Why image analysis could create a breakthrough in manufacturing automation

image analysis manufacturing
0 Shares 0 Flares ×

The manufacturing environment has always been one in which people often perform tedious, repetitive tasks. If you ask any manufacturing operative about the monotony of quality inspection, sorting or grading, and moving materials or components from A to B, you’ll soon know why these tasks are eminently suitable for automation.

However, the reason these activities haven’t been automated in the past is the need for eyesight—a sensory ability which until very recently was a gift possessed only by humans and other forms of complex organic life. This situation is now changing as computer vision technologies come of age, driven by machine learning and other artificial intelligence advances.

For example, Robotics Business Review reports that the North-American demand for machine vision solutions reached an all-time high in 2017, with sales growing by 14.6% over the previous year, according to the data released by the Association for Advancing Automation (A3).

Now let’s look at how this demand is satiated on the production floor where many of the computer vision-equipped assistants are coming to redefine manufacturing routines.  

The Step-change in Computer and Robotic Vision

If you’re not already familiar with the latest developments in computer vision, you may be wondering how it has become possible for automated systems to be deployed on tasks requiring the gift of sight.

In fact, computer vision is not completely new. Image analysis using cameras, single board image processors, and grayscale machine-vision algorithms have been used in quality assurance and packaging since the 1980s. Until recently though, the algorithms used for image processing were based on rigid rules, requiring developers to hard-code shapes and sizes of objects in order for software to recognize them. This form of image recognition is inefficient—prohibitively so when complex objects and shapes are involved.

A breakthrough arrived with the application of machine learning to solve optical recognition problems. The most successful approach uses deep learning models known as convolutional neural networks (CNNs). The CNNs are trained through exposure to vast quantities of images, from which they learn to identify patterns in visual data and become capable of recognizing complex shapes and object structures.

Computer Vision Drives a Renewed Focus on Automation

According to the market research company Tractica, the value of the annual market for computer-vision technology will reach $48.6 billion by 2022. The technology is already being adopted across the spectrum of manufacturing and production industries, bringing automation and its associated benefits to tasks that are tiresome and expensive for humans to perform.

Let’s take a look at a few manufacturing activities which are increasingly automated through computer vision, enabling companies to reduce labor costs and redeploy staff to more physically and emotionally rewarding duties.

Automated Visual Inspection

If you find it discomforting that robots and computers should be taking over the tasks once performed by humans, and are unnerved at how they evolve to perform those duties with superhuman effectiveness, this first example of image analysis software in manufacturing may only reinforce your discontent.

On the other hand, if your company is currently spending a lot on component inspection, you may welcome the idea of automated replacements for your human QA inspectors.

Automated optical inspection (AOI) solutions based on advanced image analysis are finding their way into production environments where exhaustive visual quality inspections are required. For example, the manufacture of printed circuit boards (PCBs) is one area in which AOI is especially pervasive.

AOI in Printed Circuit Board Manufacture

Automated optical inspection has become essential in the manufacture of printed circuit assemblies, as these vital hardware components are growing smaller and more complex. Even a basic PCB may have thousands of soldered joints, and this is the most likely cause of malfunctions, making visual quality inspections vital. With the high throughput of PCBs along the production line, manual inspection is no longer practicable—and fortunately, no longer necessary.

Instead, image analysis systems can be installed on production conveyors to inspect PCBs immediately after the soldering process is complete. This placement ensures that soldering faults are identified right away, making rectification less costly than when defects are discovered further along the production line.

Optical Sorting and Grading

The use of cameras to optically grade and sort produce such as fish, fruit, and vegetables is not especially new. Until recently though, the systems in use required a fair degree of human attention. This is because the accuracy of image analysis algorithms was dependent on fine-tuning that had to be performed by a human operator.  

With the integration of deep learning technology, however, optical sorting and grading systems can operate completely autonomously, becoming exponentially faster, more reliable, and more consistently accurate.

These systems are suitable for a wide range of visual quality inspection processes, lending themselves perfectly to the grading and sorting of the following products, just to name a few examples:

  • Fruits and vegetables
  • Seeds and nuts
  • Fish and shellfish
  • Recycled and virgin plastics
  • Timber products

Non-stop, Precision-assured Sorting and Grading

Grading and sorting solutions based on deep learning can distinguish and analyze products based on shape, size, and color. They can also be trained to identify product features, like knots in timber, and defects such as bruises, blemishes, and scratches on fruit and vegetables. This enables the technology to take over from human operatives, for whom grading and sorting are tedious, repetitive tasks carrying an ever-present risk of lapses in concentration and precision.

No such lapses can occur when image processing hardware and machine learning are put to work on the production or packaging line. Furthermore, precision accuracy is assured on a continuous basis, 24/7/365, at a fraction of the cost of human graders.

Robots with the Gift of Sight

The advances in image analysis and object tracking are allowing robots to break free from the constraints of primitive guidance. In short, they can finally see the world around them and take on many tasks that until recently were beyond them and hence needed the human touch.

Robotic vision is especially useful for tasks requiring mobility and navigation, such as moving materials around in warehouses. The system deployed in Amazon’s fulfillment centers is a prime example of robotic vision in the real-world use. That said, the Amazon automatons’ image-recognition technology is quite rudimentary. The robots navigate by scanning barcodes affixed to warehouse floors, though they also employ a degree of spatial sensing to calculate distances.

The next generation of warehouse robots, like those developed by Vecna Robotics (a company specializing in warehouse automation), use a more sophisticated combination of camera hardware and machine learning software, so they can actually study the environment and develop the correct reactions to the features and events they encounter.

Seeing the Way to Increased Manufacturing Automation

Production and manufacturing tasks, along with many others, require optical capabilities which previously made them impossible to automate. Now the technology is finally available to the enterprises that are ready to invest in image analysis. With deep learning algorithms improving all the time, the expectation is that they will stimulate growth in the use of computer vision systems in manufacturing.

This growth is not surprising, since advanced image analysis offers attractive benefits to manufacturers, including cost reductions, improved efficiencies, consistency of quality, and the ability to switch human attention to activities with a greater value for companies and their customers.

0 Shares Twitter 0 Facebook 0 Google+ 0 LinkedIn 0 Email -- 0 Flares ×
The following two tabs change content below.

Adriana Blum

Adriana Blum is a Senior Mobile Developer and Technical Lead at Iflexion with 13+ years of experience in designing and implementing software applications for renowned companies. She specializes in cross-platform development on React Native and Xamarin and has broad experience in native iOS and Android development. Currently, Adriana is actively researching the capabilities and applications of AR and VR in the mobile industry to create innovative mobile solutions for outstanding user experience. Leveraging the potential of Apple’s ARKit and Core ML, she’s working on the prototypes of AI-based apps for Retail and Entertainment.