Source code of the paper "Deep Learning of Latent Variable Models for Industrial Process Monitoring".

"DOI: 10.1109/TII.2021.3134251" "https://ieeexplore.ieee.org/document/9647968"

First author: Xiangyin Kong

Corresponding author: Prof. Zhiqiang Ge (Google Scholar: https://scholar.google.com/citations?user=g_EMkuMAAAAJ&hl=zh-CN)

If the paper or the code is helpful to your work, please cite the above paper. Thank you!

@author: Xiangyin Kong

@github account: kxytim, https://github.com/kxytim

Code description:

utils.py: the basic functions used in the model.

deep_PCA_ICA.py: the proposed deep PCA-ICA model.

Bayesian_Fusion.py: the proposed Bayesian fusion strategy to integrate the information at different layers.

DPI_monitor_TE.py: use the proposed model to monitor TE process.

For a fast test, please execute:

python DPI_monitor_TE.py

Then you may get the following results:

Start running

The average FAR and MDR of DB_T2 are 2.887% and 18.786%.

The average FAR and MDR of DB_SPE_T are 1.577% and 18.792%.

The average FAR and MDR of DB_I2 are 2.292% and 18.643%.

The average FAR and MDR of DB_SPE_I are 2.560% and 19.601%.

The average FAR and MDR of ODBS are 3.333% and 16.702%.

The total time of training and testing is: 80.858s.

End

If we keep all hyperparameters unchanged and only increase the number of layers, the relationship between the average MDR and FAR and the number of layers is shown in the following figure: