Estimation of particle size distribution on an industrial conveyor belt using image analysis and neural networks

Monitoring and controlling particle size distribution in crushing and grinding circuits are essential for improved energy efficiency and metallurgical performance. Machine vision is probably the most suitable approach for on-line particle size estimation because it is robust, cost-effective and non-intrusive. In the present study, size distribution of particles in crushing circuit of a copper concentrator was estimated using image processing and neural network techniques. Several images were taken from material on a conveyor belt and processed for particle identification and segmentation. A number of the most commonly used size features were extracted from the segmented images and their potential to estimate the actual particle size, represented by sieve size analysis, was evaluated. The results showed that there were substantial differences between size distributions obtained from various size measures. Maximum inscribed disk was found to be the most effective feature for particle size description. Finally, the particle size distribution of material on the conveyor belt was precisely estimated by Principal Component Analysis (PCA) and neural network techniques. The proposed soft sensors can be used for real time measurement of particle size distribution in the industrial operations instead of sophisticated and expensive instruments.