Machine Vision 101 - The Technology
The following article presents some things to consider in purchasing a vision system. Whether you buy from us or you buy from a competitor or you choose not to buy, these are thoughts to consider before you choose. Of course these considerations are generic, not every one of them may apply and there may be other issues that apply to your situation. An effort is made to avoid technical content (i.e. CCD vs. CMOS, Line Scan vs. Area Scan, or GigE vs. Camera Link vs. FireWire) and just summarize the following:
- What high level choices need to be made?
- What can it do for our business?
- What can't it do?
- What could it do?
- Does this fit our needs?
The idea is to give you a plainspoken sense of capabilities so you can determine if this is right for you before diving into the details of an implementation. If you want to talk through any of these points, or discuss a specific situation just give us a call (303)832-1111.
Machine Vision Systems:
For our purposes let's define a vision system as any system that uses cameras of any type to check features of a product in an industrial environment. These systems can take many forms depending on client needs, but in general they provide an automated and repeatable way of performing quality control and other verification tasks.
In-line vs. Off-line:
Do you want to inspect parts on the production line or off-line at a measuring station? Off-line measurement can have a cheaper sticker price because more standardized products exist and there's less integration work to be done. However, the labor savings are significantly less because employees still need to sample at regular intervals. While in-line systems automate the plant by continually monitoring production, off-line systems give QC teams another great tool to do their jobs more effectively.
"Off the shelf" Product vs. Custom System:
For common vision tasks some vendors make "off the shelf" products (i.e. comparing a print against a proof, reading a barcode, etc.). Sometimes an off the shelf product is just the thing, other times we can slightly modify an off the shelf product to suit a client need. However, sometimes a client's process and/or product is sufficiently unique, or there are few enough manufacturers in their field that no off the shelf product exists. Feel free to give us a call and we're happy to discuss what exists in your field.
Flexibility for the Future:
Products change, new product lines are added and others are removed. A vision system typically performs a precise inspection recipe for a specific product. Unless someone reprograms it, the system typically cannot inspect the new part. It knows a certain batch of features to verify and if those features change the part doesn't pass inspection because it doesn't look like the old part the system is supposed to recognize. If the new part passes the same system without modifying the inspection recipe it may be time to question the accuracy of the system if critical features of the part have been changed.
Is there a plan to change automated inspection recipes as needed?
If you make 1 inch roofing nails that haven't changed in a generation the risk of product change may be low. If you make rubber soles for sneakers that change every year almost without fail, it tends to be a good idea to negotiate a fee for changes upfront. Buyers may think they are saving money by not negotiating an "add on" plan for future changes to the system. However they may spend $50K on a system and find themselves needing a new system once product development decides next year's model will be different from last year's. It's important to consider what you're buying. Most times you buy hardware plus the necessary software to check critical features A-Z of the specific product you currently make. In many cases that's adequate and you shouldn't buy a service contract, but it's certainly something every buyer should consider. For example in exchange for $10K per year the vision system provider may agree to change the inspection recipe up to 2 times per year. Alternatively there may be a $6K fixed fee for changes, or an hourly rate. If you decide to buy, make sure to discuss what happens when there are changes to a product after implementation.
|Scrap:||Vision systems detect problems as they happen. If critical diameter A moves out of spec, the vision system will capture that immediately and if necessary stop the machine. This saves the scrap that would have been generated between a machine moving out of spec and the next sampling.|
|Downtime:||Minutes where we make scrap cost more than raw materials. They are also wasted production time. The plant was staffed, the machines running and shippable product was not produced. These minutes are costly. Thus identifying problems immediately reduces downtime and saves money. In some cases the vision system can correct the deviation from spec automatically. For instance, if the diameter problem is because the mold is too hot, the system could talk to the cooling system and dial up coolant pressure. This may avoid the need for a shutdown at all.|
|Repeatability:||Vision systems provide a consistent standard across plants, production lines, shifts and production cycles. One operator may get tired, may make mistakes and may despite his or her best efforts not apply the exact same standard every time. With multiple operators this problem only compounds itself. Throw in turnover and soon we may have significant variability. A vision system executes the exact same inspection recipe every time from the first unit till whenever we change the recipe to accommodate a new product.|
|Lower Labor Costs:||For an in line system, because we no longer need staff to sample production and perform measurements we can generate significant labor savings.|
|Less Quality Escape:||Sampling will catch statistical deviations from spec, however a vision system can catch the intermittent defects sampling can miss. From samples we estimate a process sigma and Cpk and from this estimate conclude defects will be rare enough. With a vision system there is no additional cost to just check every part and eliminate any defects found. Even for a process estimated to be achieving six sigma status or 3.4 defects per million, a vision system can catch some portion of those 3.4 defects and improve quality.
