Grade 91 (P91) Welding and Type IV Cracking

Grade 91 (P91 pipe, T91 tube, F91 forging, depending on product form; 9Cr-1Mo-V by composition) is the backbone material of modern supercritical boilers, main steam lines, hot reheat piping, and turbine connection components operating under high temperature and pressure. It is a tempered martensitic steel microalloyed with vanadium, niobium, and nitrogen, and it delivers markedly better creep strength than conventional low-alloy steels in the 550–620 °C range. That advantage comes at a price: from a welding engineering standpoint, P91 is an extremely unforgiving material. Wrong preheat, a missed PWHT or one that falls outside its window, an unsuitable filler metal, or uncontrolled interpass temperature will cause hydrogen cracking in the short term and, in the long term, Type IV creep cracking — the damage mode that cuts service life from decades to a few years. This article covers the metallurgical logic of P91 welding, the tight heat treatment window, and the failure mode that hurts most across the industry: Type IV cracking.

1. Why it is so sensitive: martensitic transformation and microstructure

At room temperature P91 is not ferritic-pearlitic but tempered martensitic. In production, normalizing (austenitizing) at roughly 1040–1080 °C followed by air cooling forms hard martensite; a subsequent temper at roughly 730–780 °C softens that martensite and precipitates M23C6 carbides and fine MX (V, Nb) carbonitrides. Those fine, stable MX precipitates are what carry the creep strength.

It is worth stressing that MX precipitates are carbonitrides: the steel must contain free nitrogen (N) for V(C,N) and Nb(C,N) phases to form. The typical N range is 0.030–0.070%, and it must be read together with the aluminum (Al) ≤ 0.02% limit. High Al forms AlN, consumes the free nitrogen needed for MX, and collapses creep life while the composition still looks compliant on paper. So check the material test report (MTR) for P91 not only for Cr-Mo-V-Nb, but also for N and Al content and the N/Al ratio.

During welding, the base metal adjacent to the molten weld pool is re-austenitized and, because the weld cools rapidly, transforms to hard, fresh martensite. As-welded hardness in this region can reach 400–450 HV, and above 500 HV if the process is neglected. Fresh martensite is brittle and extremely susceptible to hydrogen. This is exactly why two things are non-negotiable on P91: controlled preheat and interpass temperature, and post-weld heat treatment (PWHT).

2. Preheat and interpass temperature: a narrow band

On P91, preheat serves not only to ease hydrogen diffusion but also to keep the martensitic transformation under control. Values widely accepted in field practice:

Preheat (minimum): about 200 °C (construction codes mandate a 204 °C / 400 °F minimum for P-No. 15E). Raise the lower limit for heavy pipe wall thicknesses and highly restrained joints; going up to about 250 °C is common field practice.
Interpass temperature (maximum): the typical upper limit is 300–350 °C. EPRI guidance allows up to 370 °C (700 °F), yet many project specifications hold the limit at 300 °C. Pick one value in the procedure and write it into the WPS.
Continuity of preheat: hold preheat without interruption from root pass to cap. If welding must be interrupted, established practice is not to stop before a defined fraction of the joint cross section is complete — typically the root and hot pass plus roughly a quarter of the section. If an interruption is unavoidable, hold the part at preheat temperature or apply post-heating.

The upper interpass limit is critical. The martensite start (Ms) temperature is around 400 °C and the martensite finish (Mf) temperature in the base metal is around 200 °C. The main variable that sets Mf is not grain size but composition: in weld metal with a high Mn+Ni total, Mf can drop below 100 °C. The intermediate cooling target is therefore not one universal number but a value tied to the composition of the filler metal in use. If interpass temperature runs too high — say 380–400 °C — the weld metal cannot largely transform to martensite between passes and partly remains austenite. The real danger is not retained austenite as such: that partial austenite transforms to fresh, untempered martensite on cooling after the PWHT cycle, creating exactly the hard, brittle region the heat treatment was meant to eliminate.

