On-line Capacitive Moisture Measurement of Iron Ore Concentrate Construction of an on-line, contactless, prototype sensing sys- tem for industrial applications Bachelor’s Thesis at the Department of Electrical Engineering Kevin Bäckstäde, Pontus Liljeberg, Måns Lundberg, Ramazan Murtazaliev, Nils Wastegård, Axel Widlund DEPARTMENT OF ELECTRICAL ENGINEERING CHALMERS UNIVERSITY OF TECHNOLOGY Gothenburg, Sweden 2025 www.chalmers.se www.chalmers.se Bachelor’s thesis 2025 On-line Capacitive Moisture Measurement of Iron Ore Concentrate Construction of an on-line, contactless, prototype sensing system for industrial applications Kevin Bäckstäde Måns Lundberg Pontus Liljeberg Ramazan Murtazaliev Nils Wastegård Axel Widlund Department of Electrical Engineering Chalmers University of Technology Gothenburg, Sweden 2025 On-line Capacitive Moisture Measurement of Iron Ore Concentrate. KEVIN BÄCKSTÄDE PONTUS LILJEBERG MÅNS LUNDBERG RAMAZAN MURTAZALIEV NILS WASTEGÅRD AXEL WIDLUND © Kevin Bäckstäde, Pontus Liljeberg, Måns Lundberg, Ra- mazan Murtazaliev, Nils Wastegård, Axel Widlund , 2025. Supervisor: Viktor Lindström, Systems and Control, Electrical Engineering Examiner: Torsten Wik, Systems and Control, Electrical Engineering Bachelor’s Thesis 2025 Department of Electrical Engineering Division of Systems and Control Chalmers University of Technology SE-412 96 Gothenburg Telephone +46 31 772 1000 Cover: Picture of small-scale rotating conveyor with iron ore concentrate Typeset in LATEX, template by Kyriaki Antoniadou-Plytaria Printed by Chalmers Reproservice Gothenburg, Sweden 2025 iv Abstract Accurate moisture determination of iron ore concentrate, also referred to as ore con- centrate, is important to the production of iron ore pellets. Existing methods have long response times, sampling variability and/or require manual work. Capacitive methods for moisture determination show promising signs of solving several of these problems. To determine the feasibility of constructing an on-line measurement sensor utilising capacitive methods, a small-scale prototype consisting of a rotary conveyor dish, a capacitive sensor and measurement electronics was constructed. The sensor consists of two pairs of electrodes, each of which constitutes a capacitor with the surrounding material as dielectric. By placing the sensor below the rotational dish, allowing the measured iron ore concentrate to compose most of the dielectric medium, its per- mittivity, which will vary with its moisture content, can be measured as a change in capacitance. Testing was performed to determine the effects of varying frequency, mass and speed of the measured material. The results show that even though there seems to be a strong correlation between the measured capacitance and the moisture content, it is hard to draw any direct conclusions on the correlation due to high local variability in the material, leading to uneven sampling and poor verification tests. Further tests on a larger scale are suggested to remedy these problems. Keywords: capacitance, contactless, iron ore concentrate, iron ore moisture, LKAB, moisture content, on-line, resonance v Acknowledgements This project would not have been possible without the help and support of several people. Firstly, we want to thank our supervisor Viktor Lindström for his guiding expertise, providing us with equipment and designs, and our examiner Professor Torsten Wik for his patience and guiding words. Further, we want to thank Lars Pettersson, who contributed to the proposal of this project, and John-Erik Rutström at LKAB for their support and help. Without the support and guiding expertise of Rikard Karlsson and Douglas Jutsell Nilsson, both lab managers of the CASE- lab at Chalmers, who provided us with laboratory premises, this project would not have been able to leave the theoretical stage. Lastly, we want to thank our fellow students and friends for acting as a sounding board and elevating this report to a higher standard. Kevin Bäckstäde, Pontus Liljeberg, Måns Lundberg, Ramazan Murtazaliev, Nils Wastegård and Axel Widlund Gothenburg, May 2025 Glossary Agglomeration Process where particles lump together to form larger particles. In this case, the formation of raw iron ore pellets. Attenuation Reduction in the amplitude of a signal. Blast furnace Smelting furnace in the shape of a tower where ore is processed into iron with the help of hot, compressed air, coke and other additives. CDC A device that converts capacitance measurements into digital signals, commonly used in different sensing applications. Chalmers Chalmers University of Technology, Technological university located in Gothenburg, Sweden. Concentrate Fine-grained ore where the concentration of the metal has been increased. Dielectric medium An insulating, or poorly conducting material that can support electrostatic fields. Direct reduction plant Process where iron ore is reduced to iron without melting. Filtrate water Water which has passed through a filter, in this case water removed from a slurry. FMEA A systematic method for identifying potential failure modes in a system. LC-tank An electric circuit consisting of a capacitor and an inductor, which can be used as a resonator, oscillating at a certain frequency decided by its size. LKAB Lousavaara Kirunavaara Aktiebolag, Swedish state-owned mining and refining company. Lot Discrete and defined quantity of a material. On-line measurement Measurements performed automatically. Pellet A small, rounded, compressed mass of a sub- stance. vii Glossary Permittivity The relative permittivity (εr) is a unit-less mea- surement of how much of a material can be po- larized in an electric field. PID controller Proportional-Integral-Derivative controller. A feedback control system used to maintain a de- sired output. It combines proportional (present error), integral (accumulated past error), and derivative (predicted future error) to achieve a stable output. Printed Circuit Board A sheet of layered silicone and etched copper paths designed to connect multiple electronic components into a larger system. R2 value The variance in the dependent variable that is ex- plained by the independent variable in a regres- sion model, with values closer to 1 indicating a better fit. Rotary optical encoder A rotary encoder that uses light to detect position and rotation. Sintering Process of turning a loose material into a solid with the help of heat below the material’s melting temperature. Slurry A mixture of denser solids suspended in liquid, usually water, in this text referring to iron ore suspended in water. viii Contents Glossary vii List of Figures xii List of Tables xiv 1 Introduction 1 1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Problem Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.3 Delimitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.4 Conflict of Interests . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.5 Existing Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.5.1 Gravimetric Methods . . . . . . . . . . . . . . . . . . . . . . . 2 1.5.2 Microwave Methods . . . . . . . . . . . . . . . . . . . . . . . . 