A field foreman’s guide to failed exothermic welds — the five process causes, the diagnostic walk-through, and an interactive weld failure diagnostic you can run before you cut anything out.

Three weeks ago, a contractor in Cebu sent me photographs at 11 p.m. local time. Two hundred exothermic joints had been poured the previous quarter on a 33 kV substation earthing grid. Commissioning was the next morning. Half the joints looked perfect. The other half had micro-blowholes along the conductor exit and were measuring between 600 and 900 micro-ohm, against a healthy 30 to 50 on the good half. The supplier — not us — had delivered the weld metal powder on a vessel that sat on a Manila dock through three days of rain. Nobody opened a single cartridge to check.

That call is the reason for this article. After 17 years of machining exothermic welding molds for overseas buyers — and 12 years of watching their commissioning reports — I have stopped believing in sudden failures. What looks sudden is almost always a process error that nobody flagged because the spec sheet didn’t tell them to look. Here is what I now send every contractor who calls me with a “mystery” failure, in the order I walk them through it on site.

Why a Failed Thermit Weld Almost Never Fails the Way You Think

I have to challenge the framing first, because it shapes every troubleshooting decision a crew makes. When a contractor calls me with a “failed weld”, I ask one question before I ask anything else: did the joint ever pass a resistance test, or did it fail at commissioning the first time it was ever measured? The answer changes everything, because a joint that degraded in service and a joint that was defective from pour-day are two completely different problems with two completely different root causes.

Per the aging-management literature on buried electrical connections that US utilities reference (NUREG-1800, the Generic Aging Lessons Learned report used to justify decades of plant life extension), the documented degradation mechanisms for any buried joint are thermal cycling, ohmic heating, vibration, and corrosion — and the report is explicit that these are mechanisms that act on interfaces. An exothermic weld has no interface to age, because the conductors and the molten copper become one continuous metal piece during the reaction. So when a properly poured exothermic joint fails in service, the failure is almost never in the weld metal itself — it is in a section of conductor the weld never actually reached, or in a thermal-cycling micro-gap the weld left behind because the powder charge was wrong on pour day.

What “defect from pour day” actually looks like on a forensic teardown

The two hundred Cebu joints taught me one pattern that has held up across roughly every multi-joint failure I have walked since. When we cut a failed joint open and polish the cross-section, we see one of five recurring microstructures, and each maps cleanly to one of the five process causes in the rest of this article:

• Blowholes clustered near the conductor surface. Steam from trapped moisture — most often wet powder, sometimes a damp conductor, occasionally condensate in a cold mold.
• An oxide lamination parallel to the conductor axis. Contamination between conductor and weld metal — oil from a fingerprint, residual oxide from a poorly cleaned cable, or a release-agent film from a mold that was never pre-heated.
• An under-filled cavity with a dished bead surface. A graphite mold worn past its cycle count, or a powder charge undersized for the conductor cross-section the pour was meant to fill.
• A cold-shut line where two metal fronts failed to fuse. Ignition timing was wrong for the mold temperature — almost always because a crew tried to ignite a mold straight out of an air-conditioned store on a cold morning.
• A weld bead bonded to only one of two conductors in a crossover. Mold misalignment on closing, or a powder disc shifted off the tap hole during ignition.

Once you learn to read these five microstructures, the troubleshooting conversation stops being guesswork. The rest of this guide walks each cause in the order I would investigate on site, with the verification step that proves it before you cut a single joint out.

Is Moisture the Real Culprit When a Thermit Weld Fails?

Moisture is the failure cause I lead with because it is the most common and the most forgiving of excuses. The molten copper from the exothermic reaction sits at roughly 1,083 °C — the melting point of copper — and the moment it meets any water, that water flash-boils to steam and expands by roughly 1,700 times in volume. The steam forces its way back through the still-molten weld metal, leaves a trail of blowholes, and the joint is compromised before it ever cools. Per OSHA’s incident investigations on thermit-weld operations, this single mechanism has caused more ignition injuries and crew burns than any other exothermic process error.

The reason this fails so quietly is that moisture has three hiding places, and only one of them is obvious.

• Wet powder. The most common cause of bulk failures. Powder ships in sealed cans for a reason — once a can is open, the contents are hygroscopic and will absorb ambient humidity in coastal climates within hours. A cartridge that “looks dry” is not dry.
• Damp conductors. Copper strand and copper-clad steel both hold moisture in the strand interstices. A cable that has been stored uncovered on site through a rainstorm is wet on the inside even if the outside looks dry, and a torch-dried surface will re-absorb moisture within minutes in humid air.
• Cold-mold condensation. A graphite mold pulled out of an air-conditioned store at 18 °C into a 32 °C, 85%-humidity morning will condense a film of water on its cavity wall before the crew has the conductor in place. The pour then traps that film between the mold wall and the molten metal.

