Zenomiq
Back to Zenomiq

Changelog

Changelog

What changed in Zenomiq, newest first.

1.19 — 2026-09-18

Fixed

  • The button for copying a newly generated API key was almost invisible against the green panel behind it. Its outline and icon are now white. When you generate an API key, Zenomiq shows it once in a green panel with a button beside it that copies the key to the clipboard. The button's outline and its icon were drawn in dark grey, leaving almost nothing to see them against: a contrast ratio between 1.2 and 1.4 to one across the panel's gradient, where 4.5 to one is the level at which ordinary text is comfortably legible. In white they now measure between 4.6 and 5.1 to one over the same gradient. Hovering fills the button white and turns the icon green, which measures 5.1 to one, so the button stays readable in both states. Nothing about how API keys are generated, stored or used has changed, and no calculation is affected.

1.18 — 2026-09-16

This release changes how AI mode searches for a design. It does not change how any design is rated. No rating formula, factor or constant moved, and no value taken from ISO 6336, VDI 2736 or DIN 3990 changed. Take a design you already have, feed the same geometry, loads and materials back in, and every reported number is identical before and after. Re-rate a gear in Manual or Geometry mode and nothing below touches it — those two modes cannot move under this release. What changes is which design an AI run proposes: the same inputs can come back with a different gear, and therefore with different reported numbers.

Three things to read before you re-run an AI design. Reported safety factors move in both directions — up where the search was being distorted by penalties that should never have applied, down where it was overshooting the target you had set. Accepted AI designs now tend toward a lower helix angle and a larger normal module than the search used to pick. And AI runs take longer — on an all-metal case, budget up to roughly twice your current run time.

Added

  • In AI mode, Gear 2's facewidth now follows Gear 1's — and this is switched on by default. Two settings appear on the AI mode basic gear data card. Same facewidth (b₂ follows b₁) is set to true, and with it on the optimizer stops searching Gear 2's facewidth as a free variable and derives it instead, as Gear 1's facewidth minus twice the pinion overhang per side, the overhang being measured in multiples of the normal module. The second setting, Pinion overhang per side, is that overhang, and it defaults to 0 — which makes the two facewidths come out exactly equal. Raise it, to at most 1.0, to make the pinion deliberately wider than the wheel on each side, so the wheel stays fully in contact if it sits slightly off its nominal axial position. Because it is a multiple of the module rather than a millimetre value, the same coefficient gives a proportionally larger overhang on a larger gear: a coefficient of 0.5 produced a total width difference between the two gears of 1.00 mm at a 1 mm normal module and 20.00 mm at a 20 mm normal module, both measured.

    Because it is on by default, every AI run's search changes — it has one free variable fewer than before, so the same inputs can land on a different design. Set Same facewidth to false to get the previous independent search back. Do that when the two gears genuinely need different widths, for example a narrow pinion driving a wide wheel that is shared with a second mesh. Both settings exist in AI mode only: they do not appear in Manual or Geometry mode, and Gear 2's facewidth stays an ordinary input field in both.

Changed

  • AI runs now stop at the safety factor you asked for instead of overshooting it, so reported safety factors can come down — toward your target, never below it. The safety targets were enforced by a penalty that never fully switched off: a design already comfortably above a target still paid to be pushed higher, so runs habitually landed 20% to 50% above the floor you had set and bought that margin with extra material. The penalty now falls to exactly zero once a criterion meets its target, while pushing exactly as hard as before on anything still below it — below-target enforcement is not weakened. On our plastic gear verification case rated to VDI 2736, with the targeted root safety factor of Gear 2 set to 1.0, the shipped design's calculated root safety factor for Gear 2 moved from 1.2009 to 1.0669 — from 20.1% over target to 6.7% over — and Gear 1's facewidth came down 6.8%, from 10.6075 mm to 9.8820 mm. On our all-metal verification case rated to ISO 6336 the effect was near nil: facewidth −0.03% and every safety factor within 0.3%, because that case was not overshooting in the first place. If you have signed off an AI design, re-run it and work from the new numbers. For the same stated target you now get less margin and less material; if you wanted the old margin, raise the target.

