Often in news we hear about building collapse due to cracks. Lots of lives are lost in such disasters which can be prevented if timely structural audit is done. Most of these cases are reported during rainy season. Crack detection in concrete structures is a key part of structural audit. Health of the building depends on the quality of maintenance. For older buildings or public structures, it is advisable that structural audits are done periodically. However structural audit is a lengthy process and requires lots of resources and time. We can use technology to automate the process or aid the engineers carrying out the structural audit. Same technology can also be used for detecting manufacturing defects by using appropriate dataset.
I have built a tensorflow keras model using convolutional neural networks to detect whether an image contains cracks or not. Firstly I trained the keras model using a public dataset of 40000 images. Then I deployed the model using a flask web app and Red Hat OpenShift. The rest api can be triggered in real time to detect whether an image contains cracks or not. I am excited regarding how technologies like AI can help to build solutions that can solve daily life problems and create a huge positive impact in the society.
The next step would be to deploy this solution using a drone and use it for structural audits. It can also be used as a standalone REST API by other companies (powered by Red Hat 3scale).
This project clearly showcases how we can disrupt a traditional industry using custom domain specific ml models and ReBoot Customer Experience as well as save lives.