CONE CRUSHER MODELLING AND SIMULATION Development of a virtual rock crushing environment based on the discrete element method with industrial scale experiments for validation Master of Science Thesis JOHANNES QUIST Department of Product and Production Development Division of product development CHALMERS UNIVERSITY OF TECHNOLOGY Göteborg, Sweden, 2012 Report No. 1652-9243 i REPORT NO. 1652-9243 Cone Crusher Modelling and Simulation Development of a virtual rock crushing environment based on the discrete element method with industrial scale experiments for validation JOHANNES QUIST Department of product and production development CHALMERS UNIVERSITY OF TECHNOLOGY Göteborg, Sweden 2012 ii Cone Crusher Modelling and Simulation Development of a virtual rock crushing environment based on the discrete element method with industrial scale experiments for validation JOHANNES QUIST © JOHANNES QUIST, 2012 Technical report no. 1652-9243 Supervisor: Carl Magnus Evertsson, Professor Department of Product and Production Development Chalmers University of Technology SE-412 96 Göteborg Sweden Telephone + 46 (0)31-772 1000 Cover. DEM simulation of a Svedala H6000 cone crusher studied in the thesis. Repro service Göteborg, Sweden 2012 iii Cone Crusher Modelling and Simulation Development of a virtual rock crushing environment based on the discrete element method with industrial scale experiments for validation JOHANNES QUIST Department of Product and Production Development Chalmers University of Technology SUMMARY Compressive crushing has been proven to be the most energy efficient way of mechanically reducing the size of rock particles. Cone crushers utilize this mechanism and are the most widely used type of crusher for secondary and tertiary crushing stages in both the aggregate and mining industry. The cone crusher concept was developed in the early 20th century and the basic layout of the machine has not changed dramatically since then. Efforts aimed at developing the cone crusher concept further involve building expensive prototypes hence the changes made so far are incremental by nature. The objective of this master thesis is to develop a virtual environment for simulating cone crusher performance. This will enable experiments and tests of virtual prototypes in early product development phases. It will also actualize simulation and provide further understanding of existing crushers in operation for e.g. optimization purposes. The platform is based on the Discrete Element Method (DEM) which is a numerical technique for simulating behaviour of particle systems. The rock breakage mechanics are modelled using a bonded particle model (BPM) where spheres with a bi-modal size distribution are bonded together in a cluster shaped according to 3D scanned rock geometry. The strength of these virtual rock particles, denoted as meta-particles, has been calibrated using laboratory single particle breakage tests. Industrial scale experiments have been conducted at the Kållered aggregate quarry owned by Jehander Sand & Grus AB. A Svedala H6000 crusher operating as a secondary crusher stage has been tested at five different close side settings (CSS). A new data acquisition system has been used for sampling the pressure and power draw sensor signals at 500 Hz. The crusher liner geometries were 3D-scanned in order to retrieve the correct worn liner profile for the DEM simulations. Two DEM simulations have been performed where a quarter-section of the crusher is fed by a batch of feed particles. The first one at CSS 34 mm did not perform a good enough quality for comparison with experiments; hence a number of changes were made. The second simulation at CSS 50 mm was more successful and the performance corresponds well comparing with experiments in terms of pressure and throughput capacity. A virtual crushing platform has been developed. The simulator has been calibrated and validated by industrial scale experiments. Further work needs to be done in order to post- process and calculate a product particle size distribution from the clusters still unbroken in the discharge region. It is also recommended to further develop the modelling procedure in order to reduce the simulation setup time. Keywords: Cone Crusher, Discrete Element Method, Bonded Particle Model, Simulation, Numerical Modelling, Comminution, Rock breakage iv to Malin, Hanna & Helena v ACKNOWLEDGEMENT I would like to acknowledge and thank a number of people that have made this work possible and worthwhile. First of all I appreciate all support and mentorship from my supervisor Magnus Evertsson and co-supervisor Erik Hulthén. Much appreciation goes to Gauti Asbjörnsson and Erik Åberg for helping out during the crusher experiments! Also, thank you Elisabeth Lee, Robert Johansson and Josefin Berntsson for help, support and fruitful discussions at the office! Thanks to Roctim AB for initiating and supporting the project and for providing 3D scanning equipment! Special thanks to Erik and Kristoffer for helping with data collection on site! I’m very grateful to Jehander Sand & Grus AB and all the operators for letting me conduct experiments at the Kållered Quarry and use the site laboratory. Special thanks to Michael Eriksson, Peter Martinsson and Niklas Osvaldsson! Finally I would like to thank the engineering team at DEM-solutions Ltd for all the support, help and discussions. Special thanks to Senthil Arumugam, Mark Cook and Stephen Cole! Johannes Quist Göteborg, 2012 vi NOMENCLATURE BPM Bonded particle model DEM Discrete element method PBM Population balance model PSD Particle size distribution HPGR High pressure grinding roll CSS Close side setting OSS Open side setting DAQ Data acquisition CAD Computer aided design CAE Computer aided engineering CPUH CPU Processing Hours HMCM Hertz Mindlin Contact Model MDP Mechanical Design Parameter OCP Operational Control Parameter SOCP Semi-Operational Control Parameter OOP Operational Output Parameter SPB Single Particle Breakage IPB Interparticle breakage CRPS Chalmers Rock Processing Systems Critical bond shear force ⃗ Normal force Critical bond normal force ⃗ Tangential force Shear bond stiffness ⃗ Normal damping force Normal bond stiffness ⃗ Tangential damping force J Bond beam moment of inertia Equivalent shear modulus Radius sphere A Equivalent young’s modulus Radius sphere B Equivalent radius ⃗ Contact resultant force Equivalent mass ̅ Normal unit vector Normal overlap 𝑡 ̅ Tangential unit vector Tangential overlap Bond beam length Poisson ratio Bond disc radius Coefficient of restitution ⃗ Bond resultant force Coefficient of static friction Bond normal torque Normal stiffness Bond shear torque Tangential stiffness s Eccentric throw Damping coefficient Eccentric speed ̇ DEM throughput capacity Mantle angular slip speed ̇ DEM mass flow Chamber cross-sectional area Sectioning factor Mantle slope angle PSD scalping factor Estimated particle tensile strength BPM bulk density Critical force for fracture Crushing force Hydrostatic bearing area vii TABLE OF CONTENTS 1. INTRODUCTION ............................................................................................................................................. 1 1.1 Background ............................................................................................................................................... 1 1.2 Objectives .................................................................................................................................................. 2 1.3 Research Questions .................................................................................................................................. 2 1.4 Deliverables ............................................................................................................................................... 2 1.5 Delimitations ............................................................................................................................................. 3 1.6 Thesis Structure ........................................................................................................................................ 3 1.7 Problem analysis ....................................................................................................................................... 4 2. THEORY ............................................................................................................................................................ 5 2.1 The Cone crusher ...................................................................................................................................... 5 2.1.1 Influence of parameters ..................................................................................................................... 7 2.1.2 Feeding conditions ............................................................................................................................. 9 2.1.3 Rock mass variation ........................................................................................................................ 10 2.2 The Discrete Element Method .............................................................................................................. 10 2.2.1 Approaches for modelling breakage .............................................................................................. 10 2.2.2 Previous work on DEM and crushing ........................................................................................... 11 2.2.3 Hertz-Mindlin Contact model ......................................................................................................... 11 2.2.4 Contact model calibration ............................................................................................................... 13 2.2.5 The Bonded Particle Model ............................................................................................................ 15 2.2.6 Future DEM capabilities ................................................................................................................. 16 2.3 Compressive breakage ........................................................................................................................... 17 2.3.1 Single Particle Breakage .................................................................................................................. 17 2.3.2 Inter Particle Breakage .................................................................................................................... 19 3. METHOD .......................................................................................................................................................... 20 3.1 DEM as a CAE tool ............................................................................................................................... 20 3.2 Bonded Particle Model Rock Population ............................................................................................ 20 3.2.1 Generating a bi-modal particle packing cluster ............................................................................. 21 3.2.2 Calibration of the bonded particle model ...................................................................................... 23 3.2.3 Introducing meta-particles by particle replacement ...................................................................... 23 3.3 Industrial scale crusher experiments ..................................................................................................... 24 3.4 Crusher geometry modelling ................................................................................................................. 27 3.5 Crusher data acquisition ........................................................................................................................ 29 4. CRUSHER EXPERIMENT ........................................................................................................................... 