Detecting Steel Defects

We used a public dataset from roboflow and trained an image classifier model to identify the defects. We will also create a frontend application to use this trained model. With this model, we will attempt to answer two things. Given an image of some steel architecture, is it a wear and tear defect or production defect and try to give an estimate on the estimated remaining life of the steel architecture. We hope to test our model on some real world architecture to get viable results for estimates. Practical use case : Different architecture and buildings could be scanned by anyone. Say there is a bridge and it shows that it might collapse in a month. People can be informed about this and civil engineers are no longer required. Industrial use case : Steel industries could employ this model framework to improve production. We found a Kaggle dataset for 12000 images of different steel defects. We downloaded the data from roboflow and used these labels to train a model that highlights defect area. We use this processed image in our frontend to display the detected area and also use this image to get several dimensions that we will use in a regression model. The regression model is responsible for using historical data and normalized values to output the expected remaining life of the steel in image.

Built With

• python