• Font Size:
  • A
  • A
  • A

Feature Articles

Machine vision Application Analysis and Implementation -- Part 2

by Nello Zuech, Contributing Editor - AIA



This is the second of a series of articles designed to provide the framework for a successful machine vision system installation. The process described is targeted at companies that are planning the adoption of a machine vision system for the first time or that have a unique application that no one has previously attempted to implement.

As observed in Part 1 , today one can find many application-specific machine vision systems for somewhat generic applications in many manufacturing industries. Purchasing these 'off-the-shelf' solutions poses little risk to any first time buyer. In some cases, one can find application-specific software developed by suppliers of general-purpose machine vision systems, imaging board/frame grabber suppliers, software suppliers or merchant system integrators. While these are not turnkey packages, the vision experience is itself less risky. Examples of these packages include: alignment, OCR/OCV, LCD/LED inspection, BGA inspection, etc.

Even less risky are the turnkey machine vision systems that are industry specific; e.g. bareboard or assembled board inspection for the electronic industry, overlay registration/critical dimension inspection for the semiconductor industry, various industry specific web scanners, systems targeted at food sorting, etc. Virtually every manufacturing industry has these systems, many of which can be identified through the resources of the Automated Imaging Association.

Where these 'solutions' are not the answer and a machine vision application has been identified, success requires proceeding systematically and not treating the purchase as if one is purchasing a commodity item. It is not sufficient to send good and bad parts to various vendors and ask them if they can tell the difference.

Table 1 - Application Analysis and Implementation

Systematic planning
Know your company
Develop a team
Develop a project profile
Develop a specification
Get information on machine vision
Determine project responsibility
Write a project plan
Issue request for proposal
Conduct vendor conference
Evaluate proposals
Conduct vendor site visits
Issue purchase order
Monitor project progress
Conduct systematic buy-off
The above table depicts the process that should be used as one proceeds with the deployment of a machine vision system that is uniquely defined for a company. In this article we are going to cover the 'Specification Development' topic. In future articles we will cover the balance of the topics in the table.
After assessing corporate objectives, establishing a project team and establishing the profile and the objectives for the project, the next step is to develop a set of detailed functional specifications for the application.  Specifications are technical data that describe all the necessary functional characteristics that the system must have to perform the required job.  They should not state how the vendor is to do the job.  Specifications must state the required productivity and capability of the system as well as all significant operational requirements.  They should also explain the requirements of the related fixtures, material handling, and so on.  Similarly, requirements for machine control and line interfaces should be detailed as well as plant requirements and limitations.
The importance of specifications cannot be overstressed for the time and effort put into proper specifications will be more than repaid in terms of reduced start-up time and a reduction in maintenance and quality problems. The biggest advantage of this up front work will be the avoidance of disappointment and a higher probability of achieving a successful machine vision project.
By analyzing the present methods, observing the present operation, and getting input from all those involved, a detailed description of the operation and a review of all anticipated variables can be developed.  For example, it may appear that an operator is only performing a sorting function, separating containers by their size or shape.  On closer scrutiny, one will also observe that once in a while the operator throws a container into a scrap bin.  While probably not classified as an inspector, by virtue of innate intelligence, the operator knows enough to separate incomplete or misshapen containers. Mechanizing his role would require both appropriate material handling and 'automated eyes' or machine vision to assure both proper sort and elimination of rejects.
By virtue of increased sensitivity to exceptions, an operator can become more sensitive to specific conditions.  For example, because training has emphasized the importance of separating all green containers, the person has a heightened awareness when such objects pass.  Similarly, a machine vision system might have to somehow include a weighting factor (e.g., require color processing  capability or the use of a complementary color detector) that will increase sensitivity to a specific factor - color in this example.
The converse is also true.  An operator may be trained to be tolerant of color shade variations: For example, all yellows, regardless if pale gold or virtually orange, might be acceptable since it is the basic color itself that provides the distinction.  In this case, therefore, the machine vision system must be equally tolerant of these normal variations while also maintaining sensitivity to the fundamental defects the system is supposed to capture. A machine vision system intolerant of gray shade changes stemming from acceptable color variations would lead to an unacceptable number of false rejects.
Developing the specifications includes a quantification of existing procedures in terms of productivity and quality: number of shifts, scrap, warranty repairs, machine downtime, and so on.
Examining operations:
When examining the operations, the following should be considered:

