Hydrogen Embrittlement: The 1-Hour Clock, the 22-Hour Bake, and the Parts-Per-Million That Snap a 12.9 Bolt

Engineering guide to hydrogen embrittlement in high-strength fasteners: plating hydrogen sources, ASTM B850 bake tables, ISO 4042 triggers, diffusion arithmetic, zinc flake alternatives and failure analysis.

·

Hydrogen Embrittlement: The 1-Hour Clock, the 22-Hour Bake, and the Parts-Per-Million That Snap a 12.9 Bolt

A set of eight M8 × 30 class 12.9 socket head cap screws went into a machine guard on a Tuesday afternoon. They were zinc plated, trivalent passivated, bright, and dimensionally perfect — every thread gauged 6g, every head stamped 12.9. They were tightened to 40 N·m with a calibrated wrench, which is the value in the table for a lubricated 12.9 M8.

On Thursday morning, two of them were found with their heads on the floor and their shanks still in the tapped holes. No corrosion. No thread stripping. No bending. The fracture faces were flat, granular, and square to the axis, with none of the necking a 1,220 MPa steel should show before it tears.

Nothing was overloaded. The bolts were carrying about 680 MPa of the 1,220 MPa they were rated for. What broke them was hydrogen — perhaps 0.1 to 1 part per million by weight, introduced during pickling and plating, never baked out, and left to concentrate at the prior austenite grain boundaries under a sustained tensile stress. The failure arrived 36 hours after installation, which is precisely the timescale this mechanism runs on.

Hydrogen embrittlement is not a coating defect. It is a three-body problem — susceptible steel, a hydrogen source, and sustained stress — and it is decided by arithmetic that most drawings never ask for: how much hydrogen the plating bath injected, how long the part sat before the bake, and whether the bake table used was the one for 1,000 MPa steel or the one for 1,700 MPa steel.

1. What the fracture looks like, and why it is delayed

Three conditions must coexist: a susceptible microstructure, atomic hydrogen, and a tensile stress that is sustained — not cyclic, not impact, just held. Remove any one and the mechanism stops. The steel in that guard was 42CrMo4, quenched and tempered to 40–44 HRC, which is the top of the 12.9 window and firmly in the susceptible band.

The hydrogen does not need to be everywhere. It migrates to the regions of highest hydrostatic tension, which in a tightened bolt are the thread roots and the under-head fillet. There it lowers the cohesive energy of the grain boundaries, and a crack nucleus forms at a stress far below the bulk yield strength.

The stress concentration does the rest. For a rolled metric thread the elastic stress concentration factor at the root is roughly:

so a nominal 679 MPa preload becomes a local stress of 2.0 to 3.4 GPa — above the 1,100 MPa yield of the steel, which means the root yields on the first turn of the wrench. That local plastic zone is where hydrogen accumulates. The first engaged thread carries about a third of the entire joint load, which is why the fracture almost always starts there rather than in the middle of the shank.

Two signatures separate this from ordinary overload:

That delay is the whole reason the process window is a clock and not a checklist. Hydrogen is mobile at room temperature. It does not wait in the coating for someone to schedule the oven.

2. Where the hydrogen comes from — and how much

Every aqueous step that generates atomic hydrogen at the steel surface is a source. In a typical plating line there are four:

  1. Acid pickling in HCl or H₂SO₄ to remove scale and rust. Inhibitors reduce the attack on the base metal but do not eliminate hydrogen absorption.
  2. Cathodic electrocleaning, where the part is made the cathode in an alkaline bath. This is an efficient hydrogen generator by design.
  3. Electroplating itself, where the cathode reaction is split between metal deposition and hydrogen evolution.
  4. Acid activation, stripping, and some passivation steps after plating — including rework, which is why a stripped-and-replated lot is a new embrittlement event, not a correction.

