Examples of Machine Learning and Big Data Analytics Applications to Support OSH Management Processes in Real or Near-Real Time
A review of scientific literature and websites shows that the solutions for OSH process monitoring and management in real or near-real time based on advanced machine learning and big data analytics methods are still often at the stage of conceptual development or pilot testing in laboratory or industrial conditions. However, the rapid development of technologies supporting the Industry 4.0 concepts suggests that these solutions will soon become more mature and find more practical applications in various industry sectors.
The first example of a solution in this class is the system developed by Yang et al. (2016), which automatically detects and registers near-miss falls based upon kinematic data regarding workers captured by wearable inertial measurement units. The detection of near-miss falls is facilitated by analysing the kinematic data of workers by means of a semi-supervised learning algorithm (Support Vector Machine). The system will make it possible to predict risky locations or hazards in the iron-working industry based upon the detected locations of near-miss falls and will provide information that can be used for introducing proactive fall-prevention measures.
Another example is Al-based visual analytics technology that enables real-time video analysis with advanced algorithms and machine learning to verify if workers are using PPE items correctly. The system is offered by the UK-based Cortexica company under the name Al-PPE Compliant (Cortexica 2019; Peniak 2019). A similar purpose is that of another system that involves the use of intelligent vision-based analytics and hierarchical support vector machine algorithm for monitoring whether workers are wearing safety helmets at the workplace or not, and, at the same time, identifying colours of the helmets worn by workers (Wu and Zhao 2018). With regard to these examples, the high potential of Al-based video analytics in the area of OSH is underlined by Naylor (2019), who points to the possibility of applying these technologies to the detection of unsafe working practices and precursors of serious accidents in real time, as well as to auto-reporting of near misses, which will make is possible to deal with the current problem of significant incident under-reporting.
The next area where the harnessing of AI technology can significantly improve working conditions is the development of human-machine interfaces (HMI) which would be capable of automatically adapting to the current state of health and mental workload of operators controlling complex machinery or automated production systems. An example of this is the cyber-physical system supporting human-robot interaction and training to ensure safe human-robot collaboration in manufacturing environments (Tsiakas et al. 2017). The operator’s behaviour and mental state while interacting with a robot can be diagnosed using a variety of visual, physiological and linguistics modalities such as facial expressions, gestures, heart rate, temperature, language syntactic relations, etc. Based on this data and the algorithms of Interactive Reinforcement Learning the robot will be able to learn and adapt its collaboration policy to the needs and behaviours of individual operators.
Another promising direction for the development of AI systems to support OSH management processes is the introduction of non-invasive techniques of electroencephalography (EEG) to monitor stress and cognitive load of workers in real time. One of the solutions that applies EEG is a system diagnosing increased stress levels among construction workers (Jebelli et al. 2018). This system is based on the use of a wearable EEG headset (Emotiv EPOC+) and a machine learning algorithm called Gaussian Support Vector Machine. In turn, Neu et al. (2019) developed a concept of what is called the Cognitive Work Protection system, in which they proposed the use of EEG technique to measure workers’ cognitive conditions, e.g. the stress level or the ability to concentrate, in order to optimise the interaction with robots and machines in real time. The objective is to reduce the number of accidents at work that involve co-operation between humans and robots and to promote the physical and mental health of workers.
Yet another approach to using AI methods to improve working conditions is the system currently being conceptualised and developed within the EU-funded Ageing@work project. The main objective is to support ageing workers at home and at work so that they can become actively involved in working life for longer. The leading concept is the creation of a flexible working environment that will self-adapt to workers’ changing needs taking into account ergonomic principles (Giakoumis et al. 2019). To achieve this, the project consortium intends to develop, inter alia, the novel Ambient Virtual Coach, which will consist of an empathic avatar for the provision of subtle notifications, an adaptive Visual Analytics-based personal dashboard, and a reward-based motivation system targeting positive and balanced worker behaviour at work and in the home environment.
The Concept of OSH Management Based on Smart PPE and Wearable Technologies Combined with the Use of IoT and Big Data Analytics
As demonstrated in previous sections, recent developments in smart PPE systems, workplace wearables, IoT industrial networks and industrial applications of machine learning and big data analytics, as well as the growing number of other smart machines and devices being introduced into digitally transformed industries and other business sectors, provide the basis for proposing new approaches to OSH management that could be better suited to digital work environments. These approaches involve the use of a number of new functions, most of which will be delivered automatically by supporting technologies.