  • In AI mode the default targeted transverse contact ratio changes from 1.4 to 1.1, and the target is now something the optimizer can actually reach. Two things changed together. The target used to be enforced symmetrically — a design whose contact ratio sat above it was still pulled back down toward it — and because of the penalty's shape the target could never be met, only approached from below. It is now one-sided: falling short of it counts against a design, exceeding it costs nothing. With that downward pull gone, 1.4 was no longer the right default, and the Targeted transverse contact ratio field now shows 1.1. Whatever you type wins, in both cases — enter 1.4 and you get 1.4.

    Expect AI designs whose contact ratio sat between 1.1 and 1.4 to come back with reported safety factors a few percent lower. On our plastic verification case rated to VDI 2736, the calculated root safety factor of Gear 2 — the governing one for that rating — fell 3.34%, from 1.3421 to 1.2973; Gear 1's fell 4.02%; the tooth counts went from 32/64 to 33/66, the helix angle from 10° to 11°, and the normal module down 3.40%, from 0.6144 mm to 0.5935 mm. On our metal case rated to ISO 6336, where the contact ratio already sat at 1.69, nothing moved by more than 0.03%. This is a design criterion only — the separate ε_α validity floor setting, which decides when a result is reported as outside the rating's scope, is unchanged.

  • Two of the changes in this release lower reported safety factors on plastic gear ratings to VDI 2736, by unrelated routes. The two items above were each measured on their own, so do not read either percentage as the total: on a VDI 2736 case the combined drop can be larger than either. Re-run the case and work from what it reports rather than estimating it.

  • The same inputs now give the same AI result twice. A run stops early once it stops improving. The old rule compared the last two results with zero tolerance, so differences of around one part in ten million flipped its decision, and when it did stop it kept the last point it had reached rather than the best one it had seen. Identical runs could therefore land on different designs. A run now stops only when the best result seen has failed to improve by a real margin across a window of epochs, never before a minimum number of epochs has passed, and it returns the best design it saw rather than wherever it happened to stop. Two identical runs on our verification case returned identical designs, digit for digit. Productive runs last a little longer as a result — one seed ran to epoch 325 where it used to stop at 278 — and hopeless ones are still cut short, just not before the minimum.

  • AI runs take longer. The extra work is real: the search rates more material candidates, retrains a seed that produced a broken geometry, and lets productive seeds run longer. On our all-metal verification case rated to ISO 6336, the two material-objective corrections listed below nearly doubled the run, from 121 s to 236 s. Budget up to roughly twice your current AI run time on an all-metal case, and a few seconds more each time a seed has to be retrained. That figure is an envelope taken from the largest increases we measured, not a prediction for your case.

  • When the optimizer buys a stronger material, it now makes one attempt to shrink the gear — and keeps the result only if it is genuinely better. Material selection happens after the geometry has been trained, so buying a stronger pair used to leave the geometry frozen and ship the whole safety surplus instead of turning it into a smaller gear. The run now retrains once with the chosen material in place, re-rates the smaller geometry with that same material, and replaces the shipped design only if every target you set still holds and the blank really is more compact. If either condition fails, the original design ships unchanged. Read this as "no change, or a smaller gear" — not as a saving you can count on. On the one verification case we have measured it on, the retrained geometry came out larger, the attempt was correctly rejected, and the shipped design was identical to what it would have been without it, at a cost of about 50 s of extra run time. It applies to all-metal pairs rated to ISO 6336.

Fixed

  • AI mode no longer offers you a geometrically broken gear. Every seed of a run was already being checked for geometry validity — interference between one gear's tooth root form and the mating tip, undercut, a tooth come to a point, loss of tip clearance, a contact ratio below its floor — but a seed that failed a check was still listed in the seed browser, and you could select it, generate it, view it in 3D and export it. Such a seed is now discarded and its slot retrained from a different starting point, up to two retries. The result carries a warning naming the seed and the reason it was dropped, and if a slot produces nothing viable within the retries, a second message tells you how many of the seeds you asked for came back viable. No new geometry check was added — the checks that were already being computed are now acted on.

    On our plastic verification case rated to VDI 2736, the old behaviour shipped a browsable seed whose root-form interference on Gear 2 read 0.3255 mm where the limit for that design was 0.0119 mm; after the fix, every returned seed passes. The winning design can change, because the retrained seed competes for the win: on that case it won, and the shipped design's calculated root safety factor for Gear 2 came out at 1.391 instead of 1.067, with Gear 1's facewidth at 9.574 mm instead of 9.882 mm. Those two figures were measured with the target-overshoot change described above already in effect, so read them against each other and not against the numbers in that item. The retry cost about 7 s there. On our all-metal case rated to ISO 6336 no retry fired and the design was unchanged.

