Fatigue Prediction Model of Vibration-Strain Signals with Neuro-Fuzzy Parameters for Durability Assessment

This study aimed to develop vibration fatigue prediction model based on the features of
different vibrational road excitations. Conventional strain-life approaches used for
durability predictions requires huge amount of data and heavy computational load. This
rendered the durability analysis of coil spring to be a complex and time-consuming
process. Therefore, a durability prediction model based on the vibrational excitations
from road surfaces is developed in this study to accelerate the durability assessment of
coil spring. Methodology of this study include signals acquisition, feature extraction
from the road vibrational data, model development and model validation. Road tests
were conducted to acquire strain and road vibration signals under various road
conditions. Subsequently, multifractality and low-frequency vibrational energy were
extracted from the vibration signals with multifractal analysis and singularity analysis,
respectively, as the durability-related features. An acceleration-strain conversion model
was developed based on the relationship between the spring displacement and strain in
coil spring. Through multibody dynamic modelling, spring responses from road
excitations were simulated and converted to strain signals. Simulated fatigue lives were
then predicted from the generated strain signals. Next, Adaptive Neuro-Fuzzy Inference
System (ANFIS) was used to determine the optimal fuzzy membership functions based
on the durability features of vibrational road excitations. A vibration-based durability
prediction model was developed using the optimised membership functions of road
vibration features. Reliability analysis was employed to evaluate the reliability of the
developed durability prediction model. The frequency analysis of vibrational road
excitations with power spectral density revealed that the signals had frequency range
between 0-50 Hz and high frequency noise between 80-100 HZ. Singularity analysis
was able to extract low frequency features while eliminate the high frequency noise. In
addition, the formulated acceleration-strain conversion model could generate accurate
strain signals of coil spring working under random loads. The statistics of the simulated
strain signals had deviations below 30% compared to experimental strain signals. This
suggested that the use of multibody dynamics model was appropriate to predict the
responses of coil spring under random road excitations. The durability prediction model
developed based on neuro-fuzzy approach in this study provided accurate prediction of
coil spring’s fatigue life closed to the experimental results. Fatigue life conservative
analysis validated the accuracy of fatigue lives predicted with the ANFIS-based
durability prediction model in which above 90% of the predicted fatigue lives were
found within the acceptance boundary limits. Furthermore, the simulated fatigue life
had good correlation with experimental results with r values above 0.9. The reliability
analysis validated the trained ANFIS durability prediction models as the simulated
fatigue life data had similar mean-cycle-to-failure (McTF) with the experimental
results, with differences less than 10%. This study contributed to characterisation of
fatigue-related features of vibrational road excitations based on the fuzzy membership
functions for durability prediction of coil spring. The developed durability prediction
model in this study can expedite the coil spring development process to meet the
durability requirements of automotive industry.



Citation style:
Could not load citation form.


Use and reproduction:
All rights reserved