Don’t neglect your near-misses
Use Big Data to find risk assessment gaps & biases
One of the great challenges involving near-miss reporting centers on defining what a near-miss really is in the context of the workplace.
Even if we could all agree on a definition of “near miss,” we could not manage the plethora of data generated from all the near-miss reports in a day, week, or year based on the traditional reporting means we have relied upon over the past decades. And this assumes all near-misses would actually be reported.
I want to introduce you to a new means of discovering operational near-misses before they become near-misses that lead to shutdowns, economic, productivity and property losses, and potentially to major injuries.
This new and proven technology takes information that is constantly being collected, and analyzes it using a series of proprietary calculations. At the start of each day, the plant operations team sees a list of potential process concerns that may warrant attention, even though these process anomalies have not trigged an alarm. The technology has already demonstrated its ability to warn teams of higher risk levels before the situation becomes serious and very costly to address.
The software technology installation only takes a few hours and ensures all plant data remain within the plant’s firewall. The technology is designed for any continuous process facility anywhere in the world, both onshore and offshore.
Before discussing the technology, let’s discuss some of the methods we have been using to identify high-risk areas to address near-miss scenarios.
The risk assessment tool remains the most popular approach to maintain and improve safety, operability, and productivity of plant operations through discovering incident scenarios and evaluating the probability of failure leading to identifying areas for risk reduction. Typically, risk assessments rely on incident and failure data to determine high-risk areas — leaving out the use of day-to-day alarm and process control data.
Interestingly, a summary report of the Joint Research Centre and Denmark Risk National Laboratory of the European Commission revealed that risk estimates based on generic reliability/failure databases are prone to biases and could result in large deviations depending upon the data sources. Their research project involved identical chemical process units at seven different locations where they found “large differences in frequency assessments of the same hazardous scenarios.”1
Another favored approach is the ubiquitous safety audit, which typically occurs once or twice per year involving internal teams or outside consulting firms. If a facility has a near-miss reporting program, auditors will review near-miss reports submitted by employees; however, there is no guarantee that an accurate picture is being portrayed by the reports. Indeed, it is difficult to monitor risk-level changes resulting from these near-miss reports.
Finally, manufacturing operations intelligence software tools monitor performance through trending analyses, reporting, and visual analytics of pre-chosen data sets. One of the drawbacks of these software tools is their inadequacy in handling big data analysis, especially when operations need to know when a specific part of an operation is about to fail.
For my purposes, big data refers to a voluminous amount of structured, semi-structured, and unstructured data that has the potential to be mined using predictive analytics to identify relevant information (i.e., near-misses). Although big data does not refer to a specific size of a data set, it is common to see big data refer to petabytes and exabytes of data.2
The new approach I’ll describe next will add a powerful tool in our arsenal for discovering near-misses before they occur. This approach concentrates on closing risk assessment gaps by identifying risk levels and risk drivers dynamically, without significant resource consumption.
The technology solution was developed following years of research by Dr. Ulku Oktem and Dr. Ankur Pariyani of Near-Miss Management™ LLC, resulting in patented methodologies that make use of big data generated by process online sensor measurements and alarm data and stored in plant data historians. These data are collected automatically via a computer interface, which rapidly processes these data and extracts crucial risk information. The system is designed to create leading indicators of potential performance issues, such as shutdowns, accidents, incidents, and operational anomalies.
The Near-Miss Management™ approach comprises a bundled software solution suite called the Dynamic Risk Predictor Suite (DRPS). The DRPS consists of three units: the Dynamic Risk Analyzer™, the Real-Time Leading Signal Generator™, and the Alarm Fitness™. At the moment, the Real-Time Leading Signal Generator™ and the Alarm Fitness™ units are under development.
Analyzing the data
The Dynamic Risk Analyzer™ software analyzes all raw process, mechanical, and alarm data providing plant management morning reports detailing the newest risks discovered as well as their underlying causes. In essence, the Dynamic Risk Analyzer™ searches for hidden near-misses leading to identifying process deviations, which could be indicators of potential incidents. The technology provides operators and plant management with warnings of impending problems well before they happen, allowing for corrective action to be taken.
The Dynamic Risk Analyzer™ has shown success in petroleum refineries and large complex fertilizer plants. End users of Dynamic Risk Analyzer™ have said: “issues are showing up on our radar that reveal something is not right even though our lab results indicate otherwise;” “DRA allows us to catch deviations that we not have caught looking at averages;” and “DRA reports have improved shift-to-shift communications and turnover.”
The Dynamic Risk Analyzer™ provides plant management, engineers, operators, and safety professionals the opportunity to work in an integrated fashion to anticipate near-miss events BEFORE they happen and take corrective action to avert costly plant shutdowns and potentially catastrophic accidents from occurring.
1 Lauridsen, K., I. Kozine, F. Markert, A. Amendola, M. Christou, and M. Fiori. May 2002. Assessment of Uncertainties in Risk Analysis of Chemical Establishments. Final Summary Report. Risk National Laboratory, Roskilde, Denmark. Access here: www.risoe.dk/rispubl/SYS/syspdf/ris-r-1344.pdf
2 Rouse, M. big data – definition. http://searchcloudcomputing.techtarget.com/definition/big-data-Big-Data. Retrieved 15 April 2015.