Prévia do material em texto
This article is also available online at: www.elsevier.com/locate/mineng Minerals Engineering 17 (2004) 1189–1198 Dynamic simulation of grinding circuits Yi Liu *, Steven Spencer a CSIRO Minerals, Private Mail Bag 5, Menai, NSW 2234, Australia Received 1 April 2004; accepted 12 May 2004 Abstract A flexible and powerful dynamic simulation approach to grinding circuit simulation has recently been developed in CSIRO Min- erals. The MATLAB/SIMULINK graphical programming environment has been used to construct a library of dynamic mathemat- ical models of a number of key grinding and separation devices and to link them into various complex dynamic grinding circuits. True real-time dynamic simulation and visualisation of interlinked unit process operations in grinding circuits of arbitrary complex- ity can readily be achieved. The application of the dynamic simulation approach can help greatly in understanding the sometimes complex, nonlinear behav- iour and dynamic interactions in various grinding circuits. Dynamic simulation can be used to test ‘‘what-ifs’’ in grinding process operations such as circuit response to variations in feed and unit operation characteristics. It is a cheap and effective means of inves- tigating circuit optimisation without the risk of possible damage to operating units or production of a large amount of unwanted product during a physical optimisation process. Dynamic simulation is also extremely useful in developing and testing new ideas for process soft-sensors and control. The experience and knowledge gained in dynamic simulation of grinding circuits is directly appli- cable to other dynamic flowsheet modelling and optimisation problems in the minerals and process engineering industries. The advantages of building flowsheet models within the MATLAB/SIMULINK programming environment include the ability to readily develop and modify continuous, discrete and/or hybrid models of individual unit operations, with solution of the flowsheet system by a powerful in-built suite of equation solvers and analysis of results utilising extensive existing graphical capabilities. Flowsheet models of arbitrary complexity can easily be graphically developed, while individual unit models can be developed in terms of graph- ical block diagrams and/or customised block models written in computer code. � 2004 Elsevier Ltd. All rights reserved. Keywords: Comminution; SAG milling; Modelling; Simulation 1. Introduction Real time dynamic computer simulation has been a powerful tool not only in traditional high-tech aero- space and military industries, but also in other areas such as the automotive, steel making, and chemical processing industries. However, until very recently, there has been limited practical application of dynamic 0892-6875/$ - see front matter � 2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.mineng.2004.05.018 * Corresponding author. Tel.: +61 2 9710 6731; fax: +61 2 9710 6789. E-mail addresses: yi.liu@csiro.au (Y. Liu), steven.spencer@csiro.au (S. Spencer). simulation in most of the mineral processing industry, instead relying on pilot plant studies and/or steady-state flowsheet simulation for plant design, equipment dimen- sioning and pre-control optimisation. With recent progress in on-line measurement in min- eral processing, there are an increasing number of min- eral processing variables that can be measured on-line in real time (Death et al., 2002). Soft sensor models are also increasingly being developed for critical plant varia- bles that have previously been unavailable (Gonzalez, 1999). This progress has greatly improved the opportuni- ties for more advanced control techniques to be applied to mineral processing. To do this, a full understanding mailto:yi.liu@csiro.au mailto:steven.spencer@csiro.au 1190 Y. Liu, S. Spencer / Minerals Engineering 17 (2004) 1189–1198 of the dynamic behaviour of a processing circuit and the dynamic interactions between the external process varia- bles open to manipulation and internal (and perform- ance) variables of the circuit is crucial before any advanced process control can be successfully imple- mented. Real time dynamic simulation provides a power- ful tool to gain such an understanding with minimum associated cost. Many simulation packages and techniques already exist for flowsheet simulation in the mineral processing industry. They have been widely and successfully used for plant design, capacity planning (equipment sizing), circuit optimisation, problem diagnosis and costing pur- poses. However, most of these existing simulation pack- ages are based on steady state analysis (for instance METSIM, USIM PAC, Limn and JKSimMet) and may utilise empirical (and in the worst cases �black- box�) models of limited generality for