3 1.5.3 Optical Methods . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.5.4 Capacitive Methods . . . . . . . . . . . . . . . . . . . . . . . . 3 1.6 Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 2 Methodology and Theoretical Framework 5 2.1 Theoretical Basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 2.1.1 Basics of Capacitors . . . . . . . . . . . . . . . . . . . . . . . 5 2.1.2 Dielectrics and Permittivity . . . . . . . . . . . . . . . . . . . 5 2.1.3 Resonator Circuits . . . . . . . . . . . . . . . . . . . . . . . . 6 2.2 Sensor Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 2.2.1 Electrodes Construction . . . . . . . . . . . . . . . . . . . . . 7 2.2.2 Conversion Device Construction . . . . . . . . . . . . . . . . . 9 2.2.3 Controller And Interface Construction . . . . . . . . . . . . . 10 2.2.4 Data Logging . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 2.2.5 Evaluation Module Usage . . . . . . . . . . . . . . . . . . . . 11 2.3 Iron Ore Concentrate . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 2.3.1 Moisture Verification . . . . . . . . . . . . . . . . . . . . . . . 12 2.3.2 Sample Quality Assurance . . . . . . . . . . . . . . . . . . . . 13 2.4 Rotary Conveyor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14 2.4.1 Rotary Dish . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 2.4.2 Frame and Sensor Mount . . . . . . . . . . . . . . . . . . . . . 15 ix Contents 2.4.3 Sample Partitioning . . . . . . . . . . . . . . . . . . . . . . . . 16 2.4.4 Drive System . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 2.4.5 FMEA and Safety . . . . . . . . . . . . . . . . . . . . . . . . . 17 2.5 Testing Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 2.5.1 Determination of Measurement Frequency . . . . . . . . . . . 17 2.5.2 Sample Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 2.5.3 Sample Density . . . . . . . . . . . . . . . . . . . . . . . . . . 19 2.5.4 Test Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 2.5.5 Sample moisture . . . . . . . . . . . . . . . . . . . . . . . . . 20 2.5.6 Single Lot Averaging Tests . . . . . . . . . . . . . . . . . . . . 20 2.6 Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 2.6.1 Stationary Analysis . . . . . . . . . . . . . . . . . . . . . . . . 21 2.6.2 Moving Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 21 3 Results 23 3.1 Choice of Measuring Frequency . . . . . . . . . . . . . . . . . . . . . 23 3.2 Influence From Sample Mass . . . . . . . . . . . . . . . . . . . . . . . 23 3.3 Influence From Sample Density . . . . . . . . . . . . . . . . . . . . . 24 3.4 Influence From Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 3.5 Moisture Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 3.6 Nugget-To-Sill Ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 3.7 Repeatability of Results . . . . . . . . . . . . . . . . . . . . . . . . . 28 3.8 Single Lot Averaging Tests . . . . . . . . . . . . . . . . . . . . . . . . 29 4 Discussion 33 4.1 Analysis of Parameter Impact . . . . . . . . . . . . . . . . . . . . . . 33 4.1.1 Measuring Frequency Optimization . . . . . . . . . . . . . . . 33 4.1.2 Effect of Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 4.1.3 Effect of Material Density . . . . . . . . . . . . . . . . . . . . 34 4.1.4 Effect of Speed . . . . . . . . . . . . . . . . . . . . . . . . . . 34 4.1.5 Repeatability of Measurements . . . . . . . . . . . . . . . . . 34 4.1.6 Effect of Moisture . . . . . . . . . . . . . . . . . . . . . . . . . 35 4.1.7 Single Lot Averaging Tests . . . . . . . . . . . . . . . . . . . . 35 4.2 FMEA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 4.2.1 Rotary Conveyor . . . . . . . . . . . . . . . . . . . . . . . . . 36 4.2.2 Sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 4.3 Ethical aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 4.4 Social aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 5 Conclusion 39 5.1 Conclusion on Feasibility . . . . . . . . . . . . . . . . . . . . . . . . . 39 5.2 Conclusion Compared to Existing Methods . . . . . . . . . . . . . . . 39 5.3 Areas for Further Study . . . . . . . . . . . . . . . . . . . . . . . . . 40 Bibliography 40 A FDC2214 Configuration I x Contents B Sensor Mount III C Gearbox Diagram V D Motor Driver Circuit VII E Failure Mode & Effects Analysis (FMEA) IX F Test Moisture Content XI G Moisture Cycles XIII xi Contents xii List of Figures 1 Sensor-System Overview Diagram . . . . . . . . . . . . . . . . . . . . 6 2 Illustration of electrodes cross-section . . . . . . . . . . . . . . . . . 7 3 Illustration of electrodes separation . . . . . . . . . . . . . . . . . . . 8 4 Electrodes Render . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 5 Electrodes Drawing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 6 Iron Ore Concentrate in Container . . . . . . . . . . . . . . . . . . . 12 7 Samples Drying in Oven . . . . . . . . . . . . . . . . . . . . . . . . . 13 8 Iron Ore Concentrate Mixing Device . . . . . . . . . . . . . . . . . . 14 9 Conveyor Dish Drawing . . . . . . . . . . . . . . . . . . . . . . . . . 15 10 Rotary Conveyor Picture . . . . . . . . . . . . . . . . . . . . . . . . . 16 11 Test Setup for Mass Illustration . . . . . . . . . . . . . . . . . . . . . 19 12 Capacitance given Speed Box-Plot, LC-tank 150pF . . . . . . . . . . 25 13 Capacitance given Speed Box-Plot, LC-tank 330pF . . . . . . . . . . 25 14 Capacitance given Moisture Box-Plot . . . . . . . . . . . . . . . . . . 26 15 Nugget-to-Sill Ratio Box-Plot . . . . . . . . . . . . . . . . . . . . . . 27 16 Measurement Repeatability Visualisation . . . . . . . . . . . . . . . . 28 17 Repeatability test for single lot averaging . . . . . . . . . . . . . . . . 29 18 Full conveyor single moisture content tests for lots 1-8 . . . . . . . . . 29 19 Box plot of second set . . . . . . . . . . . . . . . . . . . . . . . . . . 30 20 Average capacitance and verified moisture content of the second set . 30 21 Linear Regression on the Second Set of Tests . . . . . . . . . . . . . . 30 22 Drawing of Sensor Mount . . . . . . . . . . . . . . . . . . . . . . . . III 23 Gearbox Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . V 24 Motor Circuit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VII xiii List of Figures xiv List of Tables 1 Desired Lot Moisture Contents . . . . . . . . . . . . . . . . . . . . . . 11 2 Channel Resolution and Variation given LC-tank . . . . . . . . . . . 23 3 Capacitance given Mass . . . . . . . . . . . . . . . . . . . . . . . . . 23 4 Capacitance given Compactness . . . . . . . . . . . . . . . . . . . . . 