  • Two penalties in the AI search were charging designs that had nothing wrong with them, and the reported safety factors of affected designs come out higher now. The search keeps a design clear of undercut and of a tooth that has come to a point. Both guards went on charging a design that already satisfied them in full — on the reference all-metal case rated to ISO 6336 they made up about 70% of the total search cost at a design that satisfied both completely — and because the charge varied with the profile shift, the optimizer was quietly working against a penalty instead of against your criteria. The undercut guard also could not be escaped by changing the tooth count at all, only by changing the profile shift. Both now go to zero once satisfied, and the undercut guard responds to tooth count.

    On that reference case the effect is large. Calculated root safety factors went from 1.843 and 1.655 to 2.402 and 2.474, and calculated flank safety factors from 1.207 and 1.350 to 1.243 and 1.417. No variable ends the run pinned against its own limit, where two of the thirteen did before — and no limit was widened to achieve that.

    Expect accepted AI designs to come back with a lower helix angle and a larger normal module than before, and read the geometry before you commit to it. On that same reference case the tooth counts went from 34/136 to 30/120, the helix angle from 17.00° to 11.00°, the normal pressure angle from 28.00° to 25.00°, the normal module from 1.3374 mm to 1.5679 mm, and the axial contact ratio from 1.010 to 0.640. The design is valid: the total contact ratio — transverse plus axial — is 2.046 there, above the 2.0 at which a low axial contact ratio is acceptable for the rating. But it is a materially different kind of gear than the search used to choose, and whether a coarser, less helical pair suits your application is your decision, not the optimizer's. This was measured on one reference case with two seeds.

  • AI mode could price your design against a 17-tooth wheel that does not exist. The optimization criteria card lets you choose which criteria a run optimizes. If you unchecked both root safety factor criteria, the search never worked out your wheel's tooth count and silently fell back to 17, so the profile-shift geometry it priced belonged to a gear that appears nowhere in your design. At a design point with 35 pinion teeth and a ratio of 4 — a 140-tooth wheel — this overstated the undercut penalty by 4.86 times and the penalty steering the profile shift by 6.30 times. The wheel tooth count is now derived from your pinion tooth count and gear ratio directly, and a genuinely missing value stops the run with an error rather than substituting a number. If you leave at least one root safety factor criterion checked, which is the default, nothing changes for you — those results are identical, digit for digit.

  • On all-metal pairs rated to ISO 6336, the AI search's material-usage objective was switched off on exactly the cases that needed it most. Before counting the material a candidate design uses, the search checks whether that candidate can meet your safety targets at all. That check rated the candidate in the cheapest material compatible with its surface finish — but the run then upgrades the material in order to reach your targets, so the check was judging the design in a material the run had already decided to replace. On any case where that upgrade does real work, the check failed on every candidate: on our reference case the best calculated flank safety factor seen anywhere in training was 0.556 against a target of 1.2, across 5016 evaluations. With the check failing, the material-usage objective contributed nothing at all — neither a cost nor a direction to move in. The check now rates the same geometry in the strongest material allowed for that surface finish. Every number reported to you still comes from the material the design actually ships in; the stronger-material rating is used inside that check only, and nothing else in the run reads it.

    A second error in the same objective is corrected alongside it: a factor meant to penalise poor geometry was applied so that it discounted it instead, so a degenerate design was priced as though it used less material than a sound one.

    Measured on our all-metal verification case rated to ISO 6336: with the check alone corrected, Gear 1's facewidth came out 5.83% narrower — 18.1534 mm instead of 19.2776 mm — with higher calculated flank safety factors, 1.0557 and 1.1573 against 1.0239 and 1.1297. With both corrections in place the facewidth returned to 19.2750 mm, but on a sounder design: the axial contact ratio was 0.956 instead of 0.701. Expect different AI designs on all-metal ISO 6336 cases, and a longer run — this pair of corrections nearly doubled that case's run time, from 121 s to 236 s. Plastic gear ratings to VDI 2736 are not affected by either correction.