The lower limit matters too, but its rationale is often stated incorrectly. Dropping below the minimum does not mean you have "prevented" martensite — once the region falls below Ms, martensite forms anyway. The real risk is that below the minimum preheat band the cooling rate rises and hydrogen no longer has time to diffuse out, which raises the risk of hydrogen (cold) cracking. The P91 interpass window is squeezed from both sides: from below by hydrogen and cooling rate, from above by partial transformation. Monitor it continuously with heat treatment thermocouples and contact thermometers (see EN ISO 13916 for measurement practice).

Intermediate cooling before PWHT: a two-threshold rule

Do not take the part straight from welding into PWHT. Cool it in a controlled way first, so the martensitic transformation can complete. EPRI Best Practice (1023199) gives this rule in two steps based on the Ni+Mn content of the filler metal and defines the measuring point explicitly as mid-wall (mid-thickness):

Ni+Mn < 1.2%: cool to below 190 °C (375 °F) at mid-wall.
Ni+Mn ≥ 1.2%: cool to below 95 °C (200 °F) at mid-wall.

After cooling, a short hold at that temperature — typically 1–2 hours, depending on section thickness — helps the transformation complete. Aim to start PWHT within 8 hours of finishing the weld. If that window will be exceeded — shift change, furnace queue, weather — insert a dehydrogenation (hydrogen bake-out) heat treatment: typically post-heating at 300–350 °C for 2–4 hours. Bake-out drives off diffusible hydrogen and lowers the risk of delayed cold cracking, but it does not replace PWHT and it does not remove the need for intermediate cooling. For heavy, restrained joints the correct sequence is: weld complete → intermediate cooling (to the threshold above) → bake-out if required → PWHT.

3. Mandatory PWHT and the narrow window below Ac1

On P91, PWHT is not a recommendation — it is mandatory. It tempers the hard martensite formed by welding, lowers hardness, restores toughness, and relieves residual stress. The critical point is the temperature window:

Typical PWHT temperature: about 730–770 °C; most specifications target 750–760 °C.
Lower limit: the code minimum for P-No. 15E is 730 °C (1350 °F). Nothing below that is acceptable. A cycle that falls below the lower limit, or that holds too briefly, leaves tempering incomplete: hardness stays high and toughness stays low.
Upper limit — the Ac1 line: the Ac1 temperature of P91 (the lower critical point where austenite begins to form) depends strongly on composition and heating rate. Published data give roughly 800–830 °C for the base metal and, depending on Ni+Mn content, roughly 780–830 °C for the weld metal. Ac1 is defined on heating (its cooling counterpart is Ar1) and shifts with heating rate. For critical work, measure it dilatometrically at the actual PWHT heating rate. A catalog value is not binding.

The rule is clear: PWHT temperature must stay below Ac1 — not "just below," but below by a margin that safely prevents partial re-austenitizing. Two separate limits work together here and must not be confused: (a) the code ceiling, a fixed upper limit stepped according to the Ni+Mn content of the filler metal; (b) the engineering margin, the rule of staying 30–50 °C below the measured or calculated real Ac1. For a heat of material with a low measured Ac1, the binding value is not the code ceiling but the margin computed from the measured Ac1.

The ASME stepping has three tiers (confirm against the table in the edition of the construction code in force before applying it):

Filler metal Ni + Mn	Maximum PWHT temperature
Composition unknown	1425 °F (~775 °C)
≥ 1.5%	1425 °F (~775 °C)
1.0% ≤ Ni+Mn < 1.5%	1450 °F (~788 °C)
< 1.0%	1470 °F (~790–800 °C; conversion differs by code and edition)

For weld metal with Ni+Mn ≥ 1.5%, you must additionally stay below the measured real Ac1, typically by at least 10–15 °C. The 1.2% threshold, by contrast, does not set the PWHT ceiling — it sets the intermediate cooling temperature before PWHT (Section 2). If the temperature exceeds Ac1 in a furnace or in local resistance PWHT, austenite re-forms locally; on cooling after PWHT that region reverts to hard, untempered martensite. You get the exact opposite of the softening the heat treatment was meant to deliver.

Hold time is set by pipe wall thickness. Established field practice is about 1 hour per 25 mm of pipe wall thickness, with a 2 hour minimum (EPRI practice). The code minimum (ASME Section I PW-39, B31.1 para. 132, B31.3 para. 331) may be shorter than that; contract specifications usually make 2 hours mandatory. Heating and cooling rates are controlled as well — abrupt heating or cooling creates new residual stress.

