Geopoliticsnorth

The question a restart actually asks

On 28 September 2020, a fire broke out in the air intake of gas turbine generator 4 (GTG4) at Equinor’s Hammerfest LNG plant on Melkøya. The plant was in a running-up phase after an unplanned shutdown, and GTG4 had been taken out of service to correct an oil-filter fault. The Petroleum Safety Authority Norway (PSA, now Havtil) investigated and published its report in April 2021. The report is public, specific, and unusually useful — not because the ignition mechanism was exotic, but because the findings describe a failure pattern that any Arctic LNG operator can audit for before the next restart.

The operational question this article answers is narrow: what does the regulator’s own report tell a practitioner to verify before returning an Arctic LNG train to service? The report does not support claims about security investment, defence posture, or geopolitical causation, and none are made here. What it does support is a documented account of how a known-weak manual practice survived earlier near-misses and became consequential during a restart configuration.

What the report establishes

The PSA investigation report (2020/1862, English translation, approved 20 April 2021) sets out the facts with care. The plant had been in a running-up phase following an unplanned shutdown on 10 September 2020, itself caused by frequency disruptions in the national grid during turbine maintenance. GTG4 was shut down to correct an oil-filter fault when the fire started. The air temperature was about 10–12°C, with a light south-easterly wind.

The causal chain the report supports is this: the control valve for hot-oil supply to the anti-icing panels had been opened manually while the turbine was shut down, causing the temperature in the filter housing to rise sufficiently over time to auto-ignite the pre-filters. Equinor commissioned technical fire tests and KFX simulations of temperature development in the GTG4 filter housing; these support the auto-ignition hypothesis. The filters on GTG4 had last been replaced in 2015, and the technical fire test indicated that biomass buildups — primarily insects — in the pre-filters may have lowered their auto-ignition temperature.

The most important sentence in the report is not about the valve. It is this: the manual valve-opening method was still being practised despite earlier incidents at HLNG where pre-filter cassettes had melted or deformed and filters had dropped after the control valve was opened with the turbine out of operation. That is the organisational finding. A known-weak practice survived prior near-misses and was still available to be used during a restart.

Why restart phases deserve their own barrier review

Steady-state operations and restart configurations are not the same risk object. During a running-up phase, equipment is out of service, manual overrides are in use, and the plant is being brought back through states it does not occupy in normal production. The Melkøya report describes exactly this: GTG4 was down for an oil-filter fault, the hot-oil circuit needed excess heat dissipated, and a manual valve operation was used to do it. The practice was not new. The configuration in which it was used was temporary.

The transferable lesson is not the ignition mechanism. It is that temporary configurations are where documented practice drift becomes consequential. A pre-restart review that treats the plant as if it were in steady state will miss the very conditions the report identifies. The review has to ask: which manual operations are available during this restart, which of them have prior incident history, and were those prior findings closed by procedure change or only by local awareness?

The five nonconformities as a pre-restart checklist

The investigation identified five nonconformities. They are not a narrative; they are a structured audit sequence, and they map cleanly onto a pre-restart verification.

1. Management and control

The report’s management-and-control finding concerns whether the operator’s systems actually governed the practice in question. A pre-restart check should ask whether the manual operations available during the restart are covered by current procedures, whether those procedures reflect the plant’s actual configuration, and whether deviations are recorded and closed.

2. Manning in the organisation

Manning is a barrier. The report treats it as one. A restart review should verify that the shift composition for the running-up phase matches the temporary configuration’s demands — not the steady-state establishment.

3. Risk analyses

The report found deficiencies in risk analyses. The practical question is whether the analyses in force cover the restart configuration specifically, including manual overrides and out-of-service equipment, rather than only normal production.

4. Filter maintenance

GTG4’s filters had last been replaced in 2015, and biomass accumulation may have lowered their auto-ignition temperature. The report does not prescribe a replacement interval, and this article does not invent one. The defensible inference is that maintenance intervals should be justified against the actual operating environment — a coastal Arctic intake exposed to insect ingress is not a temperate inland equivalent — rather than accepted as calendar defaults.

5. Overview of external emergency response resources

This is the finding with the sharpest siting implication. The report states that had external assistance not been available, the fire would have burnt for longer, because firewater monitors at the plant do not cover the turbine air intake. External response was, in fact, a functioning barrier. But it functioned as an assumed backstop, not as a designed barrier with verified performance requirements. For remote Arctic facilities, that distinction matters: the further the site from external resources, the less defensible it is to treat those resources as a given.

