Learning
from a system’s past performance to automate monitor and control.
ACT: Adaptive
Control Technology based on learned behavior and applied to real-time monitor
and control.
Using funding from a NASA STTR, ICS has developed new technologies which allow us to extract behavioral characteristics from past performance data and automate the real-time monitor and control of a system with a high degree of accuracy. Using Adaptive Machine Learning techniques, we can mine historical data to define relationships and discover operational signatures for groups of related sensors.
One of the greatest producers of historical performance data in the aerospace industry is the Space Shuttle. One of the most critical components of the Shuttle is the Space Shuttle Main Engines (SSME). Many interrelated SSME sensors must be monitored in real-time and compared against expected behaviors; an out of tolerance value can cause a launch scrub…in flight, a failure could be disastrous. An on-the-pad scrub of a shuttle can cost taxpayers more than $1 Million.
Since 1981, SSME faults have caused 23 scrubbed launches and 29% of total Shuttle downtime. The most serious cases typically occur in the last few seconds before ignition; a launch scrub that late in the countdown usually means a period of investigation of a month or more. For example, the scrub of STS-41D at T-4 seconds forced the cancellation of another mission, plus a two-month delay. STS-51F was aborted at T-3 seconds in July'85; two weeks later, the Shuttle had to do an abort-to-orbit when one SSME shut down prematurely. Another T-3 second SSME malfunction, this one on STS-55, delayed launch by over a month, and the same thing happened five months later to STS-51 and in 1994, STS-68 scrubbed at an unbelievable T-1.9 seconds.
ICS
and NASA decided to attack this problem by looking at well-characterized
actuators and valves used to gimbal the SSMEs.
The actuator at the left is one used to gimbal one of the three SSMEs on
a shuttle. Each actuator is extensively
tested at the bench level as well as sub-assemblies and on the shuttle
itself. Each actuator has performance
data catalogued which defines the expected operational characteristics that can
be observed during operations. Prior to
launch and post-launch, experts normally have to manually sift through
performance data to determine if any unexpected readings have been
recorded. For a few sensors, this is
within the realm of human capabilities.
Escalate the problem to hundreds of sensors for each SSME and tie in the
need for three SSMEs to operate in concert, and you quickly exceed the human
capacity. Each SSME produces 12.8M
bytes of information per second.
Real-time monitoring and correlation of data from these systems requires
an efficient real-time application capable of responding within
milliseconds. ICS has developed the SCL
real-time executive for real-time spacecraft monitor and control, but
population of the system has always proved labor intensive. This AML tool automates the population of
the Expert System toolset.
In partnership with Florida Institute Technology’s Dr. Phillip Chan and his students, ICS is developing algorithms that use Clustering techniques to automatically isolate groups of related values in historical data. Groups of related data defined in each cluster represent distinct mode changes and can be represented as a signature for that group of data items. Working in tandem with our existing intelligent technology, those mode signatures can later be recognized as nominal (or “off-nominal”) as a function of commands, time or behavior of other sensors. In real-time the SCL executive can determine off-nominal behavior and notify an operator, another system, or safe the system by sending a series of commands to gracefully shut down.
|
|
AML techniques have isolated clusters in valve actuator current data at the left. You can see that the behavior of one valve can be complex over short periods of time (milliseconds). Clustering allows characterization of nominal patterns; then a deviation from the expected behavior would represent a failure that can impact the safety of the astronauts and the mission. This is far beyond the traditional limit and range checking used in current monitoring systems. |
|
|
|
|
|
In addition to enhanced quality and safety, capturing information and leveraging its capability for real-time monitor and control has potential for costs savings in the millions of dollars. We expect to discover new relationships between sensors that are too complex to be seen using traditional manual processes and visual inspection. Adaptive Control Technologies have wide applicability within as well as outside the aerospace community. The ACT capabilities will be implemented as part of our standard data acquisition system. ACT capabilities will benefit many of our existing and future customers. This tool will complement our current software and join the recent addition of the NASA Ames Livingstone model-based reasoner and the JPL continuous replanner (CASPER) for autonomous operations such as the Air Force Research Laboratory’s TECHSAT-21 satellite constellation.
ICS is a privately owned company dedicated to intelligent command and control applications for mission critical systems.
For more information, please contact: