Detect Solar Panel Failure with SAM 3

The winter season is often harsh on solar panels. One snowfall and your panels go from generating clean energy to being suboptimal. After spending thousands on installation, it’s helpful for everyone (homeowners, landlords, factories, buildings) to know when this happens and implement a fix.

In this guide, we’ll use Meta’s SAM 3 Segmentation model and Roboflow Workflows to automatically scan an image of a set of solar panels, identify which panels are unobstructed and which are snow-covered, and return a count of them to build a tool that can be used for optimizing solar panel generation.

Let’s get started!

Detect Solar Panel Failure with Computer Vision

We’ll start by creating a new Workflow (finished version). If you haven't already, create a Roboflow account, with an API key, and a new workspace:

From here, create a new Workflow with the “Build My Own” template:

How to Build the Workflow to Detect Solar Panel Failure

Next, add the SAM 3 block, and add the input image (default):

SAM 3 excels at instance segmentation, which is perfect for what we’re trying to achieve. We can take advantage of this by adding target classes for it to detect. Inside the properties of the SAM 3 block, add the following classes to detect:

Thanks to Roboflow’s integration, running SAM 3 also does not require a separate API key, and rather, uses Roboflow credits! This implementation allows for rapid creation of segmentation projects. 

To see what SAM 3 detected, add a polygon visualization block that takes in the SAM 3 block’s predictions:

Make sure you also specify the input image from earlier, so it’s able to display visuals on it. 

Lastly, connect the visualization to the output block, and add the following outputs:

Through this, we’ll have access to the visualization image, as well as the raw predictions from SAM 3 (we’ll need this to build the tool). Saving and testing the workflow:

As you can see, SAM 3 does an excellent job at identifying both the solar panels (purple), and covered panels (red).

Now, we’re ready to deploy and build our project.

Implementation of the Workflow

Once the workflow is saved, click “Deploy”. Since this project will run on images of a solar array, it is most easily deployed via the serverless API:

Using this option provides sample code in various languages (Python for this demo) that automatically allows you to run the workflow on a test image in a local environment:

However, make sure you install the inference-sdk:

Add this to a local environment with a test image. Then, change the path of the image (“YOUR_IMG.jpg”) to match the path of your test image:

Now, update main.py to: 

import base64
import cv2
import numpy as np
from inference_sdk import InferenceHTTPClient

client = InferenceHTTPClient(
    api_url="https://serverless.roboflow.com",
    api_key="YOUR_API_KEY"
)

result = client.run_workflow(
    workspace_name="YOUR_WORKSPACE_ID",
    workflow_id="YOUR_WORKFLOW_ID",
    images={"image": "YOUR_IMAGE.jpg"},
    use_cache=True
)

predictions = result[0]["sam_3_predictions"]["predictions"]
visual = result[0]["visual"]

efficient, failing = 0, 0
for p in predictions:
    if p["confidence"] > 0.7:
        if p["class"] == "solar panel":
            efficient += 1
        else:
            failing += 1

print(f"Efficient solar panels: {efficient}")
print(f"Failing solar panels: {failing}")

img = cv2.imdecode(np.frombuffer(base64.b64decode(visual), np.uint8), cv2.IMREAD_COLOR)
cv2.imshow("Solar Panel Failure Detection", img)
cv2.waitKey(0)
cv2.destroyAllWindows()

This script runs the workflow on a solar panel array to detect panel failures. It counts efficient vs failing panels by looping through the SAM 3 predictions, prints the counts, then displays the annotated result image using cv2. The failing solar panels are the ones that SAM 3 detects are covered with snow. Additionally, it ensures the confidence of the prediction is substantial (> 70%), ensuring accurate annotation.

Running main.py (after replacing all the necessary info):

And with that, your tool is complete!

Conclusion: Predicting Solar Panel Failure with Snow-Segmented Thermal Imaging

In this guide, we used SAM 3 and Roboflow Workflows to build a solar panel snow detection tool. With just a segmentation model and a few lines of Python, we can instantly identify which panels in an array are underperforming due to snow coverage. The applications for this project scale tremendously, and has the potential to help plenty of people be more efficient.

Written by Ayran Vasudevan

Cite this Post

Use the following entry to cite this post in your research:

Contributing Writer. (Mar 17, 2026). Detect Solar Panel Failure with SAM 3. Roboflow Blog: https://blog.roboflow.com/detect-solar-panel-failure/