Dans le domaine de la sécurité des machines, l’estimation de la probabilité qu’une lésion survienne est un problème récurrent. Cet article propose et applique une nouvelle méthode pour estimer cette probabilité.
In safety of machinery, estimating the probability of occurrence of harm is a recurrent problem. This paper proposes and applies a new method to estimate that probability. Information regarding accidents involving machinery that is gathered and analyzed by experts is formatted based on a systemic-inspired model using the MELITO concept. Then, Logical Analysis of Data (LAD) is used to extract knowledge automatically to characterize accidents. MELITO describes the context in which the accident has occurred, gathering information about the moment (M), equipment (E), location (L), individual (I), task (T) and organization (O). LAD is a data mining algorithm that infers knowledge learning from a database. In this paper, a case study consisting of twenty-three fatal and serious accident reports involving belt conveyors is presented. Data on these accidents is classified according to MELITO. The inferred knowledge is presented in the form of interpretable patterns that characterize and distinguish fatalities from non-fatal harm. Each pattern consists of a Boolean equation from MELITO and covers a subset of accidents. Based on each pattern, the probability of the occurrence of harm related to a hazardous situation is estimated. Such probability is useful in monitoring risk behavior after the occurrence of a new accident, for instance.