Low-Cost Detection of Structural Behavior Changes Using Smartphone Vibration Data

Overview

This project addresses an inspection gap in structural monitoring because loosened connections or developing cracks can increase the risk of collapse. However, frequent vibration monitoring is too expensive for wide deployment, and visual inspection misses minor changes. The proposed solution is a low-cost screening system using a smartphone accelerometer and logistic-regression classifier to label a short impact test as baseline-consistent (no immediate inspection indicated) or changed-flag (inspection recommended). An aluminum cantilever beam with a 24-inch free span was used as a test structure because, like larger structures, its vibration response changes when stiffness/effective mass changes. For each trial, the beam was clamped, a smartphone was mounted, the beam was struck with a rubber mallet, and accelerometer data was exported to CSV. Changes were introduced by adding 10g, 20g, and 40g masses near the free end. The code isolates the ring-down, and computes the power spectral density to extract the first bending frequency f_1. It also collects metrics like percent shift from baseline, decay rate, and peak strength, combining these features into a probability of change and using logistic regression to apply a fixed cutoff to assign the final label. I defined performance expectations of baseline false-alarm rate on holdout <=5%, detection at 40 g >=95%, and baseline mean stability across rebuild sessions within 2%. In Version 1, utilizing only frequency as a threshold, 40 g detection was about 93%, and baseline false alarms were about 47%, with baseline mean shift about 2.7% across sessions. The approach was not reliable enough. Versions 2 and 3 improved reliability by improving trial quality control and adding normalized features, reducing false alarms and improving stability. However, detection remained below target. The final Version 4 model achieved 0/15 baseline false alarms and 15/15 detection at 40 g indoors, and 0/15 baseline false alarms and 15/15 detection at 40 g outdoors. A Welch t-test confirmed the outdoor baseline shift was statistically significant, and one-way ANOVA on f_1 showed significant separation among mass states. These results validate the system as a low-cost screening method and support future field deployment.

Competition history

• CSEF 2026 Applied Mechanics (Junior Division) · Entry J-02-14

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Engineering projects
• CSEF Engineering projects
• Every CSEF 2026 project