24 5 FDC2214 Configuration . . . . . . . . . . . . . . . . . . . . . . . . . I 6 FMEA for Rotary Conveyor . . . . . . . . . . . . . . . . . . . . . . . IX 7 FMEA for Sensor and Electronics . . . . . . . . . . . . . . . . . . . . X 8 Desired Moisture Content Calculation . . . . . . . . . . . . . . . . . . XI 9 Capacitance given Moisture, Multicycle, LC-tank 33pF, Channel A . XIII 10 Capacitance given Moisture, Multicycle, LC-tank 150pF, Channel B . XIII 11 Capacitance given Moisture, Multicycle, LC-tank 330pF, Channel A . XIV 12 Capacitance given Moisture, Multicycle, LC-tank 330pF, Channel B . XIV 13 Capacitance given Velocity and Moisture, LC-tank 33pF, Channel A . XV 14 Capacitance given Velocity and Moisture, LC-tank 150pF, Channel B XVI 15 Capacitance given Velocity and Moisture, LC-tank 330pF, Channel A XVII 16 Capacitance given Velocity and Moisture, LC-tank 330pF, Channel B XVIII xv 1 Introduction 1.1 Background In modern steel production, it is common to use iron ore pellets to improve trans- portability, the chemical composition and process efficiency. Turning raw iron ore into pellets is a multi-step process where mined ore is crushed, sorted, concen- trated and purified. This process is partly aided by water, which results in a slurry consisting of finely ground iron ore and water. In order to form raw pellets, the slurry is dewatered or filtered, resulting in a concentrate which is then agglomerated into balls, raw iron pellets, in rotating dishes or drums. The raw iron pellets are then dried, preheated, and sintered into finished pellets which can be used in both blast furnaces and direct reduction plants. The largest iron ore producer in Europe, Swedish Lousavaara Kirunavaara Aktiebolag (LKAB), relies on this process for most of its production, which is predominantly shipped within Europe, but also to the rest of the world, where it can be processed into steel. The moisture content of the iron ore concentrate is essential in ensuring the cor- rect conditions for agglomeration to form the raw pellets [1], [2]. If the concentrate is too wet, it may not be able to hold the shape of a ball, and if it is too dry, it may fail to agglomerate at all. Further, the energy-intensive process of drying and sintering the pellet emphasises the need to accurately measure the moisture content. A higher water content would lead to higher energy consumption, since more water needs to evaporate when drying the pellets. As discussed further in Section 1.5, the conventional methods of determining moisture content typically have long response times. This causes a delay in the regulation of the moisture filtration process. Im- provements to this method could potentially increase the throughput and energy efficiency of the filtering process by reducing the variability of the moisture content in the concentrate. 1.2 Problem Definition This project aims to answer the following main questions: • What is the feasibility of implementing capacitive iron ore concentrate mois- ture content measurement in real-time? • How does the capacitive moisture measurement system compare to traditional methods? 1 1. Introduction These questions will be answered and evaluated by focusing on three main areas: the design, the measurement signal and the evaluation method of the sensor. 1.3 Delimitations To limit the scope of the project, three main delimitations were set. The main goal of these delimitations was to enable testing in existing facilities without complicated logistics. This project only focused on one concentrate product quality, which was chosen after consultation with LKAB. The reasons for this were several. For example, the additives in the concentrate change with product quality. By limiting the work to the same product quality, focus can be placed on examining moisture content. The choice of product quality also ensured access to concentrate throughout the project. The nominal moisture of the concentrate is 9 % with an assumed possible variation of ±3 %. To enable tests without the need for a climate-controlled environment, this project does not include measurements of surrounding temperature and humidity in any calculations. Instead, they were recorded and will be discussed in the error analysis. No connection to the control system of the processing plant was considered. 1.4 Conflict of Interests This project was performed in cooperation with the mining company LKAB, which provided materials for testing as well as financed some of the electronics used. They also provided consultation on subject matters. One of the project members previ- ously worked at LKAB with the pelletisation process and is currently employed by LKAB but with no direct involvement in subjects of this project. LKAB has not exercised any influence on the scientific process of this project, neither economic nor by other incentives. 1.5 Existing Methods Several methods to determine the moisture content of iron ore concentrate and other materials are already well established in the industry today. The most relevant for this project will be discussed below. 1.5.1 Gravimetric Methods One of the more common methods used to determine the moisture content of a bulk material is the loss-on-drying method, which is the method used in LKAB:s processing plants today. The method involves taking a sample and drying it until no 2 1. Introduction more weight is lost. The moisture content can be calculated by subtracting the dry weight from the initial weight. This method takes several hours to return a result, but is very accurate [3]. It is also difficult to implement as an on-line measurement [4], leading to introduced sampling variability and lower accuracy [5]. The drying can be performed in an oven or a special moisture-content scale with built-in drying methods. 1.5.2 Microwave Methods There are two main methods based on microwave radiation, one uses phase shift and the other uses attenuation to determine the moisture content. When a microwave beam travels through water, a phase shift is caused. This phase shift is typically much larger in water than the material in which the water is entrained. By sending a beam of microwaves through a bulk material and measuring the phase shift, it is possible to infer a correlation between its moisture content and the phase shift, making it usable in continuous measurement applications like conveyor belts. How- ever, using the method for iron ore concentrate has proved challenging mainly due to the high attenuation of the signal [6]. The other method, using attenuation, works by analysing how much a signal is attenuated when passing through a material and, by measuring this, determining the water content. The attenuation and phase shift methods are both considered high investment and are sensitive to the surrounding environment, such as metal in the signal path [4]. 1.5.3 Optical Methods When light is reflected off a material, the absorption depends on the light’s wave- length. Through choosing wavelengths for which water has specific characteristics, it is possible to determine the moisture content by analysing the reflected light. This method has been successfully tested in the industry with a deviation of ±0.3 % [7]. Later tests have shown non-trivial calibration problems, significant variations in variability from varying production, and no better accuracy than current methods. However, the improved sampling rate and response time have been highlighted as advantages. The method also requires some mixing of the iron ore concentrate be- fore the measurement point, since it will only measure on the exposed top layer [5]. 