Fusing imperfect experimental data for risk assessment of musculoskeletal disorders in construction using canonical polyadic decomposition

Highlights

• Missing data is a common problem in data collection for WMSD risk assessment.

• A data fusion method was developed to tackle this problem.

• Canonical Polyadic Decomposition was applied in the method development.

• Two real WMSD risk datasets were used to validate the developed method.

• The method was found useful in handling missing data for reliable risk assessment.

Abstract

Field or laboratory data collected for work-related musculoskeletal disorder (WMSD) risk assessment in construction often becomes unreliable as a large amount of data go missing due to technology-induced errors, instrument failures or sometimes at random. Missing data can adversely affect the assessment conclusions. This study proposes a method that applies Canonical Polyadic Decomposition (CPD) tensor decomposition to fuse multiple sparse risk-related datasets and fill in missing data by leveraging the correlation among multiple risk indicators within those datasets. Two knee WMSD risk-related datasets—3D knee rotation (kinematics) and electromyography (EMG) of five knee postural muscles—collected from previous studies were used for the validation and demonstration of the proposed method. The analysis results revealed that for a large portion of missing values (40%), the proposed method can generate a fused dataset that provides reliable risk assessment results highly consistent (70%–87%) with those obtained from the original experimental datasets. This signified the usefulness of the proposed method for use in WMSD risk assessment studies when data collection is affected by a significant amount of missing data, which will facilitate reliable assessment of WMSD risks among construction workers. In the future, findings of this study will be implemented to explore whether, and to what extent, the fused dataset outperforms the datasets with missing values by comparing consistencies of the risk assessment results obtained from these datasets for further investigation of the fusion performance.

Introduction

Work-related musculoskeletal disorders (WMSDs) are one of the most common causes of days away from work and physical disabilities in the construction industry [1]. An increased exposure to risk factors in the workplace can enhance the likelihood of WMSDs; hence, proper identification of possible risk exposures and developing injury prevention strategies are essential to alleviate WMSDs.

Collecting risk exposure data with human subject involvement is most often accepted as the gold standard for understanding risky behaviors and conditions that may expose workers to WMSD risks on construction sites [2]. Generally, these data are collected in laboratory settings or real construction sites by technologies such as optical motion capture systems or surface electromyography sensors. However, data collected from these technologies often suffer from ‘drop out’, a phenomenon in which data is missing due to technology-induced errors (e.g., disconnection of sensors, errors in communicating with the database server, instrument failures), human-induced errors (e.g., accidental human omission) or other unknown reasons [3]. The result is incompleteness of the collected risk exposure data that may lead to invalid conclusions on the effects of the potential WMSD risk factors. Missing data is a common problem associated with data collection in ergonomic risk assessment using technologies, regardless of the quality of the research design [4]. Therefore, it should be carefully handled. In doing so, reserving the interrelation among the potential risk factors and the risk indicators across multiple datasets is necessary. Among several benefits of data fusion, one is revealing the latent pattern of the data and leveraging collaborative relationships among various factors within multiple datasets based on that pattern. This benefit can be utilized to reserve the interrelation among different factors and the risk indicators across the datasets to fill in the missing data. This study proposes a method for dealing with multiple imperfect and incomplete datasets by applying a Canonical Polyadic Decomposition (CPD) technique to treat the imperfect data for WMSD risk assessment. CPD decomposes the incomplete datasets based on the latent relationship among different risk factors and the risk indicators, then reconstructs a new dataset through fusion as a high-order tensor [5]. This newly reconstructed dataset is referred to as fused dataset, which can then be used for assessing the risk of WMSDs. To validate the effectiveness of the CPD-based method in assessing WMSDs, two WMSD risk-related datasets collected from prior experimental studies (original datasets) were intentionally modified to represent incomplete datasets. Then CPD was applied for fusion and to reconstruct the fused datasets. The risk assessment results obtained using the fused datasets were further compared to those obtained by using the original datasets to evaluate the performance of the fusion treatment.