31 4.1 Size reduction .......................................................................................................................................... 31 4.2 Power draw .............................................................................................................................................. 32 4.3 Hydrostatic pressure ............................................................................................................................... 35 5. ROCK MATERIAL MODEL DEVELOPMENT ...................................................................................... 38 5.1 Single particle breakage experiment..................................................................................................... 38 5.2 BPM calibration ...................................................................................................................................... 40 5.2.1 Calibration simulations of BPMfine ................................................................................................ 40 5.2.2 Calibration simulations of BPMcoarse ............................................................................................. 42 6. CRUSHER SIMULATION ............................................................................................................................ 45 6.1 Particle Flow Behaviour ......................................................................................................................... 46 6.2 Breakage behaviour................................................................................................................................ 47 6.1 Size reduction .......................................................................................................................................... 49 7. RESULTS & DISCUSSION ........................................................................................................................... 52 7.1 Hydrostatic Pressure .............................................................................................................................. 52 7.2 Throughput Capacity .............................................................................................................................. 52 7.3 Power draw .............................................................................................................................................. 53 8. CONCLUSIONS .............................................................................................................................................. 55 9. FUTURE WORK ............................................................................................................................................. 57 10. REFERENCES ................................................................................................................................................. 58 1 1. INTRODUCTION In this section the background and scope of the project is presented. The ambition is to give the reader an understanding of why the project was initiated, how it has been set up and what the boundaries are in terms of time, resources, limitations and deliverables. 1.1 Background Rock crushers are used for breaking rock particles into smaller fragments. Rock materials of different sizes, normally called aggregates, are used as building materials in a vast number of products and applications in modern society. Infrastructure and building industries are heavily dependent on rock material with specified characteristics as the basis for foundations, concrete structures, roads and so on. Hence this gives a strong incentive to facilitate production of aggregates at low cost, high quality and low environmental footprint. In the mining industry the same argument applies, however here, the objective is to decimate the ore to the size at which the minerals can be extracted. Crushers are usually a part of a larger system of size reduction machines and so performance has to be considered not only on a component level but more importantly on a systematic level. This means that the optimum size reduction process is not the same for the mining and the aggregate industry. Cone crushers are the most commonly used crusher type for secondary and tertiary crushing stages in both the aggregate and the mining industry. Due to the vast number of active operating crushers in the world, a very strong global common incentive is to maximize performance and minimize energy consumption and liner wear rate. These goals are aimed both towards creating more cost efficient production facilities, but also in order to satisfy the aspiration for a more sustainable production in general. Historically the same type of crushers have been used both for the aggregate and the mining industry, however this is about to change and the crusher manufacturers now customize the design towards specific applications. In order to be able to design and create more application specific cone crushers, optimized towards specific conditions and constraints, better evaluation tools are needed in the design process. Normally in modern product development efforts, a large number of concepts are designed and evaluated over several iterations in order to find a suitable solution. These concepts can either be evaluated using real prototypes of different kinds or virtual prototypes. Physical prototypes of full scale crusher concepts are very expensive and the test procedures cumbersome. This provides a strong incentive for using virtual prototypes during the evaluation and design process. If a crusher manufacturer had the possibility of evaluating design changes or new concepts before building physical prototypes, lead times and time to market could potentially be dramatically shortened and the inherent risk coupled to development projects would decrease. The methods available for predicting rock crusher performance today are scarce and the engineering methodology used for prediction has historically been empirical or mechanical analytical modelling. These models have been possible to validate through experiments and tests. However, much remains to be explored regarding how the rock particles travel through the crushing chamber and which machine parameters influence the events the particles are subjected to. Compressive crushing is a very energy efficient way of crushing rock compared to many other comminution devices. The dry processing sections in future mines will probably use crushers to 2 reduce material to even finer size distributions than today in order to feed High Pressure Grinding Roll (HPGR) circuits in an optimum way. Replacing inefficient energy intensive tumbling milling operations with more effective operation based on crushers and HPGR circuits will potentially result in an extensive decrease in energy usage in comminution circuits. In order to enable this new type of process layout, cone crushers need further development in order to crush at higher reduction ratios. It is in these development efforts that DEM simulations will play a crucial role. 1.2 Objectives The aim of this work is to develop a virtual environment and framework for modelling and simulating cone crusher performance. The general idea is to perform experiments on an industrial operating cone crusher and carefully measure material, machine and operating conditions. The experimental conditions will act as input to the simulation model and finally the output from experiments and simulations will be compared in order to draw conclusions regarding the quality and performance of the simulation model. The crusher studied is located in a quarry in Kållered owned by Sand & Grus AB Jehander (Heidelberg Cement). The crusher is a Svedala H6000 crusher with a CXD chamber. 1.3 Research Questions In order to give focus to the work a number of key research questions have been stated in the initial phase of the project. The ambition is to provide answers with supporting evidence on the following: RQ1. How should a modelling and simulation environment be structured in order to effectively evaluate performance and design of cone crushers? RQ2. To what extent is it possible to use DEM for predicting crusher performance? RQ3. How should a DEM simulation properly be calibrated in order to comply with real behaviour? RQ4. What level of accuracy can be reached by using DEM for simulation of existing cone crushers? RQ5. How does a change in close side setting influence the internal particle dynamics and crushing operation in the crushing chamber? 1.4 Deliverables Apart from the learning process and experience gained by the author during the project, the project will result in a set of deliverables:  A simulation environment for modelling cone crusher performance  A BPM model incorporating heterogeneous rock behaviour, particle shape, size distribution and variable strength criteria  A method for calibrating BPM models  New insight into how the rock particle dynamics inside the crushing chamber are influenced by a change of the CSS parameter  New insight on how to validate DEM models by using laboratory and full scale experiments  A high frequency DAQ system  Master thesis report  Final presentation 3 1.5 Delimitations The scope of this project is relatively vast and hence it is important to consider not only what to do, but also what the boundaries are. The following points aim towards limiting the scope of the project:  One type, model and size of cone crusher will be studied  No iterations of the full scale DEM simulations will be performed in this thesis  The main focus of the Methods chapter will be on the methods developed in the project and how they have been applied. Standardized test methods utilized during sample processing, statistical methods and basic DEM theory will not be extensively reviewed.  No analytical cone crusher flow model will be used in the work  The numerical modelling will be done using the commercial software EDEM developed by DEM Solutions ltd.  The work will only briefly cover aspects regarding rock mechanics and rock breakage theory 1.6 Thesis Structure The work in this thesis is based on a two parallel tracks of activities in the experimental and numerical domain, see Figure 1. When using numerical modelling tools for investigating machine performance or design it is necessary to also conduct experiments. The experiments not only act as the basis for validation of simulations but also give the researcher fundamental insight into the operation of the system being studied. In order to be able to discuss and draw conclusions a brief theory section is included in the thesis. It aims towards describing the fundamentals regarding the cone crusher, the discrete element method and some theory regarding rock breakage. In the methods chapter, the focus is put on presenting the methods developed in the project rather than describing and listing standardized procedures utilized e.g. how to conduct sieving analysis. A separate chapter is dedicated to the Material Model Development. Both the physical breakage experiments as well as the DEM simulations performed in order to calibrate the material model are presented. This chapter is followed by the Crusher Experiments and Crusher Simulation sections. In these chapters the results from each separate activity are shown and briefly commented. The Result & Discussion chapter is dedicated to a comparative study of the simulation and experimental result. Finally Conclusions are drawn regarding the results and each research question stated above is addressed. 4 Figure 1 - Project approach with two parallel tracks of activities in the experimental and numerical domains 1.7 Problem analysis In order to fulfil the stated goals a number of problematic obstacles need to be bridged. One of the difficult issues to decide is how and what to measure in the cone crusher system. The following experimental aspects need special consideration;  How to sample and measure the coarse feed size distribution as there are no standardized laboratory methods or mechanical sieves that handle sizes above ~90mm?  