1.17 — 2026-09-15

Added

  • A new AI setting decides how much finishing cost counts against material cost when the optimizer chooses a material. "Finishing vs material cost weight" is on the AI Optimization tab of Settings, alongside the epoch and seed counts. It exists because of the selection change described below: the optimizer now compares candidate designs that differ in both material and surface finish, so it has to know how you weigh the two costs against each other. At the default of 1.0 they count equally. Raise it to steer the choice toward a coarser, cheaper finish on a stronger material; lower it to steer toward a finer finish on a cheaper material; set it to 0 to ignore finishing cost altogether. It has no effect on a calculation the optimizer does not choose a material for.

Fixed

  • AI mode, plastic gear ratings to VDI 2736: when the optimizer upgraded the metal gear's material mid-run, that gear was still rated on the properties of the material it started from, and its root safety factor came out too high. It is now lower. In a pair of one metal and one plastic gear, the AI optimizer may swap the metal gear to a stronger material part-way through a run in order to reach your root safety target. Until now, only the new material's allowable bending stress was carried into the rating. Everything else the rating reads about that gear's material — its material category, its hardness, its surface condition and its notch-sensitivity data — stayed pinned to the cheaper material the run had started from. So the gear was rated partly as a material it no longer was, and the life, notch-sensitivity and root-roughness factors were taken from the wrong material class. The complete property set is now carried across with the upgrade.

    The metal gear's reported root safety factor comes down. On the example case used to verify the fix, the value the run reported was 4.7% higher than the correct rating of that very same design. Your own figure will differ — how far depends on which two materials the run moved between, since the size of the error is the gap between their material classes. Re-run any AI design in this class that you have signed off, and work from the new numbers. A design that had comfortable margin is unlikely to move enough to matter; one that sat near its target may no longer meet it.

    The plastic gear in the pair is unaffected — its safety factors are identical before and after, confirmed on the verification case, because the plastic rating never reads the properties that were stale. Pairs in which both gears are plastic are unaffected for the same reason. All-metal pairs rated to ISO 6336 were already handled correctly and do not change. Manual and Geometry mode never choose or upgrade materials, so neither is affected by this fix.

Changed

  • In AI mode, all-metal pairs rated to ISO 6336: the optimizer now keeps searching until the design actually meets your safety targets, instead of stopping at the first material that looks strong enough on raw stress alone. When a run needed a stronger material, it used to buy the cheapest material whose allowable stress merely covered the calculated stress. That is a bare ratio of about 1, and it ignores every factor the rating applies on top of that ratio — so the finished design could be handed to you sitting below the safety target you asked for, with nothing saying so. On our reference steel pair the run shipped a calculated flank safety for gear 1 of 1.0267, short of the 1.2 flank target it had been set.

    Zenomiq now rates each candidate in full and accepts only a pair that meets every target you have active. It works through the compatible surface-finish classes, rates a candidate material pair in each, and ships the cheapest one that passes — cost here being each gear's material cost weighted by its blank volume, plus a finishing cost weighted by the working flank area, the two balanced by the new setting above. On that same reference pair the shipped design moved from a through-hardened steel at 300 HB to a nitrided steel, the calculated flank safety of gear 1 rose from 1.0267 to 1.3388 and of gear 2 from 1.1287 to 1.3870, and gear 1 came out 5.9% narrower — a facewidth of 18.14 mm instead of 19.28 mm.

    Expect different material and surface-finish choices than before, and expect AI runs to take longer, because every compatible finish class is now trained and rated rather than only the first one that passes. If no material you allow can meet your targets, the run still completes and ships the rated pair with the most margin it could find — the same outcome you get today, not a refusal. It does not yet tell you that it fell short, so keep checking the calculated safety factors against your targets.

    No rating formula, factor or default changed here: the same rating is called, on more candidates. Plastic gear ratings to VDI 2736 keep the behaviour they have today, and Manual and Geometry mode never choose materials, so neither is affected.