1.5.4 Capacitive Methods Capacitive methods present an alternative approach to moisture measurement that relies on the principle that the permittivity, or dielectric constant, of a porous ma- terial changes significantly with its water content. A capacitive sensor essentially forms a capacitor where the measured material acts as the dielectric medium between two or more electrodes. The capacitance of this arrangement is directly influenced by the permittivity of the material. As the moisture content within the iron ore 3 1. Introduction concentrate increases, the overall permittivity of the material also increases because water has a much higher permittivity than air and the dry iron ore components [8]. Consequently, this change in permittivity leads to a measurable change in the capacitance of the sensor. By establishing a relationship between the measured ca- pacitance and the moisture content through calibration, this method can be used for moisture determination. Capacitive sensors are commonly employed in applications such as soil moisture measurement and liquid level sensing due to their relatively simple design, robustness, and potential for cost-effective implementation [9]. Re- search in Brazil [10] has demonstrated the feasibility of using capacitive methods for measuring moisture content in iron ore concentrate, suggesting its potential as a viable alternative to traditional techniques. 1.6 Scope Given the limited research into using capacitive methods for iron ore concentrate, which shows promising signs of being feasible compared to the other methods, the purpose of this project has been to explore the feasibility of employing capacitive sensing technology for real-time moisture measurement, specifically in iron ore con- centrate. By undertaking the design, construction, and testing of a prototype sensor system, this research aims to determine whether this method can provide reliable and accurate moisture content data that is responsive enough for real-time process control. The successful implementation of such a system could offer significant ad- vantages over existing methods, potentially leading to improved process control and enhanced energy efficiency in the iron ore pelletisation process. The focus will be on evaluating the sensor’s performance under conditions relevant to the industrial environment, considering factors such as the material’s movement, varying moisture content, and potential environmental influences. 4 2 Methodology and Theoretical Framework This chapter describes the methodology applied during the measurements and the theoretical concepts that support it. Given the novel character of the project, the method and theory are integrated to provide a unified and comprehensive overview. 2.1 Theoretical Basis The method of measuring the moisture content of a bulk material with the help of capacitance, as introduced earlier, relies on the following physical principles. 2.1.1 Basics of Capacitors A capacitor is a simple electric component that can store electrical energy as an electrical potential. When two conductors containing movable electric charges are placed with an isolator or dielectric separating them, and an electric potential is applied over them, electrons travel from one side to the other. Electrical energy is thereby stored in the capacitor. The magnitude of the charge is proportional to the potential difference of the two conductors Q = C∆V (1) where Q is the charge in C, C is the capacitance in F and ∆V is the difference in voltage over the two conductors in V. When conductive surfaces are involved, capacitance is related to the area of those surfaces, their distance from each other, and the relative permittivity of the dielectric separating them C = ε0εr A d (2) where C, as above, is capacitance, ε0 is the permittivity of vacuum, εr is the relative permittivity of the dielectric medium between the conductive surfaces, A is the area in m2 and d is the thickness of the dielectric in m [11]. 2.1.2 Dielectrics and Permittivity A dielectric material is typically an insulator that can be polarised by an electric field, although most materials display dielectric properties. The relative permit- tivity of a dielectric, εr, denotes the material’s ability to support an electric field. 5 2. Methodology and Theoretical Framework The permittivity of a material is not fixed but varies slightly with parameters like frequency of the electrical field, temperature and humidity [11]. The relative per- mittivity of air is approximately 1, while that of water is around 80, which makes it possible to determine the content of water in a bulk material by analysing its relative permittivity. This can be done by constructing a capacitor where the dielectric is made up of that bulk material [12]. 2.1.3 Resonator Circuits A circuit consisting of a capacitor and an inductor in parallel, also known as a LC-tank, has a natural resonance frequency determined by: fresonance = 1√ LC (3) Where L is the circuit’s inductance in H and C is the capacitance in F. This relation stems from the nature of how charges are alternating between being stored in the capacitor and the inductor. By measuring the resonance frequency and knowing either one of the circuit’s capacitance or inductance, the other parameter can be deduced [13]. 2.2 Sensor Construction The sensor in its entirety consists of a few different, interconnected parts, illustrated in Figure 1. Figure 1: Sensor-System Overview Diagram • Electrodes - Conductive plates placed near the iron ore concentrate, subject to electromagnetic effects of the nearby environment, including said material’s dielectric properties and proximity. • Conversion Device - Method for converting electromagnetic effects experi- enced on the electrodes into a numerically quantifiable property. 6 2. Methodology and Theoretical Framework • Controller And Interface - Device configuring the conversion device to be in the correct state, handles the logic of reading and potentially transforming conversion results and sending them to where they need to be. • Evaluation Module - A pre-assembled development board with verified func- tionality, useful for configuring and verifying the function of a custom conver- sion device. • Data - Generated stream of data that can be analysed on-line in real-time or stored for later off-line analysis. 