Autonomous diagnostic tool identifies failures in satellite
Artificial intelligence software Livingstone Version 2 (L2) successfully
identified and diagnosed more than a dozen errors in an Earth-observing
Computer scientists deployed L2 to the NASA Earth Observing One (EO-1) satellite in September to find and analyze a number of simulated failures while the satellite's autonomy software executed the satellite's imaging process.
"Everything went perfectly," Sandra Hayden, principal investigator and project manager for the Livingstone on EO-1 experiment, adding that one of the 17 test scenarios failed, but that scientists had expected that. L2 will continue flying on EO-1, trying to detect and identify any failures that occur during its normal operations.
Software like L2 that can find failures in systems before problems become critical will significantly reduce NASA mission operations costs and boost mission efficiency, and could one day save a rover, a spacecraft, or even a human life. Today, when an problem in a system occurs a large team of mission controllers troubleshoots to solve it. This method is expensive in labor costs, and it is becoming riskier as exploration systems are becoming more complex. As the space agency expands exploration deeper into space, where communication delays preclude intervention by controllers on Earth, diagnostic software like L2 will be critical to safe, reliable missions.
JPL's autonomy software, the Autonomous Sciencecraft Experiment (ASE), is in control of EO1, launched in 2000 as a platform for testing new technologies and strategies for improving missions while reducing cost and development time. NASA's Goddard Space Flight Center manages the satellite.
Diagnosis in Space
Livingstone uses a model of the system to predict its behavior. If actual behavior diverges from the model's predictions, a diagnosis is made to isolate the cause of the discrepancy to a specific failure. The first version of Livingstone could give mission operators only a single candidate for an error. L2 provides several hypotheses of what went wrong. L2 updates a diagnosis based on the histories of the sensor data and the commands that have been sent to the spacecraft. L2 continually monitors a spacecraft's state and history, providing the most accurate, up-to-date diagnosis in real time.
"Cutting-edge space technologies are growing in complexity and sophistication, as we embark on more ambitious space ventures," Hayden said. "That's why it is critical to make sure these systems behave as their designers intended, and to diagnose accurately when things go wrong. This gives us a chance to recover from errors, protect our investments in space, and continue on to achieve our mission goals. Model-based diagnosis is a means to that end."
L2 on EO1 is Ames' first demonstration in space since 1999, when the first version of Livingstone flew on the Deep Space One spacecraft with the Remote Agent Experiment, an autonomy demonstration named one of the 10 greatest achievements in artificial intelligence.
Supporting Exploration Systems
A tool like L2 would have been helpful this spring when Mars Exploration Rover (MER) Mission staff realized that rover Opportunity had spent the previous day spinning its wheels on the slippery soil of a crater slope.
During a blind drive, the rover had taken a right turn, which should have been made outside the crater, while still inside the crater. A worst-case scenario could have been a toppled-over rover, and an untimely end to the $400 million mission.
During the MER mission, more than 200 mission staff members worked around the clock to oversee the direction and safety of two MER rovers. For future, affordable long-duration rover missions, that number will have to be cut drastically.
Before sending humans to Mars, where Earth-based crews cannot monitor what is happening in real time, better automatic diagnostic tools for spacecraft and robots are needed. When people start traveling deeper into space, automatic diagnostic tools can tell crew and controllers about a potential problem in sufficient time to make repairs.
"In a future long-duration (at least two years) human mission to Mars, things will fail. Frequent and extensive spacecraft maintenance operations or overhauls will not be an option, so we need to get smarter about recognizing degradations or failures early on and mitigating their impact on the success of the mission " says Serdar Uckun, ISHM technology lead for Ames Research Center's Exploration Systems Office.
Researchers at Ames built models of the EO1 spacecraft, its instruments, its cameras, and one of its processors. L2's model-based diagnosis approach is more technically advanced than traditional approaches. The Livingstone reasoner is given a model of the system over which it is to watch. This separation between the reasoner and the object of the diagnosis means that with less effort, Livingstone can be applied to diagnose new systems. The diagnostic reasoner doesn't need to change; only a new model has to be developed for the system to be diagnosed.
Many systems have the same parts—such as valves, switches, and sensors—so a model can be built more quickly by reusing these common parts. In this fast-paced experiment, the development of the models and the integration of L2 with the autonomy software onboard EO-1 has all been done within the last year. L2 can use the same model for simulation and diagnosis. In the future, models might be developed that can be used both for planning and diagnosis.
Another planned extension is for L2 to address software failures. Current mission protocol for preventing failures is manual software testing. However, bugs inevitably slip through. Meanwhile, software code is growing in size as vehicles become more complex, introducing more likelihood for errors, creating a prime opportunity for L2 to add value.
Livingstone has proven valuable as a research tool in academia and as a diagnostic application at several NASA centers. The tool has been used in several NASA technology demonstrations. Livingstone developed a strong following in the aerospace industry through NASA's Space Launch Initiative, an effort to define, develop, and test technologies for a reusable launch vehicle.
The continued research and development of tools like L2 will make future NASA missions safer, more affordable, and more effective.