Wheel Load Reconstruction for Intelligent Vehicle Control

S.M.A.A. Kerst

Research output: ThesisDissertation (TU Delft)

22 Downloads (Pure)


After decades of incremental change in the automotive industry, we now face an era of disruption as environmental concerns and social change propel the introduction of electric vehicles and vehicle automation. Besides the clear benefit of zero-emission transport for society, there is a strong commercial incentive for automated driving, as it will lead to more efficient and safer mobility. A vast amount of research and development is therefore dedicated to its realization.
As human drivers are progressively taken out of the loop, intelligent vehicles impose increasing demands on the highly complex control loop, from measurement and perception to vehicle control. Of particular interest are limit and critical conditions, as optimal performance in these situations is paramount to maximize safety. Therefore, accurate real-time knowledge of the wheel forces is essential, since it represents the tire-road interaction of the individual wheels, determining vehicle behaviour and its handling limits. However, no commercially feasible method is available for the measurement of these important vehicle states.
Current vehicle control systems circumvent this measurement issue by focusing on downstream effects, such as wheel slip and body accelerations. Due to the focus on secondary effects these systems are overly complex and lead to sub-optimal performance. For optimal vehicle control of future intelligent vehicles, therefore, the development of wheel force measurement is considered invaluable. By providing direct access to the most important control variables for dynamics control, such measurement allows for less complex control algorithms with improved performance and robustness, and hence will lead to safer mobility.
Although various approaches for the reconstruction of wheel forces have been developed, no cost effective method is yet available. This can be explained by the fact that load measurement approaches generally require mechanical load decoupling to avoid crosstalk, something that is difficult to achieve on a wheel-end suspension setup that is already complex on itself. In this thesis, a novel method for wheel force reconstruction is proposed via load measurement at bearing level.
The concept of bearing load measurement dates back to the early ’70s and has been investigated by all major bearing manufacturers ever since. This has led to various measurement approaches based on relative ring displacement and outer-ring deformation. Despite all efforts, currently still no accurate nor robust approach for multi-dimensional load reconstruction is available. The state-of-the-art provides unsatisfying results due to the complexity of bearing behaviour and the inability of the currently applied data-driven methods to leverage unique bearing characteristics.
In this thesis a novel approach to bearing load reconstruction is proposed based on outer-ring deformation measurement and real-time simulation of bearing physics. The novel approach includes an explicit description of important physical effects as the rearrangement of rolling elements and the one-dimensional nature of their load transfer. As such it captures the bearing behaviour and allows to make use of its unique characteristics. The proposed approach is based on Kalman filtering and includes two independent physical models: a bearing strain model and a bearing load model.
The bearing strain model defines the outer-ring surface strain variation as a function of the local rolling element loading and location. The proposed model provides a simple though effective continuous and parameterized description of this behaviour. The model is implemented in an Extended Kalman Filter as a means of signal conditioning to estimate local rolling element forces from the measured outer-ring strain. By considering the change of strain due to the reallocation of rolling elements over time, a differential measurement is performed that results in invariance to thermal effects.
The proposed bearing load model is an extension of traditional rigid bearing modelling by a semi-analytical description of outer-ring flexibility. The latter is achieved by static deformation shapes and a Fourier series-based compliance approximation. The proposed model thereby provides a computationally effective but highly accurate description of rolling element forces for common bearing designs, in which significant raceway deformation occurs. Included in an Unscented Kalman Filter, the model provides the relationship between the estimated rolling element forces and the bearing loading and as such serves as a load reconstruction method. By explicit description of the individual one-dimensional element forces the approach considers the internal load decoupling effect and thereby limits crosscoupling on the estimated loads.
The wheel load reconstruction algorithm has been validated in both laboratory and field conditions on a production vehicle wheel-end bearing instrumented with
strain gauges. The study in laboratory conditions was performed on a bearing test setup at our industrial partner whereas the field validation has been performed on a dedicated test vehicle prepared as part of this thesis. Besides the proposed approach, a state-of-the-art algorithm and a variant including the model based signal conditioning method are evaluated to properly assess the results.
The experimental results show that the proposed approach leads to a considerable improvement in accuracy, reproducibility and robustness in comparison to the state-of-the-art data-driven approach. The proposed strain model-based conditioning approach leads to higher reproducibility and improved accuracy of up to 5 percent full scale due to its invariance to thermal effects and ability to discriminate in- and outboard rolling element forces. Additionally, the model-based load reconstruction method further improves accuracy by leveraging the internal bearing load decoupling behaviour to avoid crosstalk. This results in an improvement of over 5 percent full scale for combined loading conditions. Additionally, the approach is more robust, as important relationships are captured by modelling. The latter is well observed for loading conditions outside the calibration domain as an accuracy improvement of 6.8 to 18.4 percent full scale is achieved for the various reconstructed loads. The application of modelling furthermore leads to a significant reduction of parameters subject to calibration and provides physical meaning to these parameters.
Finally, an application study on anti-lock braking was performed to investigate both the load reconstruction performance in dynamic loading conditions and the advantages of load information for vehicle dynamics control. The study shows that sufficient signal bandwidth is provided and confirms the value of direct wheel force measurement for anti-lock braking control. In particular, as traditional difficulties like velocity estimation and slip threshold determination are circumvented whilst the effects of road friction fluctuations and brake efficiency are minimized.
By providing an accurate, robust and scalable solution for the processing of bearing outer-ring strain to the bearing loading, this thesis sets an important step towards a commercially viable solution for wheel-end load measurement. In addition, it is shown how this new information could push the boundaries of vehicle dynamics control. Next is the development of a suitable hardware setup to apply these results in a commercial solution, a topic currently pursued by the author.
Original languageEnglish
QualificationDoctor of Philosophy
Awarding Institution
  • Delft University of Technology
  • Happee, R., Supervisor
  • Shyrokau, B., Advisor
Award date22 Oct 2020
Electronic ISBNs978-94-6419-056-4
Publication statusPublished - 2020


Dive into the research topics of 'Wheel Load Reconstruction for Intelligent Vehicle Control'. Together they form a unique fingerprint.

Cite this