To achieve the goal of preventing serious injuries and fatalities, it is important for a mine site to analyze site specific mine safety data. The advances in natural language processing (NLP) create an opportunity to develop machine learning (ML) tools to automate analysis of mine health and safety management systems (HSMS) data without requiring experts at every mine site. As a demonstration, nine random forest (RF) models were developed to classify narratives from the Mine Safety and Health Administration (MSHA) database into nine accident types. MSHA accident categories are quite descriptive and are, thus, a proxy for high level understanding of the incidents. A single model developed to classify narratives into a single category was more effective than a single model that classified narratives into different categories. The developed models were then applied to narratives taken from a mine HSMS (non-MSHA), to classify them into MSHA accident categories. About two thirds of the non-MSHA narratives were automatically classified by the RF models. The automatically classified narratives were then evaluated manually. The evaluation showed an accuracy of 96% for automated classifications. The near perfect classification of non-MSHA narratives by MSHA based machine learning models demonstrates that NLP can be a powerful tool to analyze HSMS data.

In a mine, knowledge of rock types is often desired as they are important indicators of grade, mineral processing complications, or geotechnical attributes. It is common to model the rock types with visual graphics tools using geologist-generated rock type information in exploration drillhole databases. Instead of this manual approach, this paper used random forest (RF), a machine learning (ML) algorithm, to model the rock type at Erdenet Copper Mine, Mongolia. Exploration drillhole data was used to develop the RF models and predict the rock type based on the coordinates of locations. Data selection and model evaluation methods were designed to ensure applicability for real life scenarios. In the scenario where rock type is predicted close to locations where information is available (such as in blocks being blasted), RF did very well with an overall success rate (OSR) of 89%. In the scenario where rock type was predicted for two future benches (i.e., 30 m below known locations), the best OSR was 86%. When an exploration program was simulated, performance was poor with a OSR of 59%. The results indicate that EMC can leverage RF models for short-term and long-term planning by predicting rock types within drilling blocks or future blocks quite accurately.

Contextual representation has taken center stage in Natural Language Processing (NLP) in the recent past. Models such as Bidirectional Encoder Representations from Transformers (BERT) have found tremendous success in the arena. As a first attempt in the mining industry, in the current work, BERT architecture is adapted in developing the MineBERT model to accomplish the classification of accident narratives from the US Mine Safety and Health Administration (MSHA) data set. In the past multi-year research, several machine learning (ML) methods were used by authors to improve classification success rates in nine significant MSHA accident categories. Out of nine, for four major categories (“Type Groups”) and five “narrow groups”, Random Forests (RF) registered 75% and 42% classification success rates, respectively, on average, while keeping the false positives under 5%. Feature-based innovative NLP methods such as accident-specific expert choice vocabulary (ASECV) and similarity score (SS) methods were developed to improve upon the RF success rates. A combination of all these methods (“Stacked” approach) is able to slightly improve success over RF (71%) to 73.28% for the major category “Caught-in”. Homographs in narratives are identified as the major problem that was preventing further success. Their presence was creating ambiguity problems for classification algorithms. Adaptation of BERT effectively solved the problem. When compared to RF, MineBERT implementation improved success rates among major and narrow groups by 13% and 32%, respectively, while keeping the false positives under 1%, which is very significant. However, BERT implementation in the mining industry, which has unique technical aspects and jargon, brought a set of challenges in terms of preparation of data, selection of hyperparameters, and fine-tuning the model to achieve the best performance, which was met in the current research.

