May 6, 2022
Table of contents
Technology Research Analyst
From watchmaker's monocles to the so-called "camera lucida" used to facilitate drawing, artisans are no stranger to enhancing their sensory skills and manufacturing goods or works of art with the aid of optical technology. Following the industrial revolution and the inexorable shift from manual craftsmanship to machine-driven mass production, however, our approach to creating things has completely changed. As a result, all those optical devices designed to augment human vision have given way to new high-tech tools that could provide such machines with some kind of visual perception. And this is where computer vision comes into play.
Computer vision in manufacturing focuses on the creation of artificial systems which can capture, process, and therefore understand visual inputs from the physical world (mainly factories and other industrial sites) to elicit proper reactions and assist humans in a variety of production-related tasks.
The most primitive incarnations of computer vision, both in manufacturing and in other fields, can recognize specific objects and trigger an answer according to a rule-based principle, namely by identifying certain features in the captured visuals and verifying if they match with a set of given parameters. However, this approach is prone to creating a lot of false positives and isn't as efficient for handling the subtler nuances and variations that typically occur when dealing with unstructured data sources like pictures or videos.
The latest advances in AI, machine learning (ML), and deep neural networks have helped to solve these issues, allowing manufacturing enterprises to enhance their computer vision systems with self-improving algorithms which can identify recurring visual patterns and relate them to certain items through experience. In practical terms, ML-powered computer vision solutions can be trained with millions of images to autonomously spot the typical features of each object (including products and machinery with very complex structures or in anomalous conditions), learn to recognize them, and even fine-tune their performance over time.
This has resulted in higher precision, better context understanding, and superior flexibility and responsiveness to new visual elements without prior programming, as well as an ever wider range of applications in the manufacturing field.
Nowadays, computer vision in manufacturing is commonly deployed to:
Before delving into these use cases, let's find out the benefits of investing in computer vision services, along with some of the potential adoption barriers and the best practices to overcome them.
Over the past two years, the operational and logistical disruption caused by the COVID-19 outbreak has put a strain on a manufacturing industry already bogged down by a ten-year productivity slowdown, as pointed out by the US National Institute of Standards and Technology.
The need to seek ways to gain new momentum and make this sector more resilient in highly unstable scenarios has been a huge catalyst for the digital transformation in manufacturing, namely the implementation of new cutting-edge technologies and the resulting transition to the Industry 4.0 model. The growing deployment of computer vision in industrial processes certainly represents one of the major aspects of this transition, as it significantly contributes to strengthening the manufacturing sector in terms of:
Most of the sector analyses seem to confirm the positive impact of computer vision in manufacturing. Based on Grand View Research's 2021 Computer Vision Market Size, Share & Trends Analysis Report, the manufacturing segment led the global computer vision market in 2020 due to the extensive use of this technology for assembly line automation and predictive maintenance.
Furthermore, according to IBM's 2021 Digital transformation assessment report, computer vision represents one of the major technologies helping manufacturers achieve their business goals (77% of the companies surveyed). On the other hand, IBM's study also sheds light on some of the most common barriers to adopting new technologies in the manufacturing field.
Let's better frame the aforementioned issues and define some general guidelines that may be helpful for addressing the challenges of computer vision deployment in a manufacturing scenario:
Since we mentioned the identification of suitable use cases as a potential challenge for computer vision implementation, here’s a brief overview of key applications and successful real-life examples of this technology in the manufacturing sector.
If Leonardo had known about the future role (and capabilities) of robots in manufacturing, he would probably have been thrilled. For 21st century humans, instead, industrial robotics is something taken for granted. Nowadays, computer vision-guided robots represent a cornerstone of any assembly line. In fact, they can easily identify and manipulate objects with mechanical arms or map their surroundings to roam a manufacturing plant, which makes them a valuable tool for improving production outputs and streamlining warehouse management and logistics.
These are some of the typical tasks performed by computer vision-powered robots:
The examples of this technology on the ground are countless. Among them, we may mention Sawyer Robot, a multipurpose robotic arm deployed by Tennessee-based plastic injection molding company Tennplasco, but also BluePrint Automation’s robotic carton-loading system leveraging computer vision to grab unoriented packages. Austrian automobile manufacturer Magna Steyr, on the other hand, adopted smart drones to scan labels and facilitate inventory operations.
Computer vision-powered robots are extremely accurate, but something in the production chain can always go wrong. Fortunately, computer vision systems can also be deployed to double check the quality of a product. This advanced type of automated visual inspection involves scanning finished products with high resolution cameras, processing data with machine learning algorithms to identify anomalies, and therefore ensuring that each item (including its packaging) meets all necessary quality standards.
In this regard, take a look at Volvo Cars’ solution. Its computer vision system, named Atlas, can scan each vehicle with over 20 cameras to spot surface defects, allowing it to find up to 40% more anomalies compared to manual inspections.
The devil is in the details, both when it comes to identifying manufacturing defects and industrial asset anomalies. The good news is that computer vision systems, enhanced with machine learning for anomaly detection, can deal fairly well with details. Indeed, these tools can probe industrial machinery with cameras, infrared thermography, and other types of sensors to spot any deviations (such as anomalous temperatures and vibrations) that may be signs of malfunction and predict upcoming failures before they actually occur.
General Motors, for example, adopted a computer vision solution designed to analyze images from cameras mounted on assembly robots and detect failures affecting their components.
Computer vision can be a guardian angel for both machines and, most importantly, humans, since predictive maintenance allows manufacturing companies to repair machinery in advance and thus avoid hazardous situations. Furthermore, it can be used to constantly supervise complex manufacturing operations in a variety of industrial environments.
A similar approach was taken by Komatsu Ltd, a major UK construction equipment manufacturer that partnered with NVIDIA to adopt a computer vision solution based on artificial intelligence and video analytics. This platform can monitor and even predict the movement of workers and equipment to signal potential collisions or other dangerous interactions.
Computer vision, along with many other technologies involved in the digitalization of industrial processes, has proved to be a valuable ally for manufacturing enterprises, resulting in significant cost cuts, higher production output and quality, superior precision, and increased staff safety. Obviously, organizations should not take the adoption of computer vision lightly, since its actual deployment might end up being trickier than expected.
However, with proper investments, retraining programs, workflow harmonization initiatives, and use case identification, computer vision-powered machines will drive industrial manufacturing just as the love of beauty encouraged artisans to create their handcrafted products in past centuries. With the difference being that machines, unlike love, are not blind. Not anymore.
United States
720 207 2820Sales inquiries