ContextSensitive System for Particulate Matter
Abstract
Aerosol exposure is a major concern in a broad range of military-unique occupational conditions, such as aircraft maintenance and deployments. In this project, we integrated a particulate matter (PM) sensor into a wearable device, the U Iowa Personal Monitor, that is able to provide timely, context-sensitive information on the activities that a person experiences. This context awareness functionality was achieved through the use of a custom deep learning model based on a convolutional neural network (CNN). This system is able to predict the current task being performed by using both time and frequency domain features from accelerometery and sound data gathered by a suite of built-in sensors. In parallel with development of the U Iowa Personal Monitor hardware, we developed the foundation of our CNN-based task classification model using an existing data set. The data set included continuous, full-shift recordings of upper arm acceleration (triaxial; 20 Hz sampling rate) among eight workers. Overall task classification accuracy was poorest when using only audio data (72.1%) and improved when using only the accelerometer data (85.2%). However, as expected, the combination of accelerometer and audio data yielded the greatest overall task classification accuracy (92.4%, with 5 of 6 task-specific classification accuracies >90%). This system will help workers identify hazardous aerosol exposures, elucidating where the hazards occur during their daily activities. It will automatically link hazardous exposures to specific tasks within the workplace with no or minimal involvement of a health and safety professional.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2020
- Accession Number
- AD1092371
Entities
People
- Thomas Peters