How to measure pressure, power and CSS signals in a satisfactory manner as the crusher control equipment has a too low sampling rate and is hence negatively affected by the sampling theorem? In order to be able to simulate the rock breakage in the crusher a breakage model needs to be developed. The following breakage modelling aspects need special consideration;  What level of complexity is possible to incorporate in the model in terms of particle shape, rock strength and heterogeneity?  How should the breakage model be calibrated?  What is the most suitable way of generating a particle population with breakable particles of different sizes and shapes? A number of issues and obstacles need to be addressed when it comes to the crushing simulation especially concerning computational capacity. The following aspects regarding the crushing simulations need special consideration;  Should the whole crusher be simulated or only a segment?  How many bonded-particles is it possible to incorporate?  How should the rock particles be introduced in the DEM model?  What crusher geometry should be used in the simulation? In the experiments the liners are severely worn, hence nominal CAD geometry will not be a correct equivalent to the experimental operating conditions. DEM environment development Literature Study Crusher DAQ system setup Single particle 3D- scanning and compression BPM model calibration Full scale crusher experiments Crusher DEM simulations Sample Processing DEM data analysis Data compilation and processing Report compilation Numerical domainExperimental domain 5 2. THEORY In this section the theoretical background will be presented for a number of areas of interest in the thesis giving the reader an introduction and framework in order to follow the discussion and analysis in upcoming chapters. 2.1 The Cone crusher The current engineering process for developing crushing machines is based on minor incremental changes to a basic fundamental mechanical concept. There are two main types of cone crusher concepts available on the market, the so called Hydrocone and Symons type crushers. The main differences lies in the choice of main shaft design and how to take care of the loads and dynamics using different bearing design, see Figure 2. These design choices are coupled to various advantages as well as negative limitations for both concepts. The main shaft in the Hydrocone concept is supported in the top and bottom by plain bearings and a hydraulic piston. The attractiveness of this solution is that the shaft vertical position can be adjusted hydraulically. This enables online adjustment of the CSS for e.g. utilization in control algorithms or compensating liner wear. Also, it is relatively easy to take care of the tramp protection, i.e. foreign hard metal objects unintentionally placed in the crusher, by having a hydraulic safety valve that quickly drops the main shaft before the crusher is seriously damaged. In the Symons concept the mantle position is fixed on top of a shorter main shaft with the plain bearing on top. The CSS is varied by moving the top shell up and down instead of the mantle. The top shell can only be turned when not loading the crusher. Hence, it is not possible to adjust the CSS during operation. An advantage with the fixed-shaft design is that the pivot point can be positioned at a vertical position above the crusher enabling a more parallel mantle movement. The pivot point is governed by the radius of the plain thrust bearing. The illustration in Figure 3 shows a horizontal cross-section of the mantle and concave explaining the eccentric position of the mantle and how it is related to the gap settings (CSS), throw and eccentric movement. The engineering knowledge foundation is mainly built on empirical tests, field studies and analytical models developed by e.g. Whiten [1], Eloranta [2] and Evertsson [3]. Crusher manufacturers commonly use different types of regression models based on test data to predict performance output. These models are unique for each type of crusher and a number of correction factors are normally used to adjust for application specific aspects such as rock type and strength. As these models are only partly based on mechanistic principles they are more suited for designing circuits rather than designing new crushers. A simplified expression of the hydrostatic pressure and how it relates to the crushing force and angle of action can be seen in Eq. 1. The crushing force is a representation of the accumulated forces from each interaction between rocks and the mantle under compression. If the crusher chamber is evenly fed with material with homogenous properties the pressure should be relatively constant. However, if a deviation occurs at some position or over an angular section where e.g. there is less material or material of other size and shape, the force response changes on the mantle. This force response variation would be observable in the momentary oil pressure signal. In other words the shape of the pressure signal gives information regarding the current force response upon the mantle. Eq. 1 6 Figure 2 - Schematic illustrations of the vertical cross-sections of the Hydrocone (left) and Symons (right) type Cone crusher. A simplified representation of the forces can also be seen. Note the difference in pivot point position due to the different mechanical setups. Figure 3 - Illustration of the horizontal cross-section A-A from Figure 2 showing mechanical and operational parameters as well as the cross-sectional area. A-A 𝑜𝑖𝑙 ,1 ,2 ,3 𝑠 𝑣 𝑧 ′ 𝑥 ′ 𝑐𝑟𝑢𝑠 ℎ ,2 ,1 ,3 𝑣 z x z x Closed side setting (CSS) Open side setting (OSS) Eccentric throw (s) 𝑐𝑐 𝑠𝑙𝑖 𝑐 ,𝑧𝑖 x y 7 2.1.1 Influence of parameters Cone crusher related parameters can be defined in four groups;  Mechanical Design Parameters (MDP) – Static parameters established in the design and commissioning process not possible to influence actively in operation without substantial re-engineering.  Operational Control Parameters (OCP) – Parameters that are possible to change during operation in order to control and influence performance.  Semi-Operational Control Parameters (SOCP) – Parameters that are possible to change, however only during shutdown or maintenance stops due to the need for e.g. change of mechanical parts.  Operational Output Parameters (OOP) – Resulting parameters of the crushing operation like e.g. power draw and pressure. Close Side Setting – When decreasing the CSS the product size distribution evidently gets finer as the gap, limiting the size of rocks leaving the chamber, is reduced. In Hydrocone type crushers the CSS can be adjusted during crushing operation as the main shaft vertical position is controlled by hydraulics. It is hence an OCP parameter and can be actively used as a control parameter. In Symons type crushers however the top shell needs to be adjusted in order to change the CSS. This can to date only be done when the crusher is not under load and should therefore be categorized as a SOCP parameter for these crusher types. Eccentric speed – When increasing the eccentric speed of the crusher mantle the material will be subjected to an increased number of compressive events. As a consequence each compression event will be performed at a lower compression ratio as the ith event will occur at a higher position in the crushing zone. It has been experimentally shown that a lower compression ratio results in a better shape [4]. Also, due to the increased number of compression events, the particle size distribution will be finer [5]. However, when increasing the number of events the particles will move slower down through the crushing zone. Conclusively, higher speed results in a relative increase in shape quality and a finer product but with the sacrifice of reduced throughput. Historically the eccentric speed can normally not be changed during operation without changing belt drive and is therefore a MDP/SOCP parameter. However, by installing frequency drives the eccentric speed can be adjusted during operation and hence converted to an OCP parameter. This has been done successfully by Hulthén [6] in order to actively control the speed as an enabler for performance optimization. Eccentric throw – The eccentric throw controls the amplitude of the sinusoidal rotations around the pivot points X- and Y- axis. The geometrical motion is achieved by using an eccentric bushing, see Figure 2. The throw can be adjusted within a specific range during shutdown by turning the bushing and is defined as a SOCP parameter. Liner design – All commercially available crusher models come with the choice of a set of liner designs ranging from fine to coarse profiles. Choice of profile is governed mainly by the feeding size distribution and desired product size distribution. The liner surfaces wear and are replaced after a couple of hundred operation hours depending on the abrasiveness of the rock type. 8 Figure 4 - Schematic illustration of the cross-sectional area, see Figure 3, at every z-coordinate displaying the choke level position. Changes to the liner profile will evidently result in a changed shape of the cross-sectional area plot. Further on this means a new choke level position and a new operating condition. Choke level – The choke level is an indirect variable not possible to measure during operation. It is the level or vertical position in the crushing chamber which limits the particle flow through the crushing chamber. If considering the cross-sectional area in the 𝑥 𝑙 between the mantle and concave for all values (see Figure 3), as illustrated by Figure 4, a narrow section exists. Below this narrow level the gap decreases, however as the radius increases the cross- sectional area actually increases. This means there will be more space for particles to be crushed. In effect observations show that the choke level is a transition zone where the breakage mode shifts from interparticle breakage to single particle breakage [3]. The choke level is besides the geometrical features of the liner design, also a function of the eccentric speed, CSS and eccentric throw. Power draw – Based on how the crusher is run and how much material is introduced into the crushing chamber, a specific amount of energy will be used to break rocks every second. The electric motor will always try to maintain the set speed and will pull more or less current based on the load on the mantle and main shaft. If adding up the torque components from all particle- mantle interactions, obtained from the crushing force needed to break each rock, this would be the resistance the motor needs to overcome (plus mechanical losses). The power draw is an OOP parameter and is used for monitoring how much work the crusher is doing, often in relation to its optimum performance capability. Hydraulic pressure – Most modern crushers are equipped with hydrostatic bearings where the pressure can be monitored using pressure gauges. The pressure level gives an indication of the crushing force on the mantle according to the relationship in Eq. 1. The condition of the pressure signal also holds information regarding the crushing operation. High amplitude suggests that the mantle is performing different amounts of crushing work at each circumferential position. Reasons for this could be miss-aligned feeding of the crushing chamber or segregated feed. If the crusher chamber lacks material, i.e. is not choke fed, the pressure will drop when the mantle reaches that position. In the case of segregation the feed size distribution Cross-sectional area [m 2 ] Z -c o o rd in a te [ m ] 9 will be different at all circumference positions inevitably giving different bed confinement characteristics hence different force response. 