What's New
What's New in the Unified Fracture Mechanics Tool
Version 5.9.30 — what has changed since Version 5.9.11.
This is a correctness release, and an unusual one. Version 5.9.11 passed all 87 of its verification checks. Then an independent, line-by-line review compared the tool with the handbook solutions instead of with its own tests — and found that several calculations had never been tested against anything outside the program, and that some of them were wrong. Every one has been replaced, and the verification suite now reaches outside the code. If you use an earlier version, please replace it: some of its answers were on the unsafe side.
And then it happened again — which is the whole point of checking. While I was working without Claude, I asked ChatGPT's Codex to review Version 5.9.19. It reported thirteen problems. Before accepting any of them I had Claude check each one independently: twenty-five separate reviews and seven deliberate attempts to prove the findings wrong. Not one was rejected. Several of Version 5.9.19's answers were on the unsafe side — one of them by a factor of more than eight hundred. Version 5.9.20 corrects all thirteen, repairs the problems the checking found in Codex's own fixes, and adds new checks against published data. Section 1 lists what changes. If you use Version 5.9.19 or earlier, please replace it.
By the numbers
Review findings: eleven in the first review, thirteen in the second — all closed.
Verification checks: 87 → 402, all passing. Most of the new ones compare the tool with limits, tabulated values, second published solutions or an independent integration. Some only check that two parts of the program agree, or that a rule is obeyed. An earlier version of this page said the new checks never compared the tool with itself; that was not true, and since 5.9.20 every new check says in its name which kind it is.
Hand-calculation walkthroughs: 87 → 402 cases, with every revised case showing its old value.
User manual: rebuilt for v5.9.30, every screenshot regenerated, with a new chapter for the Y-Factor Lab.
Crack types: seven → eight — the elliptical corner crack is new.
Learning Center lessons: 23 → 25, now searchable.
A new tab: the Y-Factor Lab.
New in Version 5.9.30 — easier on the eyes
Two small things you will see at once. The Fatigue tab is now simply "3. Fatigue", so all eleven tab titles fit across the window again. And if you work on one of the dark themes — Dark, Slate or High Contrast — the titles, axis labels and numbers around every plot are now light on the dark background instead of dark on dark; a plot you tag for a report is still saved on white with dark text, so it prints the way it always did. No calculation changed.
New in Version 5.9.29 — the Y-Factor Lab
Every stress intensity in this tool is K = Y·σ·√(πa), and Y is the part that students, and engineers, most often get wrong. Version 5.9.29 adds a Y-Factor Lab, a new tab where you pick a crack geometry, drag sliders for the crack size, the width and the thickness, and watch Y, its curve and a to-scale drawing of the part follow as you drag. For a surface or corner crack it also shows Y along the crack front, because the deepest point and the surface point are different numbers and either one can govern; for a through crack it shows all four through-crack types side by side at the same crack size and width. The Lab uses the same solutions as every other tab and adds no formula of its own. It shades the range where a solution is extrapolated, stops at the physical limit, and never quietly changes a number you type. Turn on the comparison curves and it draws the classic shortcuts beside the real solution: the Brown–Srawley polynomial falling away past a/W = 0.6 (25 percent low by a/W = 0.8), the center-crack secant misused for double-edge cracks (11 percent low for short cracks), a surface crack treated as a two-dimensional edge crack. A new Learning Center lesson, "The Geometry Factor Y — Watch It Move", explains why Y exists in terms of energy and ends in the Lab.
Two smaller things. The LEFM tab has a button that opens your Setup geometry in the Lab. And at zero stress, or in compression, the LEFM tab used to show "Y factor = 1.0000"; Y depends on the geometry alone, and it now shows the geometry's value at any load.
