An information fusion framework for context-based accidents prevention

The oil and gas industry is increasingly concerned with achieving and demonstrating good performance with regard occupational health and safety (OHS) issues, through the control of its OHS risks, which is consistent with its core policy and objectives. There are standards to identify and record workplace accidents and incidents to provide guiding means on prevention efforts, indicating specific failures or reference, means of correction of conditions or circumstances that culminated in an accident. Therefore, events recognition is central to OHS, since the system can selectively start proper prediction services according to the user current situation and past knowledge taken from huge databases. In this sense, a fusion framework that combines data from multiples sources to achieve more specific inferences is needed. In this paper we propose a machine learning algorithm to learn from past anomalous events related to accident events in time and space. It also uses additional knowledge, like the contextual knowledge: user profile, event location and time, etc. Our proposed model provides the big picture about risk analysis for that employee at that place in that moment in a real world environment. Our main contribution lies in building a causality model for accident investigation by means of well-defined spatiotemporal constraints in the offshore oil industry domain.