2.1.2 Feeding conditions The presentation of rock material to the crusher, i.e. feeding of the crusher, is one of the most crucial operational factors. Normally vibrating feeders or belt conveyors are used for feeding material to the crusher feeding box. In many cases this arrangement is not sufficient in order to achieve satisfying feeding conditions. Figure 5 - DEM simulation of the feeding of a cone crusher. The picture clearly shows the segregation behaviour as well as the proportionally higher amount of material in the right section of the crusher chute. (Unpublished work by Quist) As implied in the previous section many crushers are to some degree badly fed and experience two different issues; misaligned feeding and segregation. Misaligned feeding means that the material is not evenly distributed around the circumference hence there will be different amounts of rock material at all φ-positions. When operating under full choke fed condition the misaligned feeding is less of a problem. Segregation means that the particle size distribution will be different at φ-positions around the circumference. The reasons for these issues are coupled to how material is presented and distributed in the crusher rock box. When using a belt conveyor as a feeder the material can segregate very quickly on the belt. This segregation propagates into the crusher and is amplified when the rock stream hits the spider cap, see Figure 5. The spider cap acts as a splitting device causing coarse particles to continue to the back of the crusher and fine particles to bounce to the front. As the material has a horizontal velocity component in order to enter the crusher a large fraction of mass will end up in the back and a lower fraction of mass in the front. This effect is less when using vibrating feeders instead of conveyors as the horizontal velocity component is lower. For a more thorough investigation and description of this issue and ways to resolve it, the reader is advised to see Quist [7]. The operational effects of these issues are that the crusher effectively will perform as a different crushing machine at all φ-positions. As already stated this means that the hydraulic pressure will vary as the mantle makes one revolution. The result can be fatigue problems leading to main 10 shaft failure, cracks in the supporting structure, uneven liner wear, poor performance and control as well as many other problems due to that the machine is run in an unbalanced state. 2.1.3 Rock mass variation For most aggregate quarries as well as mining sites the mineralogical content of the rock mass varies throughout the available area. This results in variation of the rock characteristics momentarily as well as on a long term basis. Meaning that the best operating parameters today may not be optimal next month, week or maybe even next hour [6]. When varying the rock competency the size distributions produced up-stream will slightly change giving new feeding material characteristics. 2.2 The Discrete Element Method DEM is a numerical method for simulating discrete matter in a series of events called time- steps. By generating particles and controlling the interaction between them using contact models, the forces acting on all particles can be calculated. Newton’s second law of motion is then applied and the position of all particles can be calculated for the next time-step. When this is repeated it gives the capability of simulating how particles are flowing in particle-machine systems, see Figure 6. It is also possible to apply external force fields in order to simulate the influence of e.g. air drag or electrostatics. By importing CAD geometry and setting dynamic properties the environment which the rock particles is subjected to can be emulated in a very precise manner. This gives full control over most of the parameters and factors that are active and interesting during a crushing sequence. Also, due to the fact that all particle positions, velocities and forces are stored in every time-step, it is possible to observe particle trajectories and flow characteristics. Figure 6 - Illustration of the DEM calculation loop used in EDEM 2.2.1 Approaches for modelling breakage As the main purpose of this work is to break rocks in a simulation environment the choice of breakage model is important. Two different strategies dominate when it comes to modelling rock breakage in DEM – The population balance model (PBM) and the bonded particle model (BPM). The population balance model is based on the principle that when a particle is subjected to a load exceeding a specific strength criterion it will be replaced by a set of progeny particles of predetermined size. The strength criteria values and progeny size distribution are gathered from calibration experiments. This method is suitable for simulating comminution systems where impact breakage is the dominant breakage mode. The method has however been 11 used for modelling cone crushers as well [8]. The BPM method is based on the principle of bonding particles together forming an agglomerated cluster. Despite the fact that the PBM approach is more computationally effective and easy to calibrate the BPM approach is chosen for this work. The first reason is that the performance of a cone crusher is highly dependent on the particle flow dynamics within the crushing chamber. When using the PBM approach the particle dynamics are decoupled as the progeny particles are introduced at the same position as the broken mother particle. Hence the model cannot take into consideration particle movement as a result of a crushing sequence. This is not a problem in the BPM approach as the meta- particles are actually broken apart into smaller clusters. This leads up to the second reason which is that the PBM model is not based on simulating a crushing sequence but is basically only making use of the possibility to calculate forces on particles in DEM. The breakage itself is governed by an external breakage function. In conclusion the PBM approach basically uses the DEM model as an advanced selection function. 2.2.2 Previous work on DEM and crushing A few publications exist on the topic of using DEM for rock crushers and cone crushers in particular. In the case of impact crushers Djordjevic and Shi [9] as well as Schubert [10] have simulated a horizontal impact crusher using the BPM approach. However in both cases relatively few particles have been used and the geometries are very simplified. A DEM model for simulating rock breakage in cone crushers has been presented by Lichter and Lim [8]. However, this model was based on a population balance model (PBM) coupled with a breakage function. This means that when a particle is subjected to a load greater then a threshold value it will be considered broken and the model replaces the mother particle with a set of progeny particles, sized according to the breakage function. This approach is very powerful in respect of computational efficiency but the actual breakage events are controlled by statistical functions, hence it is possible to tune the simulation towards performing according to experiments without knowing if the particle flow through the chamber is correct. Another aspect is the relationship between loading condition on a particle and particle breakage. Depending on a 1:1, 2:1, 2:2 or 3:1 point loading between two plates the rock will break differently. Generally, a rock particle subjected to a load will either be; undamaged, weakened, abraded, chipped, split or broken. In the PBM approach only the last effect is considered. Therefore current work is based around the more computational cumbersome Bonded Particle Model (BPM). This method has been previously utilized by the author for modelling a cone crusher [7, 11] as well as a primary gyratory crusher [12]. 2.2.3 Hertz-Mindlin Contact model The Hertz-Mindlin contact model, Figure 7 is used for accurately calculating forces for particles- particle and particle-geometry interactions in the simulation [13]. The normal force component is derived from Hertzian contact theory [14] and the tangential component from work done by Mindlin [15]. 12 Figure 7 - Schematic illustration of the Hertz-Mindlin contact model used in EDEM. Damping components are added to normal and tangential force components where damping coefficients are linked to the coefficient of restitution. The normal force is given by considering the normal overlap according to, ⃗ √ ⁄ Eq. 2 The damping force is given by, ⃗ √ ⁄ √ �⃗� Eq. 3 Where the equivalent Young’s modulus , equivalent radius , equivalent mass , damping coefficient and stiffness are given by, Eq. 4 Eq. 5 ( ) Eq. 6 𝑙 √𝑙 Eq. 7 √ Eq. 8 , – Young’s modulus for spheres in contact , – Poisson ratio for spheres in contact , – Radius for spheres in contact – Coefficient of restitution The tangential force component is defined as the tangential stiffness times the tangential overlap. In addition the tangential damping force and tangential stiffness is given by, 𝐵 𝑠 𝑡 ⃗ 𝑜𝑟 𝑙 ⃗𝑡 𝑔 𝑡𝑖 𝑙 𝑥𝐵 ,𝑙𝑜𝑐 𝑙 𝐵 ,𝑙𝑜𝑐 𝑙 𝑧𝐵,𝑙𝑜𝑐 𝑙 ,𝑙𝑜𝑐 𝑙 𝑥 ,𝑙𝑜𝑐 𝑙 𝑧 ,𝑙𝑜𝑐 𝑙 13 ⃗ Eq. 9 ⃗ √ ⁄ √ �⃗� Eq. 10 √ Eq. 11 2.2.4 Contact model calibration When using DEM for modelling breakage most of the focus is put on making sure that the contact model governing the fragmentation corresponds to a realistic behaviour. However it is very important to make sure that the contact model controlling flow behaviour is calibrated as well. If the friction parameters are not correct the particles will flow in an incorrect manner. When compressed the particles may e.g. slip and escape compression when in reality it would be nipped and broken. No generally accepted method exists for calibrating contact models towards good flow behaviour. Hence a calibration device has been designed and built by CRPS [16]. A CAD model of the device can be seen in Figure 8. The device consists of an aluminium mainframe that holds a bottom section with a removable sheet metal floor and fixed sides. The top section holds a hopper with variable aperture and angle as well as a sliding plane with variable angle. The height of the top section can be adjusted. The different adjustment possibilities enable tests with different conditions. It is very important when calibrating a DEM contact model that it is independent of flow condition. In Figure 9 an example of a calibration procedure can be observed. In the left picture the particle flow has been captured using a high speed camera. By iteratively varying parameters, simulating and comparing with the reference a decent set of values for the friction parameters can be found. 14 Figure 8 - DEM contact model calibration device developed by Quist at CRPS Figure 9 - Snapshots from high speed video camera to the left and DEM simulation to the right. 