What Technology Goes Into Manufacturing Machine Vision Lenses?
Many intricate components go into any machine vision operation, including lights, a camera with lens, sensor, and shutter, a computer with specifically designed software, and sometimes a display so someone can monitor the machine’s progress.
Every one of these is an integral part of the process, and as such, they should all be chosen with explicit care to find one that is exactly right for your situation.
Here we are going to go through several of these components to discuss their importance and what goes into building them, so you can understand how to differentiate the good from the great.
This includes the necessity of proper lighting, current breakthroughs in sensor technology, and how lens ruggedization is bringing greater efficiency to machine vision.
Why Does Lighting Matter for Machine Vision?
To some, the lighting present in your operation might be an afterthought, as you focus on the quality of other parts like sensors and software. But this key element should never be overlooked, as lighting can easily impact any sort of space and directly affect the results of your machine vision operation.
Simply put, you want to offer the best possible image to your machine algorithm so that it has accurate data to work with and can best see the object it’s trying to inspect. If you feed that algorithm an image that doesn’t fully showcase the item, it may flag it as acceptable when it is not.
As explained, lighting does not inherently cause issues in a pipeline, but it can cause delays, consternation, and cost overruns if not attended to properly. Today, modern lighting solutions are such that they can be tailored specifically toward the topography and reflectivity of your equipment by offering bright, dark, or diffusing panels for creating lighting.
While some would argue that artificial intelligence and software learning can compensate for lighting shortcomings, others would assert that offering the algorithm with the best possible information is of the utmost importance for acquiring the best results.
Quality Magazine wrote an excellent report about lighting in machine vision, including a case study review. They worked with 35 images of damaged parts and 35 images of undamaged parts and then used these in both ideal and nonideal lighting conditions; these images were also flipped on their axes to offer a total of 140 images in each lighting configuration.
By having these images analyzed by the same network configuration, it was found that the improved lighting offered a 12.85% accuracy increase, with only a minute increase in inference time.
They concluded that: “Deep learning cannot compensate for or replace quality lighting. This experiment’s results would hold true over a wide variety of machine vision applications. Poor lighting configurations will result in poor feature extraction and increased defect detection confusion (false positives).”
What Impacts Do Sensors Have On Machine Vision?
The image sensor will ultimately capture images of an in-production piece of equipment, so having one suited to the task is of the utmost importance. As IDTechEx explains, these sensors need to do more than just acquire RGB values per pixel.
Instead, these need to capture as much data as possible to pass this information to the algorithm for evaluation. If the algorithm doesn’t have the best available data, it might not produce the best available results.
This means acquiring optical information with spectral resolutions or across different wavelength ranges via custom sensors. IDTechEx identified several emerging sensor technologies that could be of benefit to machine vision, including increased dynamic ranges, spatially variable sensitivity, lower costs, and improved temporal resolution.
IDTechEx also noted that the development of cheaper alternatives to InGaAs sensors, which can be exceedingly expensive, paves the way for cost reductions and opens up potential applications for machine learning and powerful sensors.
Specifically, they cited autonomous vehicles, where SWIR imaging can assist in identifying similar objects or materials that are seen in the visible spectrum. SWIR imaging is done in the short-wave infra-red spectral region between 1000 and 2000 nm.
Particularly, IDTechEx identified an image sensor technology that seems well-suited toward machine vision: event-based vision. This is a technique that they define as “a new way of obtaining optical information that combines the greater temporal resolution of rapidly changing image regions with much-reduced data transfer and subsequent processing requirements.”
This only activates the sensor in the exact moments it is needed, leading to lesser data collection and better optimization of digital storage. As industrial machine vision operations can accumulate immense amounts of data, transferring it more quickly and easily can be invaluable.
What Does Lens Ruggedization Mean for Machine Vision?
Typically, manufactured lenses have several moving components to allow the focus to be adjusted manually through a system of concentric threaded barrels that slide forward and backward as the enclosure is rotated.
While this type of solution works well for commercial photography cameras, machine vision requires more specific lenses that may withstand varying moisture, shock vibration, and temperature conditions.
As such, lenses are steadily becoming ruggedized in order to make them sturdier and stable devices for industrial purposes. Quality Mag outlines the three primary types of ruggedized lenses: industrial, ingress protected, and stability ruggedization.
Industrial Ruggedized Lenses
These lenses are designed purely for industrial purposes, removing many moving, vulnerable components so that the lens will remain at a set position and focus. Here, the lens has a fixed aperture, and instead of an iris diaphragm, it utilizes a metal annulus to keep parts stable.
In addition, it is designed with a single-threaded focus barrel where the outer portion of the lens is stripped down, exposing the inner barrel that holds these lenses in place. This is so that they can be properly adjusted ahead of the entire metal casing being screwed in place, securing the entire housing.
Securing the housing not only minimizes the failure points by preventing the lens from adjusting but also minimizes both the cost and size of the element. As it is set to a fixed position, it no longer needs moving elements and can instead utilize a single metal chassis.
This single metal chassis reduces the metal used to help lessen the lens’s footprint, meaning it is lighter and is less likely to move under its own weight. Additionally, these lenses are naturally smaller, as a result, allowing them to be fit into smaller compartments in larger inspection devices, like in automobile manufacturing operations.
Ingress Protected
Ingress protection is simply what it sounds like: Designing the lens so that no substances like water or dust can enter it.
In a traditional lens mechanism, there are areas where components are threaded and slotted together, forming seams and access for particulates to enter the lens. Now, utilizing O-rings, the mechanics of these lenses can be streamlined to remove those entrances by placing rings between the house, front lens elements, and front lens retainer.
This ruggedization also involves a hydrophobic window on the front of the lens that prevents water from sticking to it with an antireflection coating. These are particularly useful for environments that require frequent washing and sanitation or are used outdoors on vehicles.
Stability Ruggedization
Stability ruggedization is a step above industrial, as it utilizes clamps and adhesives to firmly lock all components in place with constant points of contact.
Standard lenses will be assembled by having individual parts dropped into place using spacer lenses that are glued, meaning the lenses themselves can shift and move about. We ruggedize the lenses by applying adhesive to all of the individual parts, compacting the apparatus and making it a tighter, more durable item.
Stability ruggedization is designed for environments in which shocks can be experienced, as a standard lens can shift from one to 20 microns after experiencing a 50-gram shock. Stability ruggedized lenses are also useful in situations where the lens itself might be moving, such as tracking an object’s position or using robotics.
Conclusion
We’ve examined the fundamental need for lighting in a machine vision operation, how sensors are improving data optimization, and why ruggedized lenses might be the future for the industry. These complicated factors make assembling a machine vision pipeline a daunting proposition and an exciting one, as emerging technologies continue to help improve various elements of the process.
In doing your own research, be sure to examine results from the best part makers. Don’t overlook elements that seem simplistic or unnecessary; these small optimizations can bring a lot of long-term value. Visit Navitar for more information.
Sources:
Quality Mag - Simplify Deep Learning Systems with Optimized Machine Vision Lighting | Qualitymag.com
IDTechEx - Image Sensor Tech: Machine Vision Expansion Accelerates Adoption | IDTechEx.com
Quality Mag - Ruggedization of Machine Vision Lenses | Qualitymag.com