How does it work ?

It doesn't work. It is a numerical issue due to the 64bit limit of a double value. In Matlab, for instance, you get the following warning:

Matrix is close to singular or badly scaled. Results may be inaccurate. RCOND =  9.628486e-18. 

However, Matlab is (also) able to invert the normal equation matrix. Like Matlab, JAG3D uses LAPACK for matrix operations. Thus, the obtained results are not serious...

regards
Micha

We made several tests in JAG3D to compute in group 2 distincts stations.

In our lastest test, we adjust the network without measures on the references points. And JAG3D was still working, no error message. We have bad results in tools/indicators (sigma, etc). But the geometry should not permitted it.

How does it work ?

The data are attached here :

https://bygsend.bouygues-construction.com/bigsend/download/guid/DBsf-HXUwpt8W4W4vGKZJasJx3WfUPfp-5Pj1--pIIw3h7DFksa8-pWCoEjL4wRbOCNzxYir1aDJ4dyYF6yCvUuM
Password : bawvz8G

Thanks a lot for your return,

Best regards

Indeed, all the directions are derived within the same datum. But in this case, I do not understand how JAG3D can calculate his network without enable the group. I thought that there should have a geometric problem : no complete link between CD points.

To calculate your network, the link between the CD points is not needed here. Each station can be adjusted independently. Thus, you can also disable the distances and vertical angle group of the CD points. Adding more information doesn't change this effect - the network is still estimable. The pseudo-observations between the CD points are further information. If this information is enabled, dependencies between both stations are introduced.

But JAG3D adjustement do not consider the projection

In general, it is a projection problem and not an adjustment one. JAG3D is designed to adjust a network. Moreover, you are using UTM coordinates but UTM is not defined in 3D - there is no slope distance or vertical angle in a map projection.

regards
Micha

Whereas a direction is a un-orientated observation, a direction angle is defined between a station and two target points. For that reason, a single direction makes not sens because it only contributes to the unknown orientation. However, if really all directions between the CD points are derived within the same datum, you can enable the group. In this case, all directions contribute to the same unknown orientation redundantly. As already mention, I was not sure about this fact and, therefore, disabled such pseudo-observations.

Indeed, all the directions are derived within the same datum. But in this case, I do not understand how JAG3D can calculate his network without enable the group. I thought that there should have a geometric problem : no complete link between CD points. Why JAG3D can calculate this group ?
I thought that we need vector 3D constant between couple of points, dX, dY, dZ. But we get the option to insert raw data with high level of precision. We insert it in JAG3D, with 3 differents kind of observations ; without directions sets, we lost an essential information in 2D.

Are you referring to the unknown orientation? This parameter describes the discrepancies of the north-direction of your frame and the zero-direction of the total-station (angle-encoder). See above, a direction is not an (direction) angle. A direction of 100 gon (90 °) does not imply that the target point is located in the east.

No I do not referring to the unknown orientation of the total station. We understand well this concept.
We are referring to insert an orientation between points to generate vector 3D or eccentricity to the CD points presents in our network

No, just averaging the coordinates in Excel, for instance, and reduce each coordinate by the mean. Currently, the network extent is about 500 m x 500 m. If, for example, the origin is set to 0/0, the maximal coordinate becomes 500/500. Small coordinate values will increase the numerical stability because of the 64 bit limit of double values --> https://en.wikipedia.org/wiki/Double-precision_floating-point_format.

Ok thanks you but this subject is ok.
But it is introduce a problem of our georefencement points. Today, we are in France and we used local EPSG (for exemple, 2154 in this worksite). But JAG3D adjustement do not consider the projection : we can not integrate this information in the metadata of one project. Is it right ? Or we don’t use correctly JAG3D ?
If not, it could be an usefull update.

Thanks,

Kind regards,
Antoine

Thanks, we analyse that you differentiate "stations" and "new points" in 2 distincts groups in "new points" ?
Is it a good practice ? does it change something in the adjustment ? Should we insert it in our project ?

No, such separation has no effect on the results. However, I'm not part of your project, so I have to re-organise the project to get a better impression. It was just for me. Sorry for confusion.

All the direction are calculated between the CD, with always the same formula. We oriented it from the north. An excel file is attached where the calculation is available.
But I am not sure to understand why you disabled these observations. Why did you do this ? Indeed, the project had a particular geometry; with strains between the CD points : in directions sets ; zenith angles & slope distance.

Whereas a direction is a un-orientated observation, a direction angle is defined between a station and two target points. For that reason, a single direction makes not sens because it only contributes to the unknown orientation. However, if really all directions between the CD points are derived within the same datum, you can enable the group. In this case, all directions contribute to the same unknown orientation redundantly. As already mention, I was not sure about this fact and, therefore, disabled such pseudo-observations.

Theorically, we can not disactivate once parameters : directions sets and you can not always adjust the network.
But in practice, we can adjust the network... How does it work ? Where the strains in directions sets are usefull ? They seems to be useless...

Are you referring to the unknown orientation? This parameter describes the discrepancies of the north-direction of your frame and the zero-direction of the total-station (angle-encoder). See above, a direction is not an (direction) angle. A direction of 100 gon (90 °) does not imply that the target point is located in the east.

"I would also suggest to use some kind of trimmed coordinates to avoid numerical instabilities, i.e., reducing all X/Y/Z coordinates to their average values. "
You mean that we should do a preprocessing : approximation values (CTRL+M) ? before do the command "adjust network" ?

No, just averaging the coordinates in Excel, for instance, and reduce each coordinate by the mean. Currently, the network extent is about 500 m x 500 m. If, for example, the origin is set to 0/0, the maximal coordinate becomes 500/500. Small coordinate values will increase the numerical stability because of the 64 bit limit of double values.

Yes we know this low quality of our networks but in case of AMTS monitoring, we did not find a solution to make redundancy.

Okay, it was just a note, but if you keep it in mind - no problem.

regards
Micha