Will the system be for a new line or an old line?
Does the application involve: One object at a time? How many different objects?
          Multiple objects? What are the different part numbers?
Is it a batch operation? Or a continuous dedicated process/line?
What are the changeover times and frequency of changeovers?
Is the inspection to be online? Off-line?
Must every produced item be inspected, or is random sampling acceptable?
Will new part numbers or variations be added to the system at a later date?
Are product design or production process changes anticipated?
Where do parts come from?
              How are goods routed for the vision inspection task presently?
              Does the operation require automating loading/unloading?
              Is the operation inventoried?
              Are products stored in bins?  Magazines?  And so on?
Can rejected parts be repaired?
Can vision assist in the diagnosis?
Where do pass and fail parts go?
When is the machine vision system needed by?
How many lines/machines will the vision system be needed for?
Can a representative sample of parts be provided to system vendors or integrators for evaluation?
Can drawings be provided?
Can video of line be provided?
Can vision suppliers observe production at your facility?
If a system is in place today to perform the function, answer the following:
(a) What technology is used?
(b) What are the system's capabilities and limitations?
(c) What actual performance is being experienced?
(d) What problems are being experienced that hamper productivity and effectiveness?
(e) What is the utilization factor?
(f) If starting today, what would be done differently?
(g) What problems were experienced and how were they overcome?
What impact did the project have on the user organization?  Was it measurable?  If not, why not?
Is the impact more or less than target levels?  Why?
Does upper management perceive improvements attributable to the project?
What is currently perceived as the system's greatest contribution or benefit?  Is it the intended one?  If not, why not? Was the budget maintained?  The time schedule?  If not, why not?

Application issues:
What are the actual inspection functions that must be performed by the system?
What distinguishes a bad part from a good part?
Is it gauged by 'eyeball' or with instruments (micrometer)?
If application involves gaging -
           What are the tightest tolerances?
           On what specific dimension?
           What is the design goal for system accuracy?
           What is the design goal for system repeatability?
           Are there features that serve as references?
           Are there specific calibration requirements?
Is cosmetic inspection done by detail or is it performed by a cursory look for more gross appearance differences?
            Describe flaw types
           What is the smallest flaw?
            Does the flaw affect surface geometry?
            Does the flaw affect surface reflectivity?
            Is it more of a stain?
            Is classification of flaws required?
Does the inspection first require identification?  If so, is it by shape? By size? By color? By reading characters?  And so on?
           What are differences?
                   Is there a background pattern?
                   How many different patterns or shapes are there?
Is the inspection itself one of verifying shape conformance?  Again, with or without instruments?
What is the basis of identification or shape conformance?
Is the application specifically OCR? OCV?
                  (a) Reading or verification
                  (b) Inked, painted; molded, cast; stamped; raised; laser-etched; dot matrix; acid-etched; engraved; recessed; other
                  (c) Describe single-font style and type and multiple-font style and types for each font style:
                            (1) Stroke width
                            (2) Aspect ratio
                            (3) Character height
                            (4) Character width
                            (5) Character depth
                            (6) Center-to-center spacing
                            (7) Space between characters
                            (8) Characters per string and per line
                            (9) Numbers of strings and lines
                  (d) Read rate (characters per second)
                  (e) Character positioning, repeatability
                            (1) Vertically
                            (2) Horizontally
                            (3) Skew
                  (f) If 'no read,' what to do?
                  (g) Percentage of misreads allowed per character string
Is the operation one of just verifying that an assembly is complete before another operation is performed or that objects are oriented properly?
What are the dimensions of the assembly and associated parts/subassemblies?
Does application involve presence/absence and/or orientation verification as well?
What is the smallest piece to be verified? What are the dimensions of that piece?
What is the largest piece to be verified? What are the dimensions of that piece?
Is part correctness verification also required?
If, in fact, it is a complete assembly operation that is performed, does the operator locate parts and guide them into place?
In the case of location analysis
           What is the design goal for accuracy? Repeatability?
           What is the area over which 'find' is required?  
           Will calibration be required?
Is this operation a combination of several of the preceding tasks: location guidance, cosmetic examination, and gaging, for example? 
Does the operator perform tasks other than vision (e.g., assembly and machine loading)?

Part discussion:
Are multiple views of the part required?

Can the details to be observed be seen in the silhouette of the object?