individual unit operations. Such packages cannot simulate the dynamic behaviour and interactions of processing units within a circuit during transitional periods between various stea- dy states (including prediction of transition times), nor can they capture the real-time dynamic interactions be- tween external process variables (e.g. feed variations), internal variables (e.g. grinding mill load), and perform- ance variables (e.g. product size distribution and flow rate) of a processing circuit. Such dynamic variations and interactions can cause major problems for process control and optimisation, most notably, in the case of semi-autogenous grinding/autogenous grinding (SAG/ AG) mills in primary grinding circuits. Some dynamic simulation packages do exist (for instance Aspen Dynamics and SysCAD). In the case of the Aspen suite of products, their use may be viewed as relatively high cost and suitable largely as a �high end� solution to flow- sheet modelling needs for most of the mineral processing industry. In the case of SysCAD, the dynamic capability is available but to the authors knowledge has so far has largely been used as a means to obtain a steady state configuration for analysis. It is also known that SIMULINK has been used for simulation of the alu- mina refinery process and testing of control strategies at Nabalco–Alcan Gove Pty Ltd. However, the approach has not to our knowledge been extended for general use in the mineral processing industry. It is our intention to explore the techniques of dy- namic simulation used in other industries (under the MATLAB/SIMULINK environment) for development and application in mineral processing dynamic flowsheet simulation. The main reasons to use SIMULINK are its modular approach to model building, open model struc- ture, ease of changing circuit configurations and links, powerful real time graphic display functions for process variables, and integrated advanced nonlinear dynamic system solvers. A flexible and powerful dynamic simula- tion flowsheet modelling approach has accordingly been developed, with specific application in grinding circuit dynamic simulation. The reasons for the choice of grind- ing circuits as the initial area for model development is the relative maturity of dynamic mathematical models for some of the unit operations and the interest in dy- namic control of problematic unit operations such as SAG/AG mills. The approach exploits extensions of lit- erature dynamical mathematical models of grinding mill unit operation developed into a SIMULINK unit model graphical library and the flexibility/capacity of SIMU- LINK to link these individual units into complex dy- namic flowsheets. In this manner validated individual unit models can be linked in an arbitrary manner and used to perform true real-time dynamic simulations. The application of the dynamic simulation approach can help greatly in understanding the sometimes com- plex, nonlinear behaviour and dynamic interactions in various grinding circuits. Dynamic simulation can be used to test ‘‘what-ifs’’ in grinding process operations such as circuit response to variations in feed and unitoperation characteristics. It is a cheap and effective means of investigating circuit control and optimisation without the risk of possible damage to operating units or production of a large amount of unwanted product during a physical plant studies. Dynamic simulation is also extremely useful in developing and testing new ideas for process soft-sensors and control. The next section briefly describes the main mathemat- ical models used in our dynamic simulations. Section 3 summarises the general features of the simulation ap- proach and the specifics of the comminution model li- brary constructed for dynamic simulation of grinding circuits. Several dynamic simulation examples are given in Section 4. Section 5 concludes the paper with some re- marks on the flexibility of the approach, possible future extensions and practical applications. 2. Dynamic models for grinding circuit unit operations The key unit model in a grinding circuit is the grind- ing device itself, in many flowsheets being a SAG/AG and ball mills, respectively for primary and secondary grinding. There is a well-known mathematical model for ball mill operation based on the population balance modelling approach, with the assumption that mill dynamics can be modelled by a number of perfect mixers in series (see, Whiten, 1974; Austin et al., 1984). Let X(t) = [x1(t),x2(t), . . . ,xn(t)] T be the vector representing the mass of solids in discrete size fractions in a perfect mixer, then single mixer ball mill breakage in a mill with constant hold-up can be modelled as governed by the following equation: dX ðtÞ dt ¼ ðBðtÞ � IÞSðtÞX ðtÞ: ð1Þ Y. Liu, S. Spencer / Minerals Engineering 17 (2004) 1189–1198 1191 Here B(t) is the breakage distribution function (lower triangular matrix), S(t) is the breakage rate (selection) function (diagonal matrix), and I is the identity matrix. It is usually assumed that the breakage distribution and breakage rate functions are constant matrices, and can be estimated from batch grinding tests (see, Austin et al., 1984; Weller et al., 1997, 2000). To model a SAG/AG mill, as well as to reflect possible ore hardness changes in the feed, the following single mixer nonlinear grinding phenomenological model was developed by us: dX ðtÞ dt ¼ ðBðtÞ � IÞ bðtÞðcðtÞSðtÞ þ aSaðtÞX ðtÞÞ½ �X ðtÞ ð2Þ Here B(t) and S(t) are the same as in Eq. (1), and Sa(t) is a breakage rate (lower triangular matrix) repre- senting the effect of autonomous grinding. The constant 0 6 a 6 1 is a structure parameter. When a = 0, the model simulates a ball mill, 0 < a < 1 simulates a SAG mill, and when a = 1 and S(t) = 0, the model simulates an AG mill. Function b(t) is used to simulate changes in feed ore hardness. The model simulates ‘‘softer’’ ore when b(t) < 1, and b(t) > 1 for ‘‘harder’’ ore, and b(t) = 1 returns to ‘‘normal’’ ore hardness. Similarly, function c(t) is used to simulate the effects of ball charge in the mill. When c(t) > 1, extra balls are added, when c(t) < 1 balls are consumed, and when c(t) = 1, we as- sume no variations of ball charge in the mill. Eq. (2) is restricted to modelling breakage in a single perfect mixer. However, grinding mill operation gener- ally can be more reasonably modelled in terms of several perfect mixers connected in series. This provides a low order model for the dynamics of mass transportation through the mill. A critical parameter in this model is the mean residence time of the solids in the mill. An- other consideration is that at the discharge end of any grinding mill, there is generally a size classification effect, sometimes due to the presence of a grate or screen. In these circumstances, the following model better describes comminution in a perfect mixer: dX ðtÞ dt ¼ ðBðtÞ � IÞ bðtÞðcðtÞSðtÞ½ þaSaðtÞX ðtÞÞ�X ðtÞ þ 1 s ðf ðtÞ � pðtÞÞ pðtÞ ¼ CðtÞX ðtÞ ð3Þ Here f(t) = [f1(t), f2(t), . . . , fn(t)] T is the mass of the sol- ids feed, and mass of the mixer product is p(t) = [p1(t),p2(t), . . . ,pn(t)] T. Matrix C(t) contains classi- fication coefficients for the mixer. It is usually a diagonal constant identity matrix for all mixers of a grinding mill model except the last mixer, which will also be a diago- nal matrix but reflect the classification effects of the mill at the discharge. Here s is the mean residence time for solids in a mixer, which can be obtained by appropriate analysis of pulse injection tracer tests (see, Weller et al., 2000). Eq. (3) is the generic building block of the grinding mill models for this dynamic simulation approach. In practice, we also need to include a water phase mass bal- ance in the above mill model if a wet grinding circuit is to be simulated. The water phase model will not be dis- cussed in this paper. Other key unit operations in grinding circuits are mixing and separation devices. A dynamical mathemat- ical model of a sump unit operation can be derived in the similar fashion to Eq. (3), based on simple mixing principles with an associated mean residence time. There are many types of models for cyclone separators in the literature, which will not be discussed in this paper. The hydro-cyclone model we used in this study is based on an empirical model (Austin et al., 1984). Neither of these models will be described in any detail in this paper. 3. General features and the specifics of comminution models As can be seen from the last section, dynamical math- ematical models of grinding mills can quickly grow into some very complex, nonlinear and highly inter-con- nected differential equations. The complexity of the whole grinding circuit will dramatically increase once we start to connect different unit models into a grinding circuit and when wet grinding is considered. Any closed- loop control (even with simple PID control) will com- plicate the models further. It is clear that a powerful nonlinear differential equation solver is a must for any dynamic simulation of such complex models. In addi- tion, a modular and subsystem approach is highly desir- able to manage the complexity of the unit models and also, the simulation tool has to be sufficiently flexible to allow users to simulate a wide variety of types of grinding circuits with different connectivity. In order to fully understand the true dynamical behaviour of individual unit models and the linked flowsheet, it is also necessary to have real time graphical display capacity in the simulation tool. After a review of many commercially available dy- namic simulation packages on the market, SIMULINK (www.mathworks.com) was chosen for this work due to its strong dynamical modelling capability and flexibility. 