New in Version 5.9.28 — bending for cracks deeper than they are long
Since Version 5.9.20 the tool has told you plainly when it could not stand behind a bending result. For a crack deeper than it is long, it used the bending factor of a semicircular crack instead and flagged the run UNVALIDATED BENDING. The checking behind 5.9.20 showed that substitute was not conservative: for a crack twice as deep as it is long, it put the stress intensity up to 70 percent too high at the deepest point and 4 to 13 percent too low where the crack meets the surface. In 5.9.28 the tool uses Newman and Raju's own bending factors for these cracks, and I had the implementation checked against the curves printed in their report: it follows them to about one hundredth, where the old substitute missed by about a tenth. The warning now appears only beyond the range their equations cover, a crack more than twice as deep as it is long. In the examples we ran, fatigue lives changed by less than one percent, because a crack like this rounds out quickly as it grows; but at the point where such a crack meets the surface, the stress intensity in bending can be up to about 15 percent higher than earlier versions reported.
Two smaller changes. Where a surface or corner crack meets the free surface, the plastic-zone estimate and its small-scale-yielding screen now use plane stress, whatever you choose on the Setup tab — the stress through the thickness is zero at a free surface — which also makes that screen stricter, not looser. And the Luminaries lesson now reads the whole gallery as the two sides of the energy ledger, with three dates corrected against their sources.
New in Versions 5.9.21 to 5.9.27 — the Learning Center starts with why
When I tell engineers that I teach fracture mechanics, the answer I hear most often is: "Yes, I took that — but I couldn't tell you anything about it." That answer is why I teach the subject holistically, and now the Learning Center does the same. It no longer opens on a list of topics. It opens on the why: all roads lead back to energy, and fracture is one of the ways a system spends the energy we put into it; everything has flaws, and what matters is how a system tolerates them; and when a part fails anyway, the failure is the most honest teacher you will ever have. The list of what the lessons cover comes after that. Nothing else changes in 5.9.21 — every calculation, and all 371 verification checks, are exactly as in 5.9.20.
In 5.9.22 the first lesson, "Why Strength Isn't Enough," follows the same path. Before it shows you K, it asks where the stored energy goes: a crack grows when the energy it releases pays for the surface it creates, and a longer crack releases more — so at any stress, however far below yield, there is a crack size that pays for itself. Then K, and why K and energy are the same thing in two currencies. Then why toughness varies so much: from the library values, aluminum 2024-T3 spends about 3,100 times more energy on each square metre of new crack than soda-lime glass — not because its bonds are stronger, but because it can spend that energy on plastic flow first. And a new worked example: work 2024-T3 and a 4340 steel each at half their yield strength, and the steel — more than four times as strong, storing almost seven times the energy — tolerates a crack more than twelve times smaller. Same strength margin, very different damage tolerance, and the energy says why. Again, no calculation changes.
In 5.9.23 the Griffith lesson keeps the books the way Griffith did. Instead of asking how strong glass is, he asked when it becomes energetically favorable for a crack to grow — and the lesson now walks through his ledger: what the plate pays out as a crack relaxes it, what the two new faces cost, and why the payout wins once the crack is long enough. Glass, from the library values, pays only about 4 joules for each square metre of new surface; that is the 600-times paradox, answered in joules. Then the account extended: a metal also pays for plastic flow, and 2024-T3 aluminum spends some 3,300 times what the glass does on the same area of crack — the same ledger, a bigger bill. No calculation changes here either.
In 5.9.24 the Irwin lesson finishes that story. It is easy to leave a first course thinking that K replaced Griffith's energy. It did not. Irwin looked at the one place every joule a growing crack releases has to pass through — the crack tip — and showed that the energy released for each square metre of new crack is set by the strength of the field there: G = K²/E′. K is Griffith's energy, measured locally with ordinary stress analysis. The lesson reads its own Try It Live plate that way: at 100 MPa the crack in the 2024-T3 plate is offered about 2,180 joules for each square metre it could grow, against a bill of about 26,500. K is 29 percent of the toughness, but the energy on offer is only 8 percent, because energy goes as K squared. The plastic zone becomes the place that energy is spent instead of on new surface — and the lesson is honest that the demo's own zone, 0.43 mm on a 5 mm half-crack, is past the tool's strict small-scale-yielding screen, and that Irwin's correction moves K by about 2 percent there. I also corrected Irwin's biography: he was a physicist, with his doctorate from the University of Illinois in 1937, and at the Naval Research Laboratory from that year. No calculation changes.
