Identification Of Automotive Parts (CKD) Using Deep Learning and Image Processing

CLIENT/INDUSTRY BACKGROUND

Our client is a multinational motorcycle company in India. It is one of the largest motorcycle companies in the country. In the year 2020, the company sold more than 4.5 million vehicles and produce 2 billion US dollars.

CLIENT’S PROBLEMS

• Unable to identify and differentiate very similar products that are transported to the assembly line of the vehicles.

PROBLEM IMPLICATIONS

• Increased downtime in assembly due to incorrect product variants received at the inspection station.
• Increased rejections (from assembly to manufacturing station) are being observed with an average rate of 9-10 percent.

CLIENT REQUIREMENTS

The client requires a highly accurate automated vision inspection system to identify the automotive parts in order to reduce/eliminate the mismatch of the parts.

CURRENT PROCESS

Operators are deployed at the inspection station to perform a visual inspection for the classification of the CKD Parts. They ensure that the right products are being transported to the assembly line. SKU mismatches are being observed despite the presence of operators.

BUSINESS IMPACT

• Loss of revenue due to low throughput.

SOLUTION USING MACHINE VISION

A solution was developed to acquire the images of all the CKD parts and train them in the QEP (Qualitas EagleEye® Platform) cloud by annotations. A pre-trained classification deep learning module is used for the identification of different parts.

SETUP

A machine vision setup will be installed at the inspection station (before packaging). The setup includes a color camera of 2.3 megapixels having 15 FPS to capture the images of the stationary parts placed by the operators. The distance b/w the parts (to be inspected) and the lens i.e. working distance would be 350 mm and therefore a 16 mm lens will be configured. 150×300 mm sq. will be the Field Of View (FOV). Dome light will be used to illuminate the area of interest (where the parts will be placed).

DETECTING DEFECTS IN QEP (QUALITAS EAGLE-EYE® PLATFORM)

CONCLUSION

A POC (Proof Of Concept) is conducted and the following observations were made –

• CKD parts were identified with an accuracy of more than 97 percent.
• The average cycle time of the inspection is reduced to less than 1 second