Improving the Fatigue Life Prediction of Automotive Components Using Simulated Strain Signal Methods
This study aims to determine a suitable approach for generating strain signal leading to fatigue damage estimation using a significant acceleration model. It was hypothesised that the simulated model could reproduce a characteristic strain signal in similar to the actual strain signal. Three strain signals, all at 120 seconds, measured at the McPherson frontal coil spring of a Proton sedan had been used as a case study. The strain signals were acquired from a data acquisition involving car movements on various types of road surfaces at different speeds. The strains were caused by accelerations of the tyre while the car was being driven on rough road surfaces. Using a mathematical expression that was developed for car movements, the measured strain signals yielded acceleration signals usually used to describe the bumpiness of road surfaces. Furthermore, the fatigue-based acceleration signals were considered as disturbances acting on the automotive suspension system. These disturbances on the car body had an effect on generating strain signals via computer-based simulation, as responses of the coil spring, in the form of strains. Based on the simulations, all the simulated strain signals showed similar patterns to the actual strain signals. The simulated results also gave low fatigue damage deviations, which were less than 7.5 % for all the strain signals, with a root-mean square error of 0.011 % and a coefficient of determination of 0.9995. Furthermore, the extractions of higher amplitude cycle based on the energy of the wavelet transform were performed. From the extraction results, it was found that the wavelet transform was able to shorten the strain signal time up to 95.3 % and that 96.1 % of lower amplitude cycles were reduced, which these cycles theoretically contribute to a minimum fatigue damage. Thus, maintenance of fatigue damage by more than 92.7 % was produced. The segments that resulted from the extraction processes had been clustered using the Fuzzy C-means. The clustering results showed that the simulated strain signals had a significant coefficient of determination to the actual strain signals, reaching 0.8904 with a root-mean square error of only 0.5 %. Based on the cyclic testing results, the fatigue lives were distributed in a range of 1:2 or 2:1 correlation with a significant coefficient of determination of 0.9056. The testing time was successfully reduced by more than 85.1 % using the edited actual strain signals. In addition, using the edited simulated strain signals reduced the testing time up to 95.1 %. Indirectly, the use of modified strain signals could reduce device operating costs. The current study results are believed to provide a new knowledge towards generating simulated strain signals. Thus, the results bring greater meaning to the field of fatigue research. This work helps engineers in automotive industries involved in collecting road surface profiles, which are the main input for vehicle structures.