Given the requirements for autonomous control and human rating for the next generation of space exploration vehicles, control and diagnostic functions for these vehicles will require that data used by these functions be analyzed and qualified to represent the state of the system being measured. A Sensor Data Qualification System (SDQS) is one approach for addressing data qualification requirements. Sensor data qualification is the development of a mathematical network of constraints using analytical redundancy to assess the health of a particular sensor in a suite of sensors that are measuring the condition of a given system. An SDQS for a power distribution unit (PDU) testbed was developed and evaluated at the NASA Glenn Research Center as part of a longer term activity to develop methodologies for assessing the health of Ares I upper-stage systems.
Algorithmically, the SDQS has three primary functions: predict, detect, and decide. To predict the value of a given sensor in time, measurements from related sensors are used in conjunction with predefined mathematical equations (i.e., local models) that describe how each sensor relates to the other sensors measuring the system state. The residual, the difference between the measured value and predicted value, is then computed for each relationship. A relationship failure is detected when the residual exceeds a preset threshold for a specified number of consecutive cycles. The decision to declare a sensor failed is made when the frequency and number of failed relationships for that sensor reaches predefined limits over some sampling interval--said limits being determined by the use of sensor reliability information and the application of Bayesian probability theory to the joint probability distributions.
As part of a feasibility demonstration to support the Avionics System Requirements Review for Ares I, SDQS networks were applied to a testbed developed by Glenn’s Advanced Electrical Systems Branch. The testbed was composed of a single power supply, a prototype PDU, and three load banks. A schematic of the testbed and a photograph of the PDU hardware are shown in the following figure. The system was operated in one of three modes depending on the number of active loads--one, two, or all three. The magnitude of each load was variable and unknown during operation.
PDU testbed and the PDU hardware used during testing. CANbus, Controller
Area Network Bus (high-speed, high-integrity, serial data communications
bus for real-time control applications, http://www.mjschofield.com);
Vsense, voltage sensor; LEM, current sensor; CAN, port used
to connect to the CANbus.
Long description of figure 1.
The active sensor network consisted of 6, 9, or 12 active sensors, depending on the number of active loads. Relay states and output load requests were also available as discrete data. Three nominal runs provided data for determining the mathematical relationships between the sensors. Sensor failure data were obtained by superimposing simulated fault signatures on nominal data during data acquisition. Four primary sensor faults were identified as shown in the graphs: hard (high or low), drift (low, medium, or high), intermittent-binary, and intermittent-filtered. Using these faults in combination with various sensors, Glenn researchers simulated 13 different sensor failures for each mode. The SDQS correctly identified all 13 faults in each of the three operating modes, with no false alarms or missed detections.
Four primary fault signatures were simulated and superimposed on the nominal data to test the diagnostic performance of the SDQS. (a) Hard failure (open or short circuit). (b) Drift failure (thermal or resistance change). (c) Intermittent, binary (loose connector). (d) Intermittent, filtered (cracked solder joint).
Long description of figure 2.
This study used data playback to investigate the failure of a single sensor in a network of sensors. Future research is planned for detecting multiple sensor failures, identifying sensor failures in the presence of a failed component or in a closed-loop control system, and investigating issues associated with real-time implementation.
Maul, William A., et al: Sensor Data Qualification for Autonomous Operation of Space Systems. American Association for Artificial Intelligence 2006 Fall Symposium on Spacecraft Autonomy, Washington, DC, Oct. 13-15, 2006, AAAI Technical Report FS-06-07 (NASA/TM--2006-214475), pp. 59-66, 2006. http://gltrs.grc.nasa.gov/Citations.aspx?id=192
Find out more about the research of Glenn’s Controls & Dynamics Branch: http://www.grc.nasa.gov/WWW/cdtb/
Glenn contact:
Kevin J. Melcher, 216-433-3743, Kevin.J.Melcher@nasa.gov
Analex Corporation contact:
William A. Maul, 216-977-7496, William.A.Maul@nasa.gov
Authors:
Kevin J. Melcher and William A. Maul
Headquarters program office:
Exploration Systems Mission Directorate
Programs/projects:
Crew Launch Vehicle Project
Last updated: December 14, 2007
Responsible NASA Official:
Gynelle.C.Steele@nasa.gov
216-433-8258
Point of contact for NASA Glenn's Research & Technology reports:
Cynthia.L.Dreibelbis@nasa.gov
216-433-2912
SGT, Inc.
Web page curator:
Nancy.L.Obryan@nasa.gov
216-433-5793
Wyle Information Systems, LLC