The plating step is the one people try to quantify, and Faraday's law gives the first number. The mass of zinc deposited is:

with M = 65.38 g/mol, n = 2, and F = 96{,}485 C/mol. One ampere-hour therefore deposits 1.22 g of zinc. An 8 µm coating over one square metre is 8 \times 10^{-6}\,\text{m}^3 \times 7{,}140\,\text{kg/m}^3 = 57.1 g of zinc, which requires:

At a cathode current density of 2 A/dm² — 200 A/m² — that is 0.234 hours, or 14 minutes of plating.

Now the part that matters. If the bath runs at 95% cathode efficiency, 5% of that charge went to hydrogen:

One ampere-hour produces 3600/(2 \times 96{,}485) = 0.01865 mol of H₂, or 0.0376 g. So the bath evolved about 0.088 g of H₂ per square metre of plated surface. Most of it bubbles away. The fraction that enters the steel is the number nobody measures. Assume 1%, absorbed into the first millimetre of a 7.85 kg/m² section:

At 5% absorption the number is 0.56 ppm; at 0.1% it is 0.011 ppm. The band that matters for 12.9 steel sits inside that range. This is why the mechanism feels unreasonable: a coating that adds 57 grams per square metre of zinc is accompanied by tens of milligrams of hydrogen, and tens of milligrams is enough.

The efficiency numbers are where plating shops and textbooks diverge. Acid zinc chloride baths run at 93–98% cathode efficiency; alkaline zincate baths run at 70–85%. The naive conclusion is that acid zinc is safer because it evolves less hydrogen per ampere-hour. The practical conclusion is the opposite: acid pH promotes absorption of atomic hydrogen into the lattice, while the alkaline cathode tends to evolve molecular H₂ that bubbles off. The efficiency number tells you how much gas was made; the bath chemistry tells you how much of it went into the steel. Both matter, and neither is a substitute for the bake.

3. The strength cliff

Hydrogen embrittlement is a high-strength phenomenon. Below about 1,000 MPa ultimate tensile strength, ordinary quenched-and-tempered steel has enough ductility and enough trapping capacity to tolerate the residual hydrogen a normal plating cycle leaves behind. Above it, the margin collapses quickly.

Product · Min. tensile strength · Hardness · Electroplating risk

Class 8.8 bolt · 800 MPa · 232–304 HV · Low; bake usually not required

Class 10.9 bolt · 1,040 MPa · 320–380 HV · Real; assess per ISO 4042, bake if plated

Class 12.9 bolt · 1,220 MPa · 385–435 HV · High; electroplating discouraged, bake mandatory

ASTM A574 socket screw · 1,170 MPa (170 ksi) · ~36–40 HRC · High; same controls as 12.9

ASTM A490 structural bolt · 1,030–1,190 MPa · ~33–38 HRC · High; plating restricted

Case-hardened self-tapping screw · Surface > 550 HV · > 52 HRC case · Very high; thin case is unforgiving

Spring steel · 1,400–2,000 MPa · 42–52 HRC · Very high; avoid electroplating

The standards do not agree on a single trigger, and the disagreement is instructive rather than annoying:

The engineering answer is to use the lowest trigger of the three. If a part is at or above 1,000 MPa, 32 HRC, or 320 HV, and it sees acid or electrolytic processing, it gets a bake — or it gets a non-electrolytic coating. Waiting for 40 HRC means designing a 10.9 part with no relief and finding out later that the hardness landed at 39.

A useful conversion for the drawing review, accurate to about ±10% for quenched and tempered low-alloy steel:

which puts 320 HV at 1,056 MPa and 385 HV at 1,270 MPa. Those two numbers bracket the 10.9 and 12.9 classes almost exactly. Hardness is not a proxy for strength by accident; it is the same dislocation motion that hydrogen interferes with.

4. The bake, and the clock that matters more than the recipe

Post-plate baking does one thing: it raises the mobility of dissolved hydrogen so it can leave the steel before it does damage. It does not reverse a crack that has already started, and it does not empty the deepest irreversible traps. ASTM B850 says so explicitly — the heat treatment "does not guarantee complete freedom from the adverse effects of hydrogen degradation."

