PORTALE DELLA DIDATTICA

Ricerca CERCA
  KEYWORD

SINTETIC SEISMIC signals by GNENERATIVE ADVERSARIAL NETWORK WITH GATED CONVOLUTIONAL NEURAL NETWORK STRUCTURE

azienda Thesis in external company    


keywords DEEP LEARNING, GAN, GANS, GENERATIVE ADVERSARIAL NETWORKS, MACHINE LEARNING, ARTIFICIAL NEURAL NETWORKS, NEURAL NETWORKS, SEISMIC SIGNAL

Reference persons GIUSEPPE CARLO MARANO

External reference persons Prof. Giansalvo Cirrincone

Thesis type RESEARCH THESIS

Description The seismic analysis of buildings and structures to define the risk is an important objective problem that however requires in many types of analysis the use of seismic signals (time histories in the three directions) characterizing a given site. Given the relative scarcity of recordings of real seismic signals characterizing a given area, and in particular of seismic signals of considerable intensity (Magnitude greater than 5), it is necessary to resort to the generation of synthetic seismic signals.
Compared to the classic methods often used, which scale the intensity of real signals with respect to the required intensities, while in this thesis we intend to address and solve this problem using a Generative Adversarial Network (GAN) model for the synthesis of seismic signals of assigned characteristics. The GAN technique already shows its powerful ability to generate quality synthetic samples in multiple domains. In this thesis it is proposed to develop a GAN model with CNN capable of capturing seismic time series in an excellent way. It is proposed to use a GAN approach both on the seismic signal as it is (time series) and on its spectrogram obtained with different approaches to test its effectiveness.

See also  seismic_signal_synthesis_by_generative_adversarial_network_with_gated_convolutional_neural_network_structure.pdf 

Required skills neural network, time series analysis

Notes the thesis is multidisciplinary exploiting the potential of NNs (in particular GANs) to generate artificial seismic signals - possible stage at the INGV at Milano


Deadline 31/01/2023      PROPONI LA TUA CANDIDATURA




© Politecnico di Torino
Corso Duca degli Abruzzi, 24 - 10129 Torino, ITALY
Contatti