Improve Inspection Accuracy Of Bearings With AI-Powered Vision System

By December 30, 2021January 24th, 2022Industry Use Cases
Automated Visual Inspection - Ball Bearings, Scratch detection

Overview:

In this case, the client has a vision system installed in the inspection station and deployed 3-4 inspectors at the end of the line to validate the results. The inspection accuracy is lower than accepted and therefore a vision system is required to improve it.

Client Requirements :

There are two major requirements –

  1. To inspect bearings outer body with an accuracy of 99-100 percent with the help of an automated vision system powered with AI
  2. To make the inspection process independent of manual inspection in order to save time and resources.

How Is The Problem Being Addressed Currently?

Bearings that are coming at the inspection station through a moving conveyor belt. At the inspection station, machine vision is deployed to identify the defective bearings and reject/reroute them. Due to fewer accuracy rates, 3 human operators are additionally required at the end of the line to revalidate the bearings passed through the machine vision system.

Bearing Manual Inspection, Manual Vs Automated Inspection

How AI Can Solve This Problem?

A 4 camera setup is used to acquire images of the outer race of the finished bearing assemblies. Every camera setup has integrated lighting to provide sufficient illumination to the area of interest. A set of red panel lights are used in this case to minimize the reflection on the outer race surface. After a dataset (images) of defects is acquired, the defects are trained using an AI-enabled software (Qualitas EagleEye® Platform) with the help of annotations.

Bearings Image acquisition

Once the software is trained the machine vision system captures the images of the bearing surface, processes, and identifies defects in real-time. To identify these defects an AI-Based Anomaly Detection Module is used.

Bearing Data Annotation
Bearing defect inspection

Conclusion

POC (Proof Of Concept) is conducted and the following conclusion is observed:

  1. False acceptance is reduced to 1 percent that would help our customers to reduce the recall rates.
  2. Inspection cycle time is reduced to less than one second that would help our client to increase delivery rates.
  3. Human intervention is reduced by 66 percent that translates to reduced labor and training costs


Leave a Reply

Schedule A Demo
close slider