3.1. General features of SIMULINK SIMULINK is a general purpose, very powerful and flexible dynamic system simulation environment. It has been applied to various time-domain dynamic system simulations in a wide variety of industries, such as aero- space (e.g. F14 flight control, missile flight control, lunar module autopilot, and radar tracking), and automotive (e.g. engine timing control, anti-lock brake system, auto- matic transmission control, active suspension, power http://www.mathworks.com 1192 Y. Liu, S. Spencer / Minerals Engineering 17 (2004) 1189–1198 window control). A good exposition of the capabilities of SIMULINK can be found at the Mathworks SIMU- LINK Technical Literature Web Page. The key features of the SIMULINK can be summa- rised as: • Modular and subsystem approach to handle very complex systems; • Intuitive block-diagram (graphical) interfacing makes it easy to construct and understand; • Very rich commonly used block and subsystem libraries; • Extensive control system libraries can be readily used for closed-loop simulations; • Flexible structures and configurations and user defin- able functions; • Powerful simulation solvers to handle highly nonlin- ear and stiff systems; • Powerful graphics and visualisation tools; • S-functions for addition of customblocks to SIMU- LINK models, defined in terms of MATLAB, C/ C++, Fortran or Ada code. 3.2. Specifics of the comminution dynamic model library The comminution dynamic model library (see Fig. 1) so far contains several versions of perfect mixer dynamic comminution models of varying degrees of complexity, which are the building blocks of the grinding mill models. A hydro-cyclone size separation model (Austin et al., 1984) has also been developed in the library. There are also several versions of the sump/pump dynamic model and links to several demonstration dynamic simu- Comminution Mo SAG Mill SAG Mill with hardness input holdup output Mixer Perfect Mixer with RT and hardness input and solids holdup output Mixer SAG Perfect Mixer with RT for SAG Mixer Perfect Mixer with water and RT Hydro-Cyclone Ball Mill Ball Mill with hardness input holdup output Ball Mill Ball Mill with 3 mixers and a classifier at end Fig. 1. Dynamic model libr lations of grinding circuits. A number of other mineral processing unit operation library models are currently under development. There are several key features of this library: • The general approach to model development is to try to make a model as generic as possible to accommo- date a variety of simulation situations. • To develop a grinding circuit, it is a simple task of drag and drop of appropriate blocks (grinding units) from the library into a new workplace, and then the blocks can be linked into a grinding circuit by click and drag of the mouse. After linking with the feed and appropriate display tools in the same way, the system is ready to be simulated once the model parameters are loaded. • To change the structure and configuration of the grinding circuit, it is a simple task of substituting blocks or re-linking the blocks in different ways. • The hydro-cyclone model is a user-defined function (called S-function), which can be a very complex dynamic function. It is easy to develop the S-function by following a few general rules. • The sump/pump model in the library is treated as a perfect mixer with residence time but without break- age functions. • The SAG Mill model in Fig. 1 is in fact a general dynamic model for tumbling mills (see Eq. (3)). It has several very useful characteristics: – One parameter (a in Eq. (3)) could change the model from a ball-mill (a = 0) to SAG mill (0 < a < 1) or to AG mill (a = 1 and S(t) = 0) simu- lations. It can also be used to model stirred ball mills (see, Weller et al., 2000). del Library Demo 3: Feed Size Change Demo 2: Feed Rate Change Demo 1: RT Change Demo 5: Size and Hardness Changes Demo 4: Ore Hardness Changes SUMP Sump model with pump rate input and solids holdup SUMP Sump model with RT and pump rate input SUMP Sump Model (sfun) (sfun) ary for comminution. This SAG mill model allows any number of mixers to be connected. Ore hardness be changed by Input 2, where input_2 < 1.0 means softer feed, input_2 > 1.0 for harder feed. 2 Solids & water mass holdups 1 Slurry mass discharge Mixer SAG Perfect Mixer with RT for SAG 3 Mixer SAG Perfect Mixer with RT for SAG 2 Mixer SAG Perfect Mixer with RT for SAG 1 2 Solids hardness change factor 1 Slurry mass feed Fig. 2. A SAG mill model with three perfect mixers. Y. Liu, S. Spencer / Minerals Engineering 17 (2004) 1189–1198 1193 – Mills can be modelled by