2.2.1 Electrodes Construction Two electrodes were made from copper tape applied to acrylic blocks and aligned in a coplanar configuration. This was done to promote electric fields that extend outward radially and enable contactless measurements at a distance. Shielding planes were formed with additional copper tape opposite to the electrode signal plane and on the sides of the acrylic block for each respective electrode, creating a controlled area resistant to electromagnetic noise and capacitive objects in the direction blocked by the shield [14]. An illustration of an electrode pair is available in Figure 2. Figure 2: Illustration of electrodes cross-section The electrodes were each soldered to coaxial cables inserted through a hole in the shield plane, and the shield plane was soldered to the braided shielding of that same coaxial cable. The cables were subsequently fitted with SMA connectors on the opposite ends to make them easy to connect and disconnect as needed. The area of the electrode signal planes is a property that should be maximised to be sensitive to changes in the dielectric properties of the measured material, as follows from Equation 1. The length of the electrode, however, is constrained by the sensing geometry of interest, since electrodes that extend beyond that area would only introduce more undesired environmental variation with little or no increase in 7 2. Methodology and Theoretical Framework sensitivity. For this study, a rotary conveyor, described in Section 2.4, was con- structed as a platform to perform the tests on. A conveyor width of 185 mm was decided, and for some extra margin to the conductive steel ring edges, an electrode length of 120 mm was deemed suitable. The separation of the electrodes can be thought of in terms of the desired measuring depth. The further the electrodes are from each other, the weaker the electric field gets. But, with a larger separation, a higher proportion of the sensed signal comes from within the material. This is demonstrated geometrically in Figure 3. Figure 3: Illustration of electrodes separation Comparing the direct paths A and B with their respective paths going through a point at depth D in the sensing area, we can see that the path going through the material gets more and more equidistant to the direct path the further apart the electrodes are. The interactions of electric fields with porous and slightly conductive materials like iron ore concentrate, however, are very complex. For this reason, the electrode separation was left adjustable to be tinkered with further. In the end, separations of 23 mm and 35 mm were used on two separate channels for all tests. The electrode width was set in relation to the electrode separation according to a ratio T = 1.35g + 0.65s (4) discovered in a previous study [10], where T is signal depth, g is half the separation and s is electrode width. Copper tape easily available for this study was 19 mm wide. For a separation of around 30 mm this would correspond to a signal depth of 32.6 mm, which is reasonable. Finally, the distance from the shield to the electrode signal plane is a property sought to be maximised in order to minimise parasitic capacitances. The thick- est, easily available acrylic during this study was 8 mm thick, which is a significant shield distance, and as such ended up being the distance that was realised in the final design. A render and a drawing of this design are found in Figures 4 and 5. 8 2. Methodology and Theoretical Framework Figure 4: Render of electrodes Figure 5: Drawing of an electrode 2.2.2 Conversion Device Construction As capacitance-digital-converter (CDC), FDC2214 was selected as the conversion device of choice for its highly accurate, resonance circuit conversion method, en- abling resolution of as good as 0.4 fF [15]. However, the FDC2214 outputs neither a resonance frequency nor a capacitance directly; rather, a lower-level property re- ferred to as raw code that comes from counting the number of resonance signal peaks during a fixed number of crystal oscillator cycles. With a known oscillator frequency and known LC-tank properties, however, both resonance frequency (fSENSORx) and capacitance (CSENSORx) can be calculated as 9 2. Methodology and Theoretical Framework fSENSORx = fREF · DATAx · CHx_FIN_SEL CHx_FREF_DIV IDER · 228 (5) CSENSORx = 1 2π(Lx · fSENSORx)2 − Cx (6) with parameters as below: fREF - Crystal oscillator reference frequency DATAx - Raw code Conversion result from channel x CHx_FIN_SEL - Input frequency select setting on channel x CHx_FREF_DIVIDER - Reference frequency divider on channel x Lx - Fixed LC-tank inductance on channel x Cx - Fixed LC-tank capacitance on channel x according to the datasheet of the FDC2214 [15]. A custom PCB developed in a previous work by Viktor Lindström at Chalmers uni- versity of technology (CTH) was utilised for the implementation of the FDC2214. This was done because the design provided extensive configurability, such as swap- pable LC-tanks, integrated passive shielding, and cable connectors, as well as to save time and to expedite initial results. The FDC2214 was configured according to the instructions in the datasheet and tweaked to output highly accurate values in exchange for a slower output rate [15]. For the operating conditions described above, the configuration deemed most suit- able for tests can be found in Appendix A. 2.2.3 Controller And Interface Construction An Arduino MKR Zero was used to control and communicate with the conversion device as well as to provide a user interface for operators. The MKR-Zero is a flexible development board with an inbuilt SD-card reader, which is desired for the system to be able to store data independently, as mentioned in Section 1.2. The interface in the case of this study consisted of buttons for starting and stopping the logging of data, together with some LED diodes and an OLED display to communicate the state and status of the sensor system. 2.2.4 Data Logging A simple protocol for log files with raw code output data of the FDC2214 was developed and contains two parts: One section for constant data in JSON format used to save information such as which LC-tank was used, and another section for a 10 2. Methodology and Theoretical Framework chronological stream of output data with timestamps. With the help of the raw code conversion formulas in Equations 5 and 6, the log data could later be transformed into the preferred format and plotted and analysed in any way necessary. 2.2.5 Evaluation Module Usage The Texas Instruments evaluation module and Sensing Solution software suite were used to ensure the electrodes were functioning well and to verify that changes in the environment, such as the proximity of a hand or conductor, could be detected at a very early stage of development. They also served the purpose of finding the right configuration for a given LC-tank since the software’s graphical user interface provided detailed and comprehensive guidance about any out-of-spec values. 2.3 Iron Ore Concentrate Table 1: Selected moisture contents to test feasibility of measurement precision of 0.03 %, calculation is presented in appendix F Sample Nr. Moisture Content (%) 1 6.00 2 8.19 3 8.78 4 8.94 5 9.00 6 9.03 7 9.11 8 9.41 9 10.55 10 12.00 Sampled iron ore concentrate was delivered by LKAB in 15 containers, see Figure 6 weighing approximately a total of 150 kg to ensure sufficient material for multiple tests. This sampled concentrate is what the project used to test the measuring method and equipment, and its moisture content was decided with the gravimetric method described in 2.3.1. Out of the 15 lots, 10 were designated moisture contents, which lay in the entire testing area according to Table 1. The five remaining lots were left at approximately 8.7 % to be used in tests that required more material. 