Zastosowanie uczenia maszynowego do obrazowania stanu zawilgocenia murów z wykorzystaniem tomografii impedancyjnej

Overview & Synopsis

The subject of the considerations contained in this study is the comparison and verification of the effectiveness of selected methods of transforming tomographic measurements into images. The research centers on the application of electrical impedance tomography (ETI) as a technique for imaging moisture distributions within building walls. Both deterministic methods and modern methods based on machine learning were used in the research. The research was conducted in laboratory conditions and in the historical building of the Golden Gate in Gdańsk. A hybrid, multi-branch neural network model was developed, which was used to transform moisture measurements taken at three research stations. The research showed that the use of the multi-branch LSTM+CNN neural network model in combination with the ETI tomography technique is characterized by high effectiveness. Validation measurements confirmed the high accuracy of tomographic reconstructions, which is proof of the effectiveness and utilitarian potential of the described method