Mining Project: Identification of Key Factors Affecting Machine-related Fatalities and Injuries in MNM Mining Sectors

Research Summary

Machine-related accidents are currently one of the major causes of fatalities and injuries across the metal/nonmetal (MNM) and stone, sand, and gravel (SSG) mining sectors. The SSG sector, which has a relatively large number of small mine operators, has particularly high non-fatal incident rates. Statistically, machinery and powered haulage lead to a significant number of fatal injuries in mining. Researchers have pointed out that human-machine interactions lead to most of the fatalities and injuries [Groves et al. 2007; Kecojevic et al. 2007; Ruff et al. 2011; Zhang et al. 2014]. Mine Safety and Health Administration (MSHA) accident, illness, and injury data collected during 2010-2017 also show that 48% of all the MNM and SSG mining sectors fatalities were attributed to machinery and powered haulage, with 17% leading to non-fatal injuries.

Most past studies on machine-related accidents have relied heavily on the MSHA accident and injury dataset; however, there was a need to further interpret MSHA fatality reports from 2010-2017 and conduct an in-depth analysis to identify key contributing factors. The NIOSH Mining Program has the expertise and resources in engineering and statistical analysis for conducting such research.

As a one-year pilot project, this research guided the development of the MNM and SSG mining machine safety research portfolio for the Mining Program. The sole research aim of this project was to conduct a root cause analysis of machine-related fatalities and injuries in the MNM and SSG mining sectors in order to determine focus areas and gaps for future research. Text analytics techniques via the MATLAB Text Analytics Toolbox and R Quanteda were utilized for topic modeling and unigram co-occurrence plot analysis. The Latent Dirichlet Allocation (LDA) model was used for topic modeling in order to cluster similar text—i.e. accident reports with similar underlying themes. Natural language processing (NLP) was used for information extraction and part-of-speech (POS) tagging in the text. All accidents involving machinery and powered haulage as classified by MSHA were considered in this research.

The process of applying text mining and NLP to the MSHA fatality reports involves several steps. The first step (data acquisition) was to acquire text data from the MSHA site. Then the text data was converted to tabular data. After the tabulation of reports, text mining techniques such as topic modeling were applied to identify and assign underlying topics for the reports. Before applying the topic modeling technique, texts went through data preprocessing where they were broken into individual units and cleaned of the most frequently occurring words. Topic modeling was done for a set of reports to cluster them based on similar words found in those reports. Because the accident reports used in this study were based on MSHA accident classifications machinery and powered haulage, there were two sets of reports. The LDA topic modeling technique determined the suitable number of topics for each set of reports and clustered the reports randomly based on topic number.

Figure 1. Example co-occurrence plot for powered haulage accidents. Click for larger image.

After topic modeling, a new set of incident reports based on the topics was selected for the feature co-occurrence matrix and plot analysis. This was done using Quanteda toolkit from R. As an example, the co-occurrence plot for the topic of powered haulage accidents is presented in Figure 1. This topic was associated with 15 haul truck accidents, and unigram co-occurrence shows the word “truck” at the center of the plot. This co-occurrence plot gives some insight into the conditions surrounding the incident.

NLP was used after the text data was extracted from the fatality reports. This was done in the Python programming language using the spaCy package. Next, a  “chunking method” was used to extract key nouns and verbs and the words associated with them, aiding researchers in analyzing the causes associated with machinery-related fatalities and injuries. Text mining and NLP were used for MSHA final fatality reports analysis. Topic modeling along with the co-occurrence plots provided insight into text clusters, which were helpful in clustering reports based on similar underlying themes or topics. By comparison, the application of NLP proved to be more effective for the understanding of sentence structure and to be helpful in information extraction.

This project successfully identified some of the key causes that lead to machinery and powered haulage fatalities, including loss of control, improper guarding, lack of communication before any movement, lack of task training, lockout/tagout, and improper maintenance.  The most common type of machines involved in the accidents were haul trucks, belt conveyors, dozers, and drill machines. Text mining and NLP proved to be helpful tools for exploring text data in a short amount of time and in drawing meaningful conclusions from that data.

Related Conference Paper

Raj KV, Tarshizi EK [2020]. Advanced application of text analytics in MSHA metal and nonmetal fatality reports. Presented at the SME Annual Meeting & Expo, Phoenix, AZ, February 23-26, 2020. Preprint#20-111. pp. 1-6. 

References

Groves WA, Kecojevic VJ, Komljenovic D [2007]. Analysis of fatalities and injuries involving mining equipment. J Safety Res 38(4):461-470.

Kecojevic V, Komljenovic D, Groves W, Radomsky M [2007]. An analysis of equipment-related fatal accidents in U.S. mining operations: 1995-2005. Safety Sci 45(8):864-874.

Ruff T, Coleman P, Martini L [2011]. Machine-related injuries in the U.S. mining industry and priorities for safety research. Inter J Injury Control and Safety Prom 18(1):11-20.

Zhang M, Kecojevic V, Komljenovic D [2014]. Investigation of haul truck-related fatal accidents in surface mining using fault tree analysis. Safety Sci 65:106-117.