Virtual sensors are made for measuring values, which can only be measured extremely hard, very expensively or even impossibly by conventional physical sensors. For this purpose mathematical models and software functions are used to calculate the desired values from more simple and cheaper available sensor signals.
- Machine learning techniques are well-suited to develop and calibrate virtual sensors because they can train the desired functionality easily and efficiently.
- Corresponding Data Mining methods capable of automated and efficient identification of the best possible basic sensors and information for the virtual sensors.
- Stochastic simulation is used amongst others for the generation of the necessary data for the effective development and validation of the virtual sensors.
Applications & References
Typical examples of virtual sensors are
- the estimation/prediction of collision probabilities, risks, criticalities and situation dangers in vehicle safety systems,
- the estimation/prediction of capacity reserves and congestion risks in traffic depending on the current situation and boundary conditions,
- the estimation of damage or failure probabilities for predictive maintenance and failure prediction,