Canary Historian and Machine Learning - Electrical Engineering Research Experience for Undergraduates

Aug 2, 2026

Upload by Maeve

This poster presents an undergraduate electrical engineering project focused on tracking power quality issues in electrical breakers using a “Canary Historian” database and machine learning classification. It motivates the work by linking power sags, swells, flickers, interruptions, and harmonics to breaker damage and power disruptions. The approach covers data preprocessing, JSON-based data understanding, database creation, and training with decision tree, random forest, and XGBoost models. Challenges include limited available data and the need for automatic detection, with future plans to apply deep learning models such as Transformer and LSTM to improve accuracy.

LanguageEnglish

CategoryTechnology

FormatPDF

Pages2

Reading Time5 min

AccessFree to download

Goals and Objectives

Approach

Canary database creation steps

Preprocessed data and models

Results

Challenges

Future Plans

University of Arkansas, Fayetteville ScholarWorks@UARK | Electrical Engineering Research Experience for Undergraduates | Electrical Engineering | | --- | --- | | 2024
Canary Historian and Machine Learning
Cameron Eddy
University of Arkansas, Fayetteville
Follow this and additional works at: [https://scholarworks.uark.edu/elegreu](https://scholarworks.uark.edu/elegreu)[ ](https://scholarworks.uark.edu/elegreu) Part of the Electrical and Computer Engineering Commons | | Citation Eddy, C. (2024) . Canary Historian and Machine Learning. Electrical Engineering Research Experience for Undergraduates. Retrieved from [https://scholarworks.uark.edu/elegreu/2](https://scholarworks.uark.edu/elegreu/2) This Poster is brought to you for free and open access by the Electrical Engineering at ScholarWorks@UARK. It has been accepted for inclusion in Electrical Engineering Research Experience for Undergraduates by an authorized administrator of ScholarWorks@UARK. For more information, please contact [scholar@uark.edu](scholar@uark.edu), [uarepos@uark.edu](uarepos@uark.edu). Author: Cameron Eddy Mentors: Reeshad Khan, Wesley Schwartz, Dr. Chris Farnell University of Arkansas Canary Historian and Machine Learning |
Goals and Objectives |
Approach | | | | | | |
Results | | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | Power quality issues in electrical breakerscan cause:
• Breaker damage
• Power disruptions

Thus, there is a need to track sags, swells, flickers, interruptions, and harmonics.
[https://th.bing.com/th/id/OIP.MaFUFvCSjl6ZXpsETHFHD](https://th.bing.com/th/id/OIP.MaFUFvCSjl6ZXpsETHFHD)[ ](https://th.bing.com/th/id/OIP.MaFUFvCSjl6ZXpsETHFHD)[gAAAA?w=156&h=180&c=7&r=0&o=5&dpr=1.3&pid=1.7](gAAAA?w=156&h=180&c=7&r=0&o=5&dpr=1.3&pid=1.7) | Canary database creation steps: | | | | | | | Preprocessed data:

Three classification models were used for training:
1 ) Decision Tree
2 ) Random Forest
3) XG Boost
| | | | | Understand JSON format | | Add new breakers | | | Insert additional tags | | | | | | | | | | | | | | | | Machine learning steps:
| | | | | | | | | | | Find pretrained
models | | | Gather breaker test data | | | Match test data to train data | | | |
Challenges | | | | | | | | | | | Problems:
• Not enough available data
• No method for automatic detection Solutions:
• Create a database to track RMS Voltages • Use machine learning to classify issues
| | | | | | | | | | | | Equation to convert data:
􀝒 􀝓􀝐 = 325 . 269 × 􀝉 × sin(376 . 991 × 0 . 0002 × 􀝊 )
| | | | | | | | | | | | | | | | Swells Normal Sags | | | | | | | | | | | | |
Future Plans | | | | | | | | | | | | Use deep learning models such as Transformer NN and LSTM to improve accuracy. |

FAQs

What problem does the project address?

Power quality issues in electrical breakers can cause breaker damage and power disruptions. The project targets sags, swells, flickers, interruptions, and harmonics through tracking and classification.

How is the machine learning approach implemented?

The workflow includes creating a Canary database, preprocessing data, understanding JSON format, and mapping breaker test data to training data. Three models are used: decision tree, random forest, and XGBoost.

What challenges and solutions are highlighted?

Key challenges are not enough available data and the lack of a method for automatic detection. The proposed solutions include creating a database to track RMS voltages and using machine learning to classify issues.

What future improvements are planned?

Future plans focus on using deep learning models such as Transformer neural networks and LSTM to improve classification accuracy.

Aug 2, 2026

Upload by Maeve