Furthermore, because we are checking every part that goes out the door we no longer have to worry quite as much about an excessively tight sigma that may add unnecessary cost. A looser sigma allowing 50 defects per million may be just fine as long as the vision system can catch 46 of them and we can afford to scrap 46 parts. In this scenario a vision system and a looser sigma have effectively achieved six sigma status and possibly at lower cost. If the cost of a vision system plus some additional scrap parts are less than the cost of the stricter sigma, a manufacturer could save significant money and achieve the same level of quality. Imagine hitting the same requirements with a more cost effective grade of raw material, or getting an extra few years out of aging machinery.
|Financial:||Vision systems are often expensive. Hardware costs can vary widely from a handful of thousands for the most basic systems to tens of thousands if very specific components are needed. The cost of an integrator must be added on top of the hardware. If the manufacturer prefers to do the job in house, the cost of internal resources must be included as well. Sourcing the right hardware can minimize these costs. For instance getting only the resolution and frame rate you need on the camera and using the appropriate vision software libraries can mitigate costs, but truth be told a system will likely be a significant investment. Expect to spend at least $15K per system, it could end up being slightly less but for even a fairly basic system hardware is often on the order of $6-10K.|
|Risk of Change:||Any change to an established process involves risk. Vision systems are undoubtedly a change to the QC process. Consider whether you need to change. What symptoms are driving the change? Are defect rates too high? Do we need traceability? Do we currently have high labor costs from doing QC in house? Are scrap costs mounting? Are customers demanding automated verification?|
|Risk of Lock-In:||Any investment, even in high tech machinery, can lock us into a specific way of doing things or a specific product. A vision system can inadvertently prove a barrier to change. If we invest $100K in a system to check product X, we may become leery of changing product X. One thing you may want to ask a vision system integrator is how they will help you manage change. What flexibility will you have if the product needs to change? At Artemis we are willing to offer "pay by use" pricing. For instance think of a vision system as a labor contract, if things change in 6 months and you use the system less, you pay less. You no longer have a $100K weighing on any decision to change. If change comes and the need for the system disappears you stop paying. Again this is an option, we're flexible here and want to meet your financial needs. Some companies prefer to pay $50K upfront to save $50K annually, others would prefer to pay $10K per year for as long as they need the system and keep the $40K per year in savings for themselves.|
|Limits of a Vision System:||Vision systems do specific checks on critical features. They are not a catch all that will find any and every type of defect. Obviously non-visual problems won't be caught (i.e. is the part weak?) unless there's a specific visual manifestation of that weakness, i.e. a crack or a narrow critical dimension, etc. Vision systems also only check what they are programmed check. To give an example, if we program the system to check the barcode and make 5-critical measurements and check that two components are present, it will consistently and repeatedly make those checks on every part. If a new defect occurs that the system was not programed to look for it won't necessarily be caught. For instance, in the above example a discoloration may pass inspection. Checking "everything" is costly so vision systems often work best when we can define critical features A-X to check. It's important to understand that new defect Y may or may not pass inspection. The best candidates for vision systems are production processes where we can define some finite problems that are responsible for most of the costs.|
|Implementation on the Shop Floor:||Even if the buyer fully understands the limitations of what they've purchased, not every employee on the shop floor necessarily shares that understanding unless it's clearly explained. Employees need to be aware that a vision system may not catch a new problem that's never been seen before and so they should always remain alert for new issues. Consider the costs of getting man and machine to function effectively on the shop floor. There will need to be a training session and the new machine will need to be explained. Deployment may have a cost in internal resources in order to make the investment effective.|
What could it do?
The intent here is to cover other things a vision system is potentially capable of. A basic in-line or off-line system will inspect and identify bad parts and that's all some clients want and need. This section covers additional things that would be possible, some of which may be of interest and some of which may not. Depending on the feature, they may add substantial cost or hardly any cost at all. For some users one of these benefits alone may be the only reason they're investing (for instance traceability in the pharmaceutical industry). Some of the benefits may also help the client eliminate a different system and thus yield a cost savings in another area.