The Cebu post-mortem in three lines

The Cebu joints failed because the powder sat on a rain-soaked dock and the crew opened cans the morning of the pour. They used the powder anyway because it looked dry — and the underside of the bead, on every failed joint, showed the exact clustered blowhole pattern that moisture-driven steam leaves behind. A 20-minute desiccant check on three sample cartridges before the pour would have caught the entire batch, and the crew would have known to quarantine the rest.

Is Your Graphite Mold Worn Past Its Honest Service Window?

This is the cause suppliers handle worst, because the failure mode is invisible to a buyer and the cycle count is something only the supplier knows. nVent ERICO’s Cadweld product literature rates graphite molds at around 50 to 100 service cycles, and on our line we rate our thermit welding molds at the same 50 to 100 cycles, with 80 to 100 joints in normal field use. After that, the cavity has eroded enough that the tap hole no longer meters molten metal cleanly into the joint.

The honest signal of a worn mold is not a crack you can see — it is the weld bead losing its pillow shape and starting to dish inward at the center. A new mold produces a bead that domes gently over the conductor, because the powder charge slightly over-fills the cavity and the excess welds into the riser. A worn mold under-fills, because the cavity has enlarged enough that the same powder charge no longer has the surplus to dome. Crews compensate by adding powder, which then overheats the cavity and accelerates wear — a feedback loop that takes about twenty bad pours to destroy a mold completely.

How to grade a mold you did not buy from us

If you inherited molds from another supplier, weigh one against a known-good mold of the same geometry. A graphite mold loses about 3 to 8 grams per cycle from thermal erosion and cleaning; a mold 50 grams under its starting weight has had more cycles than its rating allows. Second, look at the weld bead shape across the last five pours — if any of them dished inward, the mold has been pushed past its honest service window. Per IEEE Std 837-2014, the qualification framework for permanent substation grounding connections, the geometry of the weld cavity is part of the qualification evidence — so a worn mold is not just a quality risk, it is a compliance risk against the standard the joint was specified to.

Could an Undersized Powder Charge Be Quietly Undermining Your Welds?

Powder comes in standardized cartridge sizes for a reason. Because the powder-to-copper-mass ratio is the variable that determines whether the reaction generates enough heat to fully melt the conductor cross-section, undersizing the cartridge creates a joint that pours solid on the outside but is under-fused at the conductor interface. The bead looks fine. The resistance tells the truth six months later when the joint heats up under the first fault current.

The honest version of this failure is that crews sometimes substitute a smaller cartridge when the right size is out of stock, on the assumption that “more or less the same amount of metal”. The dishonest version is that some suppliers will sell a #115 cartridge at the price of a #150 because the contractor’s order form said “cable-to-cable 50 mm²” without specifying the full conductor pair. Both end with the same joint: a beautiful bead, a hidden micro-gap at the conductor, and a resistance that drifts up year over year.

Are Conductor Prep Shortcuts the Hidden Cause of High-Resistance Joints?

I have watched crews spend twenty minutes cleaning the visible part of a copper strand and skip the section that goes inside the mold cavity, because the spec sheet says “clean to bright metal” without specifying how. Because the bond forms at the conductor surface that is inside the cavity at the moment of pour, a contaminated section under the cavity creates a joint that is metallurgically continuous on the outside and oxide-laminated on the inside.

The most common contamination I see on forensic teardowns is not dirt — it is the fingerprint. Crews handle the conductor section that goes inside the mold, leave a thin oil film from their skin, and the molten copper cannot wet through that film to fuse to the base metal. The result is a joint with a continuous oxide plane between weld metal and conductor, and an oxide plane is exactly the kind of interface that NUREG-1800 documents as a degradation pathway under thermal cycling.

• Use a dedicated, sharp steel brush per conductor material — copper, copper-clad steel, and galvanized steel each need their own brush.
• Wipe the cleaned section with a lint-free cloth dampened with isopropyl alcohol, not a gloved hand.
• Re-clean if the conductor sits in open air for more than 15 minutes before the pour, because fresh copper re-oxidizes quickly in coastal and industrial atmospheres.
• Hold the cleaned section with clean pliers or a clean cloth from that point on — no skin contact.

Is Cold-Mold Ignition Timing Causing Your Morning-Shift Failures?

This is the cause nobody puts in writing, because it sounds like an operator-error rather than a process error. The reality is that the exothermic reaction is tuned to a specific thermal envelope, and a mold that is below roughly 25 °C at the moment of ignition behaves differently from one that has been pre-warmed. The starting powder ignites, the reaction front propagates through the charge, but the molten copper loses heat to the cold graphite faster than the reaction can replenish it, and the metal reaches the conductor cavity below fusion temperature. The bead still forms. The fusion is incomplete.

