Degree Name

Master of Engineering - Research


School of Electrical, Computer and Telecommunications Engineering


During natural disasters, rescue teams fight against time to save as many civilians as they can. Researchers can contribute to rescue operations by investigating different ways of collecting information from the crisis area and delivering it to rescue teams. This project proposes a novel approach for collecting information from disaster areas in relation to building collapses and collapse pattern analysis. Our approach is based on classifying building collapse patterns using Wireless Sensor Networks (WSNs) and data-mining algorithms. Classification time and reliability are considered crucial factors, and one of the main objectives of this research is to improve these elements in order to deliver accurate information to rescue teams regarding building status in a stricken area. WSNs were installed in a simulated building to capture a building‟s motion during an earthquake. Four different types of collapse patterns were simulated: first column (FC), first storey (FS), mid-storey (MS) and pancake (PCK). The captured data was inputted to three different classification algorithms (PCA, VQH and HMM) to classify building collapse types.

Two real-life case scenarios were designed to examine the algorithms‟ reliability under sensor failure. The first scenario was sensor failure on impact, which was designed to simulate sensor failure caused by interfering with an object or hitting the ground. The second scenario was the complete failure of random sensors, which was designed to simulate early malfunctioning sensors resulting from a power supply failure, communication problem or manufacture error. Moreover, the second scenario was designed to investigate the limit of each classification algorithm in terms of the number of failed sensors. The Hidden Markov Model (HMM) proved the most robust and achieved 100% accuracy in least impact on accuracy (LIoA) case scenario and 60% accuracy as an algorithm limit when 33.3% of the sensors failed during a building collapse. By achieving this level of accuracy, the objective of classifying four possible collapse patterns in a short processing time was achieved.