Dimensionality Reduction Methods in Machine Learning

Ghosh, Shyamasree; Dasgupta, Rathi

doi:10.1007/978-981-16-8881-2_7

Cited by 1 publication

(1 citation statement)

References 3 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this paper, we quantify the influence of different railway sections on wheel damage from historical running records using Fisher's linear discriminant analysis (LDA). [22][23][24] LDA is a classical supervised dimension reduction method that aims to find the optimal linear combination of high-dimensional inputs to separate data of two classes. 22,23 In this paper, we use LDA to find the optimal combination of the running mileage of a wheelset on each railway section to achieve the best distinction between the subset of normal wheelsets (C 1 ) and the subset of defective wheelsets (C 2 ).…”

Section: Linear Discriminant Analysis (Lda)mentioning

confidence: 99%

Influence of different railway lines on wheel damage of high-speed trains: Data-driven modelling and prediction

Zeng

Song

Zhang

et al. 2022

Proceedings of the Institution of Mechanical Engineers, Part O:

View full text Add to dashboard Cite

Wheel damage of high-speed trains has received great attention since it is detrimental to safety and reliability and consumes huge maintenance costs. However, the mixed operation mode of electric multiple units (EMUs) in China induces great challenges in distinguishing the influence of different railway lines on wheel damage. This paper develops a data-driven approach to quantify the influence of different railway lines on wheel damage and further applies it to damage risk prediction. First, under several fundamental assumptions, linear discriminant analysis and the proposed data padding strategy are used to quantify the influence on wheel damage due to per unit mileage of running on each railway section. The result shows that the influence on wheel damage varies significantly for different railway sections. Then, a route-based predictor is defined to represent the cumulative influence of an EMU’s routes on its wheel damage. Further, a wheel damage prediction model is developed based on a support vector machine using the route-based predictor and three other predictors from the operational data. Validation and test results demonstrate that the actual damage rate is higher for wheelsets predicted with higher damage probabilities. The proposed methodology has the potential to be applied to the risk-based operation and maintenance of high-speed trains.

show abstract