The conditions that matter, in order of how often they are got wrong:

Start time. Bake within 1 hour of plating where possible, and no later than 4 hours. This is the single most important variable and the one most often missing from a lot record. Parts that sit overnight before baking have given hydrogen eight hours at room temperature to reach the grain boundaries — and room-temperature diffusion is not slow.

Time at temperature, not oven time. The clock starts when the centre of the load reaches temperature, not when the controller is switched on. A 500 kg load in a cold oven can take 90 minutes to soak; counting that as bake time shortens an 8-hour bake by 20%.

Temperature window. 190–230 °C is the standard band for zinc-plated steel. The ceiling is set by three things: the tempering temperature of the steel (typically 425 °C or higher for 12.9, so not usually the limit), the coating (trivalent chromate is damaged above about 65 °C, so bake before passivation, not after), and low-melting coatings (tin at 232 °C, cadmium at 321 °C).

Uniformity. ±5 °C across the working zone, with a load thermocouple rather than an air thermocouple. A bake at a nominal 200 °C that runs at 175 °C in the corner of the oven is a 4-hour bake wearing an 8-hour label.

The ASTM B850 table is the most useful piece of arithmetic in the whole subject because it turns strength into hours:

Tensile strength · Hardness · Bake at 190–220 °C

1,001–1,100 MPa · 31–33 HRC · 8 h minimum

1,101–1,200 MPa · 33–36 HRC · 10 h

1,201–1,300 MPa · 36–39 HRC · 12 h

1,301–1,400 MPa · 39–43 HRC · 14 h

1,401–1,500 MPa · 43–45 HRC · 16 h

1,501–1,600 MPa · 45–47 HRC · 18 h

1,601–1,700 MPa · 47–49 HRC · 20 h

1,701–1,800 MPa · 49–51 HRC · 22 h

Every 100 MPa buys two more hours. The whole table fits a straight line to within half an hour:

That is not a standard formula — the standard is the table — but it is a useful sanity check on a supplier's proposed cycle. A 12.9 bolt at 1,220 MPa lands at 11.4 hours, so the nearest table entry, 12 hours, is the correct conservative choice. A shop that offers "4 hours at 200 °C for everything" is running a 1,000 MPa cycle on a 1,400 MPa part.

Two additional process steps are frequently conflated with the post-plate bake but do different jobs:

5. The diffusion arithmetic that does not explain the table

Here is where the subject gets interesting, and where a competent engineer can tell a real process control from a copied recipe.

Hydrogen diffusion in body-centred cubic iron is fast. The lattice diffusivity is well described by the Kiuchi–McLellan relation:

which converts to 7.23 \times 10^{-8}\,\text{m}^2/\text{s} in SI. At 25 °C this gives 7.3 \times 10^{-9} m²/s; at 200 °C it gives 1.7 \times 10^{-8} m²/s. The ratio is only 2.3, because the activation energy for lattice diffusion is small — about 5.7 kJ/mol.

The diffusion length scale is:

At room temperature, in one hour, that is 5.1 mm. At 200 °C, in eight hours, it is 22 mm. On lattice diffusion alone, an M8 bolt would be clear in minutes, and the 12-hour bake would be absurd.

Real steel is not a perfect lattice. Hydrogen sits in traps — dislocations, grain boundaries, inclusions, carbides — and the effective diffusivity is one to two orders of magnitude lower. For ferritic pipeline steels, measured D_{eff} at room temperature is 1.5\text{–}2 \times 10^{-11} m²/s, with an effective activation energy near 32 kJ/mol. Scaling that to 200 °C gives about 2.4 \times 10^{-9} m²/s. The diffusion time constant for a section of radius r is:

For an M8 bolt (r = 4 mm) that is 1.9 hours; for an M24 bolt (r = 12 mm) it is 17 hours. This is the correct physical direction — thick sections need longer — but it still does not reproduce the standards' numbers for a 12.9 M8, which ask for 12 hours where diffusion predicts two.