Local PWHT geometry: band width matters as much as temperature

A significant share of field P91 failures comes not from temperature but from band width. In local resistance or induction PWHT, the operation is not valid unless all three bands are defined together (per ASME PCC-2 and AWS D10.10 guidance):

Soak band: the band that must be held at target temperature, covering the weld metal, the heat-affected zone (HAZ), and a defined allowance of base metal on each side.
Heated band: the band covered by the heating elements, wider than the soak band; calculate its width from a formula based on pipe wall thickness and diameter.
Gradient control band: the region beyond the heated band, covered with insulation to smooth the temperature gradient.

Insulation width and thermocouple count and placement follow from these bands. With a soak band kept too narrow, the outer edge of the HAZ never sees target temperature; with a narrow gradient control band, you create a reverse temperature gradient through the section and new residual stress. Depending on pipe position, gravity-driven top-to-bottom asymmetry of the heaters must also be compensated.

4. Hardness control: the keystone of acceptance

Hardness is the most tangible criterion in P91 weld acceptance. The acceptance criterion is not one-sided: put numbers on both the ceiling and the floor. Established EPRI practice gives a post-PWHT target band of 190–250 HB; on P91-to-P91 welds made with B9/B91 filler metal, up to 280 HB may be tolerated.

There is a unit trap here: most literature reports HV while most field specifications state HB. Approximate equivalents: 250 HB ≈ 265 HV, 280 HB ≈ 295 HV. That difference changes the accept/reject decision directly, so the procedure must state the measurement method, load, and unit in one single definition (for example "≤ 250 HB" or "≤ 265 HV10") and cite the conversion table used. Hardness measurement is a two-ended diagnosis:

Too hard (above the chosen ceiling; typically > 250 HB ≈ 265 HV10): PWHT was incomplete or missed, temperature was low, or hold time was insufficient. Risk of embrittlement and hydrogen cracking.
Too soft / over-tempered (below about 190 HB ≈ 200 HV10 in the base metal): PWHT temperature was too high (approaching or exceeding Ac1), hold time was too long, or the base metal was mistempered at the mill. This means loss of creep strength.

One caution is in order: locally low hardness in the intercritical band of the HAZ (ICHAZ/FGHAZ) is unavoidable and normal after subcritical PWHT. EPRI does not treat it on its own as grounds for rejection; low readings in that region are evaluated together with base metal and weld metal readings, across the whole hardness profile. Rejecting a joint on a single low point produces unfair rejections and needless repairs in the field — and every repair means another PWHT cycle.

In practice, 100% coverage of the weld with portable hardness testing is the most practical proof that the PWHT cycle really did what it was supposed to. But method selection is critical here:

Prefer UCI (ASTM A1038). Its resolution is adequate for narrow bands such as the HAZ and for thin pipe wall thicknesses; portable Rockwell is also usable.
The rebound (Leeb) method is not suitable for HAZ scanning. Its interaction volume is large and its mass and support requirements are demanding. It does not give reliable results on thin-wall pipe or on low-mass components; consider it only for scanning heavy-section base metal.
Surface preparation is mandatory: a reading taken before the oxide and the decarburized layer are ground away does not represent the true value.
Portable testing sees the surface only. In heavy-wall pipe, a hard region inside the section can be missed from the outside, so always read the hardness map together with the thermocouple records.

The timing of hardness testing also belongs in the procedure: take the readings after PWHT is complete and before the hydrostatic test.

5. Type IV cracking: the real long-term threat

The most insidious damage mode in P91 service is invisible when welding is finished and only appears after tens of thousands of hours at high temperature: Type IV creep cracking. The HAZ of a weld is not a single structure. It divides into sub-zones according to the peak temperature each part reached, moving from the weld outward into the base metal:

Coarse-grained HAZ (CGHAZ): well above Ac3; coarse austenite grains, MX precipitates largely dissolved.
Fine-grained HAZ (FGHAZ): just above Ac3; full renucleation occurs, but the low peak temperature and short time leave a very small prior austenite grain size and partially dissolve MX.
Intercritical HAZ (ICHAZ): between Ac1 and Ac3; the region that is only partially austenitized.

