What’s new in Axomek 0.2.4

Results that change when a saved project is solved again:

• The optimiser's powerLoss metric (axm optimize --require "powerLoss<=…", --minimize powerLoss, the app's optimise dialog) is the total that axm solve reports in powerLoss.total: gear mesh and belt or chain drive losses now count as well as bearing friction. A search that limited the power loss of a project with gears or drives could accept designs that lose more; it no longer does.
• axm optimize (and the optimiser behind MCP and the app) refuses a variable whose step is 0, negative or not finite, or whose bounds are not finite or not in order, with error invalid naming the variable and a hint; a negative step used to search for ever. The search also stops as soon as it can reach no new design, whatever budget is left.
• A misspelt or unknown field inside an array of a patch (a strength notch, bolt plates or shank sections, gear load spectra, weld lines, ball screw loads, housing ribs, …) is refused like one outside arrays, with its path (strength.notches[0].raidus); it used to be dropped silently, so the notch was checked with the default radius. Every unknown-field error now suggests the field meant ("did you mean 'radius'?"). Opening a file is unchanged: unknown fields there (from a newer Axomek) still make it read-only instead of refusing it. A contact's tractionDirection can now be set by a patch; it was refused as unknown.
• Unbalance response on supports that differ between the two bending planes (a support or housing stiffer one way): both planes are now solved together, coupled by the gyroscopic moments of the turning shaft and its parts, as the critical speeds already were. Before, each plane answered on its own as if the supports were round, which put the resonances of strongly gyroscopic rotors (a heavy overhung disc) in the wrong place: the coupled first critical speed could show almost no response and a resonance appear where the rotor has none. Shafts on round supports give the same results as before; slender shafts without discs change little.
• The shaft's largest deflection, bending moment and von Mises stress include their peaks between the mesh's nodes, exact for the beam element under its own weight. On a coarse mesh (a large maxElementLength) these were missed: a shaft meshed as one element between two supports showed no deflection and no moment under its own weight. On the default mesh the values rise by up to about 1e-4 (the gearbox-intermediate template's largest deflection by 5e-5); the deflection and stress plots still show the nodes' values.
• Cross-located shafts (two or more locating supports) solve their axial forces by statics on the supports' springs rather than by the bar's stiffness matrix, which lost precision when the supports were far softer than the shaft: with axial stiffnesses of 1e-6 N/mm the bearings' axial loads summed to 98 % of the load, and at 1e-8 N/mm the solve stopped with an internal error. They now balance the load exactly; a stiffness too small to compute with is an error on the supports with a hint. Ordinary stiffnesses give the same results as before (to round-off).
• An imported housing mesh that holds a node it does not have, or holds components other than x, y and z, is an error on that entry (housings.<id>.mesh.fixed.<n>) with a hint, instead of stopping the solve.
• A shaft that cannot be solved says how to fix it in the project report as well (shaftError ends with "(hint: …)"), as axm validate already did.