Surface roughness prediction by combining static and dynamic features in cylindrical traverse grinding

With a given grinding system, surface roughness of the ground workpiece depends mainly on the settable grinding parameters and is also inevitably influenced by some random factors during grinding. Based on the relevance vector machine, which can effectively avoid over-fitting and can also present a...

Full description

Saved in:
Bibliographic Details
Published inInternational journal of advanced manufacturing technology Vol. 75; no. 5-8; pp. 1245 - 1252
Main Author Guo, Jianliang
Format Journal Article
LanguageEnglish
Published London Springer London 01.11.2014
Springer Nature B.V
Subjects
CAE) and Design
Computer-Aided Engineering (CAD
Cylindrical grinding
Engineering
Grinding
Industrial and Production Engineering
Kernel functions
Machine learning
Mathematical models
Mechanical Engineering
Media Management
Original Article
Parameters
Predictions
Random vibration
Surface roughness
Traverse grinding
Workpieces
Surface roughness
Grinding
Relevance vector machine
Prediction
Online AccessGet full text

Cover

More Information
Summary:With a given grinding system, surface roughness of the ground workpiece depends mainly on the settable grinding parameters and is also inevitably influenced by some random factors during grinding. Based on the relevance vector machine, which can effectively avoid over-fitting and can also present a fast prediction, a model is proposed for predicting the surface roughness of ground parts. In the model, the grinding parameters are considered as the static features, and some features of the random vibration signal work as the dynamic features. The static and dynamic features compose a feature vector together, which is used as the input variable of the model to predict the surface roughness. A series of experiments were carried out to validate the model and the results show that for the particular grinding system and conditions in this work, when the width parameter of kernel function is set to 100 and all features are normalized on 100, both the predicted surface roughness and its variation trend are close to the measured values. It can be inferred that the model generates a precise prediction only when the width parameter and normalization parameter match the given grinding system and conditions.
ISSN:0268-3768
1433-3015
DOI:10.1007/s00170-014-6189-5