Deep Learning Computer Vision in Production

Purpose and goal:

This project aimed to enhance quality control in Scania´s cab assembly using deep learning for front plug inspections, currently done manually and prone to errors. It focused on automating this process with computer vision, assessing efficacy, benefits, challenges, and readiness for production integration.

Goals included automating inspections, improving accuracy, reducing waste, and enhancing the work environment. Expected results and effects:

The project met its main objectives, assessing deep learning´s readiness for quality control in cab assembly and identifying integration requirements. It explored benefits in waste reduction, ergonomics, and reliability while highlighting challenges like detecting similar objects and the need for human oversight. The technology proved effective in recognising component variants, adapting quickly with minimal data, improving quality check frequency and accuracy, reducing cognitive load, and enhancing the work environment.

Approach and implementation:

The project used a platform from a startup to capture images and train a deep learning model with a team of 12-15 individuals, including factory workers and engineers. Focused on front plug inspection, data was collected at Cab Assembly Oskarshamn. The model was refined iteratively with input from production and quality personnel. The implementation displayed results on a screen, marking correct, missing, or wrongly mounted parts and protective tape, guiding production personnel effectively.