Multimode dataset of Tennessee Eastman v1.0

Background

Tennessee Eastman (TE) model Is an important tool in all disciplines of systems theory, for comparative studies or validation of algorithms. Its strength is that it is based on real process modeling. This leads to a rather complex nonlinear model of a multi-component system. Due to the frequent use of the model, it is very beneficial that this model, or more precisely its code, works flawlessly [1]. On this basis, a revision and extension of TE process model is proposed and is shown in Fig. 1. Fig. 1 P&ID of the revised process model; additional measurements in red [1]

Usage

• If you want to use the datasets in the project, you can download them directly by using Python or Matlab.

• Example (Normal data of mode 1):

1) Python

import scipy.io as scio
  
  data00 = scio.loadmat('xxx/TEPmultimode/temexdmod/M1/m1d00')['m1d00']
  
  data00_X = data00[:,:53]
  
  data00_Y = data00[:,53:]

2) Matlab

load xxx/TEPmultimode/temexdmod/M1/m1d00
  
  data00_X = m1d00(:,1:53)
  
  data00_Y = m1d00(:,53:81)

• If you want to regenerate the dataset, you can run MultiLoopmode1.mdl to MultiLoopmode6.mdl.

Dataset Introduction

We have made six modes of datasets in 72 hours, and there are 28 faults in every mode. All of them contain 12 variables of input, 41 variables of measurement, and 28 variables of disturbance. The adjusted parameters are saved in Mode1xInitial.mat —Mode6xInitial.mat. The parameters of different modes are listed in Fig. 2 and Fig. 3. The process disturbances are shown in Fig. 4 ( Datasets m（1-6）d01-28.mat correspond to disturbance types IDV(28)-(01) respectively). Fig. 2 Measurements of different modes Fig. 3 Main variables of different modes Fig. 4 Process disturbances

Adjustment of mode

• Users can adjust parameters to achieve different modes by themselves.

• Example (mode 2)

1. Copy files Mode1Init.m, Mode1xInitial.mat and MultiLoopmode1.mdl and rename to Mode2Init.m, Mode2xInitial.mat and MultiLoopmode2.mdl. 2. Modify Mode1xInitial in 29 lines in Mode2Init.m to Mode2xInitial. 3. Modify Mode1Init to Mode2Init by opening MultiLoop_mode2.mdl and the steps as follow figures: 4. Running upda.m and modify as follows:

clear all;
     clc;
     load Mode2xInitial
     for i =1:35
         blockName = xInitial.signals(i).blockName;
         blockName(15) = '2';
         xInitial.signals(i).blockName = blockName;
     end
     save ('Mode2xInitial.mat','xInitial')

5. Adjust the parameters MultiLoop_mode2.mdl according to Fig. 2 and Fig. 3. Note that the value of parameters adjusted each time should not be too large, otherwise the simulation will end quickly. And run the following code after each successful adjustment.

xInitial = xFinal

6. All parameters are adjusted to the final mode and run:

save('Mode2xInitial.mat','xInitial')

Mode

We selected normal data under six modes for display, as shown in Fig. 5. Fig. 5 Normal data of six modes; mode 1 in red; mode 2 in green; mode 3 in blue; mode 4 in turquoise; mode 5 in yellow; mode 6 in black

Fault

We selected part of the measured data of some faults for display. A B C D E F G

Citation

Please consider citing the following work if you use the datasets:

@ARTICLE{10342662,
  author={Liu, Zeyi and Li, Chen and He, Xiao},
  journal={IEEE Transactions on Industrial Informatics}, 
  title={Evidential Ensemble Preference-Guided Learning Approach for Real-Time Multimode Fault Diagnosis}, 
  year={2024},
  volume={20},
  number={4},
  pages={5495-5504},
  keywords={Fault diagnosis;Real-time systems;Monitoring;Learning systems;Production;Informatics;Dynamical systems;Broad learning system (BLS);concept drift;evidential reasoning (ER);real-time multimode fault diagnosis (MMFD);Tennessee Eastman process},
  doi={10.1109/TII.2023.3332112}}

Change log

v1.0(2023/01/12)

About us

• We are from the THUFDD group of Prof. Xiao He and Prof. Donghua Zhou, Department of Automation, Tsinghua University.

• Welcome to discuss and put forward valuable suggestions. Email: liuzy21@mails.tsinghua.edu.cn, yan12@mails.tsinghua.edu.cn.

References

[1] Andreas, Bathelt, N, et al. Revision of the Tennessee Eastman Process Model[J]. IFAC-PapersOnLine, 2015, 48(8):309-314.