Year: 2024
Type: conference-paper
Abstract: Hydraulic systems are often equipped with a great number of actors and sensors to monitor and enable the control of a system.For continuous monitoring, a vast amount of data needs to be recorded and frequently stored.Prediction based on this big database is often not directly comprehensible by a human and requires preprocessing.With the development of computational power, machine learning becomes a promising approach to handle these huge amounts of data and offers to be a possible solution for the automated processing of raw machine data.In this paper, a deep unsupervised learning (UL) model, the variational autoencoder (VAE), is implemented.The VAE is used for the reduction of high dimensional time series of a simulation of a reference system, a hydraulic press.To handle the time dependency of the dataset, the VAE contains long short-term memory (LSTM) layers.The high dimensional input data matrix is compressed by the first part of the VAE, the encoder, into the latent.This low dimensional representation of the time series is then used by the counterpart of the encoder, the decoder, to reconstruct the original time series.The VAE achieves a compression rate of 99,94 % with a small deviation compared to the original input.This contribution shows the possibilities of compressing big data from a hydraulic system into a low dimensional representation.The compressed data can be used as input for River Journal, 1-17.
Authors: Faras Brumand-Poor, Faried Makansi, Jiakun Liao, Katharina Schmitz
Institution: RWTH Aachen University
Language: English
Subfield: Statistical and Nonlinear Physics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4403916909