In 5.9.25 the Paris lesson tells fatigue the same way. A fatigue crack is never offered enough energy in a single cycle to break the part: in the Try It Live plate each peak offers about 8 percent of what fracture costs. It grows anyway, because every cycle also spends a little energy. At the tip a small region yields forward and back, and that loss pays for new surface a little at a time — about a quarter of a micrometre per cycle at the start. Fatigue is Griffith's ledger paid in installments, and Paris's ΔK measures how large each installment is. The lesson reads the hockey stick in those terms: the crack spends 60 percent of its life just doubling from 5 to 10 mm, and the last 145 cycles take it the rest of the way, when a single peak finally pays in full. I also checked the history against the sources: three leading journals really did reject the 1961 paper, and two lines that had no source are gone. No calculation changes.
In 5.9.26 the Rice lesson completes the story. When a metal yields, Griffith's books still balance, but K can no longer read them: the region where the energy is spent has outgrown the elastic field that K describes. Rice's J-integral counts the energy flowing toward the crack tip on any path drawn around it, out where the material is still well behaved — and every path gives the same answer. For an elastic body J is exactly Griffith's energy release rate, G = K²/E′, and the lesson shows that on the exact crack-tip field to eight figures. It is also honest about what J means once a metal yields, and about where this tool stops: the demo's A36 plate is well past small-scale yielding, and the tool computes the elastic J and Irwin's correction, not a fully plastic J. I also corrected Rice's birthplace — Frederick, Maryland — and his work on rate-and-state friction, which he did with his student Andy Ruina, building on Dieterich's experiments. No calculation changes.
In 5.9.27 the Inglis lesson takes the story back to where it began. In 1913 Inglis worked out where the stress goes around a hole, and his answer holds its own paradox: as an ellipse sharpens into a crack, the stress at its tip grows without limit, so by a stress test every crack at any load should break its part at once. Yet cracked parts carry load every day. What stays finite is the energy — and Griffith took that energy from Inglis's own solution. The lesson also shows what a hole does to a crack that starts beside it: in the Try It Live plate the crack is offered about three times the energy it would get at a plain edge. I also corrected one line: a 10:1 slot concentrates stress seven times as much as a round hole, not twenty. With this release every Meet the Pioneers story about a single pioneer leads with energy. No calculation changes.
1. What changes in Version 5.9.20 — read this first
Most of these stop the tool from giving an answer it cannot justify. One makes lives shorter on purpose, by my decision.
Forman lives at a positive stress ratio are shorter — half at a ratio of 0.5, ten percent shorter at 0.1. Forman's equation has one material coefficient. When you give the tool a Paris constant, it converts it into that coefficient — and until now it did the conversion again at every stress ratio, which quietly cancelled the very mean-stress effect the equation exists to add. It now converts once, at the reference ratio you give ("measured at R =", zero unless you change it), as NASA's crack-growth handbook writes the equation. In one example a center crack that lasted 290,149 cycles at R = 0.1 now lasts 261,134, and 1,692,148 cycles at R = 0.5 become 846,074. At the reference ratio nothing changes. It is still a conversion from one measured Paris line, not a fit to data at several stress ratios; if your constant is a genuine Forman constant, set "Setup-tab C is" to "The model's own C" and it is used exactly as given.
A crack that has run out of plate is no longer called a survivor. A quarter-circular corner crack in a plate narrower than it is thick — 20 mm wide, 100 mm thick — grew past the edge of the plate, where the solution returns zero, and Version 5.9.19 reported that it survived a billion cycles, with a crack 20.04 mm long in a 20 mm plate. It actually runs out of width at 1,164,636 cycles: the old answer was 859 times too long. A semicircular crack in the same plate was allowed to grow to 40.9 mm long; it now stops at the plate's half-width.
Impossible geometry is refused. A center crack 110 mm long in a 100 mm plate used to return a stress intensity of zero, an infinite margin and "Stable under static load". It now says the input is invalid and gives no result.
Stress History, semicircular and quarter-circular cracks. For these two types the length always equals the depth, and the length box is greyed out — but the Stress History tab still read the greyed-out number and carried it into the break-through. In one example — the Setup tab's starting geometry under a 120-to-12 MPa history — the semicircular crack's life was 343,222 cycles; it is 413,783, which is what the Fatigue tab has always said. The quarter circle: 244,890 before, 286,552 now. With a stale length smaller than the depth — possible from an old saved session — the old lives were too long, by as much as 56 percent. The greyed-out box now follows the depth, on screen and in saved sessions.