These functions are linked to the implementation of various processes and components of the OSH management system and, given their importance and relevance to the achievement of OSH objectives established in the company, it is proposed to divide them and assign roughly to the following three levels: basic, diagnostic, and strategic. The application of these functions is discussed below, taking into account the proposed division.
The basic level considers first of all a new role and importance of smart PPE systems in preventing and reducing OSH risks, while taking into account the relevant priorities for action resulting from the risk control hierarchy, as discussed in section 5.3.4. In addition, the basic functions provided by smart PPE and workplace wearables may also include automatic detection of accidents and near misses, as well as subsequent notification of supervisors, safety managers and/or emergency services on such events.
The next, the diagnostic level, covers other functions that are not directly related to real-time assessment and reduction of OSH risks, but are useful for ongoing monitoring of other processes and making appropriate managerial decisions. Increasingly advanced smart networked PPE systems will consist of an increasing number of sensors that will generate increasing amounts of contextual data in real time. These large datasets, multiplied by the increasing number of workers wearing various types of smart PPE, need to be subject to advanced machine learning and big data analytics in order to be thoroughly analysed, and the resulting conclusions implemented effectively to manage OSH. These methods are not yet fully used in this area, but they have great potential, especially in terms of diagnostic functions, which are needed to monitor the effectiveness of OSH management processes in terms of their compliance with established objectives, enable prediction of the effects of these processes on the state of OSH, and thus support managers in making appropriate decisions. Moreover, data on the effectiveness of individual OSH management processes may be used to better define, select and use key performance indicators (KPIs) that can be applied in this area.
And last but not least, IoT technologies combined with predictive data analytics are increasingly being used for what is knowrn as predictive maintenance to optimise the use of machinery and equipment by eliminating failures and optimising maintenance scheduling based on the measurement of performance factors in real time. Cloud-based services that offer data analysis for predictive maintenance are now available and could also be also applied to smart PPE systems or other intelligent safety-related devices, especially when their proper functioning is essential for the safety and health of workers.
The highest strategic level includes functions enabling the acquisition of knowledge that can be useful for strategic planning and shaping of OSH policy in an enterprise on a longer time scale. For example, using context-based data analytics services with respect to workers’ physiological parameters measured by means of smart PPE and wearables systems enables monitoring of workers’ health and provides information on individual workers or groups of workers who are more vulnerable or who need additional support or special protection. It is also possible to detect potential unsafe behaviour of workers (e.g. by comparing current data with data patterns obtained during accidents or near misses), and identify high-risk zones by geographical location of areas with higher rates of incidents (such as slips, falls, injuries, etc.) or with more frequent unsafe behaviour of workers.
The use of big data analytics methods can also provide many other useful insights and predictions that will improve OSH management at the strategic level. For example, correlating data on the current exposure of workers to different risk factors with the medical data on the health status of workers will help to assess the actual
FIGURE 6.6 OSH management functions facilitated by the use of smart PPE and workplace wearables combined IoT technologies, machine learning and big data analytics
impact of working conditions on workers’ health and thus, based on the relationships revealed, to apply the most appropriate preventive measures tailored to the needs of vulnerable groups of workers. Similarly, the conclusions derived from big data analytics may also allow the diagnosis of the general level of safety culture across the enterprise or within individual departments or groups of workers, which in turn may lead to a better identification of workers’ needs with regard to OSH training, or with regard to their involvement and participation in OSH-related activities.
The above discussion on new functions implemented by smart PPE and wearables combined with the use of IoT and big data analytics, as well as their assignment to different levels of OSH management, can be summarised in the form of a graphical model, which is presented in Figure 6.6.
Most of the advanced functions of OSH management that are based on the use of smart PPE and wearables combined with the use of IoT technologies and big data analytics are already being introduced in some workplace applications, or are still at the stage of development or pilot tests. Some other features are still at the conceptualisation stage, but one can optimistically assume that as the mentioned enabling technologies are being rapidly developed, these new functions will probably be available for introduction into practice in the near future.
