- Tytuł:
- Forecasting of subjective comfort in tram using ordinal logistic regression and manifold learning
- Autorzy:
-
Pietraszek, J.
Grzegożek, W.
Szczygieł, J. - Powiązania:
- https://bibliotekanauki.pl/articles/246634.pdf
- Data publikacji:
- 2012
- Wydawca:
- Instytut Techniczny Wojsk Lotniczych
- Tematy:
-
rail transport
vibrations
ordinal logistic regression
principal component analysis
manifold learning - Opis:
- Comfort in a vehicle has a very important role to play as one of the most important dynamic performance characteristics of rail vehicles. It is the factor of ever-increasing importance, even creating a specialized branch of engineering associated with relation between human limitations and designing of machines: human–factors engineering. The vibration is known to be a major factor that affects and deteriorates ride comfort. For evaluating ride comfort in rail vehicles, there have been developed methods resulting in the creation of many standards and multiple criteria used and even standardized in different countries. One of the authors, J. Szczygieł designed and performed a passive experiment to collect data describing physical conditions of ride and associated subjective assessments of comfort. Panel of fourteen people during the tram ride made synchronous subjective assessments of comfort, assessing it on a discrete ordinal scale of 0 to 5, using electronic panels connected to the computer. At the same time computer through sensors recorded values of acceleration in three perpendicular axes. It made possible to correlate the fuzzy subjective evaluations with objective physical measurements. Because of the discrete type of fuzzy ratings of comfort, natural way of modelling is the ordinal logistic regression. The classic form of the ordinal logistic regression assumes that in the space of explanatory factors there are parallel activation hyper-planes slightly disturbed by unknown or uncontrolled noise factors. In fact, the assumption of linearity is a very strong idealization and leads to considerable misclassifications. The original space of explanatory factors is 11-dimensional with ten continuous dimensions and one discrete. Then the multivariate method, principal component analysis (PCA), was used to identify principal components, which are responsible most to the variability of the studied set. The scree plot was used to identify the number of significant PCA factors. The use of PCA revealed that the area occupied by the data set is approximately 6-dimensional. However, the dimensionality reduction of explanatory variables set did not lead to better forecasting accuracy. A more subtle analysis involving discretization techniques showed that activation hyperplanes are highly curved in the six-dimensional area identified by PCA but their dimensionality is much lower. The details of the procedure are described in the article. The article conclusion is that is necessary to introduce curvilinear coordinate system embedded into the shapes of activation hyper-planes to obtain better classification.
- Źródło:
-
Journal of KONES; 2012, 19, 2; 403-409
1231-4005
2354-0133 - Pojawia się w:
- Journal of KONES
- Dostawca treści:
- Biblioteka Nauki
Artykuł