15 2.2.5 The Bonded Particle Model The BPM model was published by Potyondy and Cundall [17] for the purpose of simulating rock breakage. The approach has been applied and further developed by Cho [18]. The concept is based on bonding or gluing a packed distribution of spheres together forming a breakable body. The particles bonded together will here be called fraction particles and the cluster created is defined as a meta-particle. The fraction particles can either be of mono size or have a size distribution. By using a relatively wide size-distribution and preferentially a bi-modal distribution the packing density within the meta-particle increases. It is important to achieve as high packing density as possible due to the problematic issue with mass conservation as the clustered rock body will not be able to achieve full solid density. Also, when the bonded particle cluster breaks into smaller fragments the bulk density will somewhat change as area new particle size distribution is generated. Figure 10 - Schematic representation of; (a) two particles overlapping when interacting giving a resultant force according to the contact model seen in Figure 7. (b) two particles bonded together with a cylindrical beam leading to a resultant force as well as normal and shear torques(modified from [17, 19]). Figure 11 - Schematic force-displacement plot of the different modes of loading on a bond beam. The stiffness's and critical stress levels are also shown. (Modified from [18]) The forces and torques acting on the theoretical beam can be seen in Figure 10. The schematic graph in Figure 11 illustrates the relationship between different loading modes (tension, shear, and compression), bond stiffness and strength criteria. Before bond-formation and after bond ⃗𝑖 ̅𝑖 𝑡�̅� 𝑡�̅� ̅𝑖 𝑏 2 𝑏 ⃗𝑖,𝑏 𝑏 𝑏 𝑠 (a) (b) 𝑐 𝑠 𝑐 Bond breaks 𝑏 𝑠 𝑏 𝑠 Bond breaks Compression T e n sio n Shear 1 1 16 breakage the particles interact according to the Hertz-Mindlin no slip contact model. The bonds are formed between particles in contact at a pre-set time 𝑡 . When the particles are bonded the forces and torques are adjusted incrementally according to the following equations: ⃗ Eq. 12 ⃗ Eq. 13 Eq. 14 Eq. 15 Where, 𝑡 𝑡 𝑡 𝑡 Eq. 16 The normal and shear stresses are computed and checked if exceeding the pre-set critical stress values according to the equations below: ̅ ⃗ Eq. 17 ̅ ⃗ Eq. 18 In this work the critical strength levels are set to a single value defining the rock strength. For future work it would be preferable to be able to randomize the critical bonding strength within a specified range or according to a suitable probability function. 2.2.6 Future DEM capabilities DEM is a very computational intensive method. The continuous development of CPU speed enables larger particle systems to be modelled and by using several CPU processors in parallel the computational capacity is improving. However, since 5-10 years back Graphics Processing Units (GPU) have been adopted and used for computational tasks due to the high potential for parallelization. While a normal DEM simulation commonly utilizes 2-16 CPU cores a high-end GPU consists of 400-500 cores. If utilized effectively, this has the potential to dramatically increase the computational capacity by 10-100 times. But it is not as easy as just recompiling the source code and starting running on GPUs. The algorithm needs to be adopted to be run in parallel on all the GPU-cores, a task which has been proven as difficult, but not impossible. The developer community is vivid and the library of available functions is steadily increasing. A few DEM vendors have beta versions of GPU-based DEM codes and these will probably be available on the market in a few years. 17 2.3 Compressive breakage It has been found by Schönert [20] that the most energy efficient way of reducing the size of rock is to use slow compressive crushing. Each particle can be loaded with the specific amount of energy needed to generate fracture resulting in progeny particles with a wanted size and specific surface. 2.3.1 Single Particle Breakage It is not possible to analytically calculate the internal state of stresses of a single irregularly shaped particle subjected to compressive load. Hence stress based measurements of particle strength are only valid for primitive regular shapes [21]. Hiramatsu and Oka [22] investigated the tensile strength for irregular as well as spherical shapes and showed that the tensile particle strength for an irregular shaped rock can be approximated by the following expression. Eq. 19 This simple equation is derived from a more complex expression of the stress state of a sphere subjected to compression. The numerator is defined as the critical force for failure times a factor given by; the loading condition, geometrical features and poisons ratio. The denominator is defined as a disc-area where D is the distance between the loading points. In this work this approximate substitute particle strength is used when conducting single particle compression tests in order to calibrate the DEM bonded particle model. The equation is very convenient since it is possible to extract both the critical force as well as the distance between loading points when conducting compression breakage tests, see Figure 12. This test-procedure will be further explained in the Material model development chapter. Table 1 - Contact loading point arrangements for single particle compression between two plates Type Plane A Plane B I. 1-point 1-point II. 2-point line 1-point III. 2-point line 2-point line IV. 3-point plane 1-point V. 3-point plane 2-point line VI. 3-point plane 3-point plane 18 When compressing an irregular shaped rock particle it will be pressed between two parallel surfaces experiencing loading at specific contact points, see Figure 12 and Figure 13. The number of contact points varies depending on the shape and orientation of the particle. In theory a number of contact arrangements exist as demonstrated in Table 1. Some types are more frequently observed than others. As an example consider type I where the particle is in contact at only one position for each plate. This for example, is the case for a spherical particle as described above. It is unlikely for an irregular rock particle to only have two contact points if the experimentalist is not positioning the rock specimen manually in such a way until compression begins. During experiments it was observed that type IV and V are the more common loading point arrangements. It was also observed that when compressing a particle between two plates an interesting phenomenon occurs; local positions of the particle subjected to contact are often relatively sharp. In the initial phase of the compression the local stresses are hence very high resulting in local crumbling breakage due to the brittle nature of rock material. This increases the area of contact and influences the upcoming stress state in the body and hence the breakage characteristics. This is also the reason why the otherwise statistically very unlikely type VI point loading arrangement may occur. Figure 12 - Schematic illustration of the different phases during a single particle compressive breakage test. Unloaded Loaded with a compressive force Loaded just before critical breakage Fi = 0 0 < Fi < Fc Fi ≤ Fc 0 < Fi << Fc Fi = 0 Loaded just after critical breakage Particle broken D 19 Figure 13 - Photo of an amphibolite particle from the feed sample subjected to compressive breakage 2.3.2 Inter Particle Breakage Inter particle breakage can be defined as the breakage mode where a bed of particles is compressed and broken within a confined or unconfined space. During compression, forces transmigrate though the bed from particle to particle creating a force network. The packing structure is hence of interest when studying bed breakage. Several parameters influence the packing structure of a material bed;  Particle size distribution  Particle shape  Internal friction  Wall friction  Solid density When a bed of particles is loaded the particles re-arrange slightly until a static condition is reached. The bed is then elastically compacted until particles start to fracture. Some research has been conducted in the field of interparticle breakage in order to better understand the complex mechanisms. Evertsson and Briggs [23] as well as Liu and Schönert [24, 25] have made important contributions. The discrete element method has been used as a tool for investigating interparticle breakage of mono-sized rocks and other agglomerates [26, 27]. Numerical FEM simulations of interparticle breakage in a confined space have been conducted by Liu [28]. The breakage of a bed of particles has been modelled in 2D FEM software in order to investigate the fragmentation behaviour when compressing the bed. In the beginning of the compression smaller particles are loaded in a quasi-uniaxial or quasi-triaxial compression mode. The smaller particles have fewer contact points then the larger particles hence the stress field generated has a higher local maximum stress resulting in crack propagation. Larger particles are surrounded by smaller fragments and hence experience a high number of contact points. As the displacement increases the larger particles will also experience high enough stresses to cause Hertzian crack propagation. The interparticle breakage in a cone crusher happens either in a confined or an unconfined condition depending on the operation. An interesting question is how large the angular segment is where actual confined breakage takes place as the mantle moves eccentrically if the feeding condition changes. 20 3. METHOD In this section all methods that have been applied or developed in the different phases of the project are presented with the aspiration that the reader should theoretically be enabled to reproduce the experiments and simulations. Focus will mainly be put on how methods and theories have been applied in contrast to the previous section where the theoretical background is introduced in a more general sense. 3.1 DEM as a CAE tool The role for CAE tools at R&D departments in all industries is growing steadily. With new capabilities to simulate and evaluate design decisions during concepts or detail development, time and resources previously spent on expensive prototypes and field testing are better spent. The DEM method is a fairly new tool which is in its cradle when it comes to systematic usage by large R&D departments. Hence, few methodologies or frameworks exist for how or when to use DEM. DEM is one of many computational engineering techniques so the methodology emerged in the field of e.g. FEM could be interesting to review. As this method has been around for a much longer time a lot of research has been conducted on the management around FEM analysis. During development and design of machines or processes which interact with granular material, it is commonly difficult to predict the behaviour and performance of the system. The types of situations where analysis and simulation are needed can roughly be categorized as follows;  Evaluation  Problem solving  Optimization  Fundamental understanding These four can be of interest both for new products as well as for existing products and implementations. It has been found in this work that in order to be effective and fully leverage the power of DEM it is crucial to adopt a statistical approach. When it comes to optimization and fundamental understanding where a high number of parameters need to be studied, it is recommended to use the design of experiment approach. As computational resources are usually scarce , fractional factorial analysis [29] is a good way of reducing the simulations needed in order to draw conclusions. In the case of e.g. concept evaluation or problem solving sometimes one single or very few simulations are needed in order to give enough information to make decisions. A framework for how to utilize DEM as a concept evaluation tool for design and problem solving of bulk materials handling applications, has been presented by Quist [7]. The work shows that the resolution or quality of the DEM model can be used actively for different purposes. When the objective is to do a quick concept screening a very simple model can be setup in order to give some information regarding basic flow trajectories and so on. Such quick simulations can be setup and simulated within one hour. By working in an iterative manner with the concepts and raising the resolution and quality of the DEM simulations accordingly the probability of succeeding with the development efforts is greatly enhanced. 