Machine Vision Inspection: Tools of the Trade
Automated in-process inspection of parts and products has been executed successfully using machine vision technologies for decades in a multitude of diverse applications for nearly every industrial and manufacturing process. While no hard numbers exist, it is plausible to make the claim that inspection tasks likely dominate the machine vision application landscape. Still, new machine vision technologies and software continue to emerge that further improve the value proposition and ease of implementation of automated inspection. Key to success for end users is understanding both the mature and new tools of the trade and how these tools can best be implemented.
The variety of inspection-related tasks in automation processes makes it difficult to label each possible use case. In general, some of the important categories include assembly verification, feature presence/absence, defect detection (in many forms), and product identification and differentiation. In all cases it is important to remember that machine vision can be a critical component of the concepts of big data and Industry 4.0.
When implementing machine vision inspection for industrial automation, perhaps the most significant value proposition lies in how the results of the inspection can be used. More than just inspecting for quality, the information gained by machine vision inspection can be vital to improving the efficiency of the broader process and thereby help to reduce overall production and automation costs.
Imaging — The Foundation of Every Successful Application
It always bears repeating that no matter what the use case or analysis method, a quality image is the critical foundation of any machine vision project. A quality image is defined as one that has the correct resolution and contrast to highlight features (objects, parts, defects) of interest in the desired field of view. Proper design specification involves the imaging component itself as well as related and required components including illumination and optics.
For many inspection applications that employ proper imaging design, 2D grayscale continues to be the most widely used technology. Advances in speed and resolution in sensors and cameras are enabling more use cases that previously might have been unachievable or too complex to be practical. With the availability of cameras with sensor resolutions of 12 MPixels to 50 MPixels and higher, detecting smaller features in larger fields of view becomes both easier and less expensive. Soon, image acquisition at 5–10 MPixels may be considered standard instead of high resolution.
Smart cameras, a core machine vision technology, continue to enjoy growth in inspection tasks and regularly feature higher resolutions and faster processing. In addition, there have been significant developments in standard camera architectures, which incorporate embedded processing using FPGAs and other computing platforms. These components are well suited to inspection applications that can be scaled for multiple duplicate installations to take advantage of the related reduced costs for some of these types of components.
Beyond 2D and Grayscale Images
Machine vision components that capture 3D information of a scene are very readily available with a variety of imaging methodology and implementation techniques. The 3D images provide a topographical representation of an object’s surface geometry whereas 2D imaging captures an image of the contrast (grayscale or color) found on the planar surface of an object. The 3D data enables or enhances inspection tasks involving features or defects with more geometric structure than contrast. Using 3D systems has become dramatically easier, and as with their 2D counterparts, resolution, speed, and accuracy continue to expand.
As noted, many machine vision inspection applications use cameras that provide a grayscale image of an object (also called monochromatic, since it is an image without color, or essentially a single color). Some applications, however, can benefit from, or must rely on, color images to provide the information needed for the analysis. Standard cameras are readily available for machine vision that captures an RGB (red, green, blue) image. Where properly integrated, these components can enhance the reliability of feature analysis where color is part of the defining characteristics of the object or defect. While most color cameras in the market use a filtering system on the sensor (Bayer filter), advanced camera components also are available that optically split the incoming image into three full-frame channels (typically RGB) for better resolution and color differentiation.
Imaging Beyond the Visible
While not new but more widely available in recent years, an even more powerful color imaging technique called hyperspectral imaging, and its close relative multispectral imaging, can perform more discrete and granular color analysis. These cameras collect multiple — sometimes hundreds — of images of a single scene, each with a different narrow bandwidth of spectral information from the scene. This type of component, with specialized classification software, can execute spectral inspection of materials or even detect chemical composition. Many industries such as food, pharmaceutical, and recycling benefit from this type of inspection capability.
Expanding on color imaging further, we find components that can create images using nonvisible illumination and even thermal energy. Generically, this may be described as infrared (IR) imaging. Applications that image near infrared (NIR), shortwave infrared (SWIR), and longwave IR (thermal imaging) provide views of objects that are not seen in visible wavelengths. This capability can be used to great advantage in many inspection applications ranging from detection of spoilage in food to seeing through opaque plastic containers to confirming fill levels.
Easier Inspection Implementation
The algorithms and software tools used with machine vision inspection tasks are quite mature and reliable. Overall, when working with a reliable image, many inspection tasks are simple to implement. New technologies have emerged that have potential for adding to the existing machine vision tools to provide further capability and different processing approaches.
The most publicized of these technologies is deep learning. Deep learning for machine vision is specifically targeted and well suited for inspection applications and is being applied successfully to a growing number of inspection applications in industrial machine vision. But the design, configuration, and integration of applications using deep learning requires a completely different implementation approach from that used for traditional machine vision projects.
So-called traditional machine vision implementation techniques involve creating a set of rules about a target object that are executed using algorithms that return specific information about the object or scene. Deep learning is trained with many representative images containing examples of good and defective parts or objects. It is not, however, a silver bullet for all inspection applications. The need to collect many images before being able predict the level of success that might be achieved may be cumbersome for some applications, and ongoing maintenance of the system and its classifications might not be suitable for a specific use case.
The Future of Machine Vision Inspection
With expanded demand for quality, smart manufacturing, and data collection, the implementation of machine vision technology for inspection applications continues to grow. Capabilities of advanced components and software for inspection certainly will drive additional use cases and provide additional value in the future.
- Unplugging the Future: Leveraging 5G to Interconnect the Factory Floor
- How to Address Tradeoffs in Robot Performance
- Manufacturing Labor: The Crisis Still Continues
- Conversations on Edge AI: Seven Takeaways from A3’s Intelligent Edge Webinar Series With Intel Corporation
- Driving Sustainability and Profitability Through Digital Transformation
- How Companies Can Use Open Standards to Drive Industrial Control Transformation
- View All