• Font Size:
  • A
  • A
  • A

Feature Articles

What’s Happening with Machine Vision in the Food Industry?

by Nello Zuech, Contributing Editor - AIA

As machine vision technology has become more rigorous it has become more successful in addressing applications in the food industry. While there are many applications of machine vision in the packaging side of the food industry, this article reviews activities at manufacturers of food products, fresh-pack and food processor.

Typical applications include sorting and grading. With advances in color cameras and the underlying ability of microprocessors to handle the additional data derived from color-based processing, more applications are being addressed. In some cases multispectral processing is now possible at the speeds required to keep up with processing tons of a product per hour.

Sorting applications are generally based on geometric property analysis: size and shape. Grading applications generally entail separating based on photometric properties: appearance, color, surface properties and in some cases internal properties. Today one can find machine vision-based systems used in the food industry addressing virtually every known farm-grown product. In addition one finds systems at value-adding food facilities like bakeries and other manufacturers of food products. In this case, process control is the goal to avoid the production of products whose appearance is inconsistent with the norm.

A questionnaire was developed and forwarded to virtually every known supplier of catalog machine vision systems for the food industry. As it turns out, three companies responded and they represent each of the generic classes of applications: fresh-pack, food processor and bakery.

Paul Pearl, President, Dipix Technologies – Manufacturer of food products including baked goods
George Dodge, VP Marketing, ESS – Fresh-Pack
Richard Hebel, Director of Marketing, Key Technology – Food Processors

Given the distinct perspectives, what follows is the response of each of the contributors to the specific questions.

Paul Pearl, from the perspective of the manufacturer of food products:

1. What are some specific applications in the food industry that your company addresses with machine vision technology?
High speed, on-line inspection of manufactured food products such as baked goods, biscuits, crackers, confectionary, meat patties, tortillas, pizza, chicken pieces, fish sticks, etc. for appearance, size, color, shape, etc.

2. What are critical machine vision system performance criteria?
Cost, maintainability, ease of operation and setup, ability to detect blemishes that are easily recognized by a human inspector.

3. What changes have been taking place in the technologies that are the basis of the machine vision systems used in the food industry that have resulted in improved performance?
Better illumination techniques, higher speed PCs capable of doing image processing without special purpose hardware.

4. Where do you see breakthroughs coming in the technologies that are the basis of machine vision systems used in the food industry that will result in further improvements in the near future - next three years?
Cost reduction in processing (still the highest cost component today). Also better illumination techniques.

5. Are there market changes in the food industry that are driving the adoption of machine vision?
Yes. Automation, labor reduction, productivity improvement.

6. How will machine vision systems have to change to meet emerging applications in the food industry?
Lower cost.

7.  What are your thoughts on the future of machine vision in the food industry?
Very large opportunity.  High barrier to entry because customers have little understanding of their real requirements.

8.  What advice would you give to a company investigating the purchase of a machine vision system for a food industry application?
Buy from an experienced supplier (like Dipix).  Avoid system integrators unless you are prepared to shoulder significant risk. Remember that most problems can be solved with machine vision but few are worth solving.

George Dodge, from the perspective of the fresh-pack market:

1. What are some specific applications in the food industry that your company addresses with machine vision technology?
Color sorting of fresh process tomatoes in field and in plant.  Initial quality sort of fresh citrus and juice fruit elimination.  Color sorting of various peppers.  Quality grading of nuts, in shell and meats.  Color grading of various fresh vegetables for processing.

2. What are critical machine vision system performance criteria?
Resolution, color discrimination, reject accuracy, and uptime.

3. What changes have been taking place in the technologies that are the basis of the machine vision systems used in the food industry that have resulted in improved performance?
Improved lighting sources, computing headroom, improved reject solenoids, improved camera and laser technology.

4. Where do you see breakthroughs coming in the technologies that are the basis of machine vision systems used in the food industry that will result in further improvements in the near future – next three years?
Short-term most effective breakthroughs look to be software, and camera driven.  Reduction in technology costs seem to provide largest short term gains.

5. Are there market changes in the food industry that are driving the adoption of machine vision?
Continued increase in labor costs, higher yielding crops, shorter harvest windows, and increased exposure to liability, workman's comp costs continue to see double digit increases.

6. How will machine vision systems have to change to meet emerging applications in the food industry?
Multi purpose applications, with much more adaptive and user-friendly operations.

7. What are your thoughts on the future of machine vision in the food industry?
Larger food industry applications are fairly mature although niche applications continue to flourish.  As improved quality continues to drive competition so does the need for more accurate and higher resolution sorting devices.

8. What advice would you give to a company investigating the purchase of a machine vision system for a food industry application?
Investigate your options thoroughly, speak with other users within your industry, and be prepared to look at a 'total' solution, including specific material handling requirements, etc.  Involve your operations staff in evaluations of specific technologies for ease of operation and user interface.

Richard Hebel, from the perspective of the food processor market:

1. What are some specific applications in the food industry that your company addresses with machine vision technology?
Performing high-resolution full color and IR sorting of bulk particulate product in excess of 20 tons per hour.

