Ned Lecky is owner of Lecky Integration, Albany, NY, USA; www.lecky.com, which was established in 2007. He has held numerous positions in the machine-vision and electronics industries and has taught electrical and computer engineering.
VSD: Please describe your company and its services. What is your personal background in the machine-vision industry?
Lecky: I started my career in machine vision after graduating from college in 1984. I was a systems engineer for Control Automation (CA) in Princeton, NJ, where I was lead vision-system programmer and learned a great deal about custom frame grabbers, cameras, lighting/ optics, and motion control for printed-circuit- board applications. I started Intelec Corp. soon after and continued integrating systems for CA, Cognex, and Intelledex. I got the idea to build a better mousetrap in 1992 and wrote Sherlock, the first industrially focused machine-vision software application for Windows. Intelec became part of Imaging Technology, which then became Coreco Imaging, which then became DALSA. The much evolved and improved Sherlock is still available today. Lecky Integration is a small company that focuses on application-specific integrated solutions involving machine vision, fuel cells, robotics, or custom electronics. As a small business, I seek out vendors, integrators, machine builders, and software professionals to form multi-disciplinary teams to solve complete problems.
VSD: What technologies and components do you use for your applications? How often do you evaluate competing technologies?
Lecky: The distributors or integrators with whom I work often specify cameras and hardware accessories. Many sales leads come through these channels, and it is best business practice to use the hardware and software products that these partners are already representing and supporting. Fortunately, vision hardware and software components are much more inherently compatible than they were even five years ago; this is the blessing and a curse of open system standards. It is a blessing for the consumers of machine-vision technology but can seem a curse to the suppliers who may do a great deal of work to convince a customer that vision is a good solution for their problem, only to have another lower-cost provider come in and sell a competing component for follow-on systems. Ultimately, this is good for the component vendors, since it helps them understand the market, see where costs can be cut, and recognize their own true advantages. However, this “advantage” usually causes bump after bump along the road to knowledge and can be quite challenging.
VSD: How do you approach a new application? Do you work with OEMs or other system integrators?
Lecky: Know your customer. Know your customer’s boss. Understand the customer’s business, both financial and political. Know your customer’s customers and maybe some of their vendors, too. Solving the technical aspects of the application is usually much easier than these first parts, but a good technical solution that flies in the face of business needs is a huge failure. I’ve been there, and it is not pretty, fun, or profitable. All of our solutions involve teams and partners. Teams take time to build and maintain, and a certain trust develops as applications are solved in a professional way that protects and nurtures each team member’s business interests. I have worked with component vendors, distributors, vision-system suppliers, OEMs, R&D organizations, and technical/ nontechnical end users extensively and interchangeably since 1985.
VSD: Recent software developments in image processing include pattern recognition, tracking, and 3-D modeling. What algorithms and specific software developments do you see emerging in the next five years?
Lecky: I think we’re totally stuck, actually. I worked on bio-inspired cognitive computing hardware strategies while getting my M.S. and Ph.D. in electrical engineering. I have long felt that the key to the next advance in machine-vision algorithms is to abandon algorithms altogether and to start, instead, emulating the human image-processing system. The dissatisfaction I have with machinevision systems failing to recognize occluded, fuzzy, out-of-expected-location patterns in variable lighting conditions is not really much less today than it was 20 years ago. Try giving a capabilities demo of a state-of-the-art vision system to someone who has never seen machine vision before and you’ll know what I mean.
Ten or so years ago, I would carefully hand code new algorithms in MMX assembly language to get optimal speed (on 266-MHz Pentiums!). Now, image-processing libraries are comprehensive and effectively free and offer optimized code that can run on multigigahertz, multicore processors. If anything, the hardware is now too fast and too powerful to be fully engaged.
We must find more cognitive and truly intelligent architectures for both hardware and software before major advances can be made. Admittedly, the state of the art is excellent in pattern recognition, tracking, and three-dimensional modeling. However, the accuracy and overall image-processing capabilities of a housefly dwarf those of our current systems, all just using some fraction of a housefly’s million-neuron brain. I find this frustrating.
