The automotive industry is living a revolution in the last years due to the level of electrification and increased presence of smart systems in the vehicles. The combination of both factors opens the possibility to add more and more functionalities in compact and smart components. Nevertheless, the behavior of electric actuators can suffer subtle variations in operating conditions which affect their response. One of the main reasons for these changes is the temperature, but there are also internal nonlinearities like for example magnetic saturation. A correct estimation of the changes in the actuation permit a better control of the system and the identification of risky conditions. This is especially important in safety components. The high compacity required in the components and also the price restrictions needed for a mass product makes it difficult the addition of many sensors and so it is necessary to search for other solutions to detect the parameter variations in the motor. A classical approach is the use of estimators like Luenberger or Kalman filters which are classically used in control architectures for improving the estimated value coming from sensors and identifying unknown or variable parameters. Nevertheless, these estimators update the values at each time step based on a single observation and depending on the specific conditions at a time observability issues may affect the reliability of the obtained value. In the last years new algorithms coming from robotics are proposing new algorithms for overcoming the previous limitation. For example, in the field of simultaneous location and map building (SLAM) which traditionally Kalman filters there is a trend for substituting this estimation technique and using graph based optimization. This solution uses a data record for improving the estimation and therefore is less prone to observability issues. Thanks to the developments in robotics, there are highly efficient graph optimization algorithms. In the present work we compare the use of Kalman filter and factor graph for estimating the resistance and magnetic constant of an electric motor for an automotive component. These factors are very important in control applications due to its variability and the effect in the estimation of the torque applied by the motor. This analysis shows the problems derived from the use of the Kalman filters on certain situations and the improved performance obtained with factor graph. This improvement has a direct effect in the robustness of the vehicle component. Factor graph is mostly applied in robot positioning applications, and to the best knowledge of the authors this is the first time that it is used in this sort of application.
Mr. PABLO MARIN, R&D ENGINEER, ITAINNOVA