NASA Intelligent Self-Situational Awareness ProjectPresentation Transcript
Applications of Data Mining in Automated ISHM and Control for Complex Engineering Systems Karl M. Reichard Applied Research Laboratory Penn State University Voice: 814-863-7681 Email: [email_address]
Data mining , or knowledge discovery, is the computer-assisted process of digging through and analyzing enormous sets of data and then extracting the meaning of the data.
Data mining tools are used to identify correlations and to predict behaviors and future trends from data.
Businesses use data mining to search and analyze databases for hidden patterns and may find trends and predictive information that experts may miss because it lies outside their expectations or because they never thought to look for correlations there.
Data mining derives its name from the similarities between searching for valuable information in a large database and mining a mountain for valuable minerals.
For businesses, the economics of data mining have many analogies to traditional mineral mining – pay for access to public data and make money from mining it.
A data warehouse is a repository where data located in disparate databases are consolidated. Data warehouses store large quantities of data by specific categories so users can easily retrieve, interpret and sort the contents.
Data captured by instruments Or data generated by simulator
Processed by software
Placed in a database / files
Scientist analyzes database / files
Data Mining as Scientific Evolution Jim Gray, “Online Science the New Computational Science,” Microsoft Research.
Leveraging Technology Trends
Performance/Price doubles every 18 months
100x per decade
Utility of computer networks grows as the number of possible connections: O(N 2 )
Computer storage capacity is actually beating Moore’s law
Jim Gray, “Online Science the New Computational Science,” Microsoft Research.
Premise: Most data is (or could be online) so, the Internet is the world’s best telescope:
It has data on every part of the sky, in every measured spectral band: optical, x-ray, radio..
It is up when you are up. The “seeing” is always great (no working at night, no clouds no moons no..).
It’s a smart telescope: links objects and data to literature on them.
Example: World Wide Telescope Virtual Observatory http://www.us-vo.org / http:// www.ivoa.net / Jim Gray, “Online Science the New Computational Science,” Microsoft Research.
Why are Computer Scientists interested in Astronomy Data?
It has no commercial value
No privacy concerns
Can freely share results with others
Great for experimenting with algorithms
It is real and well documented
High-dimensional data (with confidence intervals)
Many different instruments from many different places and many different times
Federation is a goal
There is a lot of it (petabytes)
Jim Gray, “Online Science the New Computational Science,” Microsoft Research. IRAS 100 ROSAT ~keV DSS Optical 2MASS 2 IRAS 25 NVSS 20cm WENSS 92cm GB 6cm
ISHM Development and Implementation Collect raw data Perform assessment Make diagnosis Reduce data Implementation Analyze data Sensing Feature Extraction Fault Detection Damage Estimation Feature Tracking & Prognostics Development Sensor Selection Develop Physical Models Transitional Failure Data Diagnostic Prognostic Algorithms Systems Interfaces Open Systems Architecture Degrader Analysis Where can we apply data mining techniques in ISHM development and implementation?
Data Mining in ISHM Development Sensor Selection Develop Physical Models Transitional Failure Data Diagnostic Prognostic Algorithms Systems Interfaces Degrader Analysis Parts usage Repair orders Maint. logs
Initial design and development
ISHM system verification and validation
Continuous ISHM system improvement (borrowing techniques from manufacturing and quality control)
Sensor reliability Operational data ISHM system data Maint. logs Biggest payoff may be in collection and analysis of transitional failure data across large numbers systems
Data Mining in Prognostics RCM Analysis Detection & Diagnostic Algorithms Prognostics Predicted Remaining Useful Life Anticipated Demand
Transitional failure data from seeded fault experiments
Transitional data from analogous systems
Load history Transformation Model Data mining may allow access to much larger sets of transitional data for assessing fault severity and predicting remaining useful life Sensors Physical Models
Managing System Health Information Platform Data Store Platform Data Store Platform Data Store Platform Data Store Platform Data Store Platform Data Store Platform Data Store Platform Data Store Regional Data Store (CRIS) Regional Analysis Net-Centric Compliant Service OSA-EAI Tech-XML Global Data Store (CRIS) 3 rd Party Analysis Net-Centric Compliant Service OSA-EAI or COTS Replication Depot Analysis Net Centric Attributes Only handle information once Tag and post before processing Store publicly and advertise Register metadata Tag applications for discovery Platform Data Store Platform Data Store Platform Data Store Platform Data Store Platform Data Store Platform Data Store Platform Data Store Platform Data Store Regional Data Store (CRIS) Regional Analysis Net-Centric Compliant Service OSA-EAI Tech-XML OEM Analysis Database Flat file or XML Database Flat file or XML PM IDEs What types of data are we talking about? Asset Register Management Work Management Diagnostic / Health / Rec (?) Static Trends / Alarms Dynamic Vibration / Sound Oil /Fluid / Gas / Solid Tests Binary Data / Thermography Reliability Data
