Optimering av sågverk med maskininlärning
(2025) In CODEN:LUTEDX/TEIE EIEM01 20251Industrial Electrical Engineering and Automation
- Abstract
- As the automation of industrial systems is becoming more and more complex, new methods are needed to further boost efficiency. Machine learning is a method where a computer learns to recognize complex patterns using large amounts of real-world data. The focus of this work is to leverage real-world industrial data from a sawmill to optimize one of the machines, the unscramblers.
The unscrambler separates planks so that they lie next to each other on the track. Sometimes, the unscrambler fails to separate the planks, meaning two planks lie on top of each other. Moreover, the throughput of planks is also important to consider. By tuning the settings of the unscrambler, different results of both efficiency and double boards can be achieved.... (More) - As the automation of industrial systems is becoming more and more complex, new methods are needed to further boost efficiency. Machine learning is a method where a computer learns to recognize complex patterns using large amounts of real-world data. The focus of this work is to leverage real-world industrial data from a sawmill to optimize one of the machines, the unscramblers.
The unscrambler separates planks so that they lie next to each other on the track. Sometimes, the unscrambler fails to separate the planks, meaning two planks lie on top of each other. Moreover, the throughput of planks is also important to consider. By tuning the settings of the unscrambler, different results of both efficiency and double boards can be achieved. The goal of the master thesis is ultimately to create a machine learning model which can predict optimal machine parameters for when the process is running in steady state.
In order to design a functioning machine learning model, rigorous studies of the actual process were performed. This affected data collection, filtration and transformation. An evaluation of different machine learning models was performed after which Random Forest was found to be most appropriate. Its model parameters were fine-tuned until a good balance between score and model size was found.
The process exists in a limiting context which means that the machine is sometimes forced to stop through no fault of its own. Because of this, a new efficiency filtration scheme was also developed with the goal of removing dynamic, non-steady-state effects. With the already existing efficiency measuring scheme no visible improvement was detected using the model. Using this newly developed efficiency filtration scheme, it was found that the model performs 5-10% better than the sawmill's reference parameters. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9203474
- author
- Bäckström, Oliver LU
- supervisor
- organization
- course
- EIEM01 20251
- year
- 2025
- type
- H3 - Professional qualifications (4 Years - )
- subject
- keywords
- machine learning, decision tree, random forest, unscrambler, sawmill, scikit-learn, data collection, data filtration, model evaluation, PLC, industry
- publication/series
- CODEN:LUTEDX/TEIE
- report number
- 5548
- language
- Swedish
- id
- 9203474
- date added to LUP
- 2025-06-23 18:37:33
- date last changed
- 2025-06-23 18:37:33
@misc{9203474,
abstract = {{As the automation of industrial systems is becoming more and more complex, new methods are needed to further boost efficiency. Machine learning is a method where a computer learns to recognize complex patterns using large amounts of real-world data. The focus of this work is to leverage real-world industrial data from a sawmill to optimize one of the machines, the unscramblers.
The unscrambler separates planks so that they lie next to each other on the track. Sometimes, the unscrambler fails to separate the planks, meaning two planks lie on top of each other. Moreover, the throughput of planks is also important to consider. By tuning the settings of the unscrambler, different results of both efficiency and double boards can be achieved. The goal of the master thesis is ultimately to create a machine learning model which can predict optimal machine parameters for when the process is running in steady state.
In order to design a functioning machine learning model, rigorous studies of the actual process were performed. This affected data collection, filtration and transformation. An evaluation of different machine learning models was performed after which Random Forest was found to be most appropriate. Its model parameters were fine-tuned until a good balance between score and model size was found.
The process exists in a limiting context which means that the machine is sometimes forced to stop through no fault of its own. Because of this, a new efficiency filtration scheme was also developed with the goal of removing dynamic, non-steady-state effects. With the already existing efficiency measuring scheme no visible improvement was detected using the model. Using this newly developed efficiency filtration scheme, it was found that the model performs 5-10% better than the sawmill's reference parameters.}},
author = {{Bäckström, Oliver}},
language = {{swe}},
note = {{Student Paper}},
series = {{CODEN:LUTEDX/TEIE}},
title = {{Optimering av sågverk med maskininlärning}},
year = {{2025}},
}