A publication from Lloyds Register, ‘How to optimize your HSE strategy with Artificial Intelligence*’ makes some bold claims for AI in health and safety. The announcement covers LR’s ‘Safety Scanner’, a natural language processing tool that accesses information ‘locked away’ in free text descriptions of HSE incidents, providing clues to immediate and root causes. The Safety Scanner can also use sensor data to monitor worker fatigue and heat exhaustion to understand risk setting and behavior, ‘providing previously unobtainable granularity of actionable insights into the HSE processes.’ We challenged LR to come up with some examples of such ‘hidden insights’ that have been revealed by the Scanner. LR’s Ran Merkazy kindly provided the following (which we have edited).
Although it’s early days, evidence is starting to collect. Here are some examples, some of which are soon to be released as case studies. Our most recent deployment is fatigue monitoring at a middle eastern airport airfield services operation where management suspected that fatigue played a role in the rising amount of accidents, such as airfield trucks colliding with airplanes, one which recently caused a fatality. We tracked fatigue levels using wearable sensors and compared the data with thousands of hours of benchmark data to find that the daytime average fatigue was between 3 and 4 times the expected benchmark. This result was a powerful motivator for management to come up with solutions.
Another example of surprising, otherwise unobtainable insight emerged when we used the scanner on HSE reports at a global B2B services organization, with 1000’s of field operatives. HSE management was well aware of the risks from ‘falls from height’, ‘slips & trips’ and ‘traffic incidents’ but our analysis also tagged many ‘unknown’ items for further investigation. Data analysis showed that the workforce was experiencing a large number of medical emergencies, such as blood pressure issues and age-related strain injuries such as back issues or sprained joints, particularly with older workers. We are now using analytics on data from multiple sources (training, geographic, weather, time), to identify other patterns. Early findings are interesting, showing, for instance, that it is the change in temperature (not how hot or cold it is) that counts. The day of the month can also be an early indicators of increased risk.
We are now engaged in a big AI-powered project with a National Oil producer from Asia (I can’t name it, but it’s as big as they get, on a global scale), who is asking us to deal with a massive amount of incident information, to mine for insights in textual descriptions. Here we are going further with the analytics to understand which safety barriers fail most often and what leading indicators can be identified in daily reports. Interestingly, this project demonstrates our capability to deploy our AI in different languages. An early finding is that Heinrich pyramid ratio may not be quite the gold standard of HSE practice as is often thought.
The Safety Scanner can be deployed as a plug-in to third party HSE software so clients can leverage this type of insight, without needing to change anything in the way they collect or track their HSE information.
* Read Lloyd’s original 2017 paper, How to optimize your HSE strategy with Artificial Intelligence and the recent announcement of the Safety Scanner.
© Oil IT Journal - all rights reserved.