Have a personal or library account? Click to login
Comparison of machine learning models predicting the pull-off strength of modified epoxy resin floors Cover

Comparison of machine learning models predicting the pull-off strength of modified epoxy resin floors

Studia Geotechnica et Mechanica
Volume 46 (2024): Issue s1 (December 2024)
By: Mateusz Moj,  Łukasz Kampa and  Sławomir Czarnecki  
Open Access
|Nov 2024

Figures & Tables

Figure 1:

Diagram of the basic factors that damage epoxy resin floors.
Diagram of the basic factors that damage epoxy resin floors.

Figure 2:

Search volume chart for publications based on keywords in Google Scholar and Science Direct on 03/07/2024.
Search volume chart for publications based on keywords in Google Scholar and Science Direct on 03/07/2024.

Figure 3:

Granite powder used.
Granite powder used.

Figure 4:

Linen fibers used.
Linen fibers used.

Figure 5:

Pull-off test.
Pull-off test.

Figure 6:

Violin plot of the parameter: a) amount of Component A [%], b) amount of Component B [%], c) amount of granite powder [%], d) amount of linen fibers [%], e) density [g/cm3], and f) fb [MPa].
Violin plot of the parameter: a) amount of Component A [%], b) amount of Component B [%], c) amount of granite powder [%], d) amount of linen fibers [%], e) density [g/cm3], and f) fb [MPa].

Figure 7:

Pearson correlation matrix.
Pearson correlation matrix.

Figure 8:

Pull-off strength of the modified epoxy resin coating.
Pull-off strength of the modified epoxy resin coating.

Figure 9:

Relationship between the predicted value and the experimental value of the pull-off strength fb for the: a) RL model, b) ANN model, c) DT model, and d) RF model.
Relationship between the predicted value and the experimental value of the pull-off strength fb for the: a) RL model, b) ANN model, c) DT model, and d) RF model.

Figure 10:

Relative errors for data sets for selected artificial intelligence algorithms.
Relative errors for data sets for selected artificial intelligence algorithms.

Figure 11:

Histograms of absolute error values for a) ANN, b) DT, c) RF, and d) LR.
Histograms of absolute error values for a) ANN, b) DT, c) RF, and d) LR.

Figure 12:

Visualization of SHAP values for the RF ML model.
Visualization of SHAP values for the RF ML model.

Elements of the decision tree and random forest algorithm_

Number of input categoriesDepth of treesNumber of trees (only for RF)Minimum subset to be dividedMinimum number of categories in the leaf
51–2020–20052

Descriptive statistics of the input and output parameters_

Min.Max.St.dev.MeanRange
Amount of Component A [%]0,4550,7520,0770,5600,297
Amount of Component B [%]0,1550,3100,0350,2520,105
Amount of granite powder [%]0,0000,3750,1120,1820,375
Amount of linen fibers [%]0,0000,0150,0050,0060,015
Density [g/cm3]1,1001,3060,0601,1960,206
fb [MPa]1,9503,5200,2232,5461,570

Summary of correlation coefficients R, RMSE, and average percentage forecast errors MAPE for selected models_

AI ModelStatistical metrics
R [-]RMSE [MPa]MAPE [%]
Linear regression0,62770,22997,4244
Decision tree0,83100,16434,0814
Random forest0,88480,13763,7156
Artificial neural networks0,87440,13123,8098
DOI: https://doi.org/10.2478/sgem-2024-0024 | Journal eISSN: 2083-831X | Journal ISSN: 0137-6365
Journal RSS Feed
Language: English
Page range: 377 - 388
Submitted on: Jul 7, 2024
|
Accepted on: Sep 18, 2024
|
Published on: Nov 10, 2024
Published by: Wroclaw University of Science and Technology
In partnership with: Paradigm Publishing Services
Publication frequency: 4 issues per year
Keywords:
resin coatings,
granite powder,
machine learning,
linen fibers
Related subjects:
Geosciences,
Geosciences, other,
Materials sciences,
Composites,
Porous materials,
Physics,
Mechanics and fluid dynamics

© 2024 Mateusz Moj, Łukasz Kampa, Sławomir Czarnecki, published by Wroclaw University of Science and Technology
This work is licensed under the Creative Commons Attribution 4.0 License.

Previous articleVolume 46 (2024): Issue s1 (December 2024)Next article