratory, Stanford University, Feb. 1994. [21] Sun. Java bean specification. http://java.sun.com/, 2004. [221 M. Wooldridge. Reasoning about Rational Agents. The MIT Press, Cambridge, Massachusetts, 2000. [23] R. M. Wygant. Clips: A powerful development and deliv- ery expert system. In Computers and Industrial Engineer- ing, volume 17, pages 546-549, Anaheim, CA, 1989. [24] D. E. Young. Lisa - intelligent software agents for common lisp. http://lisa.sourceforge.net, 2004.