The lockdown policies enacted in the spring of 2020, in response to the growing severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic, have remained a contentious policy tool due to the variability of outcomes they produced for some populations. While ongoing research has illustrated the unequal impact of Coronavirus disease (COVID-19) on minority populations, research in this area has been unable to fully explain the mechanisms that produce these findings. To understand why some groups have been at greater risk of contracting COVID-19, we employ structural inequality theory to better understand how inequality may impact disease transmission in a pandemic. We used a novel approach that enabled us to focus on the microprocesses of structural inequality at the zip code level to study the impact of stay-at-home pandemic policies on COVID-19 positive case rates in an urban setting across three periods of policy implementation. We then analyzed data on traffic volume, income, race, occupation, and instances of COVID-19 positive cases for each zip code in Salt Lake County, Utah (USA) between 17 February 2020 and 12 June 2020. We found that higher income, percent white, and white-collar zip codes had a greater response to the local stay-at-home order and reduced vehicular traffic by nearly 50% during lockdown. The least affluent zip codes only showed a 15% traffic decrease and had COVID-19 rates nearly 10 times higher. At this level of granularity, income and occupation were both associated with COVID-19 outcomes across all three stages of policy implementation, while race was only predictive of outcomes after the lockdown period. Our findings illuminate underlying mechanisms of structural inequality that may have facilitated unequal COVID-19 incidence rates. This study illustrates the need for more granular analyses in policy research and adds to the literature on how structural factors such as income, race, and occupation contribute to disease transmission in a pandemic.

The mining industry is diligent about reporting on safety incidents. However, these reports are not necessarily analyzed holistically to gain deep insights. Previously, it was demonstrated that mine accident narratives at a partner mine site could be automatically classified using natural language processing (NLP)-based random forest (RF) models developed, using narratives from the United States Mine Safety and Health Administration (MSHA) database. Classification of narratives is important from a holistic perspective as it affects safety intervention strategies. This paper continued the work to improve the RF classification performance in the category “caught in”. In this context, three approaches were presented in the paper. At first, two new methods were developed, named, the similarity score (SS) method and the accident-specific expert choice vocabulary (ASECV) method. The SS method focused on words or phrases that occurred most frequently, while the ASECV, a heuristic approach, focused on a narrow set of phrases. The two methods were tested with a series of experiments (iterations) on the MSHA narratives of accident category “caught in”. The SS method was not very successful due to its high false positive rates. The ASECV method, on the other hand, had low false positive rates. As a third approach (the “stacking” method), when a highly successful incidence (iteration) from ASECV method was applied in combination with the previously developed RF model (by stacking), the overall predictability of the combined model improved from 71% to 73.28%. Thus, the research showed that some phrases are key to describing particular (“caught in” in this case) types of accidents.

The economics of a mineral processing circuit is dependent on the numerous sensors that are critical to optimization and control systems. The sheer volume of sensors results in mines not being able to keep the sensors calibrated. When a sensor goes off calibration, it results in errors which can severely impact plant economics. Normally, these errors are not detected until the error grows and becomes a large (“gross”) error. Undetected errors are not remedied until the next calibration, which on average are a year apart Beamex (2019) [1] across all industries. Classical statistical methods that depend on linear relationships between input and response variables are typically used to find gross errors, but are not effective for highly non-linear and non-stationary operations in mining and mineral processing industries. Calibration of sensors is time-consuming and generally requires physical intervention that results in equipment downtimes causing production losses. Therefore, in situ detection methods are warranted that do not require physical removal of sensors to compare with standard measurements or well-calibrated sensors. Data-mining based techniques and algorithms have high success rate in tackling such problems. This paper presents results from a groundbreaking multi-year big data research project whose goal was to develop data-mining-based original techniques in a multi-sensor environment to identify when sensors start to stray, rather than wait for errors to grow. The carbon stripping (gold) circuit in the Pogo Mine of Alaska was selected for this project. Maintaining strip vessel temperatures at optimum levels is crucial for maximizing gold recoveries at Pogo, and hence the focus was on the temperature sensors of the two strip vessels. An automated algorithm to detect errors in strip vessel temperature sensors was developed that exploited data from multiple sensors in the strip circuit. This algorithm was able to detect with a high success rate when bias errors of magnitude as low as ± 2% were artificially induced into strip vessel temperature data streams. The detection times of about 1 month were much less than the industry average calibration frequency of 1 year. Thus, the algorithm helps detect errors early, without stopping the circuit. The algorithm alarms can serve as triggers when calibrations are done. Experimental methodology along with results, limitations of algorithms, and future research are also presented in this paper.