In this work we investigated whether conclusions can be drawn on the usefulness of specific sensors for performing friction estimation. We are using a leading-edge machine learning approach, Echo State Networks (ESNs), for computing friction estimates based on sensor data time series. This provides two advantages: The enormous learning capacity of an ESN having several thousand internal nodes ensures that all relevant information contained in sensor data time series is exploited for achieving the best friction estimation possible. For finding a sensor configuration optimizing both accuracy and costs, ESNs offer a fast and un-biased way to assess and compare different sensor configurations. We demonstrated how the optimization of sensor configurations can be tackled using feature selection techniques. We used Backward Elimination to draw conclusions on the usefulness of sensors during longitudinal driving manoeuvres. Of course, a greater amount and a greater variety of driving manoeuvres have to be conducted in order to get a sufficient statistical data basis for friction estimation. Also, much more effort could (and should) be invested to find an optimal sensor configuration. This can be incorporated into our approach described in this paper by adding local optimization at interesting points in design space. Once an optimal sensor configuration has been determined, a friction estimation solution can be engineered in detail. In particular, as it is not preferred to use black-box machine learning techniques within safety-critical systems (e.g. ADAS), the final design of the friction estimation system has to be done by engineers. In this case, ESN solutions can be further used as performance benchmarks.
Friction estimation - optimization of sensor configuration with respect to RMSE and costs
2014
15 Seiten, Bilder, 12 Quellen
Conference paper
English
Friction estimation - optimization of sensor configuration with respect to RMSE and costs
Automotive engineering | 2014
|Friction estimation – optimization of sensor configuration with respect to RMSE and costs
Springer Verlag | 2014
|British Library Conference Proceedings | 1994
|Helicopter Flying Qualities Sensitivity Analysis with Respect to Aerodynamic Configuration
Trans Tech Publications | 2013
|