Type IV damage forms mainly in the ICHAZ, together with the adjacent fine-grained band (the base-metal side edge of the FGHAZ). Why there? Because this band is heated between Ac1 and Ac3, it experiences only partial austenitizing: existing MX and M23C6 carbides partly dissolve, dislocation density falls, but the structure is not rebuilt with a new, stable precipitate distribution. The result combines a very small prior austenite grain size, coarsened M23C6, and depleted MX — all three mechanisms that carry creep strength are weakened at once. At service temperature (550–620 °C), creep nucleates cavities on the grain boundaries of this weak band; the cavities link into microcracks and then into a macro crack. The crack typically runs parallel to the weld, along the outer edge of the HAZ facing the base metal.

That is what makes Type IV dangerous: the weld can pass radiographic and ultrasonic examination cleanly, hardness can sit inside the acceptance band, and the joint can run without trouble for the first several years. The damage is a product of time and temperature.

What the welded joint is worth in design: the W factor

In P91 creep life calculations, a welded joint is assumed to have only a fraction of the creep strength of the base metal. Two concepts are frequently confused here and need to be kept apart:

Code W factor (weld strength reduction factor, WSRF): a reduction coefficient applied to the allowable stress in the design calculation. Read its value from the code table as a function of temperature. The code separately defines its scope of application — which weld types and which stress component it applies to.
Cross-weld creep strength ratio: the ratio of the rupture strength of a laboratory cross-weld specimen to that of the base metal. This is material data, not a design coefficient.

In the CrMo/CSEF row of ASME B31.3 Table 302.3.5-1, W is 1.0 below 510 °C (950 °F) and applies only above that temperature. Read against the edition in force, the approximate values are: about 0.95 at 538 °C; about 0.91 at 566 °C; about 0.87 at 593 °C; about 0.80 at 621 °C; about 0.71 at 649 °C; about 0.58 at 704 °C; and the code floor of 0.50 at 732 °C. There is a critical boundary here: the maximum use temperature the code permits for 9Cr-1Mo-V is 649 °C (1200 °F). The 704–732 °C rows are not valid design points for P91 — they represent conditions beyond the material temperature limit and should not be presented as P91 examples.

The often-repeated claim that "a welded joint holds only about 50% of the base metal" is the floor of the code table. Reading it as a cross-weld rupture strength ratio valid in every service condition is misleading. Design stress, operating temperature, and — critically — overtemperature excursions directly shorten Type IV life.

Options for mitigating Type IV

In principle, Type IV cannot be eliminated entirely as long as a weld exists. But there are known ways to reduce its impact in the field:

Full post-weld normalizing and tempering (N&T): erases the HAZ sub-zone structure completely and restores the base metal microstructure. Feasible in shop fabrication (prefabricated spools, forgings); not practical for field erection welds.
Weld location: place seams away from high-stress and high-temperature regions, especially away from points with high bending moment and system stress.
Verification of hanger and support design: on lines operating in the creep range, misadjusted hangers create system stress that never appears in the calculation and shorten Type IV life.
Discipline on service temperature: even brief excursions accelerate Type IV accumulation disproportionately; keep temperature records as an input to life assessment.
Optimizing heat input and pass count: to limit HAZ width and the thickness of the weak band.

6. Filler metal selection and compatibility

Select a 9Cr-1Mo-V type filler metal that is metallurgically matched to the base metal and low in Ni + Mn (so it does not depress Ac1). Common classifications: E9015-B9 / E9018-B9 type covered electrodes (AWS A5.5:2014 updated this suffix to -B91; for example E9015-B91 / E9018-B91), ER90S-B9 (AWS A5.28) for TIG/MIG, and E91T1-B9 / E91T1-B91 (AWS A5.29) for flux-cored wire. Designation consistency matters: the -B91 update landed on the A5.5 side first, and the old -B9 designation is still common in A5.28/A5.29. State clearly on the purchase order and in the WPS which specification and which edition you are referencing. Key points:

