NMFAD

This repository includes code and dataset in paper "Physics-Enhanced NMF Toward Anomaly Detection in Rotating Mechanical Systems"

Data set in experiment

• The data set in the experiment is from literature in "On the Evaluation of Unsupervised Outlier Detection: Measures, Datasets, and an Empirical Study"
• The CBCL facial data set used in the appendix is from http://www.ai.mit.edu/courses/6.899/lectures/faces.tar.gz
• The carriages data set in the case study is available from [email protected], upon reasonable request.

Requirements

• Python 3.9.7
• NIMFA in Python: for NMF and variants benchmark methods.

Code

• Self-defined function
  • GNMF.py: GNMF function
  • PNMF.py: self-defined PNMF function
  • framework_nmf.py : the benchmark NMF method
  • framework_svd.py : the benchmark SVD method
  • framework_svd_st.py : the benchmark SVD method with soft thresholding
  • framework_bd.py : the benchmark BD method
  • framework_bmf.py : the benchmark BMF method
  • framework_snmf.py : the benchmark SNMF method
  • framework_gnmf.py : the benchmark GNMF method
  • framework_lfnmf.py : the benchmark LFNMF method
  • framework_pnmf.py : the proposed PNMF method
• experiments
  • main_experiment_WBC.py : experiment based on the WBC.xlsx dataset, and the results are presented in Figure 5 in the paper.
  • main_face.py : experiment based on the CBCL face dataset, and the results are presented in Figure 15 in the paper.