Dimension Bug - AC26 5002 (UKI Full)

[DGSketcher]

I have just been criticised by my client as the dimensions in my drawing are inconsistent. This wasn’t a problem in the past. The latest version of AC is rounding all 0.5mm values upwards leading to an accumulated length error. Previously the rounding process arrived at the same overall length. Under the current rounding each 0.5mm is accumulating to generate a significant error depending on the number of 0.5mm values. I appreciate to some this may be a trivial error, but in the world of offsite fabrication it can lead to connection problems when components are designed to 0.5mm accuracy. In the attached image there is a 3mm error when the red dimensions are added. In the context of the attached the error is small, but for shorter lengths / components that 3mm could be a significant error.

 

 

This has been reported to GSUK support, but I felt others should be aware that this could affect their projects, particularly if there is no overall check dimension displayed and they are working to tight tolerances.

[gpowless]

The solution in my experience of the same issue is to draw and dimension accurately in the highest denominator possible. By modelling complex profiles in real-life dimensions sometimes these rounding errors are inevitable. So one should orient the reference lines in a consistent manner and only dimension (on plans) to the reference line.

[DGSketcher]

I don't disagree. Unfortunately my data source is a complex GDL object that needs to work at a 0.5mm tolerance, but that is a side issue. As I have suggested above AC used to adjust for rounding such that the indicated values in a chained dimension accumulated to the overall chain length. Consider this... I have an object 995mm long and I divide it equally into 10 and place my dimensions. What AC will now show is 10 dimensions of 100mm which is 1000mm. In some situations that may not be a problem, but if you just had a 1000 units deliver that don't fit by 5mm then it becomes an issue. Measurements of 0.1mm should be correctly handled by AC and that includes applying correct round up / down values to minimise errors in chain dimensions.    

[jl_lt]

As if we needed yet another excuse to not upgrade to 26

[Barry Kelly]

It is not just 26.

I just checked 25 and even 22 and got the same results.

I am not sure when (if ever) it would have rounded any differently.

Technically it is correct as it is rounding the 0.5 up to the next whole number - fine for each individual dimension.

But cumulatively, they do not add up to the overall, which is of course correct itself.

If there had been a few dimensions less than 0.5 and a few 0.5 and above, you would never have noticed.

If the individual dimensions are so important and the 0.5 accuracy is required, then there really is no choice but to show the extra accuracy in the dimensions.

Then let the person dealing with the plan decide to round up or down.

As I mentioned, unless there is an equal number of dimension rounding down as there is rounding up, then there will always be a discrepancy.

Archicad can only follow rules when it comes to rounding each individual dimension.

It doesn't actually add those figures together and compare it to the total.

It would be interesting if it could though.

The only other option is to manually pick a few dimensions and change them to custom figures.

It would be nice if we had an option where we could tell each dimension to round up or down - overriding the default setting.

Then we could keep our 'measured value' but control how it rounds.

Barry.

[DGSketcher]

So I did a little research with one of well respected engineering CAD programs. It would seem that there are a number of solutions, but in order to fix my example they would need to be applied on an instance basis. I'm guessing that isn't happening anytime soon.

The suggested adjustments for rounding are:

Truncate the 0.5

Always round the 0.5 down

Always round the 0.5 up

or the most advanced solution... Round up or down according to if the preceding digit is odd or even e.g. 22.5 changes to 22 and 21.5 changes to 22.

I still think this is an issue in chained dimensions that could be addressed by an alternate rounding algorithm but I won't hold my breath for a solution.

[Barry Kelly]

It is an interesting problem and I am not sure how it could be solved.

The alternate rounding system is an interesting solution, bit not sure how hard/easy it would be to implement.

Generally if you have an overall length that you split, the rounding will be fine.

Say 10.0 overall and you split it to 4.3 and 5.7

It will round to 4 & 6 = 10

But the object you show in you image seems to add a length/opening with a decimal fraction, but the next lengths is a whole number, and not the complimenting remaining decimal fraction.

So your object will always be rounding up as in your image or rounding down if less than 0.5.

I am not sure what your object is or how it works, but it seems to be adding lengths together rather than setting an overall length that it splits.

No matter, there is obviously an issue that needs a solution when adding decimal segments of a dimension that they somehow need to add to the overall length.

Barry.

[Erwin Edel]

I don't think I would want 0.5 mm rounded down ever, to be honest.

Then again, I generally round up the 12.5 mm gypsum boards to 13 mm if they are solitary and keep it at 25 mm if there are two of them, so I have 'control' over my dimensions.

I also typically add an extra 'subtotal' dimension in the rare cases where the dimensions don't add up properly and sometimes add a bit of text annotation to clarify or draw extra attention.

Your example does look like a frustrating mess of prefab accuracy. I'd probably also go for showing the extra accuracy, as you don't want those windows to be the 'wrong' size and so on, by rounding up on some dimensions.

[DGSketcher]

Thanks for the additional feedback. The source of the problem within the framing is it uses an integer grid, but the framing size is odd e.g. 45mm, so I end up with the side of the framing 22.5mm off grid. If an opening then butts up to it, then it also sets the position of the opening. It's not the design process that is the problem it remains a question of how AC (& others apparently)  handle numeric rounding. As another example in the UK we work with a 225mm brick grid, this has a similar problem when there is a corner (half brick return) or opening at a half brick length, correct modelling creates a 0.5mm dimension.

This is literally a scalable problem that needs fixed. It isn't a node problem, it is a text display problem that is caught up in chained dimensions that are not self aware that all the indicated dimensions between the end nodes should accumulate to the same distance and work with any unit system at any scale to display appropriate values.

[Laszlo Nagy]

@DGSketcher,

Just out of curiosity: What is wrong with displaying the extra accuracy?

By the way, maybe you could add a cumulative dimension as well?

If I had to solve this issue, I would simply subtract the previous cumulative value (the one to the left of the current value) from the current value, that way the rounding errors would be compensated for in the next dimension.

Anyway, I will forward this to GS HQ.