11 2. Methodology and Theoretical Framework Figure 6: Container with iron ore concentrate sampled from LKAB processing plant By designating moisture contents to different lots and keeping the lots in sealed containers, the need for adjusting moisture is reduced, thus minimising the need for wetting and drying the concentrate. However, since all lots are sampled from the production line, their initial moisture content needs to be adjusted. To ensure representative contents of minerals, filtrate water from the production facility was used to wet concentrate that needed increased moisture content, while concentrate with a lower desired moisture content were either air dried outside their airtight containers or partly dried in an oven and then mixed. The remaining five lots can be used in tests where the effects of movement need to be studied. 2.3.1 Moisture Verification This section describes the method used to verify the moisture content of the mate- rials tested. It is derived from the gravimetric method described in ISO 3087:2011 [3]. To ensure a representative sample for the process, the whole lot is first homogenised as described in the next section. Given the importance of accurate weight measure- ments, a scale with high precision was chosen. The scale, Kern EHA 500-2 with a precision of 0.01 g, has a maximum weighing capacity of 500 g [16]. To ensure a precision of at least 0.01 %, a sample of no less than 100 g of concentrate was added to a pre-weighed, oven-safe beaker as seen in Figure 7. In accordance with the specifications of the ISO standard, the material was then dried for four hours and weighed again, allowing the lost water to be calculated and thus the initial moisture content. The standard states that samples over 8 % should be dried for no less than 24 hours, but two tests were performed where the moisture content of 12 % was determined to be dry after only four hours through weighing after one additional hour and determining that the mass was not dropping further. This deviation, to only dry in the oven for four hours, was determined to be minuscule and not affect 12 2. Methodology and Theoretical Framework Figure 7: Four samples during drying in oven the measurements. To reduce waste of material, the dried samples were put back with the original lot in the containers and the lost mass was readded as deionised water to introduce as few minerals and other materials as possible which could affect measurements. 2.3.2 Sample Quality Assurance To verify the moisture content, a gravimetric analysis (SS-ISO 3087:2011 [3]) was conducted on subsamples using a POL-EKO SLW 75 smart oven and a Kern EHA 500-2 scale. However, deviation from the ISO-standard was made due to the limited lot sizes; according to the standard, it should be at least 1 kg, whereas for these tests, the subsamples were 100 g, as mentioned earlier. Another deviation from the ISO standard was that only one sample was taken from each lot instead of four samples, as in the standard. This was due to the limited lot sizes. The gravimetric analysis was used to verify the lot moisture and be able to set them according to Table 1, but also once for each sample for each test to account for evaporated water between tests. 13 2. Methodology and Theoretical Framework A custom mixer tool was made to homogenise the tests, see Figure 8. This was after realising the concentrate behaves unexpectedly and creates lumps with higher moisture content than the rest of the sample. When weighing the whole sample in the container, we used a Kern DE 60K10D, however, it has an error margin of 10 g, which resulted in some trouble when wetting the material to the predetermined moisture contents [17]. This made the gravimet- ric method test even more vital to ensure that our samples were correct. A source of error to take into consideration is the nugget-to-sill ratio. The nugget ef- fect is a phenomenon that shows when small randomised differences occur in a small sample, while the sill describes the total variability in the sample. The Nugget-to- Sill ratio is the effect of the nugget divided by the effect of the sill. This gives a perception of how homogenised a sample is [5]. Figure 8: The device designed for mixing, crushing and homogenising iron ore concentrate 2.4 Rotary Conveyor A test rig for the sensor prototype in the form of a rotary conveyor was constructed to simulate the conveyor belt in the plant on a small scale. To allow for consistent tests with movement, it is constructed as a rotary dish so that the samples stay the same each time they pass the sensor. This makes it possible to use the same samples multiple times while testing different parameters. 14 2. Methodology and Theoretical Framework 2.4.1 Rotary Dish The dish is made out of 3 mm thick acrylic supported by 2 mm thick steel rings on either side. Acrylic was chosen as an alternative to rubber because it has little impact on capacitive measurements and can easily be cut to shape using a laser cut- ter while also being stiff enough to support the weight of the ore without bending much. If the band bends, the distance to the sensor could vary, introducing a source of error in measurements. At the plant, the conveyor is 2400 mm wide and 15 mm thick. The test rig has a scale of 1:13, limited by the acrylic sheet width available to the project, making the test surface 185 mm wide. Unfortunately, it was not possible to match the scale for the thickness of the belt as it became too unstable with thinner material than 3 mm. A drawing of the dish can be seen in Figure 9. Figure 9: Conveyor Dish Drawing 2.4.2 Frame and Sensor Mount The dish rests on four ball bearings supported by a structure made from 2 mm thick steel parts held together by nylon brackets that were 3D-printed. This structure has adjustable feet to make it level. The sensors are also attached to this structure with an adjustable holder made from PLA plastic, which allows for adjustments both vertically to tune the distance from the acrylic and horizontally to tune the distance between the sensor plates. A drawing of this mount can be seen in Appendix B. All the electronics supporting the sensors and data logging are mounted in a box below them. The sensor configuration and supporting electronics are described in Section 2.2. 