Prediction of Masonry Strength and Earthquake Characteristics using Machine-Learning Algorithms

Abstract

„Gründerzeit houses“ in Vienna/Austria, which were built between 1848 and 1918, are an essential part of the Viennese building infrastructure with a number of approx. 30,000. Since 1992, the masonry properties of over 200 buildings with over 2,000 test positions have been determined and evaluated by the authors. These were determined in the context of the engineering expertise in order to carry out the load-bearing capacity and earthquake verifications for the refurbishment. Based on the results of the investigations, a comprehensive data set was created containing the coordinates of the buildings, district, construction period, use, building type, number of storeys, type of test, moisture, compressive strength of the brick and mortar, compressive strength of the masonry, etc. This data set contains approx. 40,000 independent numerical and categorical values. This data set is analysed for clusters and correlations using machine learning algorithms. Among others, the unsupervised learning algorithms “Hierarchical Clustering”, “k-means” and the dimension-reducing algorithm “tSNE” are used. The correlations found are statistically evaluated and visualized. The most important results will be presented. In the second part, various prediction models based on machine learning algorithms are used to predict the masonry strengths of existing objects as well as the earthquake resistance according to Eurocode 8. The algorithms used include “neural networks”, “random forest”, “k-nearest neighbours”, etc. With regard to the earthquake resistance, shear-tables are used, with which the compliance factor can be quickly specified using a few building parameters, such as the number of storeys, the floor plan shapes and the ground parameters. These values are calibrated with “push-over” calculations for different building geometries and are thus incorporated into the prediction models. The aim is to find and compare data-based models in order to enable an evaluation of all Viennese “Gründerzeit” buildings with regard to the expected strengths and earthquake values.