Automated Visual Inspection in the Age of Covid-19

The pandemic has accelerated the move away from manual inspection

The United States faced a growing labor shortage in manufacturing before the Covid-19 pandemic devastated the nation. While many manufacturers have been deemed essential, many of their employees fear going  back to work until a vaccine is widely available. Unfortunately, medical authorities are predicting that healthy people may not have access to a vaccine until 2022. 

Deloitte’s 2018 manufacturing survey projects that 4.6 million new manufacturing jobs will be needed in the coming decade, but that only about 2.1 million will be filled by  qualified candidates – leaving a shortage of 2.5 positions. The shortage will be even more critical in quality assurance and inspection. These positions require advanced skills and dedication beyond many factory workers. Even with the most skilled inspectors, unacceptable error rates are a fact of life. 

Over the last thirty years, there has been a growing acceptance that human visual inspection is ineffective for technical, psychological, organizational, workplace environment, and social reasons. Drury, Karawanu and Vanderwarker suggest that inspection error rates may range from 20% to 30%.2   Regardless of the dedication, humans have major detection limitations versus machine vision.  The human eye cannot make precise measurements. This is especially true for very tiny items. When comparing two similar objects, the human eye struggles to notice that one is slightly larger or smaller than the other. This issue also applies to size, surface roughness,  and other factors that need to be inspected. The chart below from Nanonets highlights the major limitations in human vision. 3

scope of machine vision with respect to visible spectrum

Beyond the limitations of manual/human visual inspection, there is also a cost factor.  Ohno and Shingo  pioneered Lean at Toyota in the 1960s,  advocating building quality into products over adding layers of inspection, some inspection will be needed for the foreseeable future.  According to Glassdoor, quality inspector salaries average about $38,000 per year. Adding 15% for benefits and the need to cover three shifts, manual inspection can total over $130,000 per year. 

Because of the systemic and growing problems with manual visual inspection, automated visual inspection is growing in its acceptance. The chart below demonstrates the diversity of industries applying machine vision to their operations.

Breakdown of Visual Inspection in Industries pie chart

The good news is that Overview’s machine vision technology provides image-based automatic inspection to details and defects for any type of serial production process.  Overview’s automated vision surpasses human vision at both a qualitative and quantitative level because of its speed, accuracy, and repeatability. It easily assesses object details too small to be seen by the human eye and inspect them with much greater reliability than manual methods. Overview uses deep learning algorithms that can accommodate more variability than traditional vision systems and be cost effective on a greater variety of parts and products. Unlike manual inspection, Overview’s solution never takes a break, or takes its eye off the action. It can also inspect parts that are moving at much faster speeds than human inspectors can. 

 

Conclusion

Automated visual inspection can overcome the limitations of human inspection and do it at a lower cost and more quickly  than traditional manual methods. The cost of deep learning and machine vision will continue to drop while its effectiveness grows – Moore’s Law is alive and well.  Given the growing labor shortage that has been aggravated by Covid-19, it makes sense to try Overview’s automated visual inspection solution.  It is easy to get started with a risk-free Proof of Concept in which you can see results in as little as seven weeks.

Refrences

  1. https://www2.deloitte.com/us/en/pages/manufacturing/articles/future-of-manufacturing-skills-gap-study.html
  2. Drury C.G., Sinclair M.A., Human and machine performance in an inspection task. Human Factors, 25, 391–399, 1983.
  3. https://nanonets.com/blog/ai-visual-inspection/

Contact:

 

Anthony Tarantino, PhD

Six Sigma Master Black Belt, CPIM, CPM

Adjunct Professor, Santa Clara University

Senior Advisor to Overview

tony@overview.com

Russell Nibbelink Overview Co-founder

Russell Nibbelink

Co-founder & Head of Engineering
Russell is a co-founder and head of engineering of Overview. At Overview, Russell helps customers implement cutting edge high speed Deep Learning algorithms. Prior to joining Overview, he was a software engineer at Salesforce where he worked on web-scale infrastructure projects. Russell received his BS from UC Berkeley’s College of Engineering, where he helped teach a class on High Performance Computing.
Chris Van Dyke Overview CEO

Chris Van Dyke

Co-founder & CEO

Chris Van Dyke is a co-founder and the CEO of Overview. Prior to founding Overview, Chris spent eight years in manufacturing engineering at Tesla. Most recently, he led the 80 person battery design team through the launch of the Model 3. During which time he was principally responsible for taking the team from battery design to high-volume production. Earlier in his career at Tesla he managed the infrastructure and equipment design for the first Gigafactory project including equipment for supporting battery cell manufacturing. Chris also launched Tesla’s Electric Vehicle Supercharger Program, which currently has more than 16,000 stations nationwide.

Chris holds several patents from his time leading custom equipment design while a Senior Engineer at H2Gen Innovations. Chris received his BS from Stanford University in Mechanical Engineering and an MS from the University of Virginia in Chemical Engineering.

Austin Appel

Co-founder & Head of Product

Austin Appel is a co-founder and head of product of Overview. At Overview, Austin leads product development and operations. Prior to Overview, Austin spent four years at Tesla in roles across battery manufacturing and R&D. He led the DFM and automation efforts for Model 3 battery pack production at Tesla’s first Gigafactory. Previously he expanded Tesla’s production of Model S and X battery packs through equipment design and implementation. He holds a dual bachelors degree in Manufacturing Engineering and Mechanical Engineering from Northwestern University.