This project used data mining in an attempt to validate the API755 standard. The hypothesis tested was that consecutive working days has no impact on the number of incidents.
Aaron Miller of Wright State Research Institute was tasked with mining data to determine if operator errors increased with increasing levels of fatigue. The latter was defined as continuous days on the job calculated from employee time records at four different refineries. The results were correlated with incident reports in which human error was listed as contributing factor. Error rates were constant for first eight days worked, then doubled for each on the next three. At day 12, the rate increase by six-fold.
Since all companies must comply with the API 755 guidelines recently published, the need for industry to better understand impact of personnel fatigue and the associated impact on incidents will help in the development of more accurate organizational fatigue risk management system (FRMS). If relationships of incidents to fatigue are found, then it impacts the FRMS that can be implemented for each organization.
Can data mining techniques be applied to incident reports to identify trends and patterns?
Within the oil industry, safety is a top priority. In the event of a near miss, an incident report is created that documents the event and meta-data surrounding the event. Each of these reports are reviewed and corrective actions are taken on a case-by-case basis but are never utilized in the future for further analysis.
To accommodate data mining, the information continuum is often referenced. The information continuum summarizes the process of transforming data to information, information to knowledge, and knowledge to prediction. For example, data can be the individual pieces of numbers or words contained within an incident report. Without field names and organization, the data is not very useful. However, by organizing and labeling the data in the form, the data is aggregated in a meaningful sense. The next step is transforming information into knowledge. This is done by looking at multiple reports and understanding the trends and associations of the reports as a whole. For example, a certain type of incident always occurs on Monday mornings or a specific weather factor is always present (rain). Finally, once we understand trends (Mondays), correlations (rain), and time (morning), we can begin to take predictive action by aggregating the elements of knowledge.
- Methodology to determine and mine data for correlations for all organizations
- Summary report with potential proof of correlation and causation or lack there of (Incidents to Fatigue)
- New or expanded ontology specific to industry