Using machine learning as an adaptive controller framework for optimizing properties of particleboard
Issued Date
2024-01-01
Resource Type
ISSN
00183768
eISSN
1436736X
Scopus ID
2-s2.0-85187656913
Journal Title
European Journal of Wood and Wood Products
Rights Holder(s)
SCOPUS
Bibliographic Citation
European Journal of Wood and Wood Products (2024)
Suggested Citation
Phetkaew T., Watcharakan T., Hiziroglu S., Chaowana P. Using machine learning as an adaptive controller framework for optimizing properties of particleboard. European Journal of Wood and Wood Products (2024). doi:10.1007/s00107-024-02059-1 Retrieved from: https://repository.li.mahidol.ac.th/handle/123456789/97717
Title
Using machine learning as an adaptive controller framework for optimizing properties of particleboard
Author(s)
Author's Affiliation
Corresponding Author(s)
Other Contributor(s)
Abstract
Fine adjustment of manufacturing parameters as a function of the experience of the technical manpower plays a vital role in any production line. The objective of this study was to propose an adaptive controller framework to improve the overall accuracy of the parameters regulating particleboard manufacturing. This framework has four main steps: (1) In the data gathering process, the production parameters and the sample test results were collected from the randomly picked and tested specimens in each round, (2) Relevance analysis was used to select high-power relevant variables influencing the overall quality of the final product. Those relevant variables will be inputs to construct the classification model, (3) A decision tree was employed to construct the classification model and reveal split points of the process parameters to determine the distinction between passed and failed panels, and (4) The production parameters in the next round will be adjusted according to the defined split points so the quality of the particleboard can be enhanced. Continuous improvement of the production parameters, within the perspective of the proposed framework, enables us to go back to step (1) again as desired, especially in the long production run. Based on the findings of this work, the experimental results revealed that the model could classify the failed particleboard with a specific rate of 92.50%. The model also demonstrated that resin characteristics, namely pH value and viscosity, impacted the overall performance of the particleboard.