Development of a Knowledge-Based System for Planning of Selective Mining in Hard-Rock Surface Mines

INTRODUCTION At present, the capability of production planning software based on Linear Programming (LP) is still limited to the optimization of the single LP-run. This is due to the LP-model itself which cannot consider the interdependencies between individual LP- runs. With regard to planning of selective mining this limited way of optimization frequently leads to situations, where the remaining and accessible ore blocks do no longer allow to produce ROM-ore in the qualitative composition required by the ore processing plant. However, many of the aspects taken into consideration when setting up production plans built from mutually dependent LP-runs cannot be modelled in a system of linear equations. They are thus unsuited for treatment with LP and have to be taken care of by the planning engineer without any assistance by the system. The KBS currently under development is intended to assist the planning engineer in designing a production plan under special consideration of the combination of consecutive LP-runs and blending beds. It is not necessarily intended to find the optimum solution within a given planning situation which is, anyway, hard to determine due to the multitude of influences. The objective is rather to work out a good and - from the practical point of view - feasible production plan. The new aspect with respect to mine planning is the integration of expert knowledge and experiences via the KBS into the planning process in order to support the planning engineer. The planning system is being developed in close cooperation with an iron-ore open pit mine. COMPONENTS OF THE PLANNING SYSTEM The software is being developed on a workstation under UNIX and comprises the components LP, CAD-module and the KBS. The applied multi-goal LP-algorithm is an in-house development of the Department of Mining Engineering at Technical University Berlin. It was already successfully implemented within other mine planning programmes and was only slightly adapted to the specific needs of the present system. Within individual LP- runs it finds the optimum qualitative composition of ore production in the sense of the selected optimizing criterion and under the given restrictions: i.e. it determines tonnages to be mined from blocks in order to optimally meet the requirements of the ore pro- cessing plant. A CAD-module based on the commercial SURPAC package in combination with a simulation device is used to graphically depict the block model and progress of mining. Both LP-algorithm and CAD-package are integrated in the KBS. It has been decided to use the shell NEXPERT OBJECT as it is a hybrid system which supports both rule-based and object-oriented knowledge representation. MINE-MODEL AND LP-MODEL KBS have to be tailor-made for specific planning problems. Therefore, it had to be decided which specifications of the iron-ore mine should be represented in the model. As the limited possibilities of a university institute do not allow to develop a KBS for mine planning which is ready to use in industry, it was decided to concentrate on those characteristics that can be regarded as typical for iron-ore surface mines and that seemed to be suited for treatment with knowledge-based techniques. The following chapter summarizes the most important features of the mine model. The description of the requirements to the mine's sales products is followed by an outline of the applied LP-model. Mine model • The model of the mine as it is used for planning consists of • the block model of the deposit, • the mobile equipment, • stockpiles and blending bed and • the requirements to the sales products. The deposit is described by a block model which contains data on the chemical composition, LOI, grain size and tonnages. Grain size was included as it is important for the two sales products of the mine. Furthermore, it is known whichs blocks require and which don't require blasting; this is relevant to the assignment of loading equipment to individual blocks. The blocks are devided in three categories: • ore, which will directly be taken to the blending bed; • waste, which will be put on the waste dump; and • blocks which will be either transported to the blending bed, to stockpiles or to the waste dump depending on the specific planning situation. This decision is made during planning. Neighboring blocks are combined in mining areas to which the loading equipment is individually assigned. Mobile equipment comprises shovels and wheel-loaders as well as trucks. The characteristics of the loading equipment are important for their ability to load different blocks and for the permissible degree of their re-positioning etc. The mine disposes of a blending bed for homogenization of the production, of a waste dump, and of several stockpiles with different ore qualities. The requirement to make only limited use of the stockpiles for economic reasons is included in the KBS. According to long term planning two commercial products have to be produced, which differ both in grain size and qualitative composition (TABLE 1). Their mass-proportions in the blending bed have to be within a fixed range. Not considered in long term planning is the occasional need for lump ore, which occurs at very short notice and has to be produced in a "campaign-like" manner. This requires the total re-arrangement of all plans for on- coming blending beds and would therefore be ideally suited for