Subcritical growth sees a short load pulse. The integration could step straight over a brief overload. A center crack held at 100 MPa for 1000 seconds, with a spike to 500 MPa lasting two milliseconds just after five seconds, was reported as surviving; it fractures at 5.0013 seconds. A load that just touches the toughness at one point of the profile is caught too.
Subcritical growth checks the whole crack front. It used to check only the deepest point. A semicircular crack 7.5 mm deep in a 10 mm plate at 100 MPa, with a toughness of 14, was reported as surviving — its surface point is at 14.20 from the first instant. Under a constant strain rate the fracture stress was overstated by 30 percent: 127.8 MPa against 98.6.
Bending that turns the cycle upside down at the deepest point. For a deep crack under bending, the deepest point can see its largest stress intensity at the bottom of the load cycle. Version 5.9.19 then computed a negative range and reported no growth at all. The two ends of the local cycle are now put in order first.
The whole crack front, not just its two ends. Under bending, the stress intensity of a long shallow crack can peak between the surface point and the deepest point. In one example the peak is 20.49 against 19.80 at the larger end — three and a half percent higher. Fracture is now checked at the peak.
Stress History lands on the failure. It used to stop at the first whole cycle past the event — in one short example at 9 cycles, with the crack 3.1 percent past its critical size, where the exact answer is 8.35. It now lands on the event itself, and its lives agree with the Fatigue tab to about a thousandth of a percent.
Plane-stress HRR at a hardening exponent of one borrowed the table row for two and ran 28.6 percent high. It now uses the exact limit.
"Leak before break" is now said carefully. At break-through the tool checks whether the through crack is stable at that instant. That alone is not a leak-before-break case, which also needs the leak rate, how the leak will be found and the time to act — none of which the tool calculates. The result now reads "STABLE at transition under this model; LBB not established".
Deep corner cracks in bending are flagged. For a crack deeper than it is long, the bending factor is still taken at equal depth and length, and the result now carries an UNVALIDATED BENDING warning — ahead of the life, on both plots and in the status bar. That is more than a formality: the checking behind this release found the assumption is not conservative. At twice as deep as long it runs up to 70 percent high at the deepest point and 4 to 13 percent low at the surface point. Putting in the published equations for that range is the next job. The reason it had been put off turned out to be an error in the published table, not in the equations.
The double-edge crack's stress field is no longer drawn. It was drawn with a single tip at the crack mouth, which is wrong. Its stress intensity and fatigue results are unaffected.
A saved session keeps everything. It now stores the settings of every tab, not only the Setup tab; an older file loads with a warning that lists what it did not contain. An exported report no longer includes a result that is out of date with the inputs.
2. New in Version 5.9.20
Reports. Every plot has a "Tag for report" button. A tagged figure keeps its image and the inputs it was made with, even if you change the setup afterwards. File, then Report Figures / PDF, collects them with their captions, lets you reorder them and add an image of your own, and exports a PDF. Saving the session keeps the collection.
A plastic zone you can measure. The Stress Fields tab now draws the elastic first-yield outline from von Mises — the familiar kidney shape — or the Irwin circle, and a new window gives its dimensions in millimetres on a dimensioned drawing. It is an elastic estimate, not an elastic-plastic solution, and it says so. In plane strain at a Poisson ratio of one half the forward radius is exactly zero, because directly ahead of the tip the stress is hydrostatic: that is physics, and the tool now explains it.
Search the Learning Center by keyword. And a twenty-fourth lesson, Leak Before Break — Local Failure, System Survival, which connects a crack that leaks before it breaks with fuses and sacrificial parts — and is careful about where the analogy stops.
Verification, explained first. Help, then Verification Tests — Learn More / Run, says what the checks do and do not prove before anything runs; choose Run and all 371 run on your machine, in about a minute.
