Optimizing presetting attributes by softcomputing techniques to improve tapered roller bearings working conditions. (September 2018)

• Record Type: Journal Article
• Authors: Fernandez Martinez, Roberto; Lostado Lorza, Ruben; Santos Delgado, Ana A.; Piedra Pullaguari, Nelson O.
• Is Part Of: Advances in engineering software. Volume 123(2018)
• Page Start: 13
• Page End: 24
• Publication Date: 2018-09
• Subjects: Data mining classification techniques -- Finite elements method -- Design of experiments -- Model parameter tuning and evaluation -- Double-row tapered roller bearing; Computer-aided engineering -- Periodicals; Engineering -- Computer programs -- Periodicals; Engineering -- Software -- Periodicals; Periodicals; 620.0028553
• DOI: 10.1016/j.advengsoft.2018.05.005
• Languages: English
• ISSNs: 0965-9978

Abstract

Highlights: A new approach to optimize tapered roller bearings. Methodology to optimize presetting conditions to avoid bearing damage. Improvement of classification accuracy by parameter tuning and feature reduction. Use of linear, nonlinear, and trees and rule-based methods to classify. Abstract: Double-row Tapered Roller Bearings are mechanical devices that have been designed to support a combination of loads that are fixed on an optimal presetting to ensure correct working conditions. The emergence of high contact stresses, fatigue spalling and pitting on the bearing railway makes it important to have a tool that enables knowing in advance whether certain presetting loads will lead to excellent working conditions or the opposite. This work proposes a methodology to classify the working condition on the basis of the values of presenting loads on four classes. To achieve this goal, a three-dimensional Finite Element (FE) model was generated. Later, a design of experiments was designed to provide the greatest amount of information by reducing the computational cost of the simulations based on FE models. Then, one of the four classes of working conditions was assigned to each of the experiments. Later, a statistical analysis and machine learning techniques were used to create classification models. Feature transformation and reduction, algorithm parameter tuning and validation methods were used to achieve robust classification models. The best results were obtained based on flexible discriminant analysis. As it provided acceptable accuracy, both the methodology and final model were validated.