Is there contrast in what must be observed, that is, can you visually see the condition without picking up the part to manipulate it in the light?  How small is the smallest detail you want to see?  How big is the object (field of view)?  As in photography one can see both large and small objects with television cameras, but the detail that can be detected reliably is proportional to the field of view.  Where necessary, machine vision systems can employ more than one camera so detail versus field of view need not be a factor that would preclude considering machine vision.

Does the operator perform three-dimensional analyses?  Is the decision based on color interpretation?

Are there normal variations in the appearance of the object that are ignored? A vision system will somehow have to normalize those conditions.  Are there variations in the appearance of the background? 
What is the material (steel, plastic, etc.)?
What is the finish (texture) like? Is the surface finish the same on all faces? Is the surface finish the same for all part numbers and/or production runs? What are the differences?
Is part plated? Coated? Are there any thin films (oil, mist, etc.)? Is part painted? Dull? Glossy? Specular?
Is part highly reflective (mirror like)? Poorly reflective? Matte?
Will the reflectivity of the part change from part to part? Over time?
Are there any machining marks on the part?
Does the part generally have scratches, nicks, burrs, dents, flashing, etc.?
Is there any porosity on any of the parts, part surfaces?
What are the object's shapes? Flat? Curved? Gently curved? Irregular? Grooved surface? Sharp radii (prismatic)? Mixed geometric properties?
What is the temperature of the part at inspection point?
What is the size of the part? Smallest? Largest?
Are there different colors for different models?
Does the color of the part change from part to part?
Color variables of the part? Single hue? Single brightness? Variations in saturation? Subtle color variations? Discrete color variations? Mixed with broad and discrete colors?
Does the part experience shrinkage, warpage variations from part to part?
Is there any change in appearance over time due to environment? (rust inhibitors, corrosion, lubricants, dirt, perishability, etc.)?
Are there any markings on the part? Are they physically always located in the same place?
Is it possible to make a reference mark on a part, if required?
Is the surface translucent? Variations in translucence?
Is the surface transparent? Totally? Partially?
Is part sensitive to heat?
Is part sensitive to light? Specific light - ultraviolet, infrared, visible?

Is the part repeatably and consistently located?  If repeatability is poor, it will require a system with location analysis capability to reconcile the image captured with respect to position.

Are parts moving or indexed where they can be held in place in front of the scanner?  If in motion, at what conveyor speed?  How well regulated is the speed?
What are the maximum positional variations that can be expected? X, Y and Z? degrees around X, around Y around Z?
How much spacing is there between parts?
           Is the spacing constant or random?
           Are parts overlapping?  Touching?  Jumbled? Can they be presented registered and not touching?
Is there more than one stable state involved? How many?
If there will be multiple inspections, will part maintain the same orientation throughout the process?

Material Handling:
Where will camera(s) and lighting be mounted and how?  Do custom mounts need to be designed and built?  What will produce the electrical trigger that initiates the vision system to capture and process the images?  How will reject products be removed from the line?  When will they be removed, immediately or downstream?

What is the volume envelope available at the inspection station?
Are there any restrictions or obstructions to viewing the product?
Is there a weight constraint?
If conveyor, what type? What color? What are the appearance variables of the conveyor (specular, uniformity, over time, etc.)?
Is a bypass mode required?
Is the object subject to damage in handling?
Will inspection be done at a station that also performs other functions? What are those functions?
How much time is there to make a decision? What is the production rate? Will rate increase during production catch up modes - what is the peak production rate that can be expected?
Are there any expected changes that could affect these rates?
Are parts static/indexed? Or moving continually If indexed:
           How long stationary? Total in - dwell - out time? What is the part settling time till part is stable? Is there any acceleration, and if so, where in the cycle?
If parts are moving continuously, what is the speed? Regulation of that speed?

Performance issues:
What of the performance of any system substituted for a person?  What percentage of the reject objects will one tolerate to pass as good?  What false reject rate (number of good units that are rejected) will be allowed?  No system is perfect!

What specific action is required when reject is detected? Is there additional action required if several consecutive objects are rejected? What should the number of consecutive rejects be that triggers that action?

Miscellaneous application issues:
How much warm-up time is allowed? How much overall start-up time will be allowed?  If the system is not dedicated forever to a specific task, how much time will be permitted to get the system ready between product changeovers? What will be the skill level of person performing changeovers?

How much training time would be allowed for new parts? For modified parts? How frequent might new and modified designs be expected?

