This website requires Javascript, to work properly.

AI Sensors

Create software sensors fast and unsupervized

Many companies do not yet benefit from the huge advantages of software sensors. Because until now, building software sensors was slow, expensive, and required a lot of data science expertise. This changes with AI Sensors.

With AI Sensors you can create software sensors much faster, easier and with much less data than before. And the best part: You don’t need much mathematical or data science expertise to do so. Because AI Sensors does the work for you!

Unsupervized generation of software sensors

The core of every software sensor is a model for calculating the desired sensor value from the input signals. AI Sensors generates these models without the need of human supervision or intervention. Working on asynchronous and unfiltered data, AI Sensors builds those models very quickly and based on relatively small amounts of training data (e.g. compared to Artificial Neural Networks).

Advanced techniques for finding signal relations

AI Sensors makes use of very advanced techniques such as modelling signals by stochastic differential equations (SDE), tuning of hyperparameters, clustering data and distance correlations. This way AI Sensors finds even the most complex relationships between signals. It also is fully deterministic: Identical input will always generate the same output.

Define your software sensors

Your request can be very basic like “generate a sensor for signal S“, or more complex since S can also represent a function over time of multiple signals, as well as a forecast of statistical values. Anyway, if the data you provide complies with your request and allows the generation of a model, AI Sensors will do so. The result is a software sensor provided as an easy to use, executable program to be used in various environments.


No data cleaning or parameter tuning

AI Sensors makes manual data preparation almost obsolete. Tasks like data synchronization and cleaning, removal of anomalies or adding sensor values for missing timestamps are completely covered by AI Sensors in an automated manner. Also, manual parameter tuning is not required.

Great performance, easy to maintain

Even in complex scenarios with learning data from thousands of physical sensors, including historical data, AI Sensors computes a new software sensor model within one hour. Also software sensor updates therefore can be easily performed.

Prediction including root causes

Considering sensor signal response times, AI Sensors also models time series. This allows to predict e.g. a future signal progression including root causes, which influence the forecast of the software sensor signal.

Three steps

Step 1: Provide history data

The time series data can be unfiltered and asynchronous. It can be a few megabytes, but also several gigabytes. The data can represent any number of signals – just a few, but also thousands of different signals.

Step 2: Create the sensor

Your request describes the target software sensor, that you are about to build. This request might be straight forward like “generate a sensor for signal S“, or more complex. Either way, the request must comply with the provided data. The sensor then is built automatically by AI Sensors.

Step 3: Integrate your sensor

Once created, your software sensor will be provided as an executable program. In order to be used efficiently, it has to be connected to the live signal data sources. You should also define, how often you want your sensor value to be updated.

A demo will be available soon!

© 2018 PerfectPattern GmbH