15 2. Methodology and Theoretical Framework 2.4.3 Sample Partitioning The circle is divided into eight sections, 45° each. These sections make it possible to run tests with eight different samples at the same time. For the initial tests, this was done by placing the concentrate in 3D-printed boxes made to fit the circle. The boxes were made to be able to easily move the samples into different configurations without having loose iron ore concentrate on the dish. These boxes were later abandoned as the distance to the sensor had a significant impact on the results, and there was uncertainty over whether the PLA plastic in the boxes would absorb too much moisture and cause errors in the results. Later tests were performed with 3D-printed walls that have slots for acrylic sheets to section off the different areas. This also allows for testing samples larger than 45°. These walls can be seen in Figure 10. Figure 10: Picture of the test rig 2.4.4 Drive System On the opposite side of the sensors, a DC motor with a 3D-printed gearbox that drives the rotation of the dish is mounted. The conveyor in the plant has a speed of 0.025 m s−1, which translates to 0.017 rps on the rig if the distance travelled is measured at the centre of the test area. At scale, this is 0.0013 rps. The gearbox has a reduction ratio of 750:1 to be able to reach this low speed. In the gearbox, there is also a rotary optical encoder used to find the current position of the dish as well as the speed. A drawing of this gearbox can be found in Appendix C. A proportional-integral-derivative (PID) controller, together with the encoder read- ings, controls the rotation speed. The motor control electronics and interface are in a separate box connected to the rig. More information about the motor control 16 2. Methodology and Theoretical Framework system can be found in Appendix D. 2.4.5 FMEA and Safety A Failure Mode and Effects Analysis (FMEA) was conducted for both the sensor and the test rig to identify potential failure points that could compromise function- ality or durability. This analysis guided the design phase to prevent future issues. The complete FMEA for the test rig is provided in Table 6 in Appendix E, while the sensor FMEA is detailed in Table 7 in Appendix E. To ensure safety, protective barriers were installed to block access to hazardous areas, such as the gearbox and bearings. An emergency stop was implemented for the motor, and the motor driver circuit was connected to a current-limited power supply, always set below the motor’s maximum rated current to avoid overloading. 2.5 Testing Methodology The key goals of the following tests were to evaluate the sensor’s sensitivity to its parameters and performance under controlled conditions, confirming expected be- haviour and validating the equipment. The chosen system, material, and environment variables that were tested are: LC- tank size, sample mass, sample compactness, sample speed during measurement, and lastly, iron ore concentrate moisture content. 2.5.1 Determination of Measurement Frequency As mentioned in Section 2.1.2, the permittivity changes with the frequency of the electrical field applied. It was also hypothesised that different measurement fre- quencies would be affected differently by noise. To change the frequency of the capacitance-to-digital converter, the LC-tank size can be varied. To determine op- timal frequencies, different sizes were tested to evaluate how the frequency affected the penetration depth as well as other complex frequency-dependent interactions, and therefore the measured output value. Each test used three samples placed 120◦ apart on the rotating dish. Capacitance data was collected during 30 s intervals, alternating between measuring samples and the empty conveyor. Samples with a dry weight of 946 g and moisture contents of 6.00 %, 9.00 %, and 12.00 %, were tested. These moisture contents were chosen to simplify the differen- tiation between samples. To prevent lumping, the lots were homogenised prior to sampling. 17 2. Methodology and Theoretical Framework The measured parameters were the proportional variation in measurement during a selected interval and sensitivity, defined as Sensitivity = Value at 12% − Value at 6% Value at 12% − Baseline (7) where the baseline is the measurement from the sensor with an empty conveyor. The sensitivity expresses the proportion between the range of interest and the full measured range. With higher sensitivity, as defined in Equation 7, the measurement range of the equipment is maximised for the data of interest. To analyse the variation in the measured intervals, we need a way to describe how spread out the data is. Each measurement interval is indexed by i, which refers to a separate sample group in the test. A weighted mean, µw, is calculated using µw = ∑ i(µi · ni)∑ i ni (8) where µi is the average capacitance in interval i, and ni is the number of samples in that interval. This gives an overall mean that takes into account the size of each interval. To find how much the values vary within a single interval, we compute the standard deviation as σi = √√√√ 1 ni ni∑ j=1 (xij − µi)2 (9) where xij is the j-th capacitance measurement in interval i, µi is the average capac- itance in that interval, and ni is again the number of samples. To measure the overall variation across all intervals, the weighted standard deviation is calculated as σw = √√√√∑i ni · (σ2 i + (µi − µw)2)∑ i ni (10) In this expression, σi is the standard deviation of interval i, µi is the mean of interval i, µw is the overall weighted mean, and ni is the number of samples in that interval. This formula combines the spread of data within each interval with the spread of the interval means around the global average, giving a complete picture of variation in the entire dataset. Finally, the coefficient of variation, V %, expresses the weighted standard deviation in relation to the weighted mean is V % = ( σw µw ) · 100 (11) 18 2. Methodology and Theoretical Framework Figure 11: Sample setup for mass testing Ten LC-tanks was selected with varying sizes and were all tested on each channel and evaluated using the parameters above. From the results, two combinations of LC-tanks were chosen. One pair consisted of the two LC-tank-sizes with the highest resolution for each channel and the other of the sizes with the lowest variation. These pairs were used in all subsequent tests. 2.5.2 Sample Mass Since the measurement method works by treating the samples as the dielectric of a capacitor, it was deemed necessary to test if the amount of dielectric medium would affect the measurement. To evaluate the effect of sample weight on the capacitance measurement, tests were conducted using three different samples. The same moisture content (9.00 %) and compactness were maintained for all samples to isolate the effect of mass alone. One sample had the dry weight of 1000 g and the other two 2000 g, which correspond to the standard sample weight and twice as much as in previous tests. The iron ore concentrate was arranged in different ways to discern potential changes due to form and electromagnetic effects. The smaller, 1000 g and one of the larger, 2000 g, sam- ples was spread over a 45◦ segment of the rotary conveyor, while the other large sample was spread over a 90◦ segment. Using this arrangement, both the width and the thickness of the samples will be varied. The resulting arrangement can be seen in Figure 11. 2.5.3 Sample Density Very early tests showed signs of variability, which were hypothesised to stem from varying density of the tested material due to the iron ore concentrates tendency to lump together. This would align with the theoretical understanding of the measur- ing method, where a less packed material would consist of more air pockets, which in turn would influence the measurements [18]. Therefore, it was deemed necessary to verify how the density of the material affects the measurement. 19 2. Methodology and Theoretical Framework Three samples with identical moisture content (9.00 %) and dry weight (1000 g) were placed in a PLA-box, described in Section 2.4.3 directly on the conveyor and homogenised. Two of the samples were pressed, one to moderate compaction (250 N) and the other to high compaction (500 N) using the force gauge Sauter FH 500 [19]. Each sample was measured for 30 s over each sensor. 