dynamic-graph-embedding-for-fault-detection's Introduction

Dynamic Graph Embedding for Fault Detection

Code for the paper "Dynamic Graph Embedding for Fault Detection"
Matlab version should be later than R2015b
The demo codes can be found in the directory "Matlab_code". They are developed to do the fault detection on the data of Fault 1. the file "myConstructW.m" is developed to obtain the similarities in Equation (6). In the file, we give the annotations according to the Eqautions in the paper. "myfunction_tensorLPP_markov_paper.m" is the main program, which can run directly. "TensorLGE.m" and "TensorLPP.m" are the codes needed for the main program. Both "TensorLGE.m" and "TensorLPP.m" are designed by Deng cai, who is the second author of the paper "Tensor Subspace Analysis", published in Neural Information Processing Systems 18 (NIPS 2005). The file "kde.m" is the code of kernel density estimation which is used to determine the control limit of T2 and SPE statistics.
"File_published_by_matlab_in_PDF.pdf" is the running results and the codes published with MATLAB® R2015b. "Files_and_results_published_by_matlab.zip" contains the html version published with MATLAB® R2015b. Since the normal data and the fault data of TE are simulated and generated randomly, the detection results maybe slightly different due to the randomness of the simulated datasets. But these differences are quite small compared with the large differences of the MDRs of PCA, LPP, and our proposed MLPP, and DGE.
"Twin_Peaks.mat" is the 3000 data points used for Figure 1.

Recommend Projects