How this version was checked. The thirteen findings came from Codex, and so did the fixes; I took neither on trust. Claude reproduced every finding on Version 5.9.19 and checked every fix — and found problems in the fixes that Codex's own 334 checks had not seen. The most striking: entering a Poisson ratio of one half emptied the whole Stress Fields tab. Every one is repaired, and each repair has a check that fails if the repair is undone: I had every fix deliberately broken, one at a time, and the checks caught all sixteen. The new checks reach outside the program where they can — including the first published check of the corner-crack solution for cracks deeper than they are long, against a 1981 NASA finite-element table — and each says in its name whether it rests on published data, a closed form, agreement between two parts of the program, or a rule. One honest limit, which the check itself prints: the published finite-element table for corner cracks in bending cannot tell the published shape of the stress-intensity curve between the two ends of the front from a simpler one, so that shape rests on the published equation alone.
Speed. The Stress History tab now checks the whole crack front at both ends of every cycle, and takes a little over twice as long as Version 5.9.19 for the same history. Codex's version took about twenty-five times as long; that was repaired without changing a single result.
3. Answers that changed in Versions 5.9.12 to 5.9.19
Cracks at a hole. The old solution ignored the hole radius altogether and under-predicted the stress intensity of a short crack at a hole by up to three times. The tool now uses Bowie's solution: a short crack at a hole sees the hole's stress concentration of three, and a long one behaves like a single crack that has swallowed the hole.
Forman and NASGRO fatigue lives. The materials library stores Paris-law constants. Earlier versions fed that Paris constant straight into every growth model — but a Forman constant and a NASGRO constant are different numbers. Forman grew cracks about fifty times too slowly. Lives from those two models are now much shorter, and correct. (At a positive stress ratio Forman lives are shorter again in 5.9.20 — see section 1.) A new control on the Fatigue tab, "Setup-tab C is", lets you say which kind of constant you have.
Double-edge cracks were using the center-crack formula, which has no free-surface effect: eleven percent low for short cracks. Now the handbook solution.
Surface and corner cracks now use the complete Newman-Raju equations. The old version used only their first term plus a correction of its own, which ran twelve percent low for long shallow cracks and far too high for deep ones. The tool also now checks the point where the crack front meets the surface — for a semi-circular crack that point, not the deepest one, reaches the toughness first.
Deep edge cracks. The old formula was only valid to sixty percent of the width and was silently used beyond it — sixty percent low at ninety percent depth. The replacement is accurate at any depth.
Fatigue life no longer depends on the "Max cycles" box. The old integration rounded the life up to a grid of one-thousandth of that setting: a quarter too long when the cap was set generously. The life is now the integral itself, to within a few cycles of the closed-form answer, whatever the cap.
Corner cracks in bending. The Stress History tab was scaling the bending stress at the depth end of a corner crack with the surface-crack factor. Newman and Raju give corner cracks their own, and against their finite-element table the old one ran ten to fourteen hundredths low for a deep crack — the unsafe direction. Now corrected, and checked against that table.
The Fatigue tab and the Stress History tab now agree. The same problem could give two different lives — nearly eight times apart for fully reversed loading. Both tabs now share one definition of a cycle and one way of growing an elliptical crack.
4. Surface cracks now change shape as they grow
A surface crack has two dimensions — how deep it is and how long it runs along the surface — and the stress intensity is different at its deepest point and where its front meets the surface. So the two grow at different rates. Earlier versions held the shape fixed. The tool now grows depth and length separately, the way Newman and Raju laid out in 1981: depth from the stress intensity at the deepest point, length from the stress intensity at the surface, slowed slightly because the crack closes more where it meets the free surface. That slowing factor is 0.9 at zero stress ratio and, as their 1984 work prescribes, fades toward one as the stress ratio rises and closure disappears.
In tension, a long shallow crack rounds up. It drifts toward a semicircle and lives more than twice as long as the fixed-shape calculation said — about 13,000 cycles against 6,000 in one example — and it breaks through the wall rather than fracturing first. A crack that starts as a semicircle barely changes.
In bending, the opposite happens — and this one matters for safety. The deepest point loses the bending stress as it moves toward the middle of the plate, and stalls. The surface point keeps it. The crack runs along the surface. In one example the fixed-shape calculation showed no failure in twenty million cycles; the crack actually fails at about 64,000.
