Advances in artificial intelligence and deep learning technologies have recently enabled automated visual inspection systems that can outperform previous human or m achine vision processes.

Based on half a century of AI research, the NEC Visual Inspection System enables manufacturers to accurately identify defects and dramatically reduce costs with an efficient solution that can be easily and affordably deployed. The system utilizes a deep learning one-class classification architecture that allows it to be trained using only images of non-defective components, as opposed to architectures that require images of both good and defective parts to learn the classifiers. This has significant advantages in many industrial settings, especially where defect rates are already low.

Virtually all manufacturing processes include some kind of method for identifying flawed components. Many manufacturers still rely completely on human inspection, as automated visual inspection has previously been too inaccurate, slow, and expensive.  Where AI automation has been deployed, it has generally required large investments of money, time, and expert resources.

Effective and Cost-Efficient Solution for Optimal Quality Control Operations

The NEC Visual Inspection System provides many valuable benefits for manufacturers:

Reduced inspection cost: New installations often pay for themselves within 12 months, through cost savings 

Easy to deploy: Minimal hardware requirements, and simple integration 

Secure: No dependence on cloud technologies

Fast to train: Optimized neural networks learn to distinguish good and bad parts quickly

Easy Training with images of good parts only: “One Class” classification deep learning algorithms do not require images of defective components in order to be trained.

Consistent, accurate inspection, as compared with human inspectors

These simplifications include:

No need for manually programed rule

No need to find and categorize defect types

No need to collect images of defective parts

While deployment requirements vary from site to site, generally the necessary hardware includes only computing and data storage, standard commercial cameras, and standard commercial lighting.

The deployment process occurs in three phases:

1. Capture Images: Capture images of “good” parts to be used to train the one-class deep learning model.

2. Train and Test the Deep Learning Model: Using the image of good parts, the software will develop a model containing the classes and recognition algorithms necessary to recognize defective parts.  Then the model can be tested to verify that false positive rates and false negative rates are within target limits.

3. Deploy to Production Line: When the system has been trained and tested,  it can be deployed to the production line.

The NEC Visual Inspection System’s simplified deployment with flexible interface and reduced need for manual inspectors makes it an effective and cost-efficient solution for optimal quality control operations.

SOURCE OF CONTENTS

sg.nec.com

www.necam.com/AI/Defectinspection/ 

Jiro Chiba 

jiro_chiba@nec.com.sg