AI-Supported Analysis of Fire and Explosion Type Maritime Accidents Using the Human Factors Analysis and Classification System for Maritime Accidents Methodology
Global Maritime Congress, İzmir, Türkiye, 28 - 29 Eylül 2026, (Yayınlanmadı)
- Yayın Türü: Bildiri / Yayınlanmadı
- Basıldığı Şehir: İzmir
- Basıldığı Ülke: Türkiye
- Dokuz Eylül Üniversitesi Adresli: Evet
Özet
Background / Objectives
Fire and explosion accidents represent a
disproportionately hazardous category of maritime casualties, with official reports indicating
they accounted for 9.5% of all injury-related events between 2015 and
2024 despite their comparatively low overall frequency. Although research
consistently links such accidents to human and organizational factors,
large-scale, systematically coded investigations of this specific accident type
remain limited. This study aimed to identify the most and least prevalent human
factor categories in maritime fire and explosion accidents by applying the
Human Factors Analysis and Classification System for Maritime Accidents (HFACS-MA)
framework to a decade-long sample of official casualty investigation reports.
Methods
A total of 226 fire and explosion casualties reported
in the International Maritime Organization (IMO) Global Integrated Shipping
Information System (GISIS) database for the period 2015–2025 were screened.
Applying purposive sampling criteria — full-text availability in English,
completed investigation status, and confirmed human factor involvement — 43
reports were retained (31 fires, 12 explosions). The analysis followed a
two-stage procedure. First, three pilot cases were manually coded by the researchers
to construct a standardized HFACS-MA reference matrix covering all five levels
and 21 sub-categories. Second, the remaining 40 reports were coded using a
large language model (LLM; Google
Gemini 3.1 Pro) guided strictly by that matrix. Inter-coder reliability
was assessed by independently re-coding three randomly selected AI-analyzed
reports; agreement was approximately 86%, exceeding the 70% threshold
recommended by Miles and Huberman (1994) for qualitative research. Frequency
and percentage distributions were calculated for all HFACS-MA sub-categories.
Results
All 43 accidents contained at least one human factor
code, yielding 278 individual HFACS-MA activations (mean 6.5 codes per
accident). At the organizational level, Organizational Process was the most
prevalent sub-category (f = 35; 81.4% of accidents), reflecting pervasive
deficiencies in operational procedures and safety management system
implementation. Physical Environment ranked second (f = 33; 76.7%), followed by
Technological Environment (f = 28; 65.1%), Rule-Based Mistakes (f = 24; 55.8%),
Inadequate Supervision (f = 21; 48.8%), Condition of Operator (f = 19; 44.2%),
and Exceptional Violations (f = 18; 41.9%). In contrast, Software (f = 1) and
Hardware (f = 2) were almost entirely absent, indicating that the dominant
failure mode is human–machine interface design rather than outright technical
breakdown. Night-time accidents predominated (67.7% of fires; 58.3% of
explosions), container vessels were the most affected ship type for fires (n =
10), and tankers for explosions (n = 4).
Conclusions
Maritime fire and explosion accidents are inherently
multi-causal, with human factors present across all five HFACS-MA levels.
Organizational process deficiencies and adverse physical and technological
environments are the most critical systemic vulnerabilities, while
individual-level failures manifest primarily as rule-based mistakes and
exceptional violations rather than skill-based errors. These findings suggest
strengthening the operational implementation of the International Safety
Management (ISM) Code, improving ergonomic standards for onboard alarm systems,
and updating emergency response training curricula are the most impactful
intervention priorities. The AI-assisted coding approach demonstrated high
reliability and offers a scalable methodology for future large-scale maritime
accident analyses.
Keywords
Maritime accidents, HFACS-MA, AI-assisted coding, Fire and Explosion.