Transformers use fault current indicators to detect faulty current. Since the fault current indicators are internal, FPL technicians must check every transformer in a particular area to find the location of the fault. To prevent this, team 304 is developing a method to find faulted pad-mounted transformers using computer vision. This goal is inspired by Florida Power and Light’s existing drone system, currently in use to locate and film different types of power equipment.
The team’s first task is to create a way to show a faulted transformer using a beacon that mounts to the transformer’s surface. The beacon is attached to a pre-existing internal fault indicator, which supplies power to the beacon when a fault happens. Once the beacon receives power, a spring will release the beacon to an upright position. Then, an LED at the top of the beacon will turn on. This change in shape and light is seen by FPL cameras, sending the images to the second part of the design.
The second part is a computer vision system that can detect the transformers, LED lights, and the levers position. Training this model involves collecting a library of transformer and beacon images. The library uses different angles, environments, and states to increase the model’s ability to make correct detections in as many circumstances as possible. The image library is then sorted and labeled before it is used to train a data object detection model. The name of the algorithm “You Only Look Once”, or YOLO, is chosen due to its ability to make accurate detections on real-time footage. Once the model has been trained, it will undergo a series of tests. This testing will compare changes in the training library with the goal of improving detections.
Sam HammermasterSoftware Integration Engineer Sam is a Computer Engineering student passionate about workflow efficiency. After graduation he will be moving to the DC area to work as a Solutions Engineer for a software company where he hopes to bridge the gap between customers and coding. |
Erin MurphyProject Manager Erin is a Computer Engineering student and an NCAA Beach Vollyeball athlete at FSU. After graduation, she hopes to learn more about machine learning and data science in graduate school before becoming an educator. |
Gage IrwinComputer Vision Engineer Gage Irwin is a Computer Engineering student exploring paths in machine learning and artificial intelligence. After graduation he hopes to work designing and implementing machine learning models to a diverse set of applications. |
Jordan WilkersonMechatronics Design Engineer Jordan Wilkerson is a mechanical engineering student from Fort Pierce, FL. After graduation he will be attending graduate school at the FAMU-FSU College of Engineering to pursue a master's degree in heat transfer and fluid mechanics. |
Kent LogueMechanical Design Engineer Kent Logue is a mechanical engineering student focusing on dynamic systems. After graduating he will be going to work for Florida Power and Light, focusing on making an impact with drones and robotics. |
Click each link to download the corresponding file
Project Charter Work Breakdown Structure Customer Needs Functional Decomposition Targets VDR1 Concept Generation Concept Selection Bill of Materials VDR2 Hazard Assessment Project Narrative VDR3 Preliminary Detailed Design Spring Project Plan Project Charter Update Abstract VDR4 High Resolution Graphic Updated Codes and Standards DR5
 |
 |
 |
 |
As of 3/25/2022, the final stages of the project are now in motion. The team has just finished presenting VDR5. The hardware beacon needs one or two more design tweaks and then the final design can be printed. There are promising preliminary test results for the object detection model. The team will continue to make the model better by implementing synthetic data, taking a variety of pictures, and testing in different conditions. Here is a list of deliverables that the team needs to complete in order to finish the project:
