Wooden Framed House Structural Health Monitoring by System Identification and Damage Detection under Dynamic Motion with Artificial Intelligence Sensor using a Model of House including Braces

Abstract

We are trying to discriminate damage areas of wood by machine learning. Last year, an experiment to identify the damage position of a piece of timber was conducted. This time, an experiment on the identification of the damage position of the house brace was performed. Only one brace was removed from the model of the house with 28 brace positions, and the damage position was assumed to be there. Vibration was applied to the model of the house, and the transferred vibration waveform was detected with a piezoelectric sensor. This vibration waveform was analyzed using a neural network. The classification on each side of the house succeeded after fixing the number of neurons in the hidden layer. After that, classification on the whole side of the house with 3-layer and 4-layer neural networks was conducted. The classification rate could be improved by changing the number of neurons in the hidden layer. As a result, the classification rate of the damage position of the entire house is 90.69%. Also, the classification rate is higher in the 4-layer neural network than in the 3-layer one.