any number of mixers to match mean residence time distributions (s�s in Eq. (3)) determined from tracer test experimental data (see, Weller et al., 2000). – Each mixer can have independent breakage func- tion and rate, mean residence time, and discharge classification coefficients. – Ore-hardness changes in the feed can be simulated by defining a time varying b(t) coefficient in the appropriate grinding model. Ball charge changes can also be simulated in a similar manner (time- varying c(t)), though this feature has not yet been implemented. – All models can be used for both wet and dry grind- ing simulation. Fig. 2 shows how three single mixer SAG blocks are connected to model an entire SAG mill. Again, the num- ber of mixers used to model a mill can be increased or decreased easily by the user to fit the real conditions. Each Mixer SAG block in Fig. 2 is the SIMULINK implementation of the mathematical model of Eq. (3). Usually, such a detailed model of a single unit operation would be masked under a single graphical interface icon and hence not shown to users. An important consideration in any simulation exer- cise is the validation of models. It should be noted that the ball mill model used in this library has been checked against the corresponding model in DYNAMILL, a Change Residence Time tau1, then the change of the rising and decay c of the ball mill with a constant vo -C- Water Feed -K-Volume offset K*u Solids Size Dist Solids Feed yin Mill_Feed Mill Feed Ball Mill Ball Mill with 3 mixer and a classifier at e Fig. 3. Example 1––Setup for simulation of a command line based dynamic mill simulation package originally developed by Raj Rajamani and John Herbst at University of Utah. It was found that under the same conditions, the models produced simulation results that differed by less than 2% in product size distribution. 4. Grinding circuit simulation examples Two examples are here used to demonstrate the use of the Comminution Model library and the key features of SIMULINK as mentioned above. 4.1. Example 1: Residence time effects in a ball mill In this example, a simple ball mill model consisting of three perfect mixers is simulated under open circuit con- dition. The slurry volumetric feed rate to the mill is kept constant with zero initial solids feed rate in order to grind-out the initial contents of the mill. Then a step change increase of the solids feed rate is then introduced and later a step change return to zero solids feed is intro- duced. Two runs of the simulation are performed with different residence times set for the first mixer in the ball mill model (the residence times for the second and third mixer are not changed). Fig. 3 depicts the simulation setup for the residence time test of a ball mill. Again, the feed to the mill can be easily changed to suit various simulation purposes re-run to see haracteristics lume feed Mixer Discharge yout Mill_Discharge Compare Two Total Solids Change RT tau1=3 Plot Total Solids & Water Plot Size Fractions Load Data n=16, tau1=1 s nd ball mill with residence time changes. 1194 Y. Liu, S. Spencer / Minerals Engineering 17 (2004) 1189–1198 and the simulation progress can be monitored by the scopes attached to feed stream and discharge stream. There are 16 size fractions at ffiffiffi 2 p size intervals used in the model. The simulation is for a wet mill, with an addi- tional water phase. In the first run of the simulation, the residence time constants for three mixers are all set to the same value (s1 = s2 = s3 = 1min). The simulation is repeated with the residence time of the first mixer set to s1 = 3min and the other residence times unchanged (s2 = s3 = 1min). Fig. 4 shows the dynamic responses of the ball mill model with residence time of all mixers equal (s1 = s2 = s3 = 1min). The total solids mass and water of the feed are shown in the first plot of Fig. 4 (total slur- ry volume is kept constant). The second plot in Fig. 4 shows the mass of solids in each particle size fractions in the discharge of the mill. It is interesting to note that the initial solids hold-up in the ball mill was ground out quickly due to no solids feed in the first 50min (there is a similarly rapid decline in solids mass at the second grind-out). Solids hold-up quickly increases with the step change addition of solid feed and stabilises at a con- stant level. The third plot in Fig. 4 shows the total mass of solids and water phases in the discharge of the mill. Due to the nature of theconstant volumetric feed, the discharge steam of the mill is also a constant in volume. Hence we can clear see that as expected, when solids in- creases in discharge, the water will decrease accordingly to keep the volume a constant, and vice versa. 