VSD: How do you design your systems for OEM product obsolescence?
Lecky:: Optics- and lighting-system product lines are quite stable, and same-or-better replacements are generally available when a vendor discontinues a product. Many of the specialty products are built to order anyway, and vendors are often pleased to build an “old” version of a lighting system or a lens. The camera market is extremely dynamic, but again, the new models tend to be same or better for lower cost, so updating to a different camera is often not very painful.
The software, of course, is the issue, especially when the system includes volumes of custom code written to integrate the standard system with a real factory-floor monitoring system. Since 1995, I have focused on Windows- based software using C++ for algorithms and time-critical functions and Visual Basic for operator interfaces and front-ends. This architecture has proven resilient, since this more-than-ten-year-old code will still compile and run on modern computers by taking advantage of the Windows development tools.
VSD: In which areas do you see the most growth? What are users demanding from you in the design of new systems?
Lecky: In North America, we know that the bulk of the machine-vision industry is in application- specific machine-vision (ASMV) systems, or turnkey solutions that sell for about $125K, on average. This ASMV market is projected at $1.2 billion for 2008, while component sales (cameras, lenses, lighting, and so forth) are projected at less than $200 million. So solving the customer’s problem with a complete ASMV is still where the bulk of the market is. In new designs, users continue to demand systems that are more tolerant of variation in lighting, product, or operator. And we continue to design systems that attempt to minimize variation in lighting and operators, since I still have never seen a reliable machine-vision system that permitted variation in either. Lower cost is always an issue, but the cost has come down so far from the good old $20K vision-system days that it is less important in most applications. Size and power consumption are rarely an issue for our clients.
VSD: How will OEM components targeted toward machine-vision applications have to change to meet future needs?
Lecky: As the cost and price of machine-vision components continues to tumble, the vendors of these components cannot afford to spend as much time training, supporting, or assisting their customers. This means that the components must be more reliable, self-calibrating, and self-training, and effectively bullet-proof. You see such product components coming out of most of the machine-vision software and system companies already. Unfortunately, one of the best ways to make a component more reliable is to reduce its feature set, which then renders the component less attractive to the broader customer base. So there is a continual rebalancing of feature set vs. ease of use that I think the component and system vendors are really struggling with right now.
VSD: Do you work outside North America? How do you think that the machine-vision market differs in different regions such as China?
Lecky: I’ve done machine-vision work in France, Germany, Ireland, Singapore, and China. China, of course, stands out due to the size and fluidity of its market in the present era. I was astounded by its competency in steel making, railroad building, and railway freight-car building.
The application I completed involved high-accuracy gauging of railway axles after grinding and prior to custom boring a pair of wheels to press onto them. This application required many large machines from several vendors, a football field full of material-handling equipment, and dozens of PLCs communicating with four PCs performing orchestration, control, and database management (see photo).
The local engineers were very adept at designing, building, and programming, or at least customizing, all of the equipment. I must say that the first time I saw ladder logic in Chinese characters I was a bit more overwhelmed than I usually am, even by ladder logic. The technical rank-and-file in China can do engineering and programming at the “Western” level, if not better—something that still surprises some of my colleagues. The Chinese people were also very kind, hospitable, playful, and patient with me. In North America, we often thought of machine vision as a labor-saving technology. In China, while labor costs are on the rise, they are still (and perhaps forever) much lower than those in the USA. However, repetitive tasks such as gauging, inspection, and verification are still unappealing jobs that people are simply not very good at, and so, like many others, I see a bright future for low-cost machine-vision components in China.
VSD: What new markets and technologies do you see for your company in the future?
Lecky: I believe in being very flexible and helping my clients solve their problems in ways that are acceptable and practical. I’m not going to use an FPGA-based solution for a customer who wants solution update control but doesn’t know Verilog or VHDL--it just wouldn’t be prudent.
I think the key to my staying power in the industry over the years has been a willingness to understand the customer’s business issues and corporate structure and to use these components just as much as the technical ones in tailoring solutions to their problems. Many of the professionals in our industry will be nodding their heads as they read this, I bet. So, at least in that way, we will continue to follow our customers, not lead them.