This is overwhelmingly a morning-shift problem on temperate or cold-climate sites, and it is overwhelmingly an afternoon-shift problem on tropical sites where crews pre-cool their molds in air-conditioned storage to “preserve” them. Either habit produces the same failure pattern: a cold-shut line where two metal fronts met but did not fully fuse, visible on cross-section as a sharp discontinuity in grain structure parallel to the cavity wall.

Forensic Checklist: What to Do When a Joint Fails Commissioning

When a pour-day joint fails commissioning, walk the investigation in this order. The first three checks take less than an hour on site and resolve roughly 80% of multi-joint failures without cutting anything out.

1. Verify powder batch history. When was the powder delivered, how was it stored, and were any cans opened before the pour? Any “yes” answer points to moisture (Cause 1).
2. Pull the mold log. How many cycles has the mold done? Cross-reference against the 50–100 cycle rating. Any mold past its rating gets replaced before the next pour (Cause 2).
3. Check the powder size against the conductor schedule. Match the cartridge size on every can against the conductor combination on the joint. Mismatches get re-poured with the correct cartridge (Cause 3).
4. Inspect three failed joints on cross-section. A bandsaw cut and a polish is enough to identify blowhole patterns (Cause 1), oxide lamination (Cause 4), under-filling (Causes 2 and 3), or cold-shut lines (Cause 5).
5. Map failures against the pour sequence. Failures clustered at the start of a shift almost always point to Cause 5 (cold mold). Failures clustered at the end of a powder batch point to Cause 1 (wet powder). Failures spread evenly across a shift point to Causes 2 or 4.

If the failures survive all five checks, the issue is almost certainly environmental — soil chemistry, stray currents, or galvanic corrosion finding an interface the weld left behind — and the conversation has to shift from process to design. That conversation is its own article.

Re-Pour Discipline: Cutting Back, Not Re-Welding In Place

Once a joint has failed, the only correct repair is to cut back past the heat-affected zone and re-prepare the conductor from scratch. Pouring fresh weld metal over a failed joint does not fuse the new metal to the old reliably, because the original interface has already formed its own metallurgy and any trapped porosity becomes a nucleation site for the next failure. The cut-back should be at least 50 mm of clean conductor on each side of the original joint, the new mold should be inside its cycle count, and the powder batch should be from a freshly opened, verified-dry can.

The second-time pour is also the right time to upgrade the mold if the failure pointed to mold wear. We ship our exothermic welding molds with a serialized cycle log so buyers can track each mold’s history across a project, and the same log should travel with any mold that moves between sites. A mold you cannot account for is a mold you cannot trust.

How to Audit Your Current Supplier Without Insulting Anyone

Most of the calls I take on this topic do not start with a failure. They start with a buyer who has just been asked to qualify a new supplier, or an engineer who has inherited a process they did not design, and they want to know what questions to ask before the next failure. The honest version of that conversation is that almost every supplier can ship a mold that works on pour day. The harder question is whether they can document why their product worked, and whether they will own the answer when it does not.

Three questions separate a real supplier from a price-discounter on this specific problem set, and I would walk every buyer through them in this order.

1. “Show me your mold cycle log template.” A supplier who tracks cycle counts is a supplier who knows that molds wear out. A supplier who does not is one bad pour away from your problem becoming theirs.
2. “What is your powder batch traceability?” Every can should carry a batch number, a manufacture date, and a sealed-container indicator. A supplier who repackages bulk powder into unlabeled cans is a moisture risk in a tin.
3. “What is your position on undersized cartridge substitution?” The honest answer is “we refuse the order until you confirm conductor size.” The dangerous answer is “we can ship a smaller can today.”

We built our export documentation around these three questions because they are the questions our own failure investigations keep circling back to. They are also the questions I would ask us, if I were a buyer.

Weld Failure Diagnostic — Pick the Symptom You See in the Field

Before you cut a single joint out, run this diagnostic. Pick the symptom that matches what you are actually seeing on site — the visual, the electrical, or the timing pattern — and the tool maps it back to the most likely of the five causes above, gives you a confidence reading, and lists the immediate steps to take before the next pour. This is the same decision tree I walk a buyer over the phone when a failure call comes in.

Frequently Asked Questions About Failed Exothermic Welds

Jane Yang · Sales Manager, Xinchang Shibang New Material Co., Ltd.
12 years in lightning protection & grounding foreign trade · 17 years of factory production behind every recommendation.
LinkedIn · Facebook · X