0 20 40 60 80 100 0 100 200 300 400 Solids Fractions in Dis M as s Fr ac tio ns ( kg ) 0 20 40 60 80 100 0 500 1000 1500 Mill Discharge - Total Solids & Wa M as s (k g) Time (m 0 20 40 60 80 100 0 500 1000 Mill Feed - Total Solids, M as s (k g) 0 20 40 60 80 100 Fig. 4. Dynamic responses of a ball mill model Fig. 5 is a comparison of discharge total solids and water phase mass as a function of time for the simula- tion described above and a second simulation with the residence time of the first mixer changed as also noted above. The second plot in Fig. 5 clearly shows that when the residence time in the first mixer of the ball mill model is changed from 1 to 3min, the response of the total sol- ids in discharge stream is as expected, proportionally slower, i.e. the total solids mass in the discharge stream takes longer time to stabilise to the step changes of the feed. This example clearly shows that the mean residence time input to a ball mill model has a major impact on the responsiveness of the dynamic behaviour of the model to feed rate changes. It is easy to see that by adjusting the number of mixers and associated mean res- idence times, one can match the residence time response of a grinding mill model with tracer study data from a real mill (see Weller et al., 2000). 4.2. Example 2: Responses of a SAG mill circuit to feed size and hardness changes In this example, we link SAG mill, hydrocyclone and sump/pump models in a closed grinding circuit in feed- forward configuration. A simulation is carried out with a step up and down change of solids feed size at the fresh feed stream to the grinding circuit. When feed solids size increases, the mass of coarser fractions in feed increases and the mass of the finer fractions decreases so as to 120 140 160 180 200 charge Stream 120 140 160 180 200 ter. Mixer 1 RT 1 = 1 (min) in) 120 140 160 180 200 Water & Volume 120 140 160 180 200 636 638 640 F ee d V ol um e (li tr e)Solids Water Solids Water τ demonstrating mean residence time effects. 0 20 40 60 80 100 120 140 160 180 200 0 200 400 600 800 1000 Mill Feed - Total Solids, Water & Volume M as s (k g) 636 636.5 637 637.5 638 638.5 F ee d V ol um e (li tr e) Solids Water 0 20 40 60 80 100 120 140 160 180 200 0 500 1000 1500 Time (min) To ta l S ol id s M as s & W at er ( kg ) Comparison of Residence Time Change of the 1st Mixer of Ball Mill (Water - dotted) RT: 1=1, 2=1, 3=1 RT: 1=3, 2=1, 3=1 Water Solids τ τ τ τ τ τ Fig. 5. Comparison of changes in residence time of the ball mill model. Y. Liu, S. Spencer / Minerals Engineering 17 (2004) 1189–1198 1195 keep the total mass of solids feed at a constant. A step up and down change (10%) of feed ore hardness is sub- sequently imposed on the mill model. Water feed is kept at a constant rate during the simulation. Note that some white noise is added to the feed rate in an attempt to make it closer to the reality. There are again 16 size frac- tions at ffiffiffi 2 p size intervals used in this example. Some dis- charge classification effects are introduced at the last mixer associated with the SAG mill model. A SAG Mill Circuit Model with feed size and ore hardness changes Double cl ick to load the data wff Water Feed Solis & Wa Mass Hol sff Sol ids feed Sol ids Feed Disturbance Sol ids Feed Size changes SAG M i l l SAG Mill with hardness input holdup output Ore hardness disturbance 1 Normal hardness ym Mi l l feed Plot Size Distribution Plot Total Sol ids & Water Plot Other Mil l Infos Hardness changes yu Cyclone return Size Switch 2 size system R Fig. 6. A closed-loop SAG mill grin The SAG mill model consists of three perfect mixers incorporating comminution effects (see Eq. (3)) with the same constant breakage function B and breakage rate S, but different residence times, s. The autonomous grind- ing rate functions Sa are constant and the same for all mixers. The AG structure parameter a = 0.2 is used in the simulations. Fig. 6 shows the closed-loop grinding circuit of the SAG mill simulation. In this closed grinding circuit Cyclone input Product stream Water in Product wsf Water Addition 24 Sump vo lume holdup setpoint SUM P Sump model qsp Sump discharge volume rate offset ys Sump discharge Sump Volume Holdup ter dup Solids Product Pump rate PID PID Controller s Hydro-Cyclone m Discharge eturn stream ding circuit simulation setup. 