Productivity Measurement through IMU-Based Detailed Activity Recognition Using Machine Learning: A Case Study of...

Area of Science:

• Construction Management
• Industrial Engineering
• Human Factors Engineering

Background:

• Measuring individual worker productivity on dispersed construction sites is a significant challenge.
• Existing methods often lack the granularity to capture detailed activities and their impact on productivity.
• Technological advancements offer potential solutions for objective productivity assessment.

Purpose of the Study:

• To develop and validate a framework for measuring construction worker productivity using inertial measurement units (IMUs) and activity classification.
• To assess the feasibility of deep learning algorithms for classifying worker activities in masonry work.
• To explore the impact of sensor configuration on the accuracy of productivity measurement.

Main Methods:

• Utilized inertial measurement units (IMUs) to collect motion data from workers.
• Applied two deep learning models: Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM).
• Evaluated three different sensor combinations for activity classification and productivity measurement in masonry tasks.

Main Results:

• Achieved a maximum worker activity classification accuracy of 96.70% with a CNN model and multiple sensors.
• Demonstrated a minimum activity classification accuracy of 72.11% using an LSTM model with a single sensor.
• Enabled productivity measurement with an accuracy of up to 96.47%.

Conclusions:

• The proposed framework effectively measures construction worker productivity through IMU-based activity classification.
• Deep learning models, particularly CNNs with multiple sensors, show high potential for accurate activity recognition.
• The number of IMU sensors significantly influences the accuracy of productivity measurement, highlighting the importance of optimal sensor placement.