Research and development process of A.I. based LED-analysis tool for Bittium CI-laboratory
Multanen, Tony (2023)
2023
All rights reserved. This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:amk-2023121737961
https://urn.fi/URN:NBN:fi:amk-2023121737961
Tiivistelmä
Various industries from financial to heavy industry use machine learning to provide innovative solutions to a variety of purposes.
Test automation and device functionality verification, an area where machine learning is not commonly utilized, were the focus of this study,
even though machine learning in test automation could offer many advantages and reduce manual human work when used correctly.
The main objective of this study was to create a computer vision implementation for the verification of device LED functionality. Typically,
this task was manually performed by people, resulting in low test cov-erage as the number of different LED combinations increased. Therefore,
two different machine learning algorithms were tested for this purpose.
In this study, the first assumption was made that a more recent Convolutional Neural Network would be much better suited for this purpose.
Verification for this assumption was achieved by testing it against the older Multilayered Perceptron algorithm. This testing was conducted using
a Patient Monitoring device solution created by Bittium for its customer, and the study involved a significant amount of theory learning, image/data gathering,
data preparation, and the practical utilization of the knowledge acquired.
As a result, three different machine learning algorithms were developed in this study. One was for learning what could be done and what needed to be done when
LED amounts and different combinations were scaled up, and two were for the analysis of Patient Monitoring Charger, which was the primary goal of this study from
Bittium's perspective.
As a machine learning project, the goals of this implementation were achieved, and the accuracies of both selected algorithms were exceptionally good. Therefore,
the project itself was considered a success. Concerning the earlier assumption, it was proven to be correct. However, even though the Multilayered Perceptron is an
older algorithm these days, the differences in accuracies were not significantly better with the Convolutional Neural Network. Nonetheless, the Convolutional
Neural Network was shown to be the right choice in similar projects due to its reduced resource requirements compared to the Multilayered Perceptron.
Test automation and device functionality verification, an area where machine learning is not commonly utilized, were the focus of this study,
even though machine learning in test automation could offer many advantages and reduce manual human work when used correctly.
The main objective of this study was to create a computer vision implementation for the verification of device LED functionality. Typically,
this task was manually performed by people, resulting in low test cov-erage as the number of different LED combinations increased. Therefore,
two different machine learning algorithms were tested for this purpose.
In this study, the first assumption was made that a more recent Convolutional Neural Network would be much better suited for this purpose.
Verification for this assumption was achieved by testing it against the older Multilayered Perceptron algorithm. This testing was conducted using
a Patient Monitoring device solution created by Bittium for its customer, and the study involved a significant amount of theory learning, image/data gathering,
data preparation, and the practical utilization of the knowledge acquired.
As a result, three different machine learning algorithms were developed in this study. One was for learning what could be done and what needed to be done when
LED amounts and different combinations were scaled up, and two were for the analysis of Patient Monitoring Charger, which was the primary goal of this study from
Bittium's perspective.
As a machine learning project, the goals of this implementation were achieved, and the accuracies of both selected algorithms were exceptionally good. Therefore,
the project itself was considered a success. Concerning the earlier assumption, it was proven to be correct. However, even though the Multilayered Perceptron is an
older algorithm these days, the differences in accuracies were not significantly better with the Convolutional Neural Network. Nonetheless, the Convolutional
Neural Network was shown to be the right choice in similar projects due to its reduced resource requirements compared to the Multilayered Perceptron.