Workplace safety has been a field of research that increasingly draws attention of the artificial intelligence scientific community. Automations are important to detect and prevent various types of incidents, ranging from accident prevention, human error detection, manual process substitution by robots, continuous surveillance, harassment reporting, crime prevention, and others. After over a decade of ground-breaking advancements in artificial intelligence (AI), relevant applications can be deployed to serve workplace safety automation. A variety of techniques such as computer vision, natural language processing, machine learning, neural networks, robotics, explainable decision making, and interconnected intelligent devices can collaborate to build accurate, modern systems that can be applied in any working environment, targeting a variety of challenges. The predictive power of contemporary AI systems is able to reduce human effort regarding monitoring, reporting, and predicting future incidents, thus enhancing employees' safety and satisfaction at work.

Autonomous and smart mines are predicted to become more prevalent. Automation has undeniable benefits in the mining industry, especially in terms of safety. However, automation has also led to unforeseen implications for individuals, organisations and communities. This study undertakes a systematic review of research on the impacts of automation in the mining context. A total of 94 documents that dealt with issues related to humans, safety and communities were found. Documents were analysed using both manual and natural language processing techniques. The review revealed the main concerns the industry must face for the successful implementation of automation, with interoperability and inadequate wireless networks identified as the most significant challenges. Key themes for individuals were workload, cognitive load, communication, acceptance of automation and trust. Task changes and culture were the most predominant issues at the organisational level. Impacts on employment and indigenous communities were highlighted at the community level. The emergence of advanced technologies and interoperability issues have implications for implementing of smart or intelligent mining. Human factors, precisely situation awareness and workload, have far-reaching consequences for safety and productivity because automation is becoming more complex. Moreover, not quantifying community impacts affects how companies can meet their corporate social responsibility commitments. Keywords Automation impacts , mining industry , human factors , natural language processing , autonomous , safety

Hydrogen’s potential as a clean energy vector has created increased research interest in recent years. The properties of hydrogen coupled with increased storage, transport and production volumes present safety hurdles which need to be understood for hydrogen to become an integral part of a clean energy system. Learning from past incidents is one method used by industry to prevent future incidents. Therefore, this work looked to published literature to decipher what can be learned from historical hydrogen incidents. The resultant systematic literature review found much published work referring to hydrogen and incidents or accidents. To gain insights from this literature, an unsupervised clustering algorithm was used to analyse the abstracts and uncover underlying patterns and groups. The algorithm is unsupervised because no prior knowledge is assumed with regards to the potential groups or patterns in the dataset. The mean silhouette co-efficient was calculated for each abstract to improve the algorithms ability to detect patterns in the short, relatively similar pieces of text. The work revealed that literature referring to incidents is quite diverse, with less than 10% of sources focusing on actual incidents. Manual coding of literature that reviewed actual incidents was also completed. Reviews of actual incidents included works which analyse incident databases and works which review single case studies. The single case studies predominantly offer a narrative of events without formal analysis using risk management or incident investigation techniques. The sparsity of published literature on actual incidents suggests that an opportunity exists to improve the reporting of hydrogen incidents. Arguably, not all incidents will have research merit. However, quality published data could be used by industry and academics to guide hydrogen safety improvement endeavours. This points to a need for further work on which incidents would be candidates for further research and how to get industry to engage.

In this chapter, the authors explore the theoretical and practical aspects of using text mining approaches supported by machine learning for the automatic interpretation of bulk literature on a contemporary issue—that of climate change risk analysis. The strengths, weaknesses, and opportunities associated with these approaches are investigated. Text mining provides a way to automate and enhance the analysis of text data. However, contrary to popular belief, text mining analysis is not a completely automated process. As with computer-assisted (or -aided) qualitative data analysis software (CAQDAS), it is an iterative method requiring input from a researcher with expert knowledge and a deliberate approach to the analysis. Given the heterogeneity that generally characterizes climate disclosures, the authors postulate that hybrid methodologies are ideal for analysing textual data related to climate change discourse. The authors also demonstrate that text mining is an open and evolving field, in the sense that it can be combined with other approaches to shed new light on the climate discourse.

There seems to be no agreed taxonomy for climate-related risks. The information in firms' climate risk disclosures represents a new resource for identifying the priorities and strategies of Australian companies' management of climate risk. This research surveys 839 companies listed on the Australian Stock Exchange for the presence of climate risk disclosures, identifying 201 disclosures on climate risk. The types of climate risks and the risk management strategies were extracted and evaluated using machine learning. The analysis revealed that Australian firms are focused on acute physical climate risks, followed by market and regulatory risks. The predominant management strategy for these risks was to use a risk reduction approach, rather than avoiding or transferring risk. The analysis showed that key Australian industry sectors, such as materials, banking, insurance, and energy are focusing on different mixtures of risk types, but they are all primarily managing risks through risk-reduction strategies. An underlying driver of climate risk disclosure was composed of the financial implications of climate risk, particularly with respect to acute physical risks. The research showed that emission reductions represent a primary consideration for Australian firms in their disclosures identifying how they are responding to climate risk. Further research using machine learning to evaluate climate risk disclosure should focus on analysing entire climate risk reports for key topics and trends over time.

