ML-SHM

This repo contains the pipeline code for the paper "Machine Learning based Novelty Detection using Modal Analysis"

Before running pipeline

The following order is suggested to run the files.

• Some documents require MATLAB. Make sure MATLAB installed. The code is tested in version 2018b.
• Make sure Anaconda(https://www.anaconda.com/distribution/#download-section) is installed. Moreover, Keras (https://keras.io/) and Tensorflow (https://www.tensorflow.org/install/) should be installed. The code is only tested for Python 3.6.7.

Analytical Example

This pipeline will generate 200 GB of data.

• Make sure data and eq folder are created with three folders (1, 2, and 3) in them. Each folder is related to a damage condition.
• Download OpenSees excecutable file and tcl/tk package from http://opensees.berkeley.edu/OpenSees/user/download.php
• Put the OpenSees.exe into folder where you will run the scripts. Add tcl/tk to path if it is not added already.
• Run temp_generator.py to generate reference temperatures. This will create T.txt file for each damage condition.
• Run blwn_noise_generator.py to generate ambient vibrations. This will create a set of ambient excitiations for each damage condition.
• Run eq_result_generator3.py to generate structural responses using the ambient excitation. This part requires OpenSees. The code is written to run only on multi-core processors.
• Run next_freq_modeshapper_grabber.py to obtain frequencies and mode shapes from structural responses. This step requires MATLAB as it utilizes NExT-ERA files written by Dyke, Caicedo, and Giraldo.
• Run pca_paper_version.py or aann_6freqmodes_era_final_paper_version.py to generate the novelty index.

Experimental Example

• The SHM data is provided here https://www.lanl.gov/projects/national-security-education-center/engineering/software/shm-data-sets-and-software.php under SHMTools Additional Datasets.
• Extract the contents of SHMToolsAdditionalDatasets.zip into folder where you will run the scripts.
• Run test_data.m to obtain frequencies and mode shapes from structural responses. This step requires MATLAB as it utilizes NExT-ERA files written by Dyke, Caicedo, and Giraldo for their Benchmark structure.
• Run pca_paper_version.py or aann_3freqmodes_era_final_paper_version.py to generate the novelty index.