Tool wear state recognition and prediction method based on laplacian eigenmap with ensemble learning model

•A multi-algorithm based features screening and LE features downscaling method is proposed.•For features with high dimensionality, feature filtering and feature fusion are applied to downscale the features step by step.•Fast selection of model hyperparameters based on GS.•Evaluate the prediction mod...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Advanced engineering informatics Ročník 60; s. 102382
Hlavní autoři: Xie, Yang, Gao, Shangshang, Zhang, Chaoyong, Liu, Jinfeng
Médium: Journal Article
Jazyk:angličtina
Vydáno: Elsevier Ltd 01.04.2024
Témata:
Ensemble method
Features fusion
Laplacian eigenmaps
Tool wear
XGBoost algorithm
Laplacian eigenmaps
XGBoost algorithm
Ensemble method
Features fusion
Tool wear
ISSN:1474-0346, 1873-5320
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:•A multi-algorithm based features screening and LE features downscaling method is proposed.•For features with high dimensionality, feature filtering and feature fusion are applied to downscale the features step by step.•Fast selection of model hyperparameters based on GS.•Evaluate the prediction model through multiple experiments and multiple metrics. Accurate prediction of tool wear status plays a critical role in the digital manufacturing industry, and its health level directly affects machining quality, production costs, and overall productivity. In response to the problems of the high dimensionality of extracted features from tool wear characterization sensors, redundant information, and large individual model errors and biases, a novel method for tool wear status identification and prediction that fuses downscaling dimensionality and ensemble models is proposed. First, a multi-algorithm feature filtering based on Random Forest (RF) and extreme gradient boosting (XGBoost) is utilized, and the laplacian eigenmaps (LE) algorithm is combined to perform fusion downscaling on the filtered features. Then, the parameters of the XGBoost algorithm are optimized using grid search (GS). Finally, the performance of the proposed method is evaluated by different tool wear experiments for both regression and classification using model prediction accuracy evaluation metrics (R-squared values and f1 values) and prediction time. The experimental results show that for different tools wear experimental data, the R-squared values of the regression model are higher than 0.98, the f1 values of the classification model are above 0.96, and the prediction speed is improved by an order of magnitude compared with other models. The results are analyzed to verify the effectiveness and applicability of the proposed method, which can provide technical support for the automated machining process.
ISSN:1474-0346
1873-5320
DOI:10.1016/j.aei.2024.102382