The gap is the point. The empirical tables are not solving Fick's second law; they are buying margin for trap emptying, for surface oxide and coating permeability, for load geometry, and for the fact that a residual hydrogen concentration that is harmless in a test bar may not be harmless in a thread root with K_t = 4. A bake table indexed by strength rather than section is telling you that trap density and local stress matter more than bulk transport at these scales. Treat the table as a minimum, the diffusion calculation as a sanity check, and never let a supplier replace the table with a diffusion model.

One more consequence: because the lattice is fast, the delay before baking is not a minor scheduling detail. Hydrogen that has had a day at room temperature has already sampled every trap in the part. That is why the 1-hour clock is written into ISO 4042 and ASTM F1941, and why a lot that was plated on Friday and baked on Monday is a quarantine decision, not a paperwork decision.

6. Coating selection is risk control

The most reliable way to prevent hydrogen embrittlement is to stop putting hydrogen into the steel. That makes coating selection a design decision, not a purchasing one.

Coating · Thickness · Salt spray (NSS) · Hydrogen risk · Typical friction μ

Zinc electroplated, trivalent passivate · 5–12 µm · 96–240 h · High above 1,000 MPa · 0.12–0.18 dry

Zinc-nickel electroplated (12–16% Ni) · 8–15 µm · 720–1,000 h · High; bake 190–210 °C · 0.18–0.25 unsealed

Zinc flake (ISO 10683) · 4–10 µm · 240–1,000 h · No electrolytic step · 0.08–0.18, grade-dependent

Mechanical zinc (ASTM B695) · 5–75 µm · 200–500 h · No electrolytic step · 0.15–0.25

Hot-dip galvanised (ISO 10684) · 45–85 µm · 500–1,000 h · No electrolytic H, but 450 °C bath · 0.16–0.25, scattered

Cadmium (legacy aerospace) · 5–12 µm · 500–750 h · High; toxic · 0.10–0.18

Black oxide · < 1 µm · None alone · Low (acid activation only) · 0.14–0.20

Three points are worth pulling out of that table.

Zinc flake is the default answer for 12.9 outdoors. It is applied by dip-spin or spray, cured at 200–250 °C, and involves no electrolytic cell, so it introduces no cathodic hydrogen. It also delivers 240–1,000 hours of neutral salt spray at 4–10 µm, which is two to five times the corrosion life of a comparable zinc plating. The caveat is the pre-treatment: if the parts are acid pickled before coating, the embrittlement argument is weakened, so specify abrasive blasting for high-strength parts.

Hot-dip galvanising is not a hydrogen solution for 12.9. The bath runs at about 450 °C, which is close enough to the tempering range to soften class 10.9 and is why ISO 10684 excludes 12.9. It also adds 45–85 µm, which requires oversized nuts and re-tapped threads.

Coating choice sets the torque table. The preload produced by a torque wrench is:

where K is the nut factor. For an M8 12.9 bolt at 40 N·m:

That is 43% more preload for the same wrench setting, and it pushes the bolt from 683 MPa to 976 MPa — through the 970 MPa proof stress of a 12.9 and into yield. A joint designed around a plated bolt and assembled with a zinc-flake bolt is an overload waiting to happen, and an overloaded bolt is a hydrogen crack's best friend.

7. What the tests actually prove

There are four test methods in common use, and they answer different questions.

ASTM F1940 — process control. A 0.4W notched square bar of AISI E4340, heat treated to 50–52 HRC, with K_t = 3.1, is processed with the production lot. It is then tested by incremental step loading in four-point bending. The pass criterion is a threshold of at least 75% of the bare notched fracture strength, achieved within about 24 hours. Because the specimen is deliberately harder than any real fastener, it is a worst-case witness: if the process does not embrittle a 52 HRC bar, it will not embrittle a 40 HRC bolt. F1940 is a periodic audit — monthly is typical — not a lot release test.

