Synthetic Sensors: Leveraging Sensor Data using Machine Learning

When we talk about sensors, we often don’t get too far into what the mechanism is that is used to sense the movement or temperature, light intensity or humidity for example. We don’t even think much about the fact that multiple sensors may be in play, increasing accuracy of measurement or used to detect a complex movement or situation.

---

Great strides have been made recently regarding sensors, with large credit given to Gierad Laput Ph.D., a former Carnegie Melon University student. He developed what he refers to as Synthetic Sensors. His design includes multiple sensors on a small IC board which can be trained using machine learning algorithms. The data can then be used after this training to detect special events in a particular room, be it a car, workshop, office etc. in short, this platform can translate data from a single low-cost IoT device into many different events. Think along the lines of faucets running, appliance usage and paper towels usage at its most basic example.

High-order synthetic sensors

High-order synthetic sensors are able to be programmed on top of the virtual sensors in order to monitor more sophisticated events. Let’s look at a paper towel dispenser as one case study. One would be able to create a synthetic sensor to measure how many paper towels have been dispensed based solely on the sound of the dispenser itself (even though there is no actual physical sensor on the unit itself). The synthetic sensor acquired data can be used to drive an app to order replacement paper when the dispenser is empty or notify the maintenance department that a dispenser is empty. 

 Synthetic sensor will revolutionize how devices sense their surroundings .

Final Thoughts

---