Title page for ETD etd-10052006-103803


Type of Document Master's Thesis
Author Tang, Fengming
Author's Email Address ftang1@lsu.edu
URN etd-10052006-103803
Title Grinding Wheel Condition Monitoring with Boosted Classifiers
Degree Master of Science in Industrial Engineering (M.S.I.E.)
Department Industrial & Manufacturing Systems Engineering
Advisory Committee
Advisor Name Title
T Warren Liao Committee Chair
Jianhua Chen Committee Member
Xiaoyue Jiang Committee Member
Keywords
  • tool wear condition monitoring
  • time series
  • classifier
  • boost
Date of Defense 2006-05-19
Availability unrestricted
Abstract
In this thesis, two data sets collected in grinding process under different cutting and wheel conditions were studied. One is the cutting forces in three directions, i.e. X, Y and Z, collected under two different cutting conditions. The other one is the acoustic emission (AE) signals collected under different wheel conditions(sharp and dull). For the goal of grinding wheel condition monitoring, the regression model with autocorrelated errors was proved to be effective and was used to extract features from signals in this study. The coefficients of the models served as the features used in the classification step that employed boosting method. Based on the AdaBoost and A-boosting algorithms which can only be used in two classes situation, two improved boosting methods called Adaboost-M and A-boosting-M, which can be used to classify multiple classes, are proposed. With the forces data set, we compared Adaboost-M and A-boosting-M against the traditional AdaBoost.M1 and the corresponding weak learners(KNN and Prototype). The accuracies of Adaboost-M and A-boosting-M are higher than that of AdaBoost.M1 and the weak learners in our application. With the AE data set, our focus is to recognize the signals collected when the wheels were dull from the signals collected when the wheels were sharp. The AdaBoost, A-boosting and the corresponding weak learners(KNN and Proto) were used. The results indicate that (i) boosting does not improve the effectiveness of k-nearest neighbor but greatly improve the effectives of the prototype classifier, (ii) depending upon the data, AdaBoost or A-Boosting might produce higher classification accuracy, (iii) the error of false positive is higher than the error of false negative for the better classifiers.

Based on the study, the combined use of AR models for feature extraction and boosted algorithms for classification are proved to be a viable approach for grinding wheel condition monitoring.

Files
  Filename       Size       Approximate Download Time (Hours:Minutes:Seconds) 
 
 28.8 Modem   56K Modem   ISDN (64 Kb)   ISDN (128 Kb)   Higher-speed Access 
  Tang_thesis.pdf 358.14 Kb 00:01:39 00:00:51 00:00:44 00:00:22 00:00:01

Browse All Available ETDs by ( Author | Department )

If you have more questions or technical problems, please Contact LSU-ETD Support.