1196 Y. Liu, S. Spencer / Minerals Engineering 17 (2004) 1189–1198 simulation, a simple and crude PID control is imple- mented to regulate the sump level to a constant (hence to prevent the sump from overflowing) by adjusting pump out rate of the sump alone (no changes to the water addition to the sump). Such simple control strat- egy is used for simulation purposes only and obviously it may be inappropriate for real SAG grinding mill operation. The first plot in Fig. 7 shows the total solids mass, the particle size (P50), and ore hardness in the fresh feed stream of the grinding circuit. It clearly shows the con- stant feed mass rate and the step up/down changes of both feed size and feed ore hardness. The second plot in the figure shows the solids and water hold-ups as well as the solids concentration in the SAG mill. The third plot in Fig. 7 shows the total solids mass and water in the product of the grinding circuit as well as the product size (P50). It is interesting to note that although the feed rate and production rate remain largely constant, the pro- duction size changes significantly with the change of feed size as well as the change of ore hardness. Also, we can see that the total mill hold-up does not change much during the simulation. This could simply be due to the fact that in the mixer model used, a further assumption on overflow in volume of the mixer is im- posed. This means that when feed rate is increased, dis- charge rate increases accordingly, such that volumetric hold-up remains constant. The first plot of Fig. 8 shows slurry volume of the sump (i.e. indirectly the sump level). It is clear that a 0 10 20 30 40 50 0 5000 10000 Total Solids and Size M as s (k g) & S iz e (µ m ) 0 10 20 30 40 50 1000 2000 3000 4000 5000 Solids and Water Holdup and Sol M as s (k g) 0 10 20 30 40 50 400 500 600 700 800 900 1000 Total Solids and Wa Time (m M as s (k g) Feed Size Solids Water Soli Water Fig. 7. Process variables of simple PID controller did a reasonably good job to keep the actual level close to the set point (a constant in this case). It is clear that when feed size increases, more coarse solids will pass through the mill and then be fed to the hydrocyclone. This in turn means that hydro- cyclone will reject more materials to the return stream. Hence the circulation load of the grinding circuit will in- crease (as shown in the second plot of Fig. 8). When this happens and if the sump pump out rate remains the same, the sump level will have to increase. In order to keep the sump level at a constant, the PID controller has to increase the pump out rate (as shown in the third plot of Fig. 8) to cope with the increased slurry feed to the sump. When ore hardness increases, again, more coarse solids will be discharged from the mill and be fed to the hydrocyclone, the circulation load increases, and the sump level will rise and again, the PID control loop increases the pump out rate as a compensation effect. It should be recalled that in Fig. 7, solids mass feed rate is constant, and hence so should be the product dis- charge rate of the circuit. The impact of an increased hydrocyclonecirculation load ratio due to changes in feed size and hardness are manifested in the product size distribution. The first plot in Fig. 9 shows the particle size (P50) of the solids at various points of the SAG mill circuit as a function of time. It is immediately apparent that all the P50 measures of particle size distribution at different positions in the circuit are in the expected order at any time. It is clear that when feed size increases, the circu- 60 70 80 90 100 in Fresh Feed 1 1.1 O re H ar dn es s 60 70 80 90 100 ids Concentration in Ball Mill 0.77 0.78 0.79 S ol id s C on ce nt ra tio n 60 70 80 90 100 ter in Product in) 120 130 140 150 160 170 180 P ro du ct S iz e - P 50 ( µm ) Feed mass Ore Hardness Concentration ds Size the SAG mill circuit. 0 10 20 30 40 50 60 70 80 90 100 678.5 679 679.5 680 680.5 Sump Volume V ol um e (li tr es ) 0 10 20 30 40 50 60 70 80 90 100 1 1.5 2 Cyclone Circulation Load Ratio C irc . L oa d R at io 0 10 20 30 40 50 60 70 80 90 100 1200 1400 1600 1800 2000 Sump Pump Volumatric Rate Time (min) P um p R at e (li tr es /m in ) Fig. 8. Internal dynamic responses of the SAG mill circuit. 0 10 20 30 40 50 60 70 80 90 100 10 2 10 3 10 4 Particle Size P50 TimeTraces at Different Points of the Circuit S iz e P 50 ( µm ) Time (min) 10 2 10 3 10 4 0 20 40 60 80 100 Comparison of Particle Size Distributions at T = 20 mins and T = 85 mins (*) P er ce nt ag e P as si ng ( % ) Solids Size (µm) Fresh Feed Mill Feed Cyclone Return Cyclone Feed Product Fig. 9. Particle sizes at different points of the SAG mill circuit. Y. Liu, S. Spencer / Minerals Engineering 17 (2004) 1189–1198 1197 lation load ratio increases, and then the product size in- creases (when the feed rate and production rate are effec- tively constant). When ore hardness increases, product size increases as well. The second plot in Fig. 9 shows particle size distributions (PSDs) at the same points in the SAG mill circuit at two time specific moments in time (at 25min with coarser feed size and at 85min under normal feed conditions). As expected, the PSD under coarse feed conditions is at all locations larger than for fine feed. This is an example of the detailed 1198 Y. Liu, S. Spencer / Minerals Engineering 17 (2004) 1189–1198 particle size information that may be obtained from such a dynamic simulation. 