Simplifying Tire Inspections Using Machine Learning

The application needed to:

• Detect tread depth using images of the tire surface
• Replace manual inspections with automated quality inspections

The solution measures the depth of tire treads using image processing for edge detection. Our data scientists trained the deep neural network to predict the relative depth of tires based on reference images. This proven approach in estimating measurements using a neural network helped implement a deep regression network.

• Automation resulted in reduced manual intervention and increased efficiency, with a renewed focus on productivity and product design
• 66% increase in accuracy and low defect rates resulted in reduced warranty claims and improved quality
• Preventive maintenance enabled by real-time tire analysis and monitoring
• Improved vehicle safety without requiring specialized operator training

• Custom CNN architecture development for depth detection using established architectures such as UNet and DenseNet as base
• Proved the concept of high-quality monocular image depth detection on minute relative distances (in the range of millimeters)
• Accuracy level of +/-1.5 mm on 90% of independent test images
• APIs to reject images due to reasons such as insufficient tire surface, overexposure, and poor quality
• Ability to acquire valid tire tread depth images from thousands of tire images

• TensorFlow
• Pytorch
• Numpy
• Python
• Django