Operator interface:
What are preferred approaches to physical interface with the system at the operator level? At engineering level? At technician level? Pendant? Touch screen? Keyboard? Mouse? Etc.?
What are preferred graphic user interfaces? Windows NT? Etc?
What are the expected skill levels of operators? Technicians?
Will access be limited? By password protection? By other means? How many levels of access?
Are there specific enclosure requirements?
Is it desirable to have display of objects under test?
Will archiving images of rejects be required? If so, how many reject images should be archived?
Is fail-safe operation required (part considered a reject unless it passes)?
What are desired program storage requirements? How many different part programs should system be capable of storing?
What are the desired data-storage requirements?
Is program preservation required in event of a power failure? Is data preservation required in the event of a power outage?
Does the system need to communicate with printers?  Computers?  Controllers?  Other equipment?
Are there reports that the machine vision system should generate? How often will they be required? Tally of production statistics?  Trend analysis?  Histogram analysis?  Or will data be uploaded to a cell controller or host computer?
What data must the system display to a CRT or communicated to an external device?
Should the system display inspection results for a run on demand?  Should  the display be a graph with icon-like indicators of trends?  Should the system monitor trends from nominal conditions to indicate shift in process that could lead to rejects, an indication that corrective action is needed before rejects are produced.  Should this be a bell?

Machine interfaces:
Are alarms required?
What other machines must the system be integrated with mechanically or electrically?
What event will trigger an inspection? How will the event be detected? How will this be communicated to the inspection system?
How will the results of the inspection be communicated and implemented?
What specific machine interfaces are required?

What is the environment? Factory floor, indoor/outdoor, laboratory, clean room, etc.?
What is the air quality? Dust/smoke? Steam? Oil? Corrosive? Etc.?
What is the ambient light? Incandescent, fluorescent, etc.?
What is the temperature range? Humidity range?
Is wash-down housing required?
How much floor space is available?  Overhead space that may be needed to mount cameras and/or lighting arrangements?  Will much equipment rearrangement be required?  Are power, air, vacuum and so on available?
What specific power is available? What is preferred? Is line conditioned or regulated? EMI/RFI in the environment?
How clean is the air available?
What shock and vibration levels can be expected?

System reliability/availability:
How much downtime can be tolerated for routine maintenance? How often would routine maintenance be tolerated?
What is the expected mean-time-between-failures?
What is the expected mean-time-to-repair?
What thought has been given to a challenge procedure to routinely validate performance?
How much time is permitted for calibration?

Other issues/requirements:
To support the specification of a system can the following be made available to the prospective vendors for bidding: job descriptions, present specifications, drawings, samples, photographs and/or videotapes of the facility and inspection area?  What kind of personnel will be available to operate and service the system following installation?

Are there other requirements - special paint colors?
Are there specific shipping instructions?
Who is to be responsible for installation?
Are there specific policy requirements? Warranty? Spare parts? Documentation? Training? Software? Safety?

Acceptance test/buy-off procedure
Can equipment functionality test be developed?
Can good, bad, and/or marginal parts be provided to validate performance of the system at the vendors?
What is the required sample size?
What is the acceptability criteria?
What specific parts will be used for acceptance testing? What part variables will have to be demonstrated? What positional variables will have to be demonstrated?
Will acceptance testing be different at installation site from that conducted at vendor? How?

Other Responsibilities:
Who will be responsible for installation? Designing and building any special fixtures? Any calibration tools? Start-up?


The following can be used as a guide in the 'system' analysis:

1. Straightforwardness of installation
           a. Parent equipment modifications required
           b. Rearrangements
           c. Floor space restrictions
           d. Material handling
2. In-house skills required and available
           a. Ability to do material handling
           b. Ability to do installation
           c. Personnel available to operate
           d. Personnel available to service
           e. Environment
3. Availability of
           a. Job descriptions
           b. Specifications as now performed
           c. Samples
           d. Photographs of floor space
           e. Management support
           f. Labor/union support
What should you use to obtain the comprehensive insight reviewed in the preceding?  This can come from the following:
           (a) Written description of the operation that might be available. These should be reviewed to determine if observed activities agree with the written descriptions.
           (b) Review any drawings: equipment layout drawings (are there critical dimensions?), those that affect functional capabilities, and part drawings.  What are the tolerances?
           (c) Job descriptions - do they agree with observations?
           (d) Part specifications.



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: