Projects / Programmes
Prediction of loading spectra and their scatter in a R&D process
Code |
Science |
Field |
Subfield |
2.11.00 |
Engineering sciences and technologies |
Mechanical design |
|
Code |
Science |
Field |
T455 |
Technological sciences |
Motors and propulsion systems |
T450 |
Technological sciences |
Metal technology, metallurgy, metal products |
T480 |
Technological sciences |
Technology of other products |
T280 |
Technological sciences |
Road transport technology |
T121 |
Technological sciences |
Signal processing |
random load histories, loading spectrum, mixture probability distribution models, operating conditions, non-parametric regression, neural networks, rainflow method
Researchers (2)
no. |
Code |
Name and surname |
Research area |
Role |
Period |
No. of publicationsNo. of publications |
1. |
00819 |
PhD Matija Fajdiga |
Mechanical design |
Head |
2004 - 2006 |
1,141 |
2. |
16334 |
PhD Jernej Klemenc |
Mechanical design |
Researcher |
2004 - 2006 |
779 |
Organisations (1)
Abstract
Load states of products are usually described by loading spectra. Loading spectra depend on combinations of different mutually dependent factors of operating conditions. That is why a relationship between the factors of operating conditions and loading spectra is generally non-linear. Due to many influential factors and the non-linearity of the relationship it is almost impossible to assess it by analytical methods.
In a preceding doctoral research we tried to empirically model the relationship between the factors of operating conditions and loading spectra by a non-parametric regression method. The nonparametric-regression model was built by representative samples of operating conditions together with corresponding loading spectra and tested by a set of different combinations of operating conditions. It was established that this method enables reliable prediction of loading spectra also for those combinations of operating conditions, for which no measured or simulated load data are available. Despite the promising results obtained in the doctoral research a few drawbacks of this method were noticed: bad extrapolation abilities of the nonparametric regression model and sensitivity of predictions to user-defined parameters. These drawbacks result also in a bad estimation of a scatter of loading spectra.
The goal of the post-doctoral project is to develop a new method for predicting loading spectra at given operating conditions on the basis of a limited set of known loading spectra of products. The new method should enable the extrapolation of loading spectra and eliminate the subjective influence of the user as much as possible. This would also result in a better estimation of the scatter of loading spectra. We will try to achieve this goal by applying a special hybrid multi-layer perceptron neural network. Learning and testing of the hybrid neural network will be performed for examples of simulated and measured loading spectra of existing products. The new method will be compared with the results from the doctoral research. The effectiveness of the new method will also be assessed.