Learning From Repeat Fires

At around 20.30 WIB on Wednesday, 1 October 2025, an operating unit at Pertamina's Refinery Unit II in Dumai, Riau, caught fire with a loud explosion. The refinery's emergency response team had the fire under control by 23.20 WIB. Two days later, the Directorate General of Oil and Gas at the Ministry of Energy and Mineral Resources said the refinery was operating normally, but that the exact cause could not yet be determined and that investigation findings would be shared once the technical examination was complete (Republika, 2025). As we write in early November 2025, we have found no published official investigation report.

The event drew attention because it was not the first at this site. News coverage noted that in April 2023 the same refinery had an explosion and fire which, according to media reports, began with a hydrogen gas leak on a six-inch pipe at a gas compressor unit (Listrik Indonesia, 2025). Two events at one facility do not mean they share a cause, and this article does not speculate about the cause of the October 2025 fire. Our review question is broader: what does the literature say about why lessons from process industry incidents are often not enough to prevent the next event?

What We Already Know

Drupsteen and Guldenmund (2014) reviewed the safety literature on learning from incidents, accidents and disasters and compared it with the organisational learning theory of Argyris and Schön. They describe learning as a chain of sub-processes: drawing lessons, sharing them, storing them and applying them. Factors that shape this chain include trust, the severity of the incident's consequences and the people involved. In their reading, the safety literature is very detailed about incidents and their analysis but much thinner on the sharing and applying stages.

Jacobsson et al. (2011) proposed a way to measure the part that is often missed. They assess the "level of learning" on three dimensions: how widely a lesson is applied geographically, how much organisational learning is involved and how long the effect lasts. Their six-step method was applied to a large number of incidents in six process industry companies and was found useful for revealing what influences learning.

Later results from the same research programme point to recurring weaknesses. Akselsson et al. (2012) reported that effectiveness in each step of the learning cycle was low because incident reports lacked information, analyses were very shallow, decisions focused on fixing the problem only where the incident occurred, implementation was late and solutions were weak. Both horizontal learning (spreading to other sites) and vertical learning (changing assumptions and systems) were weak. The study also found strong correlations between learning from incidents and safety climate, especially trust, safety knowledge, participation and compliance.

Measures and Gaps

To count process safety events, the refining industry widely uses API RP 754. Its third edition sorts indicators into four tiers: Tiers 1 and 2 for loss of primary containment events by consequence, Tier 3 for challenges to safety systems and Tier 4 for operating discipline and management system performance (American Petroleum Institute [API], 2021). These indicators help count and compare events, but they do not by themselves show whether the lessons from those events were applied.

The evidence gaps are clear. The research on levels of learning that we found comes mostly from Europe, particularly Sweden (Akselsson et al., 2012; Jacobsson et al., 2011), and describes the process more than it tests the link to fewer repeat events. We also found no public data on Indonesian refinery process safety events reported by tier as in API RP 754, so facilities cannot yet be compared using public figures.

Reading a Repeat Event

In our view, the most useful question after a second event at one facility is not "was the cause the same?" but "how far did the lessons from the first event travel?". The framework of Jacobsson et al. (2011) gives three checks anyone can apply: were the fixes extended to similar units elsewhere, did they change procedures and the management system, and are those changes still working two years later? The weaknesses reported by Akselsson et al. (2012), such as fixes limited to the spot where the event happened, are likely at many facilities because of the pressure to return to operation quickly.

We also note that the findings on trust (Drupsteen & Guldenmund, 2014; Akselsson et al., 2012) matter for people who report small events. Tier 3 and Tier 4 events under API RP 754 only get recorded when workers feel safe reporting them (API, 2021). This assessment is our interpretation of the literature, not a conclusion about any particular refinery.

Practical Steps

• Require every corrective action from an investigation to answer three questions: where else does a similar hazard exist, what system needs to change, and when will effectiveness be re-checked.
• Use the API RP 754 tiers to record Tier 1 to 4 events, and review Tier 3 and 4 trends as early signals before a major release.
• Measure the depth of analysis, for example whether reports cover organisational factors and not only equipment failure.
• Audit the closure of corrective actions after 6 to 12 months instead of relying on a "closed" status in the system.
• Share lessons with similar facilities in the same group in a structured way, including near misses.

How Nusa Safety Helps

Nusa Safety helps process companies review how they learn from incidents. Through a Process Safety Management Review, the Nusa Safety team checks whether investigation findings reach procedures and other units, while Incident Investigation and RCA training prepares internal teams to analyse organisational factors and not only equipment damage. For units whose hazards need a fresh look, HAZOP and LOPA studies are available. Nusa Safety issues internal training certificates only and is not a PJK3.

This article is general information and not legal or professional advice for any specific case.

References

1. Akselsson, R., Jacobsson, A., Börjesson, M., Ek, Å., & Enander, A. (2012). Efficient and effective learning for safety from incidents. Work, 41(Suppl. 1), 3216-3222. https://doi.org/10.3233/WOR-2012-0661-3216
2. American Petroleum Institute. (2021). Process safety performance indicators for the refining and petrochemical industries (ANSI/API Recommended Practice 754, 3rd ed.). https://www.api.org/oil-and-natural-gas/health-and-safety/refinery-and-plant-safety/process-safety/process-safety-standards/rp-754
3. Drupsteen, L., & Guldenmund, F. W. (2014). What is learning? A review of the safety literature to define learning from incidents, accidents and disasters. Journal of Contingencies and Crisis Management, 22(2), 81-96. https://doi.org/10.1111/1468-5973.12039
4. Jacobsson, A., Ek, Å., & Akselsson, R. (2011). Method for evaluating learning from incidents using the idea of "level of learning". Journal of Loss Prevention in the Process Industries, 24(4), 333-343. https://doi.org/10.1016/j.jlp.2011.01.011
5. Listrik Indonesia. (2025, October 2). Terulang lagi, Kilang Dumai Pertamina sudah dua kali kebakaran [Again: Pertamina's Dumai refinery has now caught fire twice]. https://listrikindonesia.com/detail/18065/terulang-lagi-kilang-dumai-pertamina-sudah-dua-kali-kebakaran
6. Republika. (2025, October 4). Update kebakaran Kilang Dumai, ESDM: Operasional sudah normal [Dumai refinery fire update, energy ministry: Operations back to normal]. https://ekonomi.republika.co.id/berita/t3kx2h502/update-kebakaran-kilang-dumai-esdm-operasional-sudah-normal