Paint fault detection

Client: Large car manufacturer.

Analysis and further testing.

Industry: Automotive.

Project Duration: 6 months.

Goal: Classify fault in paint on cars in a factory.

Tech: Tensorflow, Keras, feature creation.

A positive result

The challenge:

Classify faults in car coatings and to automatically label future faults.

The approach:

Discovery workshop with experts

Simplify problem to make it solvable (from 13 categories to 2)

Humans have an error of approximately 20%.

Using votes from the experts to create labels

Create a baseline solution

Improve solution using Deep Learning

The solution:

The solution of the deep learning methods was the best

The solution implemented the scanning robots

Robots give advice to employees

The results:

Reduced classification time and more time for correct labeling

The deep learning method has a 97% correct classification

85% error reduction