I've added two new Langevin transducers to the version 3 examples section. The first is a simple cleaning transducer based on the earlier version 2 model. The second is a basic welding transducer created from scratch in version 3.
The welding transducer serves to demonstrate two common issues, but first, some background on element types.
SonoAnalyzer (both version 3 and version 2 Pro STEP-elements) uses 10-noded tetrahedral ("quadratic") elements. The presence of mid-side nodes allows for much more accurate modelling of stresses and displacements than elements with only corner nodes and linear distribution between them. Midside nodes also allow elements to follow curves much more accurately. However fine detail in geometry can cause these elements to become overly distorted, causing the Finite Element Analysis to fail with a "non-positive Jacobean" message.
In version 2 this would cause the analysis run to fail and the user would need to select one of the "NPJ geometry compensation" options - "second order linear" or "refine by splitting". Second order linear simply places all mid-side nodes exactly half way between the corner nodes; the advantage of following curves is lost but the advantage in distribution of displacements and stresses is retained - this is usually the best compromise. Refine by splitting converts all elements to linear (with significant loss of accuracy) but to some extent compensates by increasing the mesh density significantly.
In version 3 the issue is handled more smoothly - SonoAnalyzer detects the condition and automatically falls back to a safer element type: first Quadratic modified (all mid-side nodes exactly half way between the corner nodes, identical to second order linear in version 2), then Linear - no midside nodes at all. Note that in version 3 SonoAnalyzer leaves it to the user to deliberately refine the mesh size but a clear warning is given whenever the meshing fall-back occurs.
So back to the example of a simple welding transducer. At a rather coarse 5.7mm element size the default Quadratic element type fails and SonoAnalyzer v3 automatically falls back to Quadratic (modified). The same happens at 4mm, but at 2.8mm the mesh is fine enough for the analysis to run normally. Comparing the predicted frequencies (welding transducer coarse mesh vs welding transducer fine mesh) there is little difference showing that despite the rather rough look of the coarser quadratic (modified) model it still produces reasonable results.
But the results show another interesting effect: There's an unwanted mode involving lateral bending of the clamping screw, and it appears very close to the 20kHz axial mode. So much so that the axial mode also shows the bolt bending inside - use the Display menu to choose Section plane xy to see this clearly. The effect is similar for both the coarse and fine mesh models.
The screw-bending mode is obviously strongly affected by the exposed length (between the head and the front block) so the next step was to change this length. I increased the depth of the counterbore in the back-block by 2mm, and also entered a -2mm offset to the location of the bolt (the offset field is to the right of the Align Y control), so that the bolt head remains in contact with the base of the counterbore. The result can be seen in the third example: "Improved welding transducer". It's a good outcome - the bending mode frequency has increased by about 1.5kHz, leaving clear space for the 20kHz Axial mode.
See the development for yourself in the transducer examples on the Free server (Help -> Examples). Loading the examples doesn't require sign-in.