5. Conclusions A powerful and flexible library of mineral processing dynamical flowsheet unit operation models has been developed in the SIMULINK programming environ- ment. A generic, phenomenological mathematical model based on an extended form of population balance meth- od has been developed for grinding mills. This unit model is inherently nonlinear for SAG/AG mills and is suitable for dynamic simulation purposes. The power of the dynamic simulation approach based on SIMU- LINK environment has been demonstrated by two examples which illustrate the importance of dynamic ef- fects associated with variations of the key parameters of unit mean residence time, feed size distribution and hardness. The dynamic simulation approach developed here has great potential not only for grinding circuit dy- namic simulation, optimisation and control, but also for many other dynamic flowsheet modelling and optimisa- tion applications in the mineral processing industry. A number of further conclusions can be drawn from this work. SIMULINK is a powerful and flexible simu- lation tool. It offers a modular and subsystem approach for complex system handling. It has an intuitive block- diagram interface for model building and model config- uration, and rich libraries for basic model building and development of control algorithms. The powerful solver of the package is very easy to use and includes options for easy handling of complex and stiff systems of equa- tions associated with large flowsheets of complicated connectivity. The graphical display capabilities of the SIMULINK environment are similarly powerful and flexible to use. In simulations, all relevant variables are readily accessible in real time. This is particularly useful to fully understand the dynamic interactions of the external and internal variables of grinding circuits. In real industrial situations, however, on-line measurement of some those variables are far from trivial, if not totally impossible. Hence, it is clear that dynamic simulation can be used to answer many ‘‘what if’’ hypothetical questions for industrial grinding circuits. Application of this modelling approach to real plant dynamic simu- lation is potentially of great value to industry. As can be seen from an example in this paper, a sim- ple PID control loop is not optimal for mill operation. More studies on appropriate on-line measurement and control strategies are needed. Further extensions of the comminution model library would be very useful and a more thorough process of validating the model library using real plant data should be carried out. We believe that the dynamic simulation approach used here will as- sist in devising more appropriate control strategies for achieving maximum throughput while keeping very tight control on product size for grinding circuits. References Austin, L.G., Klimpel, R.R., Luckie, P.T., 1984. The Process Engineering of Size Reduction: Ball Milling. AIME, New York. Death, D.L., Cutmore, N.G., Sowerby, B.D. The role of on-line analysis and control in sustainable minerals processing. In: Green Processing 2002: International Conference on the Sustainable Processing of Minerals, Cairns, Qld. AusIMM, pp. 299–303. Gonzalez, G.D., 1999. Soft sensors for processing plants. In: Proceedings of the Second International Conference on Intelligent Processing and Manufacturing of Materials IPMM�99, Hawaii, July. Mathworks Inc, Simulink Technical Literature Web Page: http:// www.mathworks.com/products/simulink/technicalliterature.jsp. Weller, K.R., Liu, Y., Campbell, J.J., 1997. A new method for estimating the full-scale grinding target for floating from labora- tory batch grinding and flotation and reference full-scale tests. In: Sixth Mill Operators� Conference, Madang, Papua New Guinea, AusIMM, pp. 185–191. Weller, K.R., Spencer, S., Gao, M.W., Liu, Y., 2000. Tracer studies and breakage test in pilot-scale stirred mills. Minerals Engineering 13 (4), 429–458. Whiten, W.J., 1974. A matrix theory of comminution machines. Chemical Engineering Science 29, 580–599. http://www.mathworks.com/products/simulink/technicalliterature.jsp http://www.mathworks.com/products/simulink/technicalliterature.jsp Dynamic simulation of grinding circuits Introduction Dynamic models for grinding circuit unit operations General features and the specifics of comminution models General features of SIMULINK Specifics of the comminution dynamic model library Grinding circuit simulation examples Example 1: Residence time effects in a ball mill Example 2: Responses of a SAG mill circuit to feed size and hardness changes Conclusions References