Actionable feedback right through edge computing

ASHOK AGARWAL
Thursday, April 7, 2016

Continuing our discussion on Edge Computing and Analytics ….. Remember WE SAID that a key benefit of Edge was Local Decision Making. Typically, that will preclude access to the install base data. However, there is a wealth of information which can be gleaned from the install base data (such as machine learning output). It seems a shame to not be able to utilize that on the edge.

Lets take an example of predictive analysis based on a machine learning model. Lets say there is an IoT device (perhaps a medical device) which contains a servo motor (moving a catheter through the arteries). The motor has a finite life and is dependent on hours of “typical usage”. Lets take a simple case where “Typical usage” is dependent on real usage, location, temperature and humidity of the environment in which the machine is being used.

Clearly the Edge does not have the data for all the devices in the install base, but the core does. All the elements mentioned above are captured on the edge and sent to the core. The core also gets the historical servo motor replacement data from a CRM system.

Combining these data sources into a tool like THINGWORX MACHINE LEARNING, a multivariant regression analysis model can be created which will map the remaining life to the predictor variables listed above. End of the day, all statistical models are represented in the form of an equation. The equation for a multivariant regression model is:

Y = m + n1X1 + n2X2 …….

Where Y is the “remaining life” and X1, X2 are the predictor variables. m is the intercept and n1, n2 etc are the slopes for each predictor variable. Based on historical data, computing the equation for any machine learning tool is a piece of cake. But wouldn’t it be nice if this equation was available on the edge, so that for each incoming observation you can locally decide how much life is remaining? How easy would it be to plan for replacements if you could do that?

Right, that’s where Glassbeam comes in. As I mentioned before, Glassbeam’s Rules Engine is a Message Bus based Streaming engine which applies the rules as data streams are processed. So, this is how we do it:

  1. Define a rule which implements the equation derived by the machine learning algorithm
  2. As data comes in, pass it to the FSM (Finite State Machine) applying the rule
  3. If all conditions match, an alert will be triggered if the remaining life is less than a threshold.

We just implemented a decision taken by the core at the edge. This is part of Glassbeam’s feedback mechanism – an inherent component of Glassbeam’s Edge solution.

Home Section banner