TurbineSense

What it does

TurbineSense is an AI-powered system that analyses images of wind turbine blades to detect cracks and surface damage. It replaces slow, dangerous manual inspections with an automated, accurate, and scalable software solution for preventive maintenance.

Your inspiration

Our inspiration came from the challenges faced in maintaining wind turbines, especially in remote or hazardous environments. We noticed that manual inspections are not only time-consuming and risky but also require shutting down the turbines, causing energy loss. As mechanical engineering students interested in automation and AI, we saw an opportunity to apply computer vision to make the inspection process safer, faster, and more efficient. The lack of accessible wind turbines in Malaysia led us to creatively adapt local cooling towers for testing, proving the feasibility of a software-driven solution.

How it works

TurbineSense works by analysing images of wind turbine blades using artificial intelligence. First, a drone captures high-quality images of the blades. These images are uploaded and annotated (using Roboflow) to highlight areas of potential damage. We then use a computer vision model called YOLOv8, which is trained to recognise specific types of blade damage like cracks and surface wear. The AI scans each image and draws boxes around damaged areas. This detection process runs on a regular computer using tools like PyCharm and OpenCV. The final output is a set of images showing where damage is located, helping maintenance teams take action quickly — without climbing the turbine or stopping its operation.

Design process

Our design process started by addressing the risks and inefficiencies of manual wind turbine inspections. We aimed to develop a safer, smarter solution using AI and drone technology. After market research and expert input, we explored three concepts and chose to borrow a drone and focus on software development due to time and budget limits. Using a DJI NEO drone, we collected visual data and substituted cooling towers for turbine blades. We annotated images with Roboflow and trained a YOLOv8 model to detect blade damage. Testing with PyCharm and OpenCV showed reliable crack detection. We refined the model through multiple iterations and simulated the system using Unity. While thermal imaging was not included due to hardware constraints, the final software is accurate, adaptable, and compatible with most drone platforms.

How it is different

Unlike commercial services such as FORCE Technology, which use industrial-grade drones and thermal imaging for full-scale turbine inspections, our design focuses purely on a cost-effective software solution. We do not rely on expensive hardware or enterprise platforms. Instead, we developed an AI model using open-source tools (YOLOv8, Roboflow, OpenCV) that works with standard drone images, making it accessible for academic, small-scale, or rural users. Our system also supports offline processing and can be deployed using common hardware like PyCharm or even Raspberry Pi. It is uniquely suited for local testing environments like cooling towers and offers a flexible, scalable approach without needing full turbine access — making advanced blade inspection more inclusive and adaptable.

Future plans

In the future, we plan to expand our image dataset to improve the AI model’s accuracy and reduce false detections. We also aim to integrate thermal imaging once hardware becomes available again. Real-world testing on actual wind turbine blades is a priority, to validate performance beyond our current cooling tower substitutes. We hope to develop a user-friendly interface and possibly deploy the system on low-cost embedded devices like Raspberry Pi for field use, making it accessible to maintenance teams in remote or resource-limited environments. We also plan to make our digital twinning concept a reality if we have the resources and time.

Awards

We participated in our faculty's Engenius Week which is a week where all third year engineering students present their projects and we represented the Mechanical Engineering Department. Our project was different from others as we developed our own software and used the drone as a tool to complete our project.