(117e) Application of Machine Learning Algorithm for the Analysis of OSHA Accident Database to Identify Trends in Performance Indicators of Process Safety Incidents

Many workplace injuries and fatalities occur in facilities dealing with hazardous chemicals due to unexpected leaks, exposure to dangerous process conditions and other loss of containment events. The U.S. Occupational Safety and Health Agency (OSHA)’s Process Safety Management (PSM) program has been in effect for almost three decades in the United States, requiring companies to establish comprehensive management programs to deal with such hazardous chemicals with the target to assure safety and health in the workplace. Analysis of the trend of process safety incidents in these industries can be a good performance indicator for the effectiveness of the PSM program.

For a long time, OSHA has been maintaining and publishing accident data collected from various industries since the 70’s and currently has recorded information of more than 130,000 incidents involving fatalities and injuries arising from various causes. Analysis of such a large database to identify process safety incidents manually is extremely difficult and time consuming. Current study focuses on the development of a machine learning algorithm to screen the database and identify process safety incidents through use of the keywords and incident descriptions associated with each accident provided by OSHA.

In order to validate the machine learning process, the large OSHA database is filtered to analyze incidents from 1984 to current time and only chemical manufacturing and petroleum industries are considered through use of industry classification codes. More than 2,200 incidents are screened out in this manner and are analyzed to determine those that were process safety related incidents. The analysis was done both manually and through use of the developed machine learning algorithm to validate the capability of the latter. The keywords are classified into various categories to aid the algorithm in the learning process. This is the beginning step to the application of artificial intelligence to analyze big databases such as the OSHA database to determine common patterns behind injury and fatal incidents and provide industry with performance indicators that will provide an understanding of where the current laggings are and enable proactive measure to be taken to prevent future incidents.