1.16 — 2026-09-13

Changed

  • The dynamic factor for ISO 6336 and DIN 3990 ratings is now computed with the standard's full excitation chain, and safety factors move for most gear qualities. Three corrections land together in how Zenomiq computes K_V, the dynamic factor that accounts for internal dynamic tooth loads: the tooth-passage excitation term is now resolved per accuracy grade as the standard prescribes, the running-in allowance uses the standard's own formula, and a related stiffness constant was corrected. If your gears are accuracy grade 6 to 11 — which covers most industrial gearing — expect the dynamic factor and the safety factors downstream of it to shift; the direction and size depend on the design and speed regime. The corrected chain has been verified against independent reference calculations at healthy line loads, where it agrees with the reference to within a few percent; the previous values were the ones out of line with the standard. If a signed-off design sits close to its target safety factor, re-run it and work from the new numbers. Plastic gear ratings to VDI 2736 are unaffected.

1.15 — 2026-09-11

Fixed

  • The "Auto tip shortening" switch is now honored — switching it off actually switches it off. Until now, turning the switch off had no effect for calculations submitted from the main form, in any mode: the setting never reached the calculation, and automatic tip shortening was applied regardless. It is now read and honored end-to-end. If you calculate with the switch off, your results will change — without the automatic shortening, the tip alterations, the transverse contact ratio and the safety factors all move to reflect the geometry you actually asked for. Calculations with the switch on, which is the default, are unchanged. If you switched it off in the past and signed off a design, re-run it: that earlier result was computed with shortening applied against your instruction.

1.14 — 2026-09-11

Fixed

  • Manual mode: with automatic tip shortening on, the reported contact ratio now reflects the tip shortening that was actually applied — and some inputs that were wrongly refused now calculate. When automatic tip shortening is active, Zenomiq derives the tip alteration itself, and the safety factors have always been computed from that derived value. The transverse contact ratio shown with the results, however, was computed from the tip alteration as you typed it, so it could disagree with the tip diameters printed in the same result. Left at the default of 0, it showed a value that was too high; typed strongly negative, it showed one that was too low — low enough, in some cases, to refuse the calculation outright even though the actual applied geometry was fine. Both directions are now repaired: the shown contact ratio is recomputed from the tips as applied, and previously refused inputs complete normally.

    This cuts both ways, so expect some results to change their messages. A design that previously showed no message may now — correctly — carry the low-contact-ratio advisory, or be marked not standard-compliant when the true contact ratio is below 1.0. The message is right; the earlier silence or refusal was the defect. If you signed off a design that uses automatic tip shortening and sits near the contact-ratio limits, re-run it and read the messages. Calculations where automatic tip shortening never fires are unchanged, and AI and Geometry modes already reported the correct value.

1.13 — 2026-09-11

Added

  • You can now decide how patient the AI optimizer is before it stops a run early. When a run stops improving, Zenomiq ends it early instead of spending the remaining epochs. The number of recent epochs it examines to decide that — the early-stop window — was fixed at 5 for every account. It is now a setting of its own on the AI Optimization tab of Settings, next to the epoch and seed counts. The default stays 5, so nothing changes unless you change it. Raise it to let long runs sit through flat stretches before giving up, or set it to 0 to switch early stopping off entirely and always run the full epoch count. Values up to 10000 are accepted, matching the maximum epoch count.

Fixed

  • In AI mode, the "Use default values" switches set by a CSV import now survive a page reload. Importing a case file flips those switches to match the file, but until now the flips lasted only until you reloaded or reopened the page — the switches then quietly went back to how they were before the import, and it was easy to keep working without noticing. They now stay where the file put them, exactly as if you had clicked them yourself. If you rely on an imported case, you no longer need to re-check the switches after a reload; files exported before this change are unaffected on export and simply import correctly now.

Changed

  • Gears calculated with the rim thickness left at its default were treated as thin-rimmed, and their safety factors came out too high. They are now lower. A gear cut into a thick, solid disc is stiffer than the same gear cut into a thin rim, and Zenomiq uses that stiffness when it works out how the load is shared across the teeth and how the mesh behaves at speed. Until now, leaving the Rim thickness fields untouched made the app treat the blank as a thin rim. A more flexible blank evens out the load sharing, which raises the reported safety factors — so the numbers you were given were optimistic. A gear whose rim thickness you have not stated is now rated as a solid blank, which is the correct treatment, and the safety factors come out lower.

    This affects every calculation in which you did not type a rim thickness yourself. That is the app's default, so in practice it is most calculations, not a rare corner. On one example pair the root bending safety factor of gear 1 fell by 4.6% in Manual mode and by 10.1% in AI mode, and the pitting safety factor fell by 2.5% and 7.0%. Your own figures will differ — how far depends on the design.

