← Back to Sessions

Toward Neural Network Based Automated Structural Health Monitoring With MyShake Smartphones

Description: 

Effective Structural Health Monitoring (SHM) is crucial for maintaining the safety, resilience, and integrity of buildings and infrastructure. However, traditional SHM methods often rely on costly, labor-intensive sensor installations, limiting scalability. The widespread availability of smartphones with built-in accelerometers offers an innovative, cost-effective alternative for SHM. The MyShake smartphone app, developed at UC Berkeley, harnesses this potential by collecting acceleration waveforms during seismic events and ambient vibrations to monitor the dynamic properties of buildings.

MyShake, part of California’s Earthquake Early Warning initiative, has been downloaded over 2.9 million times globally. It delivers earthquake early warnings and autonomously records three-component waveforms during significant shaking or ambient vibrations. Since smartphones are often stationary within buildings, these recordings provide critical insights into ground motion and structural behavior. Previous research has shown MyShake’s ability to capture natural frequencies of buildings during seismic and wind loading events, proving its viability for SHM.

To scale this approach, we developed an artificial neural network (ANN)-based automated workflow for analyzing natural frequencies and damping properties of buildings. We curated a dataset of 3,000 ambient vibration records, manually labeled for measurement quality, to train the ANN. This workflow automates the association of smartphones with buildings, evaluates waveform quality using statistical parameters, and applies power spectral density (PSD) analysis to compute dynamic building properties. Controlled deployments in the San Francisco Bay Area validated the method’s accuracy across structural types.

Our results confirm the feasibility of using MyShake-enabled smartphones for reliable SHM under strong and ambient dynamic loading conditions. This integration of crowdsourced smartphone data and machine learning provides a scalable, efficient solution for global SHM, paving the way for proactive maintenance and improved infrastructure safety.

Session: Why Ignore the Structure? Soil-structure Interaction and Site Response at Local and Regional Scales [Poster]

Type: Poster

Date: 4/17/2025

Presentation Time: 08:00 AM (local time)

Presenting Author: Utpal

Student Presenter: No

Invited Presentation: 

Poster Number: 10

Authors