Preventing the failure of large-scale facilities such as nuclear power plants! The SIAT analytics technology that analyzes vast quantities of sensor data to find anomalies
- YoshihiraTo be exact, it involves more than just vertical fluctuation, as a certain mathematical expression is used. SIAT attempts to find relationships between all sensors, no matter what the data type, as long as there are values associated with them. Then, by extracting and compiling these relationships, or in other words the invariants, we can quickly build a model for when the system is operating normally using just observation data.
- MitaI see.
- YoshihiraOf course, it isn't possible to create a perfect representation of a system's characteristics with this invariant model, but covering the overall system thoroughly makes it possible to perceive anomalies that people would be less likely to notice. Instead of portraying a single characteristic in precise detail, SIAT's unique approach is to give a comprehensive picture of the many relationships across an entire system, so no stone is left unturned.
"Invariant Model" created from plant sensor data
- MitaWhat do you do after you've found lots of those invariants and built a model?
- YoshihiraOnce the model for normal operation is complete, you compare this with the steady stream of observation data coming from the sensors. If this observation data deviates from the forecast values obtained from the model, you can tell there is an anomaly.
- MitaSo you compare the model for correct operation with the data coming from the sensors, and if they don't match, it's an anomaly?
- YoshihiraExactly. For example, imagine a water pipe. If sensors were installed at the entry and exit points, you could detect that under normal circumstances when the flow of water becomes stronger at the entry point, it also gets stronger at the other end. When the entry flow becomes weaker, the exit point flow also weakens. This is an invariant. If the flow strength relationship between the entry and exit points changes, there is a chance that an irregularity has occurred. This could indicate some kind of anomaly, such as a rupture between the entry and exit points. The difference from the conventional method is that instead of looking at the water flow itself, we examine the relationship between the entry and exit points.
- MitaThat's interesting.
- AsakuraLet me use this diagram to explain how anomalies are actually found.
Invariant model mapped to plant block diagram
- MitaI see a lot of dots and lines linked together.
- AsakuraThis is a relationship diagram created using data from the sensors installed in a certain plant. This maps the relationships between sensors that act the same (invariants) to an illustration of the plant, with the dots on the diagram representing sensors, and the lines showing the relationship between sensors that demonstrate the same behavior.
- MitaSo you can see how sensors that behave the same are distributed throughout the factory.
- AsakuraFirst, a model is created using the process that Mr. Yoshihira explained earlier. Then, it is implemented on a diagram like this to monitor whether there are any anomalies. When the relationship between sensors breaks, the lines turn red. Like this...
When indication of an anomaly is detected
- MitaOh, a few of the lines have turned red!
- AsakuraThis shows at a glance which sensors have a broken relationship. For example, when many lines from a sensor in one place indicated by the red circle are broken as shown in this diagram, there is likely a problem in that area. This enables you to narrow down where a fault may have occurred, or where problems may surface in the future.
- MitaI see. I think I have a much better understanding of SIAT now. What else can it be used for?
Menu starts here.
End of menu.