Cyberphysical systems refer to systems that have both software and hardware components, such as simple robots, unmanned aerial vehicles, and satellites. As they operate, these systems may experience faults or anomalies due to hardware degradation, erroneous software, or unexpected interactions with their environment. Detecting and identifying these issues is challenging, and it is difficult to find the right balance: overly sensitive detection will flag normal behaviors as issues, burdening system operators in analyzing false positives, while under-sensitive detection risks missing actual problems before they cause bigger and more expensive failures.
Traditional ways of detecting these issues often use threshold-based methods, where an issue is flagged when sensor readings cross above or below manually pre-defined thresholds. Fault detection can also be done with physics-based models, where sophisticated models of the system rely on physics principles to estimate how the system will perform during operation. This model then compares its estimates to the monitored sensors’ measurements and raises an alert if the difference between the expected and actual performances is too large. Both methods require extensive subject matter expertise and significant resources to develop and utilize them properly.
At Global Technologies, we’re using data-driven techniques to simplify fault management for these systems. By feeding a system’s recorded data into our signal processing and machine learning frameworks, we can abstract a digital twin of the physical system that can accurately estimate the system’s normal behavior in different scenarios. This allows us to automatically generate an operational baseline to start monitoring the health of the system rapidly. The process requires very little subject matter expertise and can save a lot of time and resources for system design and development teams.
Further, the abstracted system can be visualized as a graph of interconnected operational variables or nodes. These visualizations can help system operators to understand and isolate anomalies very quickly, allowing fixes to be developed and implemented faster.
Below you can see an electrical aerial vehicle propulsion test bed and its abstracted graph as an example.
Want to know how this could benefit your systems and missions? Connect with us at aradiss@globaltechinc.com or on our Contact Us page.
