Evonik Digital Research

A joint research project of EVONIK and the University of Duisburg-Essen

Locating Workers by Floor Vibration Sensing

In an emergency case, rescue helpers must know where workers are. We measure the floor vibration using an array of piezoelectric sensors. Signal processing, feature extraction, pattern recognition, and machine learning techniques are applied on the sensor data to locate workers in need of help.

The security state of a given scenario will be recognized based on the floor vibration caused by the flow of people in this scenario. The person in the scenario can be located. Signal processing, feature extraction, pattern recognition, and machine learning techniques are applied.

Floor vibration is caused by multiple factors, especially in an industrial facility machines and pumps cause vibrations in their surrounding environment. To distinguish these so called normal vibrations from abnormal vibrations caused by an emergency we do the following:

  1. using recorded vibrations from multiple days of normal operating conditions and machine learning techniques we build a model that represents normality
  2. we continuously measure the current floor vibration and match it against the model
  3. we activate stage two based on the probability that the currently measured vibration was abnormal

We conducted an initial experiment using the accelerometer in a smartphone. The following figure shows the measured acceleration when knocking a wooden table with the smartphone placed on the table’s surface and the four knocks are clearly visible with accelerations of up to 6 m/s².

By using the machine learning algorithm, as well as some necessary signal preprocessing and feature extraction approach, the location information of the person in the scenario with sensors is obtained. Whether the person in the scene is walking normally, running, or rushing can be detected.