Providing extremely high-quality vision-based quality monitoring for process data acquisition and control purposes, for:

  • Fruits & vegetables (fresh and fresh-cut, frozen, dehydrated, canned) – whole & diminuted
  • Potato products (strips/french fries, chips, slice/dice frozen/dehydrated/byproduct)
  • Nuts (tree nuts and peanuts)
  • Coffee (primarily green beans)
  • Grain & seed
  • Snacks (corn chips, potato chips, pretzels)
  • Confections
  • Food ingredients

2. What are critical machine vision system performance criteria? 
There are three fundamental, interdependent criteria that define performance in any sort:
1. Defect Removal Efficiency: The percentage of the incoming 'defective product' that the sorting system removes from the process stream.  This is the measure of the system's ability to remove product that does not meet the specifications for acceptable product.
2. Recovery Efficiency: The percentage of the incoming 'acceptable' product that leave the system in the 'pass' stream.  This is the metric that quantifies the 'false accept/false reject' performance.
3. Throughput: The maximum amount - typically measured in units of weight (lbs., kg, ton, etc.) - of product that the system can sort without degrading specified performance in defect removal efficiency and recovery efficiency.

None of these have meaning unless the others are stated.  A statement of all three is definitive for any sort, although more complex statements are typically required since most product exhibits more than one defect type and/or severity.  Note also that the term 'defect' is used from habit:  in practice, it is sometimes more efficient to sort for 'acceptable' product.  The metrics stay the same; sorter operation just changes.  In some applications, this is done to effect a 'fail-safe' sort.

Of course, what constitutes a 'defect' or feature of interest drives other, usually multiple feature vectors as machine vision descriptors.  For instance, a given defect may be differentiated from 'acceptable' product and noise on the basis of color, shape, and dimensional measurement.  With successive generations of machine vision technology, the dynamic range of Key's sorter feature extraction has dramatically expanded.  In the end, though, it all boils down to the three metrics above.  Those three metrics are also what define the economics of the sort, and by extension, the value of the sorter and the financial justification for its purchase or lease.

There are some other criteria that are somewhat more removed from the machine vision sort, but are nonetheless important:

  • Availability:  This is the percentage of total time that the customer wishes to use (realize financial benefit) the system that the system is able to operate at the expected performance as specified by the three metrics above.  100% availability is the ideal.  In practice, failures - whether system-based or operator-induced - and scheduled maintenance make availability somewhat less that 100%.  In the customer’s eyes, however, he's invested in a major capital asset, so to realize maximum return on that invested capital, he wants availability to be 100%.  This isn't trivial:  in most sorter applications nowadays, the line cannot run without the sorter in operation 'at spec'.  Key places much emphasis on system design and operating practices that maximizes availability.
  • Mean-time-to-repair of 'MTTR':  One instance of loss of availability is a failure in the sorting system.  For years it has been unacceptable to have a sorter languish in an inoperative state because of some failure internal to the sorter.  Fortunately, with successive generations of Key's sorters, and a combination of increasingly powerful but inexpensive processors and high-bandwidth connectivity, sorter 'self-awareness' is now the norm.  This awareness drives benefit in two ways: 1) Key's latest sorter technology can anticipate certain failure states and alert the operator, thus allowing repair at a scheduled outage rather than suffering through an unscheduled outage; and 2) Because of this awareness, Key's sorters allow the customer's operator or maintenance people to rapidly identify the failure syndrome and drill down to a failed component in a minimum amount of time.  The connectivity adds another benefit:  In complex troubleshooting situations, the customer can elect to have Key technical personnel access their Key sorter via the Internet from locations around the world.
  • Livability.  Here, there is a constellation of factors that define what the customer experience is beyond the performance criteria mentioned above.  One example is how robust the system is within the operating environment.  Many of Key's sorters must function in elevated temperature and humidity, hot caustic high-pressure washdown environments typical in the sanitary conditions of food processing facilities.  So, Key engineers not only need to design sorters to meet performance requirements, they must also meet those requirements in these hostile environments.  At the same time, while a sorter is often the most technologically sophisticated piece of equipment in a food processing line, it might well be run by an untrained or unskilled operator.  It doesn't matter if the sorter is capable of meeting all of the performance criteria unless the operator can easily operate the sorter at its optimum.

3. What changes have been taking place in the technologies that are the basis of the machine vision systems used in the food industry that have resulted in improved performance?
In sorting these break down into three main areas:
1. Improved sensors, generally camera or laser-detector, with greater dynamic range, much higher speeds (and resolution), and standardized connectivity.
2. Dramatic increases in image processing power vs. cost, and the ability to deploy that muscle in relatively 'plastic' embodiments (FPGAs) with high-bandwidth interconnectedness.
3. Fundamental advances in material handling technology that move increasingly larger amounts of product through the points of inspection, with optimum product presentation so that the sorter can take full advantage of the advances in sensor and processing capability.  In fact, the 'gating' factor in ALL sorter performance nowadays is the material handling system. 