Ni + Mn control: the limit applies to every filler metal — covered electrodes, TIG rod, solid and flux-cored wire, and in SAW the wire + flux combination (the flux contribution shifts the weld metal composition). The specification ceiling is on the order of 1.40–1.50%, depending on the relevant AWS class and edition; that is an upper limit, not a "good practice" figure. To keep the PWHT window open and to be able to use the 1470 °F top tier of the code, however, require Ni + Mn ≤ 1.0% in the purchase specification and verify it with a heat/lot-based certificate. As Ni + Mn rises, the Ac1 of the weld metal falls, the permitted maximum PWHT temperature steps down one tier, and the window narrows.
Hydrogen control: use strictly low-hydrogen consumables. Bake covered electrodes per the manufacturer's instructions and keep them in a heated quiver; for TIG, pure argon and a clean surface are essential. On P91, moisture means hydrogen cracking, plain and simple.
Root purging: backing gas is mandatory on the root pass. Bring residual oxygen typically below 0.1%, apply a tighter limit on critical lines (for example 0.05%), and measure and record it with an oxygen analyzer. Otherwise you get root oxidation and the discontinuities that follow from it.
Dissimilar metal welds (DMW): where P91 joins austenitic stainless or low-alloy steel, carbon migration and interface creep damage form a separate risk class; a nickel-based filler metal and a purpose-designed transition may be required.

7. Heat treatment errors: the most common root causes

Field root cause analyses trace most P91 damage back to heat treatment:

Missed PWHT: PWHT after an erection weld is forgotten, or deferred with "we'll do it later" and never done. Hard martensite goes into service.
Insufficient temperature or time: the thermocouple is placed wrong, the real metal temperature stays below target, and hardness comes out high.
Exceeding Ac1: temperature passes Ac1 in an uncontrolled furnace or local resistance PWHT, and the material re-hardens.
Wrong thermocouple placement and single-point control: heavy-wall sections develop a gradient through the section; measuring only at the outer surface is misleading.
Inadequate local PWHT band width: insulation and heaters wrapped without calculating the soak band, heated band, and gradient control band — the outer edge of the HAZ never sees target temperature.
Base metal mistempered at the mill: missing or wrong normalize-and-temper history in the supply chain; MTR verification is essential, including N and Al content and the N/Al ratio.
Skipping N&T after hot forming above Ac1: if hot bending, local heat straightening, or flame forming went above Ac1, the part requires re-normalizing and tempering. Skipping it is one of the most expensive P91 mistakes in the field.
Leaving temporary attachments and tack welds without PWHT: lugs, guide pieces, and tack welds are also welds made on P91. Even if they are removed, the area requires heat treatment and examination.
Overlooking that a repair after PWHT requires a new PWHT: grinding and welding, weld build-up, or any repair carried out after PWHT invalidates the heat treatment cycle; the repair area must go back through PWHT.

8. Control and examination: short term and long term

Examination on P91 has two layers. Post-weld (fabrication) examination: visual, volumetric examination by RT or UT, MT/PT for the surface, and without exception hardness testing plus verification of the PWHT record (the thermocouple chart). Timing is an acceptance condition here: perform the final volumetric and surface examination after PWHT is complete. Otherwise, delayed hydrogen cracking — and cracking generated by the PWHT cycle itself — slips straight through the examination net. Take the hardness readings after PWHT as well, and before the hydrostatic test.

On some critical lines, PMI (positive material identification) confirms the composition of both base metal and filler metal in the field — the wrong material or filler metal is one of the most expensive mistakes there is. But specify the instrument capability: an instrument that reads only Cr and Mo cannot distinguish P91 from other 9Cr class materials. The PMI instrument must be able to read V and Nb, and where possible, verify its calibration against a P91 reference specimen.

In-service examination focuses on Type IV. On high-temperature lines, target the outer edge of the HAZ periodically: scan for microcracks and track their size using surface replication (metallographic replicas) and phased array ultrasonic testing (PAUT). Two limitations must be understood clearly:

Replication alone is not sufficient on Grade 91. In 9Cr steels, creep cavitation develops late in life and mainly in the inner to mid portion of the pipe wall thickness; damage largely progresses without reaching the surface. Do not expect the "early warning" performance familiar from low-alloy steels.
Ultrasonic methods are limited during the cavitation stage. PAUT generally finds Type IV reliably only once microcracks have linked into a macro crack.

