Artificial intelligence methods like expert systems and self-organizing maps have proved to be excellent tools for the control of mineral processes. This technology is currently being embedded directly into process equipment like dewatering filters. AI methods can be seen most applied in industrial applications since 1991 compared to other methods. They are used in nearly 40 % of all applications reviewed, and therefore represent the most important methods applied in the control and monitoring of MM processes.
Separation of solids from a liquid by a porous medium or screen, which retains solids while allowing the liquid to pass, is called filtration. The pressure filter can be divided into eight subsystems: slurry feeding system, water pressing system, air drying system, cake washing system, cloth washing system, hydraulic system, discharging system and control system. Since in filtration the slurry volumes are extremely large, any downtime of the process is expensive.
An intelligent control system has been developed for a pressure filter. The application has been programmed with platform independent object oriented programming language, Java. The remote operating service system is a part of the fault diagnosis module. In addition to the implemented system the plant site needs a database server (SQL server connected to a Larox pressure filter), an IIS server, workstations, and a local area network. At the client end there are workstations, and both ends are connected to each other via the Internet. Data are transported using a browser and HTTP as the interface. Due to security reasons, all data transmissions on the Internet between the plant and the client are encrypted with a DES algorithm, and access to the local network is permitted only through the firewalls. The encryption key is pre-shared, and encryption/decryption of the data limits the capacity to approximately 8 Mb/s. The databases include measurements and the learning part. The aim of the learning database is to suggest possible optimal control parameters to the operator. Suggestions are based on data from previous operation cycles. The structure of the system is modular in order to keep it easily expandable and maintainable. The overall system consists of the modeling, classification, economic, fault diagnosis, control and database modules. The modeling module contains models of the different operating stages of the filter.
The aim of the optimisation is to maximise the capacity and to decrease cycle time. Economic aspects are followed on the basis of operating costs calculated from on-line measurements. The determination of the maximum capacity is based on online classification of feed type, predicting models of filtration behaviour and a self-learning database. The main aim of the classification module for the pressure filter control system is to train a SOM using information about different feed compositions, and to determine for each neuron the correct values of the variables used in control. Measurements for feed type classification in filtering depend on the installation and could be for example the particle size distribution and the density of the slurry. The preliminary testing of the modelling module has been performed in the pilot plant. Testing period of the whole integrated system in the industrial scale has been started.