This is because while there are fundamentals of sensor and processor technology development that drive multiples of performance on relatively short time frames - physics is still physics, or 'physics is cruel' - when it comes to moving, orienting, and stabilizing 20,000 lbs or more an hour of heavy, delicate, wet and slippery, product through the point of inspection, and then ensuring positive and efficient removal of individual defects from that stream.  So while Key has a heavy investment in sensor and processor R&D, Key also capitalizes on its 50-year strength in material handling.

4. Where do you see breakthroughs coming in the technologies that are the basis of machine vision systems used in the food industry that will result in further improvements in the near future – next three years?
Traditional sensor technologies such as cameras and lasers will continue to improve, albeit incrementally.  The key to effective sorting is to develop technology that can extract features that are extremely sensitive to differences between acceptable and defective product.  Early on, this was simple gray-scale to binary pixel counting.  Then color was added.  Then size and shape.  Then dimensional measurement.  So we expect that the advances will include additional feature vectors that may include 3D/volumetric imaging, densitometric imaging, and tactile/textural detection.

The cost for processing bandwidth will continue to decline, and that means pushing increasingly away from dedicated-purpose monolithic vision 'engines' and more toward highly flexible 'libraries' of modular processors that can be mixed and matched - tuned - to deliver required performance at minimum cost.  That is, the quanta of sorting capability will become increasingly fine, and this will allow Key to continue to optimize performance per application in a cost-effective way.

5. Are there market changes in the food industry that are driving the adoption of machine vision?
Most of those changes have been apparent for a number of years.  These lines just cannot run without this technology.  In the course of your daily repasts, most of what is on your plate has been through systems like Key's.

Certainly globalization continues to heat up the market: as food markets go global, the standard for competitive quality is buoyed by sorting technology such as Key's.  That's bodes well for Key as megamarkets begin to open in China, the Indian subcontinent, and Latin America.

6. How will machine vision systems have to change to meet emerging applications in the food industry?
Key sees this breaking down into the following:

  • Improved performance per unit cost, moving sorters from 'big-ticket' deployment constraints to economical tools that can be deployed in lower return - but still important/useful - deployments.
  • Dramatically reduced 'total installed cost'.  In most sorting applications, the collateral installation costs are an appreciable fraction of the cost of the sorter.
  • Even easier operation.  This is especially true as deployments in emerging geomarkets grow.
  • More ways to use the information the sorter extracts from product, and thus, more ways for the customer to realize return on their technology investment.  Traditionally, this has been limited mostly to simple two way 'accept/reject' operations, but this is a waste.  Key is investing heavily to develop OPC accommodations to allow external automation and data systems access to the universe of inspection data in Key's mainframe sorting systems.  At the same time Key is taking advantage of their expertise in material handling to provide systems for multi-class sort outputs.
  • Inherent migration support.  The cost to replace whole sorters due to obsolescence - either in component part availability or performance - is onerous to customers, yet any vision company designing with state-of-the-art technology knows component obsolescence is a real problem.  The collateral investments the customer has made in ancillary equipment and structure, operations and maintenance protocol and training, and spare parts inventories are significant, and Key feels it is an inherent part of good engineering design to provide Key's existing and legacy sorter customers with migration paths that let them continue to realize returns on their sorting investments through migration and upgrade paths.  Key is the only sorter manufacturer that will be providing migration paths for essentially all their existing sorting platforms.

7. What are your thoughts on the future of machine vision in the food industry?
The opportunities are huge, but access to those opportunities depends as much on deployment, form factor, and process economics as it does on advances in machine vision. 

Key wrestles with all of these.  Lot's of tough challenges and unknowns, but I think there is at least one very clear trend:  the cost per point of application will continue to fall.  And with that, we see substantial volume increases.  At the same time, we anticipate that continuing leaps in processing power will enable Key to bring statistical and massively parallel image processing techniques to an application space that has been traditionally compute-bound when faced with the tremendous object loads a sorter typically encounters.

8. What advice would you give to a company investigating the purchase of a machine vision system for a food industry application?
Look at the track record of the vendor, especially in satisfying other customers in your space: that speaks louder and clearer than any glossy brochure or fast-talking sales rep.  Make your vendor stand behind their performance claims - every one of them - in writing.  Ask your vendor to prove to you that they aren't going to forget you the day they get your order. 

What have they done to ensure your continued support?  What have they done to obviate or mitigate concerns about system obsolescence? And when it comes to price, certainly get a clear quote from them, but most importantly put together a 'total installed cost estimate' and use that as the basis for cost comparison to other vendor's total installed cost.  That make's sure you are comparing 'apples to apples.'

 

 

 

Comments:

There are currently no comments for this article.


Leave a Comment:

All fields are required, but only your name and comment will be visible (email addresses are kept confidential). Comments are moderated and will not appear immediately. Please no link dropping, no keywords or domains as names; do not spam, and please do not advertise.

First Name: *
Last Name: *
Your Email: *
Your Comment:
Please check the box below and respond as instructed.

Search AIA:


Browse by Products:


Browse by Company Type: