Evidence-constrained probabilistic assessment of dust-explosion precursors in tapioca-starch storage and handling using an integrated FTA-Bayesian network-ETA framework

Combustible-dust hazards in starch storage and handling systems are difficult to quantify when accident records are sparse and causal evidence is incomplete. This study developed an evidence-constrained framework integrating Fault Tree Analysis, a conditional Bayesian network, Event Tree Analysis, and hierarchy-of-controls screening for a tapioca-starch facility. Historical records from 2017 to 2024 represented 14,840 operating hours (1855 8-h shifts) and included five recorded explosions and two fire-only events. Nine abnormal-condition inputs were parameterized using Gamma–Poisson updating with a Jeffreys prior, while five human/electrostatic inputs were retained as scenario assumptions. With oxygen, combustible dust, and confinement specified as present, the modeled co-occurrence of recorded abnormal dust-dispersion and ignition-source precursor conditions was 6.97 × 10 −5 per shift (0.006966%). This is a conditional precursor-screening output, not a calibrated explosion frequency, and the dust-dispersion node does not represent routine airborne-dust prevalence. Within the ignition branch, mechanical precursors had the largest local model influence (elasticity 0.821), followed by direct human (0.133) and electrical/electrostatic (0.043) precursors. Three of five recorded explosions were followed by fire or smouldering, including one documented warehouse-fire escalation. The findings support dust-dispersion prevention, equipment-condition monitoring, ignition-source control, and consequence limitation; no post-intervention risk reduction was quantified because barrier-performance data were unavailable.