3.2 Bonded Particle Model Rock Population A rock material consists of a number of different minerals and crystalline structures with different mechanical properties. When considering the properties of a rock type the proportion of the various minerals is subject to analysis commonly by doing a petrographic analysis. The petrographic composition of the rock material in Kållered can be seen in Table 2. An example of the microstructure of granite rock can be seen in Figure 14. As can be seen it is constituted by 21 a number of different minerals. The mechanical property of the rock mass depends on the properties of each constituent, proportion, the grain architecture and size as well as weathering effects, cracks and defects. Figure 14 - Illustration of the heterogeneous microstructure of a typical granite rock Table 2 - Petrographic composition of the rock material in Kållered Fraction (mm) Quantity Proportion (%) Meas. Uncert. (±%) Mineral type 2-4 171 17 2.3 Quartz 457 46 3.1 Feldspar 117 12 2.0 Mica (Biotite) 187 19 2.4 Amphibolite 67 7 1.5 Pot. alkali-reactive material 1 0.1 0.2 Ore-minerals incl. sulphides 3.2.1 Generating a bi-modal particle packing cluster Different types of size distributions give varying packing performance as well as number of contact points as illustrated in Figure 15. Particles can be arranged in a number of different bravais lattice systems; [30]  Simple cubic (SC)  Face-centred cubic (FCC)  Body centred cubic (BCC)  Hexagonal closed packing (HCP) These packing structures mainly apply to crystalline structures made up of mono-sized or double mono size particle structures. The arrangement of the particles or atoms, together with the nature of bonding forces characterizes many of the mechanical properties of a material. When building a synthetic rock in the DEM environment the ambition is to capture as many of the features of real rock material as possible. If there was no computational constraint, one would try to model every atom, molecule or mineral grain. However, currently there is a trade- off between the number of meta-particles we want to model and how many particles we put in each meta-particle. Most of the work and simulations done on rock breakage using bonded particle models focus on the breakage of a single particle in e.g. a uniaxial strength test. In this case it is possible to capture a fairly accurate breakage mechanism using all the available particles for one rock specimen. This approach is of course irrelevant when trying to create a rock population for crushing. 22 Figure 15 - Schematic illustration of three different packing structures given by different types of size distributions. When applying a bonding model to these clusters the contact lines seen in the illustration would become boding beams. Hence the strength characteristics of a particle built up by a packed set of spheres strongly depends on the packing structure and size distribution. In this work it was found that the most suitable approach to model the breakage of rock particles is to use a bi-modal distribution with relatively large particles in the high end with smaller particles acting as cement in between, as demonstrated in the illustration to the right in Figure 15. An example of the contact network generated from a particle bed with bimodal distribution can be seen in Figure 16. When activating the bonding function in the simulation these contacts are converted to bonds. Figure 16 - Contact network generated by a particle bed with bimodal distribution. The colours represent the length of the contact vector and show that the body is highly heterogeneous. The following procedure has been developed for creating a particle packing cluster with a bimodal distribution suitable for a BPM in EDEM; i. Create two cylinders with appropriate diameter to fit the wanted end particle size. One should be the container and the other the particle factory. The container cylinder should be placed around origo so that when a 3D particle geometry later is imported it will be fully surrounded by particles. ii. Define a material with low static friction and a higher stiffness then used later. iii. Define a spherical particle named Fraction with a nominal particle size between the coarse and fine modals of the distribution. The contact radius should be set slightly higher than the physical radius. iv. Define a particle factory for the coarse end of the distribution with a capped normal distribution with e.g. µ=2 and capped in the range 1<µ<3. Set the time stamp to tstart=0s. v. Define a particle factory for the fine end of the distribution with a capped normal distribution with e.g. µ=0.8 and capped in the range 0.6<µ<1. Set the time stamp to tstart=tstep. ii: Gaussian distributioni: Mono distribution iii: Bimodal distribution - - -- - - - - - - - - - - - - - - - - - - - - - - - - -- - - - - - - - - - - - - - - - - - -- - -- - - - - - -- - - - --- - - - - - - --- - -- - - - -- - - - -- - - - - - - - - - - - - - - - 23 vi. Let the particles settle forming a loosely packed bed see Figure 17. Due to a higher stiffness the overlaps will be reduced compared to if using the actual stiffness later. By doing so the risk of a preloaded bed is lowered. vii. Define a selection space by importing rock shaped geometry, see Figure 17. viii. Export the particle positions (X,Y,Z) and radius for all particles within the selection space ix. Reorganize the exported data in the following way; 271 0.0165135 0.0097095 -0.00677124 2.418 0.00664328 -0.0377409 0.00264027 2.123 -0.0288484 0.00226568 -0.00600659 1.594 … The first position is how many particles the cluster contains. The first, second and third rows are X, Y, Z coordinates and the fourth is the scaling factor. Figure 17 - The picture to the left show a bimodal particle bed created in step (vi) above. The middle picture shows the 3D rock selection space imported in step (vii). The picture to the right shows the selection of particles within the selection space as done in step (viii) 3.2.2 Calibration of the bonded particle model As mentioned in previous chapters rock breakage experiments are often conducted on primitive shapes such as a cylinder in the uniaxial strength test due to the possibility to calculate the internal stress state. In this work single particle breakage tests have been conducted on a set of rock particles from the test material sample. The critical force for failure and the rock size was recorded. The particle strength given by Eq. 19 was applied in order to find a strength distribution. The calibration work performed in this work will be presented in detail in the Material Model Development chapter. 3.2.3 Introducing meta-particles by particle replacement In EDEM particles are generated to the simulation environment by using particle factories. Usually geometry such as a box, cylinder or a plane is defined as a particle factory and the user can define what particles should be created at what rate. This approach is practical if the purpose of the simulation is to e.g. continuously generate material to a conveyor or create 100’000 particles at once in a mill. However this way of introducing particles is not sufficient when working with multiple dynamic BPM models. It is possible to define custom factories in EDEM. In this work a special approach is used for creating the meta-particles. First a set of dummy particles is created for each meta-particle size class using standard box geometry as factory. These particles are single spheres and have to be larger than the meta-particle cluster. When a set of dummy particles has been generated each 24 dummy particle is used as a custom factory. A custom factory, called Particle Replacement Factory creates fraction particles according to the coordinates and sizes defined in the meta- particle cluster coordinate file. Fraction particles are placed inside the dummy particles according to the local coordinate system of the dummy particle. This is why it has to be larger than the cluster. When the fraction particles are in place, the dummy particle is removed. An example of this procedure can be seen in Figure 18. Figure 18 - Snapshot from EDEM showing a stage in the particle replacement procedure. In the picture a set of meta- particles has been created and a new set of large dummy particles can be seen for the next replacement action. 3.3 Industrial scale crusher experiments Industrial scale experiments have been conducted at a quarry owned by Jehander Sand & Grus AB located in Kållered, south of Göteborg. The site has several cone crushers in process for both secondary and tertiary operations. The secondary crusher, a Svedala H6000 cone crusher, was chosen for the experiments. The ambition with the tests has been to fully capture all possible data concerning both the feed material, machine operation and product material. When conducting tests on a secondary crusher operation normally it is problematic to sample the feed material and do sieve analysis due to the large sized rocks. Rock particles are up to 250 mm in size with a considerable mass for each rock. This has consequences on the statistical significance as a minimum number of particles should be sampled for each size class. If following the recommendations in European standards several tons of feed materials need to be sampled. This is not feasible hence as much material as possible has been sampled and sieved. While digging off material from the belt is a relatively simple task, sizing the rocks bigger then 45mm is cumbersome. It is very rare with mechanical sieves with aperture size larger than 90mm. The lab on site has a vibrating sieve with largest aperture of 45 mm. In order to size particles larger than this a set of square sizing-frames was designed and manufactured, see Figure 19. 25 Figure 19 - Feed sizing with sizing-frames designed in the project In Table 3 a test plan for the full scale experiments in Kållered can be seen. Five different runs have been performed at CSS ranging from 34 to 50 mm. Belt cuts are extracted from the product belt for each run and the feed sampled for the first, third and fifth run. Table 3 – Industrial scale experiment test plan SVULLO EXPERIMENTS Time frame with feed cut Run CSS Samples D. Time [min] Tot. Time [min] Activity Quist Activity Åberg Time Up-Time D. Time RUN1-34 34 F+P 12,5 22,5 1 Set CSS 0,5 0,5 RUN2-38 38 P 10 15 2 Lead CSS calibration 5 5 RUN3-42 42 F+P 12,5 22,5 4 Start DAQ Measurement 1 1 RUN4-46 46 P 10 15 5 Run Crusher 3 3 RUN5-50 50 F+P 12,5 22,5 6 Make time note 0,5 0,5 57,5 97,5 7 Stop DAQ Measurment Stop Belts 1 1 8 Lock OFF belts 0,5 0,5 Samples Expected weight Item Quantity 9 Do belt cut (Feed) Do belt cut (Product) 10 10 S1-34-P 40 Sampling Equipment 10 Lock ON 0,5 0,5 S2-38-P 40 Buckets (20l) 30 11 Start belts 0,5 0,5 S3-43-P 40 Sack 10 22,5 10 12,5 S4-46-P 40 Brush 1 S5-50-P 40 Shovel 1 S1-34-F 40 Spade 2 Time frame without feed cut S3-43-F 40 Tape 2 Activity Quist Activity Åberg Time Up-Time D. Time S5-50-F 40 Tape measure 2 1 Set CSS 0,5 0,5 320 kg Scale 1 4 Start DAQ Measurement 1 1 Sampling Processing Equipment 5 Run Crusher 3 3 Oven 1 6 Make time note 0,5 0,5 Sieve Shaker 1 7 Stop DAQ Measurment Stop Belts 1 1 Corse Sieves 6 8 Lock OFF belts 0,5 0,5 Shape Index meter 1 9 Do belt cut (Product) Do belt cut (Product) 7,5 7,5 Scale 1 10 Lock ON 0,5 0,5 11 Start belts 0,5 0,5 15 5 10 26 A test-sequence was created in order to manage the experiment. This was done due to several reasons such as personal safety; minimize the risk of data and sample loss; quality of samples and time management. Before the first test the process was operated until it reached a steady condition. Then a dry run was performed in order to test each action. The tests followed the following sequence of actions; 1. Set CSS 2. Start feeding 3. Run until choked condition 4. Start data logging and run for 3 minutes 5. Stop incoming feed 6. Stop data logging 7. Stop conveyors 8. Perform lock-out on conveyors 9. Do belt cut 10. Rendezvous at station and lock on All product samples were handled in plastic buckets with handles and lids that prevent moisture from escaping, see Figure 20. Each bucket was weighed after the experiments as a control measure and as a reference for moisture content. The feed samples were handled in tough reinforced polymer bags due to the large sized rock particles. Figure 20 - All the material sampled during the experiments placed in the lab before sample processing. The product samples have been processed in accordance with European standard EN933-1. First each sample was sieved using the large vibrating sieve with an 8mm bottom deck and 63 mm top deck. Each sample was hence split at 8 mm. The large size fraction was simply weighed due to the low amount of moisture in the large size fraction. The minus 8 mm material was split down to 2+2 kg and dried for 2 hours. Each 2 kg sample was then sieved in a conventional cylindrical vibrating screen in order to retrieve the total size distribution from 63 µm to 63 mm. One of the product samples after the coarse sieving can be seen in Figure 21. 