    The old numbers were too generous; the new ones are not too harsh. Re-run a saved calculation and it will show less margin than the saved record does. The new figure is the one to work from. If you have signed off a design that only just cleared its target safety factor, re-run it and check that it still does. A design with comfortable margin is unlikely to move enough to matter; one that sat right on its target may.

    If you entered a rim thickness, nothing changes at all — those results are identical to before. The change applies to ratings to ISO 6336 and to DIN 3990. Plastic gear ratings to VDI 2736 are unaffected, Geometry mode is unaffected because it reports no safety factors, and if you supply your own K-factors instead of letting Zenomiq calculate them your results are unchanged.

1.11 — 2026-08-11

Fixed

  • Changing your password from the Profile page now works. The Change Password fields on that page had no effect: filling them in and pressing the button only showed a "coming soon" message, and the password was never changed. The block is now a form of its own with its own button — enter your current password, then the new one twice, and the change takes effect immediately. The new password must be at least 8 characters, the same rule that applies when you register, and it has to differ from your current one. If your current password does not match, the page says so and nothing is changed; after five wrong attempts in a row you have to wait five minutes before trying again. Accounts that sign in with Google have no password of their own — use the "Forgot password" link on the sign-in page to set one. The rest of that card — first name, last name, email, username and company — still does not save. Those fields are being worked on separately.

1.10 — 2026-08-11

Added

  • Manual mode now tells you when the transverse contact ratio is too low for the rating to apply. Until now Manual mode returned safety factors for any contact ratio without comment. Two levels are reported. Below 1.0, the ISO 6336 formulae do not apply at all, so the result is marked infeasible and the safety factors shown are not standard-compliant — the geometry is still calculated and exported so you can see what went wrong. Between 1.0 and 1.1 the rating is valid, but the pair has little tooth overlap and the result carries an advisory. Consider raising the tooth counts, reducing the normal pressure angle, or increasing the profile shifts if you see either message (ISO 6336-1:2019 §1, Scope).
  • The contact ratio floor used by the AI optimizer is now yours to set, as "ε_α validity floor" on the AI Optimization tab of Settings. It defaults to 1.1 and can be set as low as 0.7 for special-purpose gears that deliberately run with little overlap. Values below the limit are corrected to it and the app tells you when that happens. Set it below 1.0 and Zenomiq will still optimise and report results, but the safety factors for such designs are not standard-compliant.

Changed

  • The AI optimizer no longer refuses designs with a contact ratio between 1.0 and 1.1. Its floor was fixed at 1.1 and, because of how the limit was applied, designs were in practice pushed above roughly 1.155. The floor now sits where you set it, defaulting to 1.1 and adjustable down to 0.7, and it applies the same way under every rating standard rather than only for plastic gears. If you have run AI optimizations before, the same inputs may now return a design with a slightly lower contact ratio than they used to. The previous 1.1 limit was a Zenomiq margin, not a requirement: ISO 6336 validates its method for transverse contact ratios "from 1,0 to 2,5" and states the formulae are not applicable below 1,0 — there is no 1,1 anywhere in the standard, and DIN 3990 sets no contact ratio limit at all (ISO 6336-1:2019 §1, Scope).

1.9 — 2026-08-10

Fixed

  • Importing a CSV no longer reports settings changes that did not happen. When a file carries settings values that already match your account — as the downloadable template does, since it lists them at their defaults — the import announced them as updates anyway. It now says nothing unless a setting actually changed.
  • When settings do change on import, the message is easier to act on. It names the settings that changed, links straight to them on the Settings page, and stays on screen until you close it instead of fading after a few seconds.