Images from a vision system can easily be stored on a hard drive. In today's world, storage space is fairly cheap so it can be cost effective to keep extensive records. You may never use most of the images, but it can be very convenient for diagnostic purposes. What went wrong during the shift change yesterday? Why are parts jamming on the conveyor? If you buy a vision system you can fairly easily get a photographic record of what goes on at a specific point on the production line. The same way a store's security camera can give a manager visibility into what's going on even while he's not around, a vision system can let management see the production line at any time. It can also replay critical events, so you can see them yourself and diagnose them. To take the above example a step further, if parts continue to jam on the conveyor you can take the video to the conveyor belt vendor so he can see the problem himself and offer a more informed solution.
Better Customer Communications:
Rather than providing a handful of sampled gage measurements with an order, you can now provide images with annotated dimensions or feature verifications. You can summarize the data and upon request provide all the files. Whether you ever miss a spec or not, every manufacturer sometimes has tense conversations with clients who feel rightly or wrongly dissatisfied. Unfortunately these conversations can carry a tone of "you didn't check the shipment", "there was insufficient QC", "you didn't meet the spec". Showing the client all the annotated images and measurements from the vision system can change the tone of the conversation. No longer does the client doubt that the parts were checked. They can see extensive QC took place. Also it's clear from the data that the spec was hit. The conversation can then move beyond veiled accusations and towards "maybe we need to change the spec because this shipment doesn't seem to meet your needs" or "maybe a higher grade of X would better match what you want". Having a documented, computer based system can quickly take the emotion out of these sometimes difficult situations. Reviewing the records together and determining how we may want to meet a different standard is a much better conversation to have than an unwinnable argument with a rightly or wrongly unhappy customer. Even if the manufacturer is in the right, they'll likely need to concede on price and may even lose the customer.
Live Production Data:
Vision systems generate an enormous amount of data. Instead of samples with operator errors you'll now have measurements on critical features for every part that comes off of a production line. The opportunities for analysis are endless if you can get the dataset. How do defects vary based on factor X, Y or Z? How can we improve yields? etc. Be sure to ask what your integrator can do with this data. It's often possible to dump the measurements into a database or an SPC software package. Payback and enormous returns can come from the analyses. In fact through analyzing the dataset you may uncover what was causing most of the defects in the first place.
Traceability is another potential high return question to ask. Vision systems by definition photograph every part that comes off of the production line. If each part has a barcode/serial number or some other identification, the image can be stored with its barcode number. This opens up a few potential options for manufacturers:
When did the problem occur?
Each product is photographed and stored with its time of production so it's easy to trace customer complaints to specific hours/shifts of production. Do most complaints come from products made during the evening shift, made during the shift change, made with supplier X's material?
Whose problem was it?
If a customer complains about a specific product we can now see if the defect was apparent when the good was produced. Many times complaints come back and we have no way to determine when the problem was introduced. More importantly, could we catch it next time? If a customer says their product has a crack, who is to blame? With a traceable vision system we can map a complaint about unit 573492 to its image. We can sort out which problems in the field relate to a visible feature we should have found or should be finding (i.e. the rim was slightly bent) and which occurred downstream. If the problem relates to something apparent at the time of production, we can change our tolerances to find it in the future. A traceable vision system allows us to better target our tolerances with real data from the field.
Alternatively, if the problem occurs downstream this dataset may allow us to re-negotiate with logistics providers and retailers over returned products. If we can show that most damage occurs in transport, it's unclear we should bear that cost.
Despite our best efforts, sometimes a problem only becomes apparent later. Innovative manufacturers inherently make new products that have never been made before. These products are extensively tested but not every issue can be foreseen every time. For instance we thought the tolerance on the caliper should have been 3 thousandths but it really needs to be 2.5 thousandths because of some unforeseen interaction in cold climates with older disks. With a vision system we have a record of every part and we can recall specific batches and serial codes as needed. Depending on how much information we have from the point of sale we may be able to call up the 50 customers whose caliper doesn't meet the new tolerance instead of recalling all 500,000 units. The cost implications of targeting these events are huge, not to mention the negative publicity we likely avoid.
Is it right for us?