Quality Control using AI

By May 10, 2018January 24th, 2022Quality Control Insights
Quality Control using Artificial Intelligence, Neural Network

Artificial Intelligence (or AI) has evolved where the intelligence and capabilities of machine intelligence have exceeded human accuracies. One significant milestone was when Google’s Deep Brain was able to defeat the reigning champion in the historic game of  GO. Read more about why this is such a significant feat.

The impetus behind such steep advances in Machine Learning and AI are the following:

  • Access to huge amounts of digital data thanks to mass digitization and user-generated image content (mostly due to widespread smartphone penetration)
  • Access to computing power thanks to GPU computing driven by Gaming
  • Innovations in Algorithmic advances in Neural Networks and AI training models

Another milestone observed is in the ImageNet Large Scale Visual Recognition Challenge (ILSVRC). This challenge demands that a machine classifies as many of the images in a random database with the least error. The categories of images are hard (like “standing elephant”).

Until 2015 machines were not as good as humans in this challenge. That threshold was crossed when Microsoft was able to break the sub 5% error rate in this challenge. In 2018, most of the teams were under the human error rate. The stage is undoubtedly set for Artificial Intelligence to enter an industry that relies heavily on the human labor force for its process control.

Artificial intelligence (AI) is just now finding its way into manufacturing due to technological advances and cost reduction in implementation.

However, the main challenge in deployment has been in skilling the manufacturing workforce in understanding the capabilities of AI and adapting it into processes that can prove to be beneficial. In India, especially, this technology is still new and most of the industrial workforce haven’t been exposed to its capabilities.

Here at Qualitas Technologies, we’ve been working with AI and Deep Learning since 2015 as we believe this is where the future of machine vision is headed. As a Preferred System Integration (PSI) partner of Vidi Systems (now Cognex Vidi). We’ve deployed over 10 successful AI-based systems all over India and the world. Some notable applications are shown below:

Here’s an example of how AI could be used to decipher complex OCR text. In this example, we’re using a Deep Learning engine that has been trained to read the tare weight of an LPG cylinder.

In the automotive paint shop, AI and Deep Learning have been used to identify defects in the primer application process. By training a deep neural network on the patterns of what constitutes an acceptable pattern and with a few examples of defective patterns, the system is able to identify even the tiniest of miss-outs of primer paint.

In the critical areas of cast parts manufacturing. Surface defects are common with parts coming out of the casts. Various defects such as blow holes, overruns, dents, and chips are inspected with a large amount of human effort. In spite of deploying large human labor, errors do still happen. With image machine classification accuracies surpassing human capabilities, deep learning and machine learning technology are ideally suited for such applications. Not only making them more accurate, but it can do the job faster and in a highly consistent fashion.

You can download an application note of various surface defects that have been solved using Artificial Intelligence.

The model training process Before a machine learning system can be deployed it has to be taught to specifically look for defects. This process is known as “training”. The training process involves sending large amounts (typically hundreds) of labeled image data to a machine learning system.

The labeled data is split up into a training set (data which is used to teach the system) and a validation set (that which is used to validate once the system has learned the model). This overcomes any biases in learning wherein the training and validation are performed on mutually exclusive data sets.

Once the model is trained, it can be exported into a “run-time” configuration which is then used for prediction or inferencing. The training process typically involves a large amount of complexity in computation and is performed on high-speed PCs with powerful GPUs.

Neousys Technologies is one such company that’s a pioneer in industrial GPU computing. With its fanless wide temperature range industrial PCs, one can configure a system of your choice with the flexible and patented MezIO technologies.

Please get in touch with us to understand how Machine Learning and Artificial Intelligence can benefit your industry.

Leave a Reply

Schedule A Demo
close slider


hbspt.forms.create({ region: "na1", portalId: "726123", formId: "bd9f048a-7e04-4512-9179-c50855c961ea" });