Continuous monitoring of mineral processes with special focus on tumbling mills :a multivariate approach
Abstract (Summary)Increasing emphasis on productivity and quality control has provided an impetus to research on better methodologies for diagnosis, modelling, monitoring, control and optimisation of mineral process systems. One of the biggest challenges facing the research community is the processing of raw sensor data into meaningful information. Information that to some extent express quality parameters such as chemical assays, size distribution and other metallurgical variables in the different process streams. This thesis shows how multivariate statistical methods can be used with great advantage to model process data as well as robust sensor data. The modelling approach has been applied on a large process section, a cobbing plant, as well as a single unit operation, a tumbling mill. The knowledge of conditions inside a tumbling mill are of limited extent, due to a very harsh and wearing environment. A strain gauge sensor that measure the deflection of a lifter bar when it hits the charge inside a tumbling mill is studied for different operating conditions in a pilot scale ball mill. The deflection of the lifter bar during every mill revolution gives rise to a characteristic signal profile that is shown to contain information on both the charge position and grinding performance. The results presented for prediction of grinding performance suggest that the strain gauge signal in combination with wavelet transformation and multivariate data analysis provide promising means for monitoring and control of process fluctuations. The low prediction error achieved for grinding performance clearly highlights the importance of well-planned experimental strategy including experimental design, signal pre-processing, multivariate modelling and validation. Results also demonstrate that different operating conditions is well distinguishable from each other and by that the finding of proper operating regimes are highly feasible. Grinding parameters that are normally measured in the laboratory are now readily modelled from the on-line signal. As a consequence this opens new possibilities for real time monitoring and control of the grinding process. A further objective of this work is to link computational results to the experimental data obtained from the instrumented pilot ball mill. The approach taken is to simulate the behaviour of a rubber lifter when it is exposed to forces from the grinding load in a two-dimensional Distinct Element Method (DEM) mill model. Typically walls in a DEM model are made up of rigid bodies where the equations of motion are not satisfied for each individual wall - i.e., forces acting on a wall do not influence its motion. Here the instrumented rubber lifter is represented as an assembly of bonded particles rather than walls in order to simulate deflection. The deflection profile obtained from the DEM simulation shows a reasonably good correspondence to the pilot mill measurements. The difference is attributed to the fact that time-dependent behaviour of the rubber lifter is ignored, resulting in rapid relaxation of the lifter when the exerted force is released. Mill charge features such as toe and shoulder position of the charge are well marked. However, DEM prediction shows lower values compared to measurements which is most likely an effect of the two dimensional model used and the inability to model the effect of slurry present in the mill. The news value of the thesis work is in the method for analysing the signal profile as well as the experimental verification in both pilot and full scale operation. The result is a contribution to improved mill lifter design and continuous monitoring of the grinding process.
School:Luleå tekniska universitet
Source Type:Doctoral Dissertation
Date of Publication:01/01/2005