Analysis of Experimental Measurement based Time-Series Dataset with Deep Learning
- 0 Collaborators
Explore use of various machine learning tools to drive usable insight from time-series based experimental datasets acquired typically from engineering systems, which in return could be used for better validation and uncertainty quantification of computational models. ...learn more
Project status: Concept
oneAPI, HPC, Artificial Intelligence
Intel Technologies
DevCloud,
Intel Opt ML/DL Framework,
Intel Python,
MKL
Overview / Usage
Time-series data is frequently encountered in diverse range applications. More often data acquisition from experimental setup and sensors provide high frequency time-series data which is typically analyzed with statistical methods. The project aims to employ machine learning tools to gain more insight from time series information acquired as the statistical techniques often loose content which could be helpful to drive better insight in the dataset.
Pressure transducer data obtained at 100 Hz from several experiments performed in a lab will be utilized as source dataset for proof of concept.
Methodology / Approach
Project aims to go through the traditional time-series analysis techniques such as AutoRegressive Moving Average (ARMA) and ARIMA to establish the baseline and then explore the use of deep learning such as Recurrent Neural Network (RNN) for the analysis and insights gained.
Technologies Used
Python and libraries such as pandas, scikit-learn, DL libraries like PyTorch and TensorFlow2