Tolerances - How To Choose Approriate Values

If the dimension is not critical to function, should it be on the drawing or inspected?  I would argue that the .02 radius should not be on the print, perhaps you could specify it as a max radius permitted.

In your example I would open the tolerance based on a sample size of a minimum of 35 pieces used to calculate the six sigma tolerance if I were to leave it on the print.

We tolerance based on mechanical design for six sigma yields of parts and assemblies. Not all parts and assemblies have a rolled yield of six sigma but it is always the goal. A continuous process improvement approach is required of all production suppliers.

Functional design requirements are determined and then manufacturing processes are selected that are capable of maintaining the required tolerances.

In the design for six sigma context all dimensions included for inspection should be toleranced based the process capability used to create the part dimension. These are hard limits and suppliers are required to submit reports with the CP and CPk data for all inspection dimensions. All parts can be within tolerance but if the CP or CPk is out of spec then a corrective action report is issued to the supplier and they must respond with a plan to fix it quickly. If any inspection dims do not meet the required criteria then the entire lot can be rejected. We typically only include critical to function dims on the inspection drawings along with a note requiring that the parts be manufactured to database along with a tolerance block that is based on the manufacturing process capability (i.e. die cast, injection molded, formed metal etc.) this block applies to the CAD model globally. The tolerance table typically has many values based on a range of the nominal dimension. The format is something like 0-1mm +_.027 mm 1-2 mm +_ .054 mm etc.

The tables have differing classes for general, precision and the like with different tolerances values for the nominal ranges. A lot of work up front to set this up but it makes for much less ambiguity with suppliers about what is required.

If we have a dimension that is out of spec but not affecting function, we have a process to issue a waiver to accept the parts. In reality if a wider tolerance does not affect the function then the tolerance should be widened. Use of this type of waiver is quite rare.

A lot of this has to do with the qty's you are talking about. I've worked in manufacturing where they build production line qty's of a product but my current position is much more of a built to order, with qty's, at least for my parts, that are closer to 1 than 10 most of the time.

We require 100% inspection (usually from the vendor). It is pretty common at my facility that if a part is "out of tolerance" that engineering gets involved and we will help determine if the part is rejected or accepted. When you are only building 1 or 2 of something, it's a different mindset.

We have written processes (ISO compliant) for all of it. 

I guess I should provide a little more context.  The picture above is a piece of tooling.  Typically every piece is unique and only one is required in the final assembly.  (If it's perishable then we will order spares, but the total quantity of each will still be very low, typically 5 or less.)  We then assemble all this tooling and use it internally to produce the actual parts that we sell to our customers.  The tooling is part of our internal production process, it's not being sold to anyone.

While tool designers are certainly looking at the fits between mating pieces of tooling, no one is performing any type of six sigma tolerance analysis on anything.  Every design is a 'one off'.  The goal is to get the design through engineering as quickly as possible so the tools can get ordered and built.  Think of this process more like custom job shop work where both the design and build are unique, and once finished everyone moves on to the next thing.

As far as leaving non-critical features undimensioned, that would violate ASME Y14.5.  Every feature must must have a tolerance:

I toyed with the idea of using reference dimensions, but that doesn't fit with the Y14.5 definition either since that radius (for example) can't be derived from anything else:

It seems like we really have two different tolerances - one that we prefer and one at the very limits of what will function.  The tolerances currently applied to the print reflect the preferred size, but are not necessarily indicative of what will actually function.

All of this is really coming to light because of a new inspection machine we are implementing.  In the past people would just check tools manually (micrometer, trace form, etc.)  Now we are adding a vision system that can inspect an entire piece of tooling in only a few seconds.  Suddenly many of these dimensions that didn't get a lot of attention before are now shown as being out of spec.  Will they work as is?  Probably, but no one is excited about opening the tolerances up because everyone thinks about them as the 'preferred size' or target to shoot for.

My real struggle is trying to decide where the cliff is to reject something.  At some point a decision has to be made.  If the tolerance says a dimension can't be larger than .625, do I reject it if it's .62501?  What about .6251?  .6259?  .626?  If .626 really is okay, and I adjust the tolerance, do I then reject it if it comes in at .6261?  That is what I'm trying to figure out.