Towards a Proactive Real-Time Data-Driven Musculoskeletal Disorders Prevention Approach in Construction
Open Access
- Author:
- Zhao, Junqi
- Graduate Program:
- Architectural Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- February 26, 2020
- Committee Members:
- Esther Adhiambo Obonyo, Dissertation Advisor/Co-Advisor
Esther Adhiambo Obonyo, Committee Chair/Co-Chair
Somayeh Asadi, Committee Member
Sven G Bilen, Committee Member
Andris Freivalds, Outside Member
Sez Atamturktur, Program Head/Chair - Keywords:
- Construction Safety Management
Proactive Injury Prevention
Wearable Sensing
Machine Learning
Deep Learning
Ergonomics
Construction Safety Management
Injury Prevention
Wearable Sensing
Machine Learning
Deep Learning
Ergonomics - Abstract:
- Safety is a significant challenge in the construction industry. The U.S. construction sector employs approximately 4.5% of the total labor. Between 2008 and 2017, it accounted for 18.9% of fatal injuries and 7.6% of nonfatal injuries of all industries. Labor-intensive construction activities leave workers susceptible to developing Musculoskeletal Disorders (MSDs), which are soft tissue injuries, such as sprains, strains, and cumulative trauma disorder. MSDs account for nearly one-third of the total costs spent on workers’ compensation in the U.S. Employers spend as much as $20 billion annually on direct cost for MSD-related compensation, which is higher than most of the other sectors. MSDs develop through repetitive exposure to risk factors, particularly the physical-related factors. Conventional construction safety management practices tend to focus on early identification of injury risk factors in safety planning and enhancing workers’ safety awareness through safety training. Such approaches predicate on the workers’ compliance with safety practices and the extent to which they can modify their behavior to minimize being overexposed to risk factors. Many of such efforts rely on observation-based inspection. Because this approach cannot be used to effectively monitor the emerging risk factors, it leaves workers vulnerable to developing MSDs. Approaches that can identify emerging injury risk in a proactive manner could be used by workers and their supervisors for proactive injury prevention. Emerging data-sensing technology, when enhanced with Machine Learning (ML) and Deep Learning (DL) techniques, can enable proactive injury prevention in construction. The captured data can be converted to actionable safety information for use in automated risk monitoring – the benefit here is that this approach can trigger one to take preventive actions before the injury occurs. The current body of knowledge concerning capturing the MSD-related data from workers in an effective manner, transferring the captured data into actionable safety information, and leveraging the safety information for proactive injury prevention cannot be readily deployed into solutions for the construction jobsite. This doctoral research focused on addressing these barriers. The overarching goal of the research was designing and developing a Real-Time Data-Driven approach that leverages Wearable Sensing (WS), ML, and DL technologies to facilitate the identification and mitigation of MSDs risks. The deployment context was proactive MSDs prevention in construction. The specific research objectives were: i) identifying the leading risk factors and related ergonomics assessment rules for MSDs; ii) assessing the performance of emerging sensing techniques in capturing MSD-related motion data; iii) identifying data processing approaches for monitoring and assessing risks of MSDs; iv) exploring a concept for a Decision Support System (DSS) that can be used to integrate the enabling technologies for proactive MSDs prevention; and v) evaluating the developed proactive MSDs prevention concept through obtaining the targeted end-users’ views and opinions on the system performance. Multiple research approaches were applied in this study to achieve the research objectives. The Literature Review identified the leading risk factors and applicable ergonomics assessment rules for evaluating MSDs risks. The technologies that can be used to capture, analyze, and communicate safety information were reviewed and critiqued with respect to the suitability for use in proactive MSDs prevention. Findings from the comprehensive review were used to design a concept for the proposed MSDs prevention approach. A Rapid Prototyping approach was applied to allow developing the concept further through an iterative “design-develop-test” process. Four subsystems underpin the functionality of the proposed MSD-prevention approach - the wearable Inertial Measurement Units (IMUs)-based Motion Data Capturing System, Data-Driven Posture Recognition Models, posture-based MSDs Risk Assessment Models, and Assessment Information Delivery System. The resulting prototype was validation through field interactions with 21 construction workers as participants. The developed Motion Data Capturing System and Posture Recognition Models were evaluated using subjects’ motion data captured during their daily tasks on jobsites. The developed MSDs Assessment Models were validated using captured workers’ postures. The usability, acceptability, and functionality of the developed proactive MSDs prevention approach were evaluated through an End-user Evaluation Survey of 30 subjects (18 workers and 12 construction managers). The data collected also included end-users’ feedback, expectations, and suggestions for further improvement. This doctoral research has established that awkward working posture is a leading risk factor for MSDs. The proportion of awkward postures and continuous posture holding time can be used for quantitative MSDs risk assessment with identified ergonomics rules, including Ovako Working Posture Assessment System (OWAS) and Maximum Holding Time (MHT). The research demonstrated the feasibility of using low-cost (up to $64 a unit) wearable IMUs sensors for motion data capturing on construction sites. The findings from the research indicate that data-driven posture recognition models can be used to detect the MSD-related postures from IMUs output with high accuracy and computational efficiency. The personalized ML-based model developed in this research achieved recognition accuracy levels of 0.74 and 0.81 for upper body and lower body postures in the laboratory test. ML model operation time was negligible. The DL-based recognition models developed in this research resulted in the recognition performance (Macro F1 score) of 0.82 and 0.87 under personalized and generalized modeling approaches, respectively. There was an observed performance improvement over conventional ML models of 6.3%. Additionally, the developed DL model used one second to recognize a worker’s postures conducted in 256 seconds, showing a high computational efficiency towards real-time deployment. When posture recognition models were applied to MSDs risk assessment, the detected awkward posture usage proportion was close to the actual proportion. There was also zero misdetection of postures breaching the MHT thresholds when using postures recognized by developed DL-based recognition model. A concept for a DSS for interfacing with the construction workers and their supervisors was developed. The targeted end-users rated this concept as being highly useful for MSDs prevention. Their suggestions for improvement included enhancing the sensor robustness/durability for ease of deployment on construction jobsites. This doctoral research’s main contribution to knowledge is the design, development and validation of a concept for a near real-time Data-Driven MSDs prevention approach in construction that leverages emerging wearable motion sensing, DL and DL based posture recognition, automated injury risk assessment, and information communication techniques. The proposed approach has validated the feasibility of identifying and evaluating MSDs risks from workers at selected construction jobsites, delivering the actionable insights that can be used to proactively take action to minimize the risk of MSDs, and supporting evidence-based safety management decision making by construction managers. As was previously indicated, MSDs emerge because of repetitive exposure to physical strain. Because the underlying risk factors can be correlated to over-exertion of the same set of muscles, the Data-Driven MSDs prevention approach from this doctoral research can be adapted for use in other labor-intensive sectors such as manufacturing, healthcare, and agriculture.