Titel: Machine Learning based Approach to include Production-Related Variations in the Simulation of Magnetic Sensors
Autoren: Hagen Schmidt, Tim Becker, Jörg Seewig
Abstract: Simulations are an important tool in the development and design of magnetic sensors. The disadvantage of these simulations lies in frequent assumption of perfect conditions, without considering production-related variations. Depending on the system, these variations can have a significant effect on the sensor output signals, leading to an offset between simulation and reality. Running multiple simulations to cover all variations is often too time consuming. This weakness is addressed in this study. Machine learning techniques are used to train a model on production-related variations with simulated data. The model is then able to predict sensor signal deviations in real time based on the ideal signals. This allows designers to check their sensor design within the specified production variations without the need for additional time-consuming simulations. The approach is implemented using a virtual module of a supporting magnet and a GMR sensor array called GLM712 from Sensitec GmbH. This type of sensor is used to detect the angle of rotation and speed of tooth structures without contact. A characteristic of this type of sensor technology is the need for a specific design depending on the geometry of the tooth structure. In the manufacturing of these sensors, there are variations in sensor position, orientation and the properties of the supporting magnet, which have been shown to affect the signal. These are the variations for which the machine learning model is trained.
Keywords: Virtual Measurement, Simulation, AI, Magnetic Sensors
+49 (0)6441 5291 0
sensitec@sensitec.com
Ansprechpartner finden
Anfrage stellen