You can see it. The crack-size plot on the Fatigue tab and the Stress History tab now draws both curves, depth and half-length, and the results list the final shape.
You can still hold the shape fixed — one setting on the Fatigue tab, which the Stress History tab follows — to reproduce earlier numbers or to compare.
It is verified from outside the code. The shape path and the life agree with an independent numerical integration of the same equations to a thousandth of a percent, and a shallow crack in tension settles at exactly the aspect ratio the equations predict in closed form — just over a semicircle.
One honest limit: the two crack types that are semicircular or quarter-circular by definition keep their shape — choose the elliptical type with equal depth and length to let the shape evolve.
5. Corner cracks change shape too
There is a new crack type on the Setup tab: the elliptical corner crack, with its own depth down the side face and its own length along the top surface. It grows in both directions separately, in both fatigue tabs, just as the surface crack does — with one difference, which was settled from the source before a line of code was written.
The 0.9 factor applies at both ends. A surface crack meets a free surface only where it breaks the top face. A corner crack meets one at both ends of its front. Newman and Raju's 1984 NASA report on crack-growth patterns says it in so many words: the factor is applied at every point where a crack intersects a free surface, and for a corner crack that is both points. The factor then cancels out of the shape and simply lengthens the life — by a little over a third when the Paris exponent is three.
A corner crack in a thick plate tends to a quarter-circle. In the shallow limit the equations reduce to something you can do on paper: the quantity c to the power two and a half, minus a to the same power, stays constant as the crack grows. The tool follows that curve to six figures. It is also the justification for having a quarter-circular type at all.
In a real plate it flattens. As the crack deepens, the back face drives the surface end harder than the depth end, and the length runs ahead. A crack 2 mm deep and 3 mm long in a 10 mm plate reaches the back face 11.5 mm long — and carries on as an edge crack of that length, with the verdict on whether that edge crack is stable.
Holding the shape fixed is short again. In one example the fixed-shape calculation predicts fracture at 324,000 cycles where the crack, allowed to find its own shape, lasts 494,000.
It is verified from outside the code. Against the closed-form shape curve, the closed-form Paris life of a quarter-circle, an independent numerical integration, the handbook edge-crack solution at the hand-over, and Newman and Raju's own finite-element table for bending. Before release I had each engine change reversed in a scratch copy to make sure the new checks catch it. They do — thirteen of the twenty-three fail.
The quarter-circular corner crack takes the factor too. Its depth end lies on a free surface just as the elliptical one's does, so its growth rate now carries the same 0.9 factor, in both fatigue tabs. In the shallow limit, where the quarter-circle is the shape a corner crack settles into anyway, the two crack types now give the same life to a thousandth of a percent — before, they differed by more than a third. If you ran the quarter-circular corner crack in Version 5.9.17 or earlier, its fatigue lives are now longer — by about 37 percent when the Paris exponent is three. Enter 1 for the free-surface factor on the Fatigue tab to reproduce the earlier numbers.
The 0.9 now depends on the stress ratio. The factor exists because a crack closes more where it meets a free surface — and closure fades as the stress ratio rises. Newman and Raju's 1984 fit takes it from 0.9 at a stress ratio of zero to 0.94 at one half and to one at the top, and the tool now follows it, at each end of the crack front and for each range of a stress history. For surface cracks as well as corner cracks. Lives at positive stress ratio are a little shorter than before — under one percent at a ratio of 0.1, about thirteen percent for a corner crack at one half — and a surface crack at high stress ratio grows a little longer for its depth. At zero or negative stress ratio nothing changes. One tick-box on the Fatigue tab holds the factor constant, as before. It is checked against the closure measurements the fit was made to, against the shape a shallow crack settles into — which the equations give in closed form — and against an independent integration.
