Virtual Sensor [Current State]

Physical sensors could be expensive, prone to fail or to drift. With the Ekkono SDK, you can create a virtual sensor based on related physical sensors. The virtual sensor can then replace a physical sensor, provide support on sensor failure or act as a reference when monitoring drift and anomalies. A virtual sensor “measures” signals without the need of a physical sensor. The sensor readings are instead generated from a model of the physical one. Virtual sensors can replace expensive sensors in the field and add robustness to the system, by acting as the sensor’s digital twin.

Edge Opportunities
Ekkono’s machine learning SDK makes it possible to create a virtual sensor on your edge device, which will act just like a physical sensor. Typically, the virtual sensor is a machine learning model that uses its surrounding physical sensors’ output data as input. The virtual sensor could be scheduled to read all sensor values at a given frequency and output the desired response in real-time. If the model is used as a replacement or a digital twin for an actual sensor, the physical sensor is used as a blueprint as to how the virtual sensor should respond to its surroundings.

When the model is fully trained, the virtual sensor will output almost the exact same values as the physical sensor. This means that an expensive sensor that has been virtualized can be removed completely. Another area of usage is to keep the physical sensor and compare its readings to the output of the virtual one. If the responses of the physical and the virtual sensor start to deviate, a technician can be sent to the site to investigate the machine. Until the technician arrives, the virtual sensor can override the faulty sensor’s output.

With the Ekkono SDK’s incremental learning capabilities, virtual sensors can be trained in real-time out on the field, with no need for additional historical data. The incremental learning process collects data over time, and learns to replicate the physical sensor.

Problem
Physical sensors may be expensive, challenging to install and require frequent calibrations. In addition, physical sensors can be subject to sensor drift and failure.

Solution
Create a virtual sensor based on related physical sensors. Use a reference sensor to train the virtual sensor, in a lab environment or incrementally on the field.

Application
Replace the physical sensor with a virtual sensor permanently, or at the time the physical sensor fails. Compare the physical sensor with the virtual sensor to monitor drift and anomalies.

Related content

More Use Cases

Health Monitoring [Current State]

Health Monitoring [Current State]

Health monitoring provides insights about system health beyond raw sensor values, and helps you prevent severe failures and unplanned production stops by exposing the true health of your machine. When using incremental learning, the need of historic data is replaced...

Predictive Alarming [Predictive]

Predictive Alarming [Predictive]

Predictive alarming lets you apply alarm levels on future sensor values, enabling the possibility to act in advance. Alarms on predicted values gives valuable input to operation teams, preventing potential problems from happening, thereby increasing uptime and...

Predictive Maintenance [Predictive]

Predictive Maintenance [Predictive]

Predictive maintenance minimizes downtime and optimizes periodic maintenance by moving from scheduled to smart maintenance planning, letting the health state of the machine, rather than a periodic time interval, decide when to carry out maintenance. Edge...