1.8 — 2026-08-07

Changed

  • Eight stored material strength values were corrected to ISO 6336-5:2016, Table 1. During a verification pass against the standard document itself, eight allowable-stress values across seven steel and cast-iron entries were found not to match any line of the table and have been corrected. The most visible changes: for C45 through-hardened (186 HB, MQ), the allowable contact stress rises from 455 to 540.4 N/mm² — pitting safety factors for this material increase by about 16%, while bending safety factors decrease slightly (about 2%), because this value also feeds the standard's running-in estimate. For grey cast iron the allowable bending stress drops from 107 to 61.8 N/mm², and for black malleable cast iron from 211 to 159.8 N/mm² — bending safety factors for these two materials decrease accordingly, and a design that previously met its target may now fall short. The corrected values are what the standard supports; the previous ones overstated the material. Smaller corrections: pearlitic nodular cast iron contact 590 → 569.5, nitrocarburized contact 1000 → 950, ferritic nodular bending 175 → 182, nitrided bending 370 → 420, and C45 bending 208 → 209.8. Re-running a saved calculation uses the corrected values. If you have released designs in grey or malleable cast iron rated for bending, we recommend re-running them (ISO 6336-5:2016, Table 1).

1.7 — 2026-08-06

Added

  • Scope warnings after a calculation, in Manual, Geometry and AI mode. ISO 6336 states the conditions under which its rating applies at all. When a design falls outside them — a contact ratio below 1.0, interference between one gear's tooth root and the mating tip, or a tooth that has come to a point — the results now say so, instead of returning safety factors with no comment. Conditions the standard asks to be confirmed by experience rather than taken from the calculation alone — a normal pressure angle outside 15°–25°, a helix angle above 30°, a transverse contact ratio above 2.5 — appear as softer advisories. Which checks apply depends on what the mode calculates. The calculated numbers themselves are unchanged. What is new is that the app now tells you when the standard says they need extra care.

Changed

  • The two root-form interference values changed meaning, in results and in exports. One per gear, they now report the plain signed distance in millimetres between the mating gear's tip circle and the root form circle: negative means clearance and no interference, positive means real interference. They previously carried an internal quantity that compressed the magnitude and was not a measurement — on one example pair the value now reads −0.1996 mm where it read −0.0948 mm before, for exactly the same gears. If a script or spreadsheet of yours reads these two columns, update it. The number is now directly comparable with a check taken off a drawing.
  • The undercut warning in Geometry mode now uses the correct helical form of the minimum tooth count, built from the transverse pressure angle and a helix-angle factor (ISO 21771:2007 §7.7). The old form ignored the helix and over-estimated the limit on helical gears — by about 19% at a 20° helix angle and by nearly 50% at 30°. Some helical designs that used to show an undercut warning will no longer show one: a 15-tooth gear at a 20° helix angle with no profile shift was warned against a limit of 17.10 teeth, where the correct limit is 14.41. The suggested minimum profile shift in the warning text drops with it. Spur gears are unaffected, and no calculated geometry value changes — only the warning.
  • Reloading a saved calculation now uses the settings that were saved with it. Your settings are stored with each calculation you save. Load that calculation later and both the recalculation and your Settings page return to the values in force when it was saved, so the saved record and a fresh run of it agree instead of drifting apart as your settings change. Calculations saved before this release carry no stored settings: they load exactly as they did before and leave your Settings untouched.

1.6 — 2026-08-06

Added

  • Default-value toggles in AI mode. Each parameter card (criteria, optimizable parameters, constants, targets, weights) now has a "Use default values" switch, plus one master switch that controls all of them together. Leave a card on defaults and it stays out of your way; switch it off to take manual control of just that card.
  • Two optional compatibility settings, on the General tab of Settings: a lower limit on the specific load used in the mesh-stiffness calculation, and a cap on the dynamic factor. Both reproduce conventions used by some other gear-calculation tools so results are easier to compare. Both are off by default — with them off, calculations follow ISO 6336 exactly as written. Turning one on is your explicit choice and applies to both Manual and AI calculations.
  • Importing a CSV that contains settings parameters now updates the Settings page to match, so the file you import and the settings you see can no longer disagree. Settings are stored when you press Save, and each Settings tab has its own "Reset defaults" button.

Changed

  • The free plan now includes more. 3D model generation (15 per day) and CSV export are part of the free plan, alongside geometry calculations (50 per day) and manual calculations (10 per day, 40 per month). STEP file download and project management remain part of the paid plans — the app now says so clearly instead of failing with an error code.

Fixed

  • Choice-type parameters could appear in CSV exports as an internal number instead of the chosen option — for example, the K-factor method showing "0" instead of "C". Exports now show the option you actually selected.

