Selective data mining for material selection

In order to achieve the separation of particles that are very similar, such as those deriving from demolition of end-use goods, it is necessary to reveal a large number of characteristics. Each characteristic can then be evaluated not only by using the simple means of scaled parameters, but also by registering definite signs of functions that are largely continuous. The next problem is to determine these signs, extract a vectorial database, and select from this large quantity of data that which is easily manipulated, in order to characterize the particles on the basis of predefined composition classes. This situation is characterized by the extreme complexity of the data that is utilizable for classifying and selecting the items. This holds particularly true in the case of sorting based on analysis of the images, especially when the source of illumination emits a continuous signal and when the sensor, or sensors, are capable of monitoring such a signal by sampling it with an elevated number of spectral value intervals. The data-mining approach is proposed in this work, with a specific case analysis of glass fragment selection.