NASA Logo, National Aeronautics and Space Administration

Testbed Simulator

The testbed simulator has been developed to aid in the design of PDM algorithms for the testbed. It captures both nominal and faulty behavior, with the controlled ability to inject faults. In this way, it serves as a virtual testbed through which algorithms can be initially tested and validated. Faults in the simulator are modeled as undesired changes in system parameters or configuration. In addition to serving as a virtual testbed, the simulator may also be utilized in guiding the decision making process. A graphical user interface was developed for interacting with the simulator and injecting faults.

Diagnostics

Diagnosis is the process of detecting, isolating, and identifying faults in the system. Diagnostic approaches can be broadly divided into two types: model-based and data-driven. Model-based methods rely on a system model built from a priori knowledge about the system. Data-driven methods, on the other hand, do not require such models but instead require large, diverse sets of exemplar failure data, which are often not available. Currently a model-based approach is adopted for providing a diagnostic system for the rover, as the sensors and fault modes lend themselves to physics-based modeling. Once sensors measuring more complex dynamics (e.g. accelerometers) are added to the system, data-driven diagnosis methods may be required. Additionally, model-based and data-driven algorithms can be synergistically combined to improve upon either approach implemented individually.

Prognostics

Prognostics is defined as the process which predicts the time when a system variable or vector indicating system health no longer falls within the limits set forth by the system specifications (End-of-Life or EOL). The prediction is based on proposed future usage. In some cases the trajectory of the aforementioned variable or vector through time is predicted as well. Similarly to diagnostic methods, prognostics methods are generally classified as either data-driven or model-based.

Generally, the inputs to a prognostic algorithm include information on the fault provided by the diagnostic algorithm (e.g. fault type, time and magnitude). Output of a prognostic algorithm could be then presented to a PDM algorithm expressed as a probability density function (pdf) or as moments derived from the probability distribution. The distribution may change from one time of prediction to the next as more information about the behavior of the system becomes available.

In particular, we are interested in performing prognostics for mechanical jam and windings insulation deterioration, battery capacity deterioration and charge tracking, and electronics faults.

Decision Making

One of the main objectives of the K11 testbed is to investigate PDM algorithms in order to enhance an aerospace vehicle’s capability to achieve its high-level goals – be it under a faulty condition, degraded operation of a subsystem, or an anticipated catastrophic failure. There has been an increasing amount of research conducted over the last several years in prognostic methodologies for various types of components or systems.

Several factors are being used to select the appropriate system level (or levels) on which to respond to an off-nominal condition. These factors include the severity of a fault, its criticality, and predicted time-to-failure interval. A faulty electronic component in an electric motor driver could prompt the decision-making system to trigger a controller reconfiguration - so as to ensure the dynamic stability of the system and a certain level of retained performance. At a different level, a control effort reallocation can be triggered by a supervisory mid-level controller in order reduce the torque required from a faulty drive motor and compensate for the reduction with the other motors. Reallocating the load could, potentially, extend the remaining useful life of the affected component long enough to ensure achievement of the mission objectives. At the highest level, the rover mission can be re-planned based on prognostic health information so as to achieve maximum possible utility and safety. The above examples call on different system components in their response; there are, however, commonalities for all of them. There is always an objective (or a set of objectives) to be met and a series of actions to be selected by the decision making process in order the meet those objectives. Therefore, the decision making process is, essentially, an optimization process which tries to achieve specified objectives by considering system performance and health constraints.

The scope for the decision-making module in the current implementation is defined as the following: getting vehicle health information from the prognostic health reasoner and the simulator, the decision-making module evaluates the best course of action to take (e.g., controller reconfiguration or mission replanning), while stopping short of performing the actual reconfiguration or re-planning. Instead, the decision-making module adjusts goals and constraints for other software components. To use the planner as an example, the module could set a constraint on rover speed or on the total mission duration, then request the planner to come up with a detailed new plan. In the future, however, it may be necessary to consider whether making PDM-specific modifications to the planner or the adaptive controller, for instance, would improve performance.

Task Planning and Execution

Once the high level goals and constraints are determined by the prognostics-enabled decision making module, the detailed task planning for the rover will be generated using NASA’s Extensible Universal Remote Operations Architecture (EUROPA). EUROPA provides the capability of solving task planning, scheduling, and constraint-programming problems.. In a complex system, such as a rover, scheduling specific tasks to be executed is often a non-trivial problem. There are resources that are shared by different processes that may not necessarily be available at all times, so EUROPA supports generation of a schedule of activities. Plans and schedules generated by EUROPA (either nominal or those generated in response to a fault) will be passed for automated execution via Plan Execution Interchange Language (PLEXIL).

First Gov logo
NASA Logo - nasa.gov