A study of vehicle fatigue damage predictions based on state estimations
Time: Fri 2021-05-21 13.00
Subject area: Solid Mechanics
Doctoral student: Ulrika Lagerblad , Hållfasthetslära, Scania CV AB
Opponent: Professor Costas Papadimitriou, University of Thessaly
Supervisor: Professor Artem Kulachenko, Hållfasthetslära; PhD Henrik Wentzel, Scania CV AB
Individual health monitoring of heavy vehicles is one of the keystones on which an optimized utility of a vehicle fleet is based. For many components mounted on the chassis of a heavy vehicle, failures due to fatigue is a major concern. The fatigue is caused by vibrations in the chassis, originating from road irregularities and the driver’s manoeuvring of the vehicle. The variation in vibration characteristics is very large, and a reliable prediction of the fatigue damage must therefore be based on vibrations measured on each specific vehicle. Even so, the prediction of fatigue damage is far from trivial. To minimize the cost of both measurement equipment and data processing, the vibrations must be measured very sparsely. Furthermore, it will in general not be possible to measure the vibrations in the locations where failures are anticipated.
In Paper A, an augmented Kalman filter extended with a fixed-lag smoother is presented (hereinafter referred to as the smoothing algorithm). The smoothing algorithm is used to estimate the vibrations everywhere in a dynamic system based on sparsely measured accelerations and on a numerical model describing the system. Measurements from a vehicle component mounted in a vibration test rig are used to validate the smoothing algorithm. The numerical model is constructed from an operational modal analysis of the dynamic system. It is shown that the smoothing algorithm accurately estimates the dynamic responses.
The smoothing algorithm is then investigated in more detail in Paper B. When analysing measurements from complex dynamic structures, both measurement noise and modelling errors will be present. This paper presents a study of how the magnitude of these errors affects the accuracy of the estimates. Furthermore, some of the assumptions made within the estimating process are analysed. From sensitivity analysis, it is concluded that a tuning methodology for the smoothing algorithm, presented in Paper A, results in stable estimates with a good trade-off between estimator adaptability and noise sensitivity.
In Paper C, the fatigue damage in a vehicle component is predicted based on strains estimated with the smoothing algorithm. Accelerations and strains are measured on a vehicle component mounted in a vibration test rig, and together with a numerical model based on a finite element analysis, the strains in critical regions are estimated. The estimated strain is then used for computation of the accumulated fatigue damage. The predicted damage corresponds well with the failures occurring in a vibration fatigue life test of the analysed component.
Finally, in Paper D the fatigue damage prediction methodology presented in Paper C is used to predict the fatigue damage in a complete vehicle. Based on a finite element model of the vehicle and on several hundred kilometres of driving, the fatigue damage in some critical regions on three different chassis-mounted components is predicted. The smoothing algorithm estimates the strains with the acceptable result, even though the model error is large and the number of acceleration measurements is low. The subsequent damage predictions correspond well with failures seen on operating trucks.