Novel pattern recognition algorithm for real-time measuring coal dust with bimodal peak distribution

Real-time measurement of coal dust concentration is vital for colliery safety in production. To improve the precision, a novel inversion algorithm for dust distribution with bimodal peak is presented. A three-parameter was brought forward and the eigenvectors of 360 patterns are worked out. The pattern classification was performed according to diffraction angular with dust information. Simulation indicates the minimum recognition time is reduced to 0.05 times of that before. Thereupon, transitional patterns were supplemented and the precision increased markedly. But sometimes there was gross error. Therefore the pattern amendment function was introduced and the eigenvectors of amendment patterns were calculated. The normalized eigenvectors of amendment patterns ranked were stored in advance. During measurement the optimal patterns were recognized in the whole and amended in the local area according to the principle of the minimum of variance sum. Experiments proved the error of total dust and respiring dust declined from 6% to 2% and from 9% to 3%, respectively. It is concluded that the novel algorithm has improved the precision and real-time performance of dust sensor remarkably.