For that reason, never interpret the scan results on their own. Read them together with the remaining life calculation (fitness-for-service assessment for creep), service temperature and excursion records, hanger and support condition, and stress analysis. The goal is to quantify remaining life before a crack reaches critical size.

9. Related standards

ASME BPVC Section IX — welding procedure qualification (WPS/PQR) and welder performance qualification; P-No. 15E, Group 1 grouping. Preheat and PWHT are managed here as essential variables (QW-406/QW-407); the mandatory numerical temperature and time requirements come from the construction code.
ASME BPVC Section I / B31.1 / B31.3 — boiler and pressure piping design; mandatory preheat and PWHT temperature/time tables (Section I PW-39, B31.1 para. 132, B31.3 para. 331) and the W factor tables.
ASME BPVC Section II Part D — allowable stress values and temperature limits.
ASME PCC-2 / AWS D10.10 — local heat treatment practice; definitions of soak band, heated band, and gradient control band.
API 579-1 / ASME FFS-1, Part 10 — fitness-for-service (FFS) assessment for creep damage and remaining life calculation.
EN ISO 15614-1 — welding procedure qualification for metallic materials (WPQR).
EN ISO 15609-1 — content of the welding procedure specification (WPS).
EN ISO 9606-1 — welder qualification testing (steels).
EN ISO 3834-2 — comprehensive quality requirements for welding.
EN ISO 17663 — quality requirements for heat treatment in welding.
EN ISO 13916 — measurement of preheat, interpass, and post-heat temperatures.
EN ISO 5817 — quality levels for imperfections in welds (steel).
EN ISO 6520-1 — classification of imperfections in welds.
EN ISO 2553 / AWS A2.4 — welding symbols.
ASME BPVC Section V — nondestructive examination methods.
Material standards: ASTM A335 (P91 pipe), ASTM A213 (T91 tube), ASTM A182 (F91 forgings), ASTM A234 WP91 (fittings), EN 10216-2 / X10CrMoVNb9-1 (1.4903) as the European equivalent.
Consumable standards: AWS/SFA-5.5 (covered electrodes, B9/B91), SFA-5.28 (solid wire/rod), SFA-5.29 (flux-cored wire), EN ISO 3580 (covered electrodes).

From the field

On a supercritical boiler project I was the inspector reviewing the PWHT records for the P91 joints on the main steam line. Every thermocouple chart showed the 755 °C target met. RT was clean, and the hardness readings sat inside the acceptance band. Then we noticed something: the thermocouples had all been attached to the outside surface of the pipe wall, with nothing verifying the inside surface. You cannot measure a temperature retroactively, so we ran a representative repeat cycle on the same pipe wall thickness with the same heater layout — this time with a thermocouple added on the inside surface. Through the whole cycle, the inside surface ran about 30 °C below target. That matched the hardness we then measured in the same area: 280 HV10, roughly 265 HB, against a project specification ceiling of 250 HB (≈ 265 HV10). The chart looked green. The metal did not. The lesson is plain — on P91, the paperwork is not the metal. Check thermocouple placement against the through-thickness gradient. Calculate the band widths for local PWHT. Pair every PWHT record with an independent hardness map. And set the PWHT upper limit from the material's actual Ni + Mn composition and its measured Ac1, never from a catalog value.

Disclaimer

This article is for educational purposes and does not replace the official standard, the code, or the written procedure. Always work from the current edition of the applicable code or standard, and from the project specification — the final decision rests with the code in force and the responsible inspector.

Take it to the field: To keep the welding standards, procedure logic, and field control steps in this article in your pocket — completely offline and free — take a look at the Doawise Welding Eng Guide app.

At DoaWise we carry out welding engineering and inspection services to international standards: WPS/PQR development for creep-strength-enhanced materials such as P91, preheat and PWHT control, hardness and PMI verification, and in-service Type IV damage monitoring. Every step turns into a recordable, auditable result — thermocouple charts, hardness maps, and examination reports. On P91 the margin for error is thin, and the record is what proves you stayed inside it.