ASTM F519 — plating process qualification. Notched specimens are plated and held at 75% of notched fracture strength for 200 hours. Any fracture is a fail. This is the test that qualifies a new bath, a new line, or a new supplier, and it is the one to ask for when a plating vendor claims "we bake everything."

ISO 15330 — in-process preload test. Actual fasteners are tightened against parallel hardened plates to yield or to the breaking torque and held for at least 48 hours, retightened every 24 hours. It is cheap, it uses production parts rather than specimens, and it is excellent for detecting a change in process conditions. It is explicitly not an acceptance test for a lot.

ASTM F606 — mechanical properties. The wedge test and proof-load test measure tensile strength, ductility, and head integrity. They do not detect hydrogen embrittlement, because an embrittled bolt can pass a short-term tensile test and fail the next day under the same load. A supplier who offers F606 as evidence of embrittlement control is answering a different question.

The audit trail that matters is mundane: plating completion time, oven entry time, oven chart with load thermocouple, time at temperature, and the F1940 result for the period. If a shop cannot produce those four records, it does not have process control regardless of what the certificate says.

8. The forensic signature

When a bolt does fail, the diagnosis is usually decided by four observations.

Observation · Hydrogen embrittlement · Overload · Fatigue · Stress corrosion cracking

Fracture path · Intergranular, prior austenite boundaries · Ductile dimples · Striations, beach marks · Intergranular or transgranular, corrosion products

Necking · None or minimal · Pronounced · Localised at origin · Minimal

Origin · Thread root, under-head fillet · Thread root or shank · Thread root, corrosion pit · Pit or crevice

Time to failure · 24–72 h typical · Immediate · 10³–10⁷ cycles · Days to months, with electrolyte

Environment · Air is sufficient · Any · Any · Requires corrosive medium

Two supporting tests can close the case. Thermal desorption spectroscopy (TDS) heats a sample at a controlled rate and measures the hydrogen released; a diffusible-hydrogen peak in the 100–400 °C range is consistent with process hydrogen. SIMS or GD-OES depth profiling shows whether hydrogen is concentrated at the fracture surface. Neither is routine in a job shop, but both are available at failure-analysis labs and both are cheaper than replacing a batch twice.

The batch pattern is the strongest clue. Hydrogen embrittlement does not usually take out one bolt; it takes out the ones from the same plating lot, in the same heat-treat condition, at the same torque. If three of eight fasteners from one lot fail within a week and the other five are fine, the question is not "why those three" but "what did that lot share" — and the answer is usually a missed bake, a late bake, or a stripped-and-replated batch that went back into service without a second bake.

9. Ten rules that remove the risk

  1. Decide the finish from the strength, not from the drawing template. At 1,000 MPa and above, electroplating is a controlled process, not a default.
  2. Prefer zinc flake for 10.9 and 12.9 outdoors. No electrolytic cell, better corrosion life, and a controlled friction window.
  3. Never acid pickle a hardened part. Use abrasive blasting or an inhibited, short, alkaline clean.
  4. Do not cathodically electroclean above 1,000 MPa. Anodic cleaning or a soak cleaner does the job without generating hydrogen at the part.
  5. Bake within 1 hour, never later than 4. Log both timestamps on the lot record.
  6. Use the B850 table, not a house recipe. Twelve hours for 12.9, not four.
  7. Bake before passivation. Trivalent chromate is damaged above about 65 °C, and a post-bake acid dip is a new hydrogen source.
  8. Stress-relieve before plating if the part is heavily cold-formed or machined, but do not count it as the post-plate bake.
  9. Re-plating is a new embrittlement event. A stripped and re-plated lot needs a fresh bake and a fresh qualification, not a touch-up.
  10. Quarantine suspect lots. Baking a lot that has already cracked does not heal it; sorting by proof load and holding for 72 hours is the only honest disposition.

