A perfect automated anomaly detector
Key takeaways:
- Through a deep learning architecture, Falkonry has developed an efficient, understandable, and accurate AI that is incapable of missing anomalies.
- Falkonry's automated anomaly detection lets you optimize the value of your attention by highlighting emerging hotspots in your operations that are deviating from normal.
- As this is achieved without significant setup efforts and in an automated cost-effective manner, it is a great way to start your 'Smart' initiatives.
Ever since machinery and production processes became complex, we have been on a quest to accurately and automatically identify unexpected or unusual behavior from industrial automation data. These unusual behaviors – or instances where the data differs significantly from the expected – are what we call anomalies. Now, you may ask, why does anomaly detection matter in the first place? The answer is simply that we cannot possibly pay attention to all the data we generate today. We need something that will narrow down the data to only the parts that are important – the parts that are unexplained and therefore likely to provide us with a learning opportunity. These anomalies are what we can spare some attention to in order to keep production in our control.
It is said that the most valuable information is contained where you will find the most uncertainty. One way of looking at this uncertainty is “unusualness”. A perfect anomaly detection system is capable of never missing unusual behavior and also accurately measuring the degree of unusualness. Through a deep learning architecture, Falkonry has developed an efficient, understandable, and accurate AI.
This AI has such an extreme level of competence that Falkonry is incapable of missing anomalies. And because we are able to reliably measure the degree of unusualness, you have the option of defining how unusual a behavior has to be to get your attention. It then becomes a function of how much attention you want to give to a certain class or source of data, and you can dial up or down how much you want to know or participate in learning about your data by simply choosing what level of unusualness you want to work with. Effectively, this perfect anomaly detection system optimizes the value of your attention.
What does this mean in practice? It means that you can find all of the behaviors that need your attention distilled down from all the information that is being produced. Let’s say that you look for anomalous behavior close to three standard deviations from the mean within a normal distribution. Statistically, that means 99.95 percent of data won’t be unusual. So by choosing the level of unusualness, you can narrow down your interest or your attention to just 0.05 percent of the data. In other words, you could ration your time and your attention and achieve 99.95 percent improvement over exhaustive analysis without getting overwhelmed.
We are now at a point where perfect, automated anomaly detection in industrial automation data doesn’t require any people and soon enough pattern recognition will not require any either. Already, it only requires supervision by domain experts and soon they will only be responsible for validation, with the rest managed by computers. Achieving all of this effortlessly (requiring no setup) makes it a great place to start your Smart initiatives.
If you would like to see how our perfect anomaly detection system can benefit your operations, watch this video of Falkonry Insights in action, or reach out to us for a live demo