The three improvement points

The investigation also identified three improvement points: log-keeping and whiteboarding; overview of performance requirements in the emergency preparedness plan; and follow-up of leaks from anti-icing panels. These are lower-severity findings, but two of them bear directly on restart risk. Log-keeping determines whether a temporary configuration is reconstructable after the fact. Performance requirements in the emergency preparedness plan determine whether external response is a verified barrier or an assumption. The third — follow-up of anti-icing panel leaks — sits close to the hot-oil system that was central to the incident.

What the report does not say

It is worth being explicit about the boundaries, because the Melkøya case is sometimes recruited to arguments it does not support. The report attributes the fire to auto-ignition following manual valve operation. It does not attribute it to sabotage, cyber action, or geopolitical intent. It does not quantify security investment, defence spending, or NATO posture, and no retrieved source links the incident to such decisions. It states that the plant had been shut down since the fire and was due to resume operation on 1 October 2021; it does not provide the cost or the detailed schedule of the rebuild. Those are the report’s limits, and they should be respected.

There is also a more recent document worth noting for completeness: Havtil’s public listings include a 2026 investigation report titled Equinor – Hammerfest LNG – investigation of incidents involving chemical exposure, published 24 September 2026. This article does not characterise its findings, because the report itself was not retrieved. Its existence is relevant only as a reminder that the regulatory record at Melkøya is continuing, not closed.

Who enforces what

For practitioners, the institutional question is as important as the technical one. The PSA/Havtil is the supervisory authority for safety in the petroleum sector, and its investigation reports are the primary public record of what went wrong and what the operator must correct. The report’s nonconformities and improvement points are addressed to the operator, and follow-up runs through Havtil’s supervision. The report does not set a public deadline for closure of each finding; it establishes the findings and the expectation of follow-up. That is the institutional gap worth naming: the public record tells you what failed, but the closure of each finding is tracked through supervisory dialogue rather than a published schedule. If that follow-up stalls, the barrier that failed is the one that depends on institutional memory — the same memory that should have caught the manual valve practice after the earlier incidents.

A pre-restart verification sequence

The following is a practical sequence derived from the report’s categories. It is a recommendation, not a finding of the investigation, and it is deliberately framed as questions rather than as a compliance checklist.

1. Configuration inventory. List every manual operation available during the restart, and every piece of equipment out of service. For each, ask whether prior incident findings exist and how they were closed.
2. Procedure currency. Verify that procedures covering the restart configuration reflect the plant as it actually is during running-up, not as it is in steady state.
3. Manning match. Confirm that shift composition for the running-up phase matches the temporary configuration’s demands.
4. Risk analysis coverage. Check that the analyses in force cover the restart configuration, including overrides and out-of-service equipment.
5. Maintenance justification. Confirm that filter and intake maintenance intervals are justified against the actual operating environment, not calendar defaults.
6. External response as a designed barrier. Verify that external emergency response has stated performance requirements, that those requirements are tested, and that the site’s own fixed coverage gaps are documented rather than assumed away.
7. Log-keeping. Confirm that the temporary configuration is reconstructable from records after the fact.

FAQ

What caused the 2020 Melkøya fire?

The PSA investigation report states that the control valve for hot-oil supply to the anti-icing panels was opened manually while GTG4 was shut down, raising the filter-housing temperature sufficiently over time to auto-ignite the pre-filters. Equinor’s fire tests and KFX simulations support this hypothesis.

Was the fire linked to security or sabotage?

No. The report attributes the fire to auto-ignition following manual valve operation. It does not attribute it to sabotage, cyber action, or geopolitical intent, and no retrieved source supports such a link.

When did the plant restart?

The report states the plant had been shut down since the fire and was due to resume operation on 1 October 2021. It does not provide the cost or detailed schedule of the rebuild.

Which regulator investigated?

The Petroleum Safety Authority Norway (PSA), now Havtil, investigated and published its report in April 2021. Havtil’s public page also notes that a digital presentation was produced to communicate the findings to the industry.

What is the most transferable lesson?

Not the ignition mechanism, but the survival of a known-weak manual practice past earlier near-misses. A pre-restart review should ask whether prior incident findings were closed by procedure change or only by local awareness.

Sources

• Petroleum Safety Authority Norway, Report of the investigation of a fire in the air intake of GTG4 at Hammerfest LNG, Melkøya, 2020/1862, English translation, approved 20 April 2021. havtil.no (PDF)
• Havtil, Investigation of fire at Hammerfest LNG, published 10 May 2021. havtil.no
• Havtil homepage listing, Equinor – Hammerfest LNG – investigation of incidents involving chemical exposure, published 24 September 2026. havtil.no