Two design changes are worth more than any process control: roll the threads after heat treatment rather than cutting them, which puts the root in compression and raises fatigue life, and increase the thread root radius where the design allows it. Both reduce the local triaxial stress that hydrogen needs. Neither replaces the bake, but both buy margin.

10. The cost of getting it right, and the cost of getting it wrong

Baking is cheap. A 10 kW oven running an 8-hour cycle on a 300 kg load consumes 80 kWh; at ₹8/kWh that is ₹640, or about ₹2.1/kg of parts. Add half an hour of load and unload labour at ₹400/hour and the direct cost is roughly ₹2.8/kg. Job shops typically charge ₹10–20/kg for a bake cycle because oven time is the real constraint, not electricity.

Zinc flake costs more than zinc plating — typically 1.5 to 2.5 times the per-kilogram coating cost, and in India roughly ₹8–15/kg of fasteners for a standard basecoat plus topcoat, against ₹3–6/kg for barrel zinc plating. On an M8 × 30 bolt weighing 15 g, that is a difference of a few paise per piece. It is almost never the reason to choose the riskier coating.

The failure cost is not a few paise. A failed lot has to be stripped, re-plated, re-baked, re-inspected, and re-sorted — typically two to three times the original finishing cost, with a week of schedule lost. A field failure adds the cost of the machine downtime, the warranty claim, and the investigation. The arithmetic that matters is not ₹15/kg versus ₹5/kg; it is a ₹2/kg bake against a ₹200/kg recall.

11. What to put on the drawing and the purchase order

A finish specification that a plating shop can execute and an auditor can check:

ELECTROPLATE ZINC PER ISO 4042, Fe/Zn 8, TRIVALENT PASSIVATE
SUBSTRATE HARDNESS 40-44 HRC (CLASS 12.9) — HYDROGEN EMBRITTLEMENT RISK
PRE-TREATMENT: ABRASIVE BLAST ONLY. NO ACID PICKLING. NO CATHODIC ELECTROCLEAN.
BAKE PER ASTM B850: 200 ± 10 °C, 12 h MIN AT PART TEMPERATURE,
  START WITHIN 1 h OF PLATING AND BEFORE PASSIVATION
OVEN UNIFORMITY ±5 °C, LOAD THERMOCOUPLE RECORDED
PROCESS CONTROL PER ASTM F1940, MONTHLY, THRESHOLD ≥ 75% NFS(B)
LOT RECORD: PLATING END TIME, OVEN IN TIME, TIME AT TEMPERATURE
NO STRIPPING OR RE-PLATING WITHOUT RE-BAKE AND RE-QUALIFICATION

Five lines do most of the work: the strength class, the pre-treatment restriction, the bake temperature and time, the clock, and the F1940 record. The rest is there so that a purchasing officer who has never seen a fracture surface cannot quietly substitute a cheaper line.

If the part is a 12.9 fastener in a safety-critical joint, add the sentence that ends the argument: "Zinc flake to ISO 10683 with controlled friction window μ = 0.12–0.18 is an acceptable alternative." In most cases it is also the cheaper long-term answer, because it removes the bake, the clock, and the failure mode in one decision.

The arithmetic, once more

The eight bolts in that guard were made to a drawing that specified a coating and a class, and nothing about the interaction between them. The plating shop ran a standard zinc cycle, the parts were bright and gauged correctly, and the bake — if there was one — was a house recipe of four hours at 200 °C applied to a 1,220 MPa part that needed twelve.

The hydrogen that killed them weighed less than a microgram each. It entered during fourteen minutes in a plating bath, migrated for a day at room temperature, and concentrated at a thread root that was already yielding at 679 MPa nominal. No inspection could have caught it, because the bolts were perfect when they left the rack.

That is the nature of the mechanism: it is not a defect you can see, it is a process window you either close or leave open. The strength class sets the bake time. The bake time sets the clock. The clock starts when the part leaves the tank. Get all three right at the drawing stage, and a 12.9 bolt stays a 12.9 bolt.

More FabFlow blog posts