Software sensors, often referred to as virtual sensors, provide an indirect measurement by computation of a measurand. The core of every software sensor is a model for calculating the desired sensor value from a variety of input signals, which typically are values of other sensors in the system.
Software sensors are mostly used where deploying a physical sensor is not possible, failure-prone, or just too expensive. Despite their obvious advantages, only few companies are yet making use of software sensors – building the underlying data models typically requires very large amounts of data and is very time-consuming. In addition, it requires manual intervention and deep involvement of data scientists with domain expertise. This made the generation of software sensors slow and expensive.
PerfectPattern now introduces AI Sensors, enabling production experts to create software sensors much faster, easier and with much less training data than before. Mathematical or data science expertise is not required – the system generates the underlying models unsupervised without the need for human intervention.
Working on asynchronous and unfiltered data, AI Sensors builds those models very quickly and requiring less training data compared, for example, to Artificial Neural Networks. Beyond that, the system 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 in an automated manner.
Even in complex scenarios with learning data from thousands of physical sensors, including historical data, AI Sensors computes a new software sensor model usually within one hour. Using conventional methods, this process could take several days or weeks. Updating the sensor is also quick and easy.
The software sensor is provided as an executable program. Its execution time is typically just a few milliseconds. It can be used in closed-loop and real-time scenarios, for example to control and optimize complex production lines.
AI Sensors is based on PerfectPattern´s AI technology platform PYHTIA, enabling pattern recognition, time series prediction and anomaly detection in real-time data streams. By combining methods from deep learning, stochastic calculus, infinite dimensional geometry and quantum field theory, PYTHIA offers decisive advantages compared to typical procedures of data analysis. It independently finds even the most hidden patterns, and the system learns how to predict and control any desired variable.
PerfectPattern was established in 2012 as a technology and software company based in Munich, Germany. The team creates software solutions that combine unique, powerful algorithms for process optimization with artificial intelligence. The solutions enable the fully automatic planning of virtually any production process in real time, making the “smart factory” a reality. The three key objectives are flexibility, punctuality and the reduction of production costs. More information is available at www.perfectpattern.de.