Ore mineral particles identification, and recognition by neural nets: strategies, practical implementation and possible applications

The identification of the procedure and instruments to carry out a preliminary evaluation of the possibilities of a crude ore beneficiation, starting from the exam of a considerable number of samples, has been carried out since ever, through the setting of comminution models, that can foresee, besides the granulometric distribution of products, also their composition. In the formulation and successive setting of a comminution product, a correct schematization of the textural and structural characteristics of the crude ore has a great importance; this is in relation to the influence of these parameters on the following mechanisms of fracture and particles' freedom. The models used up to now, do not take into consideration the "real" textural and structural characteristics of the crude ore, on the contrary, especially in the case of complex crude ores, they supply results sometimes unreliable. The aim of this paper is that of suggesting a structure (both hardware and software) of expert system that can carry out the mentioned classification of crude ore sample, in a structural and textural way, by means of algorythms of "image textural analysis". The developed and set procedure has been planned and implemented in order to be easily interfaced with numerical simulation programs, where the textural characteristic modelling represents one of the fondamental "step" of the input parameters. In particular, the possibility of "interface" with comminution models, is examined. The analysis has been based on a certain number of ore samples characterized by textures and structures of different type and with a different complexity grade. The content of the paper is as follows. After the Introduction, in Section 2 the theoretical bases of the method are described, shown and discussed with a particular reference to the type of samples object of investigation (ores). In Section 3, the whole procedure of analysis and the structure of numerical algorythms used, is described. In Section 4, the types of crude ores in the study, both from the petrographic, mineralogic and numerical (textural features) point of view, are described and characterized. In Section 5, the results obtained from the various alternative features of used extraction procedure, are described. In Section 6, on the basis of experimental results, the possibility of developing software architectures that enable the management of dedicated data base, and the development of expert systems that work out an identification and an automatic classification of the different textures, have been discussed. In Section 7, an example of "interface" between the proposed software structure and the model of liberation, is shown. In Section 8, the advantages and limits of this method of approach are indicated, both as regards the performance of investigations and subsequent processing of results.