27 Figure 21 - Picture showing each size class during coarse sieve processing as well as the minus 8 mm material. In Figure 22 one of the feed samples can be seen. All rocks larger than 45 mm have been individually tested in the sieve-frames and put in the corresponding box. The picture also gives an indication of the size distribution of the feed. Figure 22 - The picture shows each size class from the manual sieve analysis of one of the feed samples. 3.4 Crusher geometry modelling One of the most difficult obstacles to overcome when trying to simulate and replicate the behaviour of a real crusher is to create a good geometrical model. The easy method is to use nominal CAD geometry. However these geometries do not take wear or liner design changes into consideration. Even if it is known what type of mantle and concave should be installed it is very difficult to know for sure when looking at the liners in operation. Also it may be very difficult to get hold of the CAD geometry for each specific liner profile. In this project this was solved by 3D-scanning both the mantle and the concave two weeks after the experiments had been performed. The scanner used is a FARO FOCUS3D laser scanner provided and owned by Roctim AB. The ambition was to perform the scanning inside the plant workshop in a controlled environment. However due to operational issues on site, the liners were never moved. Hence the scanning was performed outdoors without possibility to arrange the liners in a suitable way, see Figure 23. 28 Figure 23 - Left: test scan of a mantle in the mechanical workshop. Right: position of the concave and top frame when scanning. The concave was positioned in a slope hence the scanning procedure was problematic. In Figure 24 a planar view of the 3D scan of the concave is shown. The scanner was placed inside the mantle at two positions in order to capture the full concave geometry. Due to the position on the ground it was difficult to get a high quality scan. If the concave would have been placed inside the workshop on a support structure it would have been in level and possible to clean before scanning. Figure 24 - Snapshot from the 3D-scanning post-processing software showing the unwrapped model of the concave and spiderarms with a color map applied to it. Ideally when scanning a mantle it should be positioned as seen in Figure 23. However the mantle of interest had to be scanned on its position after maintenance hence only a section was captured as shown in Figure 25. 29 Figure 25 - Snapshot of the mantle from the 3D-scan post-processing software. Since it was difficult to capture the full mantle and concave geometries an alternative approach was used for creating a representative liner profile. From both the mantle and concave scan data a set of section samples was extracted and imported to CatiaV5. By drawing spline curves on these sections and finding a best mean a representative profile has been found. The final mantle and concave surfaces were generated by revolving the spline profile around the centre axis. 3.5 Crusher data acquisition A data acquisition system has been developed for sampling data at high frequency from the available crusher sensors. Pressure, power draw, shaft position and temperature signals have be sampled by using opto-isolators splitting the signal from the installed crusher control system. In this work only the pressure and power draw signals have been analysed. The motive behind using a secondary data acquisition system instead of extracting data from the installed control system is based on the suspicion of signal aliasing. The installed system samples data at 10 Hz which is a too slow frequency to capture the true nature of the signals as will be shown in the next chapter. Figure 26 - NI USB-6211 data acquisition card 30 A multifunctional data acquisition card (model: NI USB-6211) from National Instruments was used for sampling the signals, see Figure 26. The card is connected via USB 2.0 interface to a laptop with the NI software LabVIEW. A simple program was developed using block programming language. The program is based on three functionalities;  Data capturing and conversion – A function is setup to acquire the signals from the DAQ card and make them available for the program. Then the signals are separated and converted/calibrated from 1-10 V to the correct unit. The calibration factors are based on sensor specific parameters.  Data logging – The calibrated signals are coupled to a logging function that, when enabled, continuously writes data to a log file until disabled.  Graphical interface – In order to enable online monitoring of the crusher signals a simple interface was designed. The interface also contains fields for setting the scaling parameters for each signal as well as a data log trigger button. The graphical interface can be seen in Figure 27. Even though the DAQ system design was relatively straight forward there were a number of practical difficulties that had to be solved before the system operated as anticipated. Figure 27 – top: LabVIEW graphical interface with functions for displaying and logging data. Bottom: Data acquisition system setup at the crusher control room 31 4. CRUSHER EXPERIMENT The aim of the following section is to present the results from the industrial scale experiments performed in the project. The data shown will be presented and commented on independently from the simulation results. 4.1 Size reduction The product particle size distribution for the five different tests as well as the feed particle size distributions can be seen in Figure 28. As anticipated the product gets finer when reducing the gap setting apart from the CSS42 sample that deviates from expectations. The reason for this deviation is unknown but could be either related to a mistake in the sampling, sampling processing or the post processing. It could also be due to stochastic variation in the feed. As can be seen the feed samples differ relatively much in the CSS42 feed sample which could also be the reason for the deviation. As previously mentioned a very large feed sample is preferred in order to achieve statistical significance, hence the three different samples have been combined as a representation of the total feed sample. Figure 28 - Feed and product particle size distributions for the five different CSS settings. The throughput capacity for the five different tests can be seen in Figure 29. The data collected during a previous survey on the same process is also shown as a reference. Both data sets suggest a non-linear relationship between close side setting and capacity. An interesting feature of the curve shape is the mid peak at 42 mm for the current tests and 44 mm for the old survey. The 2 mm difference may be due to difference in feed material or a gap calibration deviation. A possible explanation to the raising trend for higher CSS is that the cross-sectional area at the choke level gets slightly larger when increasing the gap setting. 0,0% 10,0% 20,0% 30,0% 40,0% 50,0% 60,0% 70,0% 80,0% 90,0% 100,0% 0,1 1 10 100 C u m u la ti v e P a s s in g [ % ] Aperture Size [mm] CSS-34-P CSS-38-P CSS-42-P CSS-46-P CSS-50-P FEED TOT CSS-34-Fd CSS-42-Fd CSS-50-Fd 32 Figure 29 - Capacity for the five different CSS settings. Also capacity data from a previous survey conducted on the same machine is shown for reference. Even though the particle size distribution plot shows that the material is finer for lower CSS it is more easily displayed by looking at the reduction ratio, see Figure 30. The reduction ratio is defined as the 50th percentile for the feed divided by the 50th percentile for the product. For example the F50 equals 67 mm and P50 for CSS at 34 mm is 19.5 mm which gives a reduction ratio of 3.44. The data shows a strong negative linear trend when increasing CSS. The correlation coefficient value is relatively high even though the CSS 42 mm deviates from the trend as described previously. If combining the insights from both the capacity and reduction ratio plots we can see that for lower CSS the rock material is subjected to more crushing to the expense of lower throughput capacity. Figure 30 - The reduction ratio for the five different CSS settings showing a negative trend when increasing the CSS. 4.2 Power draw The power draw signal has been sampled at 500 Hz by using the developed DAQ system. The sampled signal during 120 seconds of operation for the five tests can be seen in Figure 31. The 0,00 50,00 100,00 150,00 200,00 250,00 300,00 350,00 400,00 450,00 500,00 34 36 38 40 42 44 46 48 50 52 C a p a c it y [ T /h ] Close Side Setting [mm] Current Survey Previous Survey y = -0,0806x + 6,2385 R² = 0,9458 0 0,5 1 1,5 2 2,5 3 3,5 4 32 34 36 38 40 42 44 46 48 50 52 R e d u c ti o n r a ti o [ -] CSS [mm] Reduction ratio F50/P50 vs. CSS 33 signal amplitude is very high for all tests which normally indicate poor operation. The initial ambition was to sample the power draw signal from the plant control system, however this data was lost. When the test was conducted the power draw signal displayed by the plant control system did not show this large amplitude. It is strongly suspected that the sampling frequency of the control system is too low and that signal filtering is applied in such a way that the peaks are effectively removed. Figure 31 - Power signals for five different CSS over two minutes of operation. Even though the amplitude is very high the linear trend lines show a distinct difference in mean power draw. As the plot in Figure 31 is very compact it is difficult to see what is actually going on. However the ambition is to show the clear difference in the average power draw when applying a linear regression line on each signal. Due to the large number of data points it is impossible to see what the signal looks like in detail. In Figure 32 the power draw signal for one second of operation is shown. Here it is clearly seen how the signal fluctuates at a specific frequency. The frequencies of the fluctuations are approximately 5 Hz which is the same as the mantle eccentric speed. This means that the variation is somehow related to the movement of the mantle. Recall from the theory chapter that the feeding of material is a vital aspect of a crusher operation. The probable cause of the fluctuations observed is hence miss-aligned feed and segregation. At the 34 peak angular position there is probably both a larger amount of material as well as a finer feed size distribution that requires more energy to be broken. Figure 32 - Pressure signal measured during one second showing the highly fluctuating signal for all CSS settings. The specific time span chosen for each test is randomly picked. When operating any type of process it is of the essence to run it under statistical process control [29]. This generally means that variation from both stochastic as well as systematic sources should be limited. When the variation is suppressed the challenge is to keep the process stable and hence predictable. If the process is stable and predictable then it is possible to control it. The standard deviation of the power draw signal can be seen in Figure 33. The lowest variation can be seen for CSS 38mm. Figure 33 - The standard deviation of the pressure signal is shown for the five CSS settings. As previously mentioned and also seen in Figure 31 