The honest limits. The quarter-circular type keeps its shape, and a real corner crack in a plate of finite thickness does not: it flattens, reaches the back face a little sooner, and hands a longer crack to the edge-crack phase. In one example the quarter-circular type breaks through about six percent later than the elliptical type started at the same size, and lasts about six percent longer overall. The earlier versions, without the factor, erred on the short side; this one does not — where the difference matters, use the elliptical type. The rise of the factor with stress ratio is a smooth fit that sits up to four percent from the closure measurements behind it, and those were made on one aluminum alloy. And for a corner crack deeper than it is long, the bending factors are taken at equal depth and length. The published equations for that range had not been used because they disagreed with the published finite-element table; in 5.9.20 the table was shown to be the one in error, and the equal-depth assumption was shown not to be conservative — see section 1. Until the equations go in, such results carry an UNVALIDATED BENDING warning.
6. Break-through, and whether the through crack holds
Until now, a surface crack that grew to the back face simply stopped there, and the tool reported that the crack had run out of material. A real part does not stop. The crack breaks through and carries on as a through crack — and whether it does so quietly or catastrophically is one of the most practical questions in damage tolerance.
The run now continues. At the back face a surface crack carries on as a through crack of the same surface length, and a corner crack as an edge crack. That is the standard hand-over, and it is conservative. Growth continues until the stress intensity reaches the toughness or the crack runs out of width.
You get a stability verdict at break-through. The results give the cycles at break-through, the crack length there, and the through-crack stress intensity — and say plainly whether that through crack is stable at that instant. Unstable means it fractures at break-through. Stable is necessary for leak before break, but it is not the whole case: that also needs the leak rate, how the leak will be found and the time to act, which the tool does not calculate. Since 5.9.20 the result says so: "STABLE at transition under this model; LBB not established".
Read the margin, not just the verdict. In one example the through crack at break-through sits within two percent of the toughness: two of three starting shapes are still stable when they break through — with fewer than thirty cycles left — and the third fractures at break-through. In another, a fifth of the crack's whole life comes after it has broken through.
Both tabs, one switch. The Fatigue tab and the Stress History tab make the same hand-over; a checkbox on the Fatigue tab turns it off if you want the run to stop at the wall.
Verified against the oldest closed form in the book. The through phase reproduces the closed-form Paris life to the cycle.
The honest limits: the hand-over is instant, where a real crack's back-face length takes a little while to catch up with the front; and in bending the through phase applies the full surface stress across the thickness, which is conservative and can call a crack unstable when a finer model would not.
7. The elastic-plastic tab tells you more, and more honestly
The HRR constants were wrong. When no citable table could be reached to settle the question, I had them recomputed from the governing equations themselves — a nonlinear eigenvalue problem, one solution for every hardening exponent. The results agree to about a tenth of a percent with the values of Hutchinson and Shih as I have them — those published figures are still to be checked against the printed tables — and plane stress now has its own table instead of borrowing the plane-strain one — and, since 5.9.20, its exact value at a hardening exponent of one, where it had borrowed the row for two.
The famous plot is now like for like. It compares the opening stress ahead of the crack tip computed both ways. The curves cross well inside the plastic zone, where the opening stress is about three times yield in plane strain — the classic constraint result, now visible in the tool.
A better J under moderate yielding. Beside the elastic J the tab reports the Irwin plastic-zone-corrected value.
An honest verdict. When small-scale yielding fails, the tab now says plainly that the problem needs a fully plastic J analysis, and that this tool does not perform one.
8. Stress-field plots you can trust closer to the crack
Straight cracks are now drawn from the exact Westergaard solution. The crack faces are stress-free, as they must be; the field far away is the applied stress; and half a crack length ahead of the tip the opening stress reads 115 percent of the applied stress — the exact answer — where the old picture showed 170.
9. Smaller things you will notice
Range notes on the LEFM tab. When your geometry lies outside the range a solution was fitted for, a line beginning "Range:" says so. The number is still computed; the note tells you how far to trust it.
Surface cracks report two stress intensities — deepest point and surface point — and which one governs.
The Learning Center tells the story. The NASGRO lesson and the Assumptions lesson were rewritten, including a plain account of what went wrong and what the review taught: a passing test suite proves only that the code agrees with its tests.
Verification is one click away, as always. Help, then Verification Tests — Learn More / Run, runs all 371 checks on your machine with nothing to install.
Unified Fracture Mechanics Tool v5.9.30 · Joseph P. McFadden Sr. · www.mcfaddencae.com. Library values are representative screening data for education and preliminary analysis — not certified design allowables.
