1. Michelle Harris 407-306-6693 Michelle.L.Harris@lmco.com
ATML on LM-STAR®
Michelle Harris
Alicia Helton
Steven O’Donnell

2. Introduction Implemented a set of ATML schemas on LM-STAR®
Schemas used –
TestDescription ML (draft 5.0)
TestResults ML (version 0.15)
Diagnostic ML
Bayes
Common Element Model (CEM)
Dynamic Context Model (DCM – version 0.07)

3. Task Definition
Convert a legacy CASS ATLAS TPS into ATML TestDescription.
Use TestDescription as input to the SELEX TPS Wizard™ and generate TestStand™ sequences.
Execute the TestStand™ sequences on the LM-STAR®.
Collect measured values using ATML TestResults.
Interface with diagnostic reasoner to isolate to the fault more quickly and more accurately.

4. Initial Approach
Use an externally developed tool to convert ATLAS to Intermediate XML
Use XML tools to transform the Intermediate XML to TestDescription
TestDescription will provide the “what” to do information for the TPS
Use the TPS Wizard™ to generate TestStand™ sequence files capable of being run on LM-STAR®.
ATLAS
Intermediate
XML
ATML Test
Description
TestStand
Sequence Files

5. Issues Legacy ATLAS TPS was not designed to maximize portability
Intermediate XML generated from ATLAS was very flat
Difficult to understand test flow and translate into TestDescription
Legacy ATLAS TPS didn’t adhere to style guide which would have enforced specific design rules
Multiple fault callout permutations based on data evaluations made without test numbers created problems in the diagnostic model development

6. Revised Approach
An application was developed to extract the “what” to do information from the ATLAS and save it to a spreadsheet.
Human intervention verified the information and added missing values.
An application was written to convert the spreadsheet to TestDescription.
Test
2000
Test Group 1
Next on Fail
DIAGNOSTIC1
Next on Pass
2010
Callout on Fail A4
High Limit 25
Low Limit NA
Comparison EQ
Units
Ohm
Entry Point No

7. TestDescription Sample
<Outcomes>
<Outcome ID="0_1" value="Passed"/>
<Outcome ID="0_2" value="Failed"/>
<Outcome ID=" DIAGN1" value="Failed">
<ReplaceComponents>
<ReplaceComponent uutComponentId="UUT-0"/>
</ReplaceComponents>
</Outcome>
<ReplaceComponent uutComponentId="UUT-1"/>
------snipped
<Step xsi:type="Step_Test" ID="Step_2" testId="2000">
<Results>
<Result xsi:type="Result_Test" testOutcomeId="2000A">
<NextStep stepId="Step_3"/> <! >
</Result>
Using the information from TestDescription, the Selex TPS Wizard™ builds the frame of the new TPS with initiated variables, test criteria, simulation mode, pre and post conditions, and calls to “how-to” sequences.

8. TestDescription to LM-STAR®
Needed to create the “how-to” TestStand™ Sequences
Highly intensive manual task
Simplified through the use of Custom Steps
Graphical interface to LM-STAR® system software

9. Diagnostic Model Description Development of Model
Model is based off the Bayesian and Common Element Models from the AI-ESTATE standard
Stored in XML format derived from the AI-ESTATE models
Development of Model
Start with the fault tree of the TPS
Use historical test results and maintenance data to add more intelligence to the Model
Learning algorithms are used to continuously feed back newly discovered test results (in TestResults ML format) and maintenance data

10. Diagnostic Reasoner
Provides run-time environment for using the diagnostic models
Implements the AI-ESTATE interface to the diagnostic models
Uses the Dynamic Context Model to track session information
Allows for back-tracking through session
Allows restart of Session from previous stopping point
Provides a set of “higher-order” interface functions to minimize required calls for accessing model/reasoner data
Web-service based interface (using WSDL)
Utilizes a Bayesian Network Analyzer called SMILE
By Decisions System Laboratory – Univ. of Pittsburgh

11. Lessons Learned
Our current process is still heavily dependent on manual intervention.
Very time-consuming
Current legacy TPSs are implemented with tight coupling making it difficult to separate the “what” and “how” information
Other ATML schemas such as UUT Description and TestAdapter could aid in the porting process
They were not mature enough at the time the task started
Would be more cost effective to implement UUT test requirements on new systems as opposed to re-hosting the application
Not always a one-to-one test mapping from TPS to Diagnostic Model

12. Conclusion
Industry needs tools that can generate and consume ATML that could be exported to C, ATLAS etc
Using IEEE-1641 for Signal and Test Definition appears promising and further study by Lockheed Martin is planned
Lockheed Martin is embracing ATML
TestResults ML is deployed on LM-STAR® systems supporting the JSF program
As ATML matures, Lockheed Martin is prepared to implement this technology into our legacy and future programs

13. DEMONTSTRATION

14. ? Questions Michelle Harris Alicia Helton 407-306-6693 407-306-1592
Alicia Helton
Steven O’Donnell