the signal has distinct average values. In Figure 34 the average power draw can be seen. The linear trend is very strong with a correlation coefficient of 0.9837. The data indicates that the crusher is working harder i.e. putting more energy into the rock bed per unit time, at lower CSS values. This also aligns with the previous 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 180 190 200 210 0 0,2 0,4 0,6 0,8 1 P o w e r d ra w [ k W ] Time [s] Power draw during 1 second for 5 CSS Power_CSS34 Power_CSS38 Power_CSS42 Power_CSS46 Power_CSS50 0 5 10 15 20 25 30 35 40 32 34 36 38 40 42 44 46 48 50 52 P o w e r d ra w S ta . D e v . [k W ] CSS [mm] Power draw Std vs. CSS 35 results regarding higher reduction ratio at lower CSS as the electrical energy in the motor, via kinetic energy, is transformed to surface energy when the rocks break. Figure 34 - The average power draw measured for the five CSS settings displaying a negative trend when increasing the close side setting 4.3 Hydrostatic pressure The pressure signals sampled at 500 Hz for 120 seconds are plotted in Figure 36. The same argument as in the case with the power draw signals can be made for the pressure signal. The signal has very high amplitude and appears to be noisy. However as previously described the amplitude is not a result of noise but due to cyclic behaviour in the crusher and super positioned discrete crushing events. A detailed view of two of the signals for one second of operation can be seen in Figure 35. The data reveals that the frequency of the variation matches the eccentric speed of the crusher. Hence this provides yet further evidence that the crusher is not fed in a satisfactory manner or there may be uneven wear on the concave. Figure 35 - In order to get a closer look at the measured pressure signal one second of operation is displayed in the plot for two of the CSS settings. The time span between each peak corresponds to the eccentric speed of the mantle. y = -1,5462x + 181,82 R² = 0,9837 0 20 40 60 80 100 120 140 32 34 36 38 40 42 44 46 48 50 52 A v e ra g e P o w e r D ra w [ k W ] CSS [mm] Average Power draw vs. CSS 0 0,5 1 1,5 2 2,5 3 3,5 0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1 P re s s u re [ M P a ] Time [s] Pressure vs. time for two CSS Pressure_34 Pressure_50 36 Figure 36 - Pressure signals for five different CSS over two minutes of operation. Even though the amplitude is very high the linear trend lines show a distinct difference in mean pressure. The standard deviation of the pressure signal can be seen in Figure 37. It is a bit difficult to draw conclusions regarding the shape of the curve. However assuming it is representative, it shows that the variation is slightly higher when pushing the crusher at lower CSS values. This may indicate that the small gap setting is more sensitive to segregation resulting in a larger variation. Hence if the ambition is to run a crusher at small setting in order to have a high reduction ratio, the feeding conditions are very important to consider in order to run the operation under statistical process control. 37 Figure 37 - The standard deviation of the measured pressure signal for the five CSS settings. As in the case of the power draw the pressure shows a strong decreasing linear trend for increasing CSS, see Figure 38. The correlation coefficient is remarkably high at 0.9956 which indicates a successful experiment. An interesting detail is that there is no deviation from the trend for CSS 42 mm as in the case of the product size distribution and the reduction ratio. Figure 38 - The average pressure for the five CSS settings showing a negative trend for increasing CSS. The very strong linear correlation should be noted. 0 0,2 0,4 0,6 0,8 1 32 34 36 38 40 42 44 46 48 50 52 P re s s u re S td . [M P a ] CSS [mm] Pressure Std vs. CSS y = -0,029x + 2,9453 R² = 0,9956 0 0,4 0,8 1,2 1,6 2 2,4 32 34 36 38 40 42 44 46 48 50 52 A v e ra g e P re s s u re [ M P a ] CSS [mm] Average pressure vs. CSS 38 5. ROCK MATERIAL MODEL DEVELOPMENT The aim of this section is to: - Describe the process of building a rock material model in DEM - Present the experiments and simulations performed in order to calibrate the rock model Five different 3D scanned rock geometries with random shapes have been chosen as models for the meta-particles. When building a rock population in the DEM environment it is necessary to cut the particle size distribution at some level. Here the smallest size class was chosen to be 31.5 mm. The total population consists of six different types of particles all with different size and shape. During the development of the feed population packing clusters were made for the six different sizes. Unfortunately only the two largest cluster sizes were later used in the DEM simulations due to practical problems with the particle replacement procedure. A substantial amount of time was spent finding the optimum bi-modal size distribution for each size class. When designing the meta-particles a few aspects need to be balanced. Generally the quality of the cluster stands against the computational load a meta-particle will place on the overall system. High quality is generally achieved by having a high number of small fraction particles. This leads to an optimization problem as a high number of particles in each meta- particle reduce the total number of meta-particles possible to use in the total feed population. A number of different particle beds have been tested iteratively in order to find a suitable bi- modal size distribution that complies both with breakage quality whilst minimizing the number of particles. The key finding/idea that led to a breakthrough during these development efforts was to use relatively large top size in the bi-modal distribution. By having a large gap between the large and small particles a high packing density can be obtained while keeping the number of particles down. Also it results in a bonding structure that is complex and gives heterogenic behaviour. It is important to emphasize that the ambition with the BPM model is not to capture the intricate complexity of rock mechanics but more to build a rock particle that responds with similar behaviour when interacting with the crusher. As long as the particle fractures at similar force levels and breaks apart in progenies it is expected that it can be used for the stated purposes in the thesis. 5.1 Single particle breakage experiment In order to calibrate the BPM model a large number of Single Particle Breakage (SPB) tests have been conducted on the sample material. In each test the particle is weighed and placed in the hydraulic compression test rig. The particle is then compressed 10% which normally is enough to break the rock. Particles commonly break in two different ways. When loaded a stress is built up in the particle body until reaching a high enough level to cause crack propagation through the particle in a single main fracture. The other case can be characterized by sequential fracture where e.g. a sharp corner is chipped off and the particle is reloaded with subsequent local damage as a result. The breakage mode is mostly a matter of loading condition as previously described in Table 1. Prior to breakage tests all particles have been weighed in order to test the hypothesis of a correlation between particle strength and particle mass. The LabVIEW software used for logging the data from the compression rig automatically registers the distance between the compacting plates when experiencing 50 N of load. This gives an estimate of the particle size. The particles are then semi-continuously, as a manual hydraulic pump is used, loaded until broken or reaching 10% compression. From the force-displacement curves the critical force for the main breakage can be logged. The critical force values have been plotted against rock height in Figure 40 and against rock mass in Figure 41. The graphs show that the strength is highly scattered in both plots. Some potential reasons for this are listed below: 39  Stochastic orientation of particles  Anisotropic structure and strength  Particle shape difference and hence variation in loading condition which means totally different stress fields within the rock body.  Stochastic variation in the rock strength  Variation in the compression movement due to manual hydraulic pump  Variation due to robustness problems with the compression rig frame Figure 39 shows the force displacement curves for 7 of the 116 performed SPB tests. As can be seen there is a staircase pattern due to the use of a manual hydraulic pump. Ideally these tests should be performed using a compressive rig with a motorized hydraulic pump where the rate can be controlled. As can be seen in the plot the force is built up until the particle breaks. The peak force at which breakage occurs is defined as the critical force. Figure 39 - The figure shows the force-displacement plot for 7 of the 116 performed SPB tests. The staircase pattern seen is related to the manual hydraulic pumping motion. Figure 40 - Scatter plot of critical breakage force vs. rock height. The height is defined as the distance between the plates when the rock is loaded by 50 N. 0 2000 4000 6000 8000 10000 12000 0% 1% 2% 3% 4% 5% 6% 7% 8% 9% 10% F o rc e [ N ] Compression ratio [%] SPB experiments - Force vs. Displacement S1 S2 S3 S4 S5 S6 S7 R² = 0,3154 y = -7,6805x2 + 1119,8x - 13076 R² = 0,5278 y = 18,817x2 - 339,69x + 6793,1 R² = 0,5609 0 10000 20000 30000 40000 50000 60000 70000 10 15 20 25 30 35 40 45 50 55 60 F c [N ] Rock height [mm] Peak Force - Rock Height Granite Amphibolite Linear (Granite) Poly. (Granite) Linear (Amphibolite) Poly. (Amphibolite) 40 Figure 41 - Scatter plot of critical breakage force vs. rock mass. Each rock specimen was weighed before being crushed. Eq. 19 has been used to calculate the approximate rock strength of all rock samples. The average strength was calculated to 13.29 MPa. 5.2 BPM calibration The general objective with the calibration is to find the bond parameters that give relevant breakage behaviour similar to the experimental results. The general process for creating meta- particle clusters has been previously described in the Methods chapter. When beginning to put together different cluster particle structures it was found that it is not feasible to use the same fraction size distribution for all the size classes. The biggest size class particle can be fitted with relatively large fraction particles, in fact bigger than the smallest size class. Hence two different bi-modal fraction particle beds were created, one fine and one coarse. This was done in order to minimize the number of particles used for each meta-particle. Onwards the two bed configurations are denoted as BPMfine and BPMcoarse. 5.2.1 Calibration simulations of BPMfine In the case of the BPMfine the meta-particle representing the smallest size class was chosen due to its close resemblance in shape and size to the particles tested in the single particle breakage experiments. A 3D rock shape was randomly chosen as subject of breakage, see Figure 42. The strength parameters simulated for the BPMfine calibration can be seen in Table 4. The start values have been calculated based on previous work done by Potyondy and Cundall [17]. Table 4 - Strength parameters simulated for the calibration of BPMfine Run Normal stiffness [GN/m 3 ] Shear Stiffness [GN/m 3 ] Normal Critical Stress [MPa] Shear Critical stress [MPa] 1 2500 1000 10 5 2 2500 1000 10 12,5 3 2500 1000 10 20 4 2500 1000 45 5 5 2500 1000 45 12,5 6 2500 1000 45 20 7 2500 1000 80 5 8 2500 1000 80 12,5 9 2500 1000 80 20 R² = 0,2546 y = 1612x0,4437 R² = 0,328 R² = 0,5039 y = 651,86x0,6327 R² = 0,5587 0 10000 20000 30000 40000 50000 60000 70000 0 100 200 300 400 500 600 F c [N ] Rock mass [g] Peak Force - Rock mass Granite Amphibolite Linear (Granite) Power (Granite) Linear (Amphibolite) Power (Amphibolite) 41 Figure 42 - Pictures of the 3