Development of Predictive Machine Learning Model using Neural Network for Threshold Value Determination of Buildings

Abstract

Machine learning (ML), a subset off artificial intelligence (AI), is now part of people’s everyday lives. It is now applied in many fields and industries like the automotive industry, medical field, e-commerce and many more. Some examples of this can be found in the self-driving cars, medical diagnosis, recommendation engines, patient sickness prediction and many more. In the past years, engineering had been showing growing interest over the application of AI in the field. In fact, several studies had been conducted to see what advantages it can bring to the engineering discipline. It is evident that ML is now being applied in lots of field of engineering. However, ML as applied to structural health monitoring (SHM), specifically to the determination of threshold for buildings has not yet been established. The threshold plays a very important role in SHM as it will be the basis for evaluating the integrity of a structure after it ages as time goes by or even after earthquake events. This study focuses on developing a predictive machine learning model that will be incorporated in an earthquake recording instrument that will give the threshold value specifically for a building given specific input parameters. To do the predictive model, structural data of thirty (30) buildings were collected. It consisted of acceleration data, maximum displacement on non-linear and linear state, lower and upper limit of moderate damage state, and its threshold. The proponent was able to gather 3750 rows of data to be used for the training of network. Creating of the neural network model was done using the MATLAB neural network tool, and trained using the Levenberg-Marquadt algorithm which yielded the best performance among the training algorithms in MATLAB neural network tool. After training, a MATLAB function was generated and run compatibly with python to allow integration with the earthquake recording instrument. Furthermore, an accuracy test was done wherein it produced a 91.77% accuracy. Through the predictive ML model, structural engineers are expected to experience a great amount of savings in terms of time and effort on determining the threshold value for a specific model