Classifying tasks performed by electrical line workers using a wrist-worn sensor: A data analytic approach.

Area of Science:

• Occupational Safety and Health
• Human Activity Recognition
• Wearable Technology

Background:

• Electrical line workers (ELWs) face challenging work conditions, including long hours and hazardous tasks.
• Wearable devices offer potential for monitoring ELW productivity and safety, but require accurate task identification.
• Existing human activity recognition research has limited application to the complex tasks of electrical line maintenance.

Purpose of the Study:

• To investigate feature engineering from wrist-worn accelerometers for classifying electrical line worker tasks.
• To evaluate the effectiveness of different feature sets and window lengths for task recognition.
• To assess both individualized (intra-subject) and generalized (inter-subject) model performance.

Main Methods:

• Utilized data from 37 participants performing ten common ELW tasks in a lab setting.
• Engineered features from a single wrist-worn accelerometer across time, frequency, and time-frequency domains.
• Compared three classifiers using 4-second and 10-second window lengths for task classification.

Main Results:

• Achieved classification accuracies of ≥93% for both intra-subject and inter-subject scenarios.
• Increased accuracy to ≥96% when using 10-second window lengths.
• Identified and explained the importance of specific features for accurate task prediction.

Conclusions:

• Wrist-worn accelerometers can reliably classify electrical line worker tasks.
• The developed method supports the development of wearable systems for enhanced worker safety and productivity.
• This research contributes to risk mitigation strategies for electrical line workers through advanced wearable technology.