Underground coal dust real-time measurement based on cluster analysis and pattern recognition

To prevent coal dust explosion, dust concentration measure is very important. The dust pattern recognition was performed according to the diffraction angular distribution of dust samples, but there was a conflict between measure precision and real-time demands. To resolve the contradiction, an improved method of dust pattern cluster analysis and recognition was put forward. The pattern cluster analysis was performed by the difference of diffraction angular distribution, so patterns could be recognized easily and rapidly with the principle of minimum variance sum between pattern and dust sample eigenvectors. The simulation indicates the maximal recognition speed improves observably, which can ensure single-chip operating real-time inversion method. Number of transitional patterns was increased reasonably and hierarchical cluster method was adopted. The sensor error is controlled within 4%. We conclude that the advanced algorithm of cluster analysis and pattern recognition improves the sensor accuracy in measurement remarkably.