1.5.1 — 2026-08-06

Fixed

  • Downloading the STEP file on the free plan now explains itself. STEP export is part of the paid plans; clicking the download button used to fail with an unhelpful "Export failed (403)". It now shows a clear message saying the feature is not in your current plan and how to upgrade, and the button carries a small lock icon so you can see it before clicking. Paid plans are unaffected — the download works exactly as before.
  • One of the CSV export routes always failed. Exporting a calculation to CSV through the programmatic interface (the route scripts and integrations use, not the in-app export buttons) returned an error for every user in every mode. It now returns the CSV file. Rows for parameters the optimizer varies are labelled "Variable", as intended.

1.5 — 2026-07-28

Added

  • Running parameters preview. Gear 1 speed, both torques and Gear 1 load cycles are shown in the input card and update as you type the power, the wheel speed, the ratio or the wheel load cycles. Available in AI and Manual mode.
  • One Settings page instead of two. Every settings entry in the sidebar, including the new Geometry entry, opens the same page, which is split into a General tab and an AI Optimization tab.
  • A choice of column separator for exported CSV files, on the General tab of Settings. Comma or semicolon, remembered per account.
  • Help bubbles on the K-factor method row and on the three shaft inputs (shaft diameter, bearing span and pinion offset), explaining what each input means and where its default comes from.
  • A warning after an ISO 6336 calculation when the mesh is running at very light load or close to mesh resonance. It reports the dynamic factor and the specific load that triggered it, and lists ways to move the design out of that regime. It is advisory only and does not stop the calculation.

Changed

  • Lightly loaded gears: the dynamic factor changed, and with it the root and flank safety factors. In an ISO 6336 rating, the specific load that divides the dynamic-factor terms is now held at a lower limit of 25 N/mm. Below that load the old formula had no lower bound, so the dynamic factor grew without limit as the load fell and reached values far outside anything physical. The new lower limit brings the dynamic factor down, which raises the safety factors.

    A design carrying 25 N/mm or more of specific load is completely unaffected — its numbers are unchanged. Specific load here means the tangential force multiplied by the application factor and divided by the face width. If you are not sure where your design sits, re-run it — only lightly loaded gears moved.

  • CSV files exported from a saved calculation now use a semicolon between columns, matching the other two export routes. This one route used a comma before. You can change it in Settings.

  • The CSV export and the on-screen results summary now hide the same rows the PDF report already hid. With tooth-tip topping switched off, the two topping tool-radius rows no longer appear; with it switched on, the tip-shortening rows no longer appear. Those rows only ever showed zero. No calculated value changes.

  • Grey helper text throughout the app is darker and easier to read, and the descriptions under the settings fields are larger.

  • The Running Parameters card now sits directly below Design Settings in AI and Manual mode.

  • The permanent yellow note on the K-factor method C row is gone. The same explanation is now in that row's help bubble, where it does not take up space on every screen.

Fixed

  • The left-hand / right-hand choice now really flips the helix. Choosing left-hand used to produce a 3D preview and a STEP export identical to right-hand. The rating results are unaffected, because the strength calculation only uses the helix angle through its cosine.
  • A saved calculation now reloads the material and lubricant, the ISO 53 basic rack profile, the K-factor method and the Basic/Pro tier exactly as they were saved. All four were silently dropped, so loading an older case and recalculating it could give different safety factors from the ones on the saved record.
  • Saving a calculation no longer throws away the working state. Exporting the CSV or the STEP file after a save works again. When the state really is gone, the export now says so and stops, instead of quietly falling back to generic defaults and writing a file that looks plausible but does not describe the calculation on screen.
  • In Manual mode with K-factor method C selected, the five manual K-factor input rows no longer appear on the ISO 6336 card. Method C computes those factors, so anything typed into them was ignored.
  • Switching into AI mode no longer overwrites values you have typed with the defaults.
  • The ISO 6336 card no longer jumps on load. Rows that are meant to be hidden now start hidden instead of appearing for about two seconds first.
  • Warnings that a value was rounded or limited now appear straight after the calculation. They used to sit unseen until the next time the page was fully reloaded.
  • The material table in the ISO 6336 help bubble is no longer cut off. All four columns fit.
  • The 3D result and gear-selection panels now appear and disappear correctly in several cases where they did not: a leftover single-gear result no longer hides the pair view, and the selection frame no longer shows when there is no pair result to select from.