Leveraging Platform Asset Health Information SME Supply Smart Maintainer Maintenance Life Cycle Mgmt PM MC What is the status of all my platform Cs? What is broken or about to break on this vehicle? I need to report a problem. Give me battery health data for all of platform type As. What are the new maintenance requests? Do I have the capacity? Operations Logistician Planner Do I have assets required for the upcoming mission? What risks do I mitigate or introduce? When do I need to buy material for systems that are predicted to fail? Vehicle Type A Block 1 Vehicle Type A Block 1 Vehicle Type A Block 1 Vehicle Type A Block 1 Vehicle Type A Block 1 Vehicle Type A Block 2 Vehicle Type A Block 2 Vehicle Type A Block 2 Vehicle Type A Block 1 Vehicle Type A Block 1 Vehicle Type A Block 1 Vehicle Type A Block 1 Vehicle Type B Block 2 Vehicle Type A Block 1 Vehicle Type A Block 1 Vehicle Type A Block 1 Vehicle Type A Block 1 Vehicle Type A Block 1 Vehicle Type A Block 2 Vehicle Type A Block 2 Vehicle Type C Block 2 Vehicle Type B Block 1 Vehicle Type C Block 1 Operator How can I automatically report when the remaining useful life of an LRU drops below threshold? Integrated Architecture Transfer
USMC Autonomic Logistics Architecture Internal Message Manager GCCS GCSS AL Operator / Maintainer Presentation & Interface EXternal Message Manager Message Manager Adapters Platform Level e Book Configuration BIT/BITE Sensor Maintenance Operational Embedded e Data Tech Data Fault Tree RPSTL e PMCS Serial UID/AIT Configuration Remote Diag. Data Sync Remote Diagnostics Alerts Trending Algorithms HUMS Data Download Manager Platform LRU LRU LRU LRU LRU LRU LRU
US Army Common Logistics Operating Environment
US Army Common Logistics Operating Environment Maintain on-platform persistent health data store using MIMOSA OSA-EAI CRIS or Tech-XML Bulk health data transfer using MIMOSA OSA-EAI Tech-XML Manage off- platform persistent health data store using MIMOSA OSA-EAI CRIS Emergency Resupply Request using JVMF K07.xx messages Replicate to CONUS using COTS tools Query persistent health data store using MIMOSA OSA-EAI Tech-XML
CACE - Coherent Analytical Computing Environment CACE is an example of a data mining application which taps into databases of maintenance requests, repair schedules, and training schedules to optimize flight operations
Access to Data
Need access to data for all information customers
Embedded ISHM systems
Latency of information will determine value to different customers
Who is responsible for maintaining the data warehouse?
Open Systems Architecture for Condition Based Maintenance
Standardized architecture for health and condition monitoring systems
Breaks monitoring system into functional layers
Defines inputs and outputs required for each layer
Modules not confined to one locale
Uncertain Healthy Faulty Advisory Generation Prognostic Assessment Diagnostic Health Assessment Presentation Alarm & Event State Detection Envelope Detector Band-Pass Filter Half-wave or Full-wave Rectifier Peak-Hold Smoothing Accelerometer Data Band-Pass Filtered Signal Rectified Signal Envelope Detected Signal 2 1 1 2 1 4 1 4 m j N i j ij N i i r r m r r N NA Data Manipulation Smart Sensor Distributed Computing Technology Data Acquisition Sensor
Standards for Enabling System Management Machinery Information Management Open Standards Alliance (MIMOSA) MIMOSA Technology Types REG (Physical Asset Register Management) WORK (O&M Agent Work Management) DIAG (Diagnostics / Prognostics / Health Assessment) TREND (Operational Scalar Data & Alarms) DYN (Dynamic Vibration/Sound Data & Alarms) SAMPLE (Oil/Fluid/Gas/Solid Test Data & Alarms) BLOB (Binary Data/Thermography Data & Alarms) REL (RCM/FMECA/Model Reliability Information) ALGORITHM (Algorithm Management Information) AGENT (Intelligent Agent Management Information) TRACK (Physical Asset GeoSpatial Tracking Info.) FORECAST (Capability Forecasting & Projections)
Beyond System Health Monitoring Need techniques to capture decisions and actions in addition to data and permit machines to mine those high level control “lessons learned” Intelligent Monitoring Fault Detection Safety Fault Location & Severity Maintenance/ Repair Diagnosis & Remaining Life Condition- Based Maintenance Prognosis Health Management Intelligent Control + Relating Subsystem Health Intelligent Systems Management Vehicle Monitoring & Control Vehicle Management Intelligent “ System” Automation Mission Management
Other Data Mining Challenges
Indexing data for efficient computer access
Ownership of data, data rights, and security
Distributed processing architectures
Data summarization, trend detection anomaly detection are key technologies
Data mining is the search for hidden relationships and meaning in data
Applications in ISHM include modeling, diagnostics, and prognostics
Useful for system development and for closed-loop improvement of systems
Biggest payoff could be in creation of virtual transitional failure data sets
Information customers are driving for network-centric applications so databases and data warehouses will be there – we need to be sure the data is useful for ISHM