Q&A: Selecting Lenses for Machine Vision Systems
V&S: What are the common mistakes or misconceptions when designing a machine vision system?
NJ: One of the most common mistakes is that a customer will choose a camera and a sensor before choosing a lens. This could lead to a situation where there isn’t an affordable lens available. One of the key things is to select them at the same time.
V&S: Why does this happen?
NJ: The lens is often viewed as an accessory, and the camera is considered the primary component. In the days of VGA resolution, it was more acceptable to assume that just about any lens would maximize the sensor resolution. As we’ve moved to higher resolutions, performance has become lens-limited as opposed to sensor- (pixel-) limited. Though a customer could purchase a five megapixel camera, they might achieve less than two megapixels with a poor lens. Most customers do not want to buy a camera and not get its full resolution potential. It’s critically important when you have higher-than-VGA resolution to decide on a lens at the same time as the camera. This became more important over the last 5 years with the increased availability of lower cost, high resolution sensors. If you use the highest-end sensors on the market, it has become crucial to consider lenses in tandem with the camera.
From the standpoint of resolution, lenses in machine vision systems have progressed, helping assembly lines achieve a much higher level of accuracy and throughput. When using a wide field of view, operators or algorithms might not be able to see the smallest details that may be required for the system to achieve 100% accuracy. But machine vision systems with the correct components are able to detect smaller defects and details on parts with greater accuracy.
Accuracy is certainly important, especially in the pharmaceutical industry. A machine must read labels on packaging and barcodes on actual bottles correctly. It is an example of a regimented supply chain where mistakes are not acceptable. Anything less than zero-defect precision can have deadly consequences.
V&S: What is the machine vision lens selection process like?
NJ: The process involves identifying customer needs and making sure they understand key specifications. Once the four key specifications are known, we can suggest a lens and camera. They are:
Desired Field of View
Resolution, or size of the smallest part or detail to see and identify
Working Distance, or the range that can potentially be used
Depth of Field, or height of objects under inspection (It is important to understand the resolution throughout the needed depth for “tall” parts such as circuit boards or bottles.)
V&S: What trends are you seeing in lenses?
NJ: Lens design in machine vision continues to evolve. Some of the things we’re seeing a need for are higher resolution in both the smaller and larger sensor formats. Many traditional vision applications have stayed with cameras that use the 2/3-inch format and smaller, but we are starting to see the one-inch sensor format become increasingly popular. We have also seen customers moving outside the visible spectra into the short-wave infrared (SWIR) spectra for optics. Sensors for SWIR are becoming more readily available, and more people are beginning to integrate them.
V&S: Is it difficult to find the right fit for each customer?
NJ: Usually there are many options that could potentially work. The most difficult part is educating customers about the tradeoffs between field of view and the smallest size that they need.We let them know what’s possible and when compromises may have to be made. It’s a balancing act. The larger the field, the more difficult it will be to see the small details they want to detect.
Obviously, everyone wants the highest resolution at the lowest price. Usually, after we get the customer’s desired specifications and understand what they are able to settle for, we can offer “good”, “better”, and “best scenario” options. The best system hits everything they want and maybe offers more. The others offer cost-effective options that may have compromises to keep costs low. We usually like to give customers as many options as possible to show them how different tradeoffs affect the final system so that they can make an educated decision.
V&S: What does the future of machine vision look like?
NJ: I would definitely say that things are slowly progressing to larger formats for certain applications so more detail can be observed with one system. Ideally someone doesn’t want to have five camera systems side-by-side to get what they need. They would like to have one camera, one lens and go. Moving to larger sensor formats is certainly a relatively slow process, taking many years. For example, with the newer Sony sensors, there aren’t as many lens options for one-inch sensor formats as there are for the more traditional smaller-than-one-inch formats.
In terms of new lenses, some companies have lenses that match the sensor format but not the resolution. We’ll see more in the next six to 18 months that match sensor format and resolution.
In speaking about technology, the other area that vision systems are really pushing into is the SWIR region. We’ve seen the advent of more InGaAs cameras that have sensitivities from ~0.7 to 2.0 microns and are available in both line-scan and area-scan formats. This region is excellent to use, as many materials are transparent in this waveband, and certain defects can be seen much more clearly. Many lens manufacturers simply put a new coating on their vision optics to use in the SWIR, though several companies, including Edmund Optics, optimize lenses for this waveband from the initial design standpoint.
From a broad application standpoint, we’re certainly seeing more demand for lenses that integrate into benchtop applications, especially biomedical and life sciences instrumentation such as blood analyzers and flow cytometers. As systems continue to have smaller form factors and production volume of these products increases, it is important to select lenses that maximize your cost to performance ratio. V&S
Nicholas James is a product line manager at Edmund Optics (Barrington, NJ). For more information, visit www.edmundoptics.com/imaging .
Tech TipsA common machine vision mistake is choosing a camera and sensor well before choosing a lens.
This could lead to someone picking a sensor where there isn’t an affordable lens available.
Instead, try to pick the camera, sensor and lens at the same time.