Industrial accident-related records are continually collected and stored by public and private organizations creating big data repositories. Along with others, narrative-based documents of various industrial accidents are becoming continuously more valuable for computer-assisted analysis of unwanted events owing to the advances of machine learning in recent years. Nevertheless, vital narrative data comprised in text documents, for example, detailing unwanted event descriptions and descriptions of more subtle circumstances along with novel computer-based detection methods still require scientific exploration. Therefore, this study aims to improve data processing techniques to provide unsupervised anomaly detection that can be utilized by safety experts to efficiently evaluate accident risk factors. Variational autoencoder, a deep learning-based method, is introduced to obtain the CSB accident reports anomaly factor ranking. In the next step, an analysis of key terms is performed to review the underlying causes of selected accidents. As a result, vital narrative-based information buried in a myriad of accident report documents is analyzed, providing quantitative insights on risk factors. The anomalous events can be investigated further by safety experts to fully understand and explore the unique causes for their occurrence. The anomaly detection solution presented in this paper is an adaptive method that can be extended to other use cases to enable the analysis of complex narrative data in predictive and prescriptive implementation scenarios.

Accurate inferences of the emotional state of conversation participants can be critical in shaping analysis and interpretation of conversational exchanges. In qualitative analyses of discourse, most labelling of the perceived emotional state of conversation participants is performed by hand, and is limited to selected moments where an analyst may believe that emotional information is valuable for interpretation. This reliance on manual labelling processes can have implications for repeatability and objectivity, both in terms of accuracy, but also in terms of changes in emotional state that might go unnoticed. In this paper we introduce a qualitative discourse analytic support method intended to support the labelling of emotional state of conversational participants over time. We demonstrate the utility of the technique using a suite of well-studied broadcast interviews, taking a particular focus on identifying instances of inter-speaker conflict. Our findings indicate that this two-step machine learning approach can help decode how moments of conflict arise, sustain, and are resolved through the mapping of emotion over time. We show how such a method can provide useful evidence of the change in emotional state by interlocutors which could be useful to prompt and support further in-depth study.

Objective: To evaluate the application of a Deep Learning based emotion recognition system for detecting operator stress, where operator stress is a proxy for Situation Awareness (SA) changes during abnormal/contingency situation management and decision making. Background: When operators are overwhelmed by stress, their perceptions, thinking, and judgments are impaired, increasing the chance of misinterpretation of events and increasing the potential for human error. The "intelligent control room" has been proposed as a possible solution for helping operators to deal with such stress. The control room comprises a variety of components used to monitor the operator and infrastructure under his/her control in an effort to optimize the performance of the human-technical system as a whole. A critical component of this control room solution is the provision of human monitoring and assessment data in order to determine the operator's situation awareness. Methods: An emotion recognition system is designed based on two Deep Learning models, the Bidirectional Long Short Term Memory network (BiD-LSTM) and the Deep Convolutional Neural Network (DCNN), in order to process audio and facial data respectively. The system is first validated against a standard corpus of expert-coded emotion data. Post-validation, a dataset of expert-coded user stress data is coded by the system for emotional valence, and these system-generated emotional readings are compared to the expert-coded stress markers to determine any significant correlations. Contribution: This research contributes to developing the idea of intelligent and automated decision-making support in situational awareness measurement systems. Such systems support users by real-time collecting and processing data, and assist decision-making based on operator behavioral patterns.

Robots that facilitate touch by children have special requirements in terms of safety and robustness, but little is known about how and when children actually use touch with robots. Tools and techniques are required to sense the variety of children's touch and to interpret the volumes of data generated. This explorative user study investigated children's patterns of touch during game play with a robot. We examined where the children touch the robot and their patterns of touch over time, using a raster-based visualisation of each child's time series of touches, recording patterns of touch across different games and children. We found that children readily engage with the robot, in particular spontaneously touching the robot's hands more than any other area. This user study and the tools developed may aid future designs of robots to autonomously detect when they have been touched.

Pretotyping is a set of techniques, tools, and metrics for gauging the interest in a product, prior to full-scale development [1]. This late breaking report describes a pretotyping case study of an ethnodroid -- a robot that functions as an ethnographer -- intended to engage with young children and record their learning progress. The central requirement for the project is that the robot will be able to interact socially with children aged 1-6 years in tablet-based tasks. We developed a simple robot made of MDF (thick cardboard), added tablets for the face and torso, and controlled a scripted interaction using Wizard of Oz (WoZ). Children's engagement with the robot was tested in an early learning centre which provided a relatively structured environment ("in the lab") and at a science fair which provided a relatively unconstrained setting ("in the wild"). The rapid testing revealed distinct effects in the children's attitudes and behaviors in the two user contexts and provided insights into form, sensors and analyses for the design process.