2.5.4 Test Speed Since conductive materials moving through an electric field create eddy currents, and currents, in turn, generate opposing electric fields, it was deemed necessary to investigate the sample speed and how much it influences the measurement. To investigate the influence of the rotating acrylic dish’s speed on capacitance mea- surements, the rotational speed was varied from 0.001 revolutions per second (rps) to 0.090 rps. The program consists of: • Stationary measurements for each sample • Two revolutions at 0.001 rps • Five revolutions at 0.008 rps • Five revolutions at 0.016 rps • Five revolutions at 0.090 rps The program is intended to allow a systematic assessment of the influence of speed on capacitance. 2.5.5 Sample moisture Once all parameters were accounted for, tests were performed on varying moisture contents to answer the problems defined in Section 1.1. Initial tests focused on a moisture content range of 8.19 % to 10.55 % to determine whether capacitance measurements could reliably detect an upward trend corre- lated with increasing moisture. All samples, prepared with a consistent dry weight of 1000 g, were thoroughly stirred to ensure uniform moisture distribution. To start, testing consisted of stationary tests, where eight samples were equidistantly placed on the conveyor, each being measured for 30 s over each sensor. Continuous tests were also performed according to the program described in Section 2.5.4. 2.5.6 Single Lot Averaging Tests Due to high uncertainty in the results, a final set of tests was performed, designed to reduce the sources of error. If this sensor were to be implemented on the real conveyor, an average of the capacitance over a large area and time could be used to handle variations. Using the same concept, these tests take the average capacitance of one revolution from 0 to 360◦ at a speed of 0.015 rps with 5 kg of iron ore concen- trate from a single lot spread evenly across the dish. 20 2. Methodology and Theoretical Framework To ensure that the ore was homogenous for these tests, a sift was 3D-printed with 3 mm wide holes. This eliminated the lumps that form as the ore is mixed, further reducing variance. Another tool was 3D-printed to evenly spread the ore around the dish, making sure the height is similar all around. All single lot averaging tests were performed with 330 pF LC-tanks on both channels due to time limitations. 2.6 Data Analysis Depending on the type of test, two different methods were used to interpret the data. Mean capacitance was calculated for each sample and documented for later analysis. 2.6.1 Stationary Analysis For tests where samples were measured while stationary, measurement intervals were selected manually from the plotted data. Attention was given to not counting extremes and abnormal readings that would result in spikes in the data and deviate from the mean capacitance value. 2.6.2 Moving Analysis In the continuous movement tests, data from the test rig’s encoder was used to as- sociate each data point with a specific angle (in degrees). Given that a full rotation is 360◦, and the rig contained eight equally spaced sample positions, as mentioned in Section 2.4.3, the rotation was divided into eight segments of 45◦ each. This seg- mentation allowed us to estimate the sensor readings corresponding to each sample position. For every 45◦ interval, the mean capacitance was calculated. This method minimised manual processing but could not detect and exclude anomalies in the recorded data. 21 2. Methodology and Theoretical Framework 22 3 Results This chapter will present the results of the tests described in Section 2.5. 3.1 Choice of Measuring Frequency Table 2: Resolution and variation per LC-Tank configuration for Channel A and B. LC-tank (pF) 10 20 33 56 75 100 150 220 330 Resolution A (%) 11.3 9.7 14.1 11.0 11.6 7.9 11.3 12.0 12.8 Resolution B (%) 9.4 8.5 9.8 9.6 9.4 8.6 10.4 9.1 8.5 Variation A (%) 0.05 0.06 0.06 0.04 0.04 0.04 0.04 0.02 0.02 Variation B (%) 0.05 0.06 0.06 0.04 0.05 0.04 0.05 0.02 0.01 Among the tested LC-tank configurations, a setup with 33 pF on Channel A and 150 pF on Channel B provided the highest resolution (see Table 2) equivalent to 2.773 MHz and 2.166 MHz for channel A and B respectively. Additionally, tests indicate that the 330 pF LC-tank exhibits the lowest proportional deviation within the target intervals, meaning 1.712 MHz is optimal for variation, being 0.01 to 0.02 % for both channel A and B. 3.2 Influence From Sample Mass Table 3: Capacitance given Mass for Different LC-Tanks Mass (g) Capacitance (pF) 33 (pF) 150 (pF) 330 (pF) 330 (pF) 1000 (45◦) 165.832 156.553 179.731 161.616 2000 (45◦, double height) 167.966 157.765 181.718 162.886 2000 (90◦, double width) 166.977 156.879 181.127 162.062 Table 3 illustrates the effect of varying sample weight and spread angle on capaci- tance measurements. Data shows approximately 2 pF in difference between 1000 g 23 3. Results and 2000 g in dry-weight. The same 2000 g spread over a wider area has less of an impact. 3.3 Influence From Sample Density Table 4: Capacitance given Compactness for Different LC-tanks Compactness (N) Capacitance (pF) 33 (pF) 150 (pF) 330 (pF) 330 (pF) 0 165.428 156.358 179.396 161.441 250 165.257 156.093 179.246 161.179 500 165.176 156.151 179.178 161.151 Capacitance measurements varied only marginally, 0.2 to 0.3 pF, with compaction force of 0, 250 and 500 N, as shown in Table 4. The capacitance appears to decrease slightly with increasing compactness. 3.4 Influence From Speed As shown in Figure 12 and 13, the capacitance exhibits a reduction of approximately 0.2 pF with increasing speed. Variations are quite high at each speed, with the me- dian varying by 0.1 to 0.3 pF. The box plot used to represent these measurements is described as follows: • The Box - Represents the interquartile range (IQR), spanning from the 25th percentile (Q1) to the 75th percentile (Q3) of the measured values. The me- dian (50th percentile) is typically indicated by a line within the box. • Whiskers - Extend to the smallest and largest values within 1.5 × IQR from Q1 and Q3, respectively. • Outliers - Data points outside the whiskers are considered outliers and are displayed as individual points. 24 3. Results 0.0 0.001 0.008 0.016 0.09 speed (rps) 166.0 166.1 166.2 166.3 166.4 166.5 166.6 ca pa cit an ce (p F) Capacitance given Speed, Ch A 0.0 0.001 0.008 0.016 0.09 speed (rps) 156.9 157.0 157.1 157.2 157.3 ca pa cit an ce (p F) Capacitance given Speed, Ch B Figure 12: Capacitance given speed, with sample with 9% moisture, 33 pF LC- tank on channel A and 150 pF LC-tank on channel B. 0.0 0.001 0.008 0.016 0.09 speed (rps) 180.6 180.7 180.8 180.9 181.0 181.1 181.2 181.3 ca pa cit an ce (p F) Capacitance given Speed, Ch A 0.0 0.001 0.008 0.016 0.09 speed (rps) 162.30 162.35 162.40 162.45 162.50 162.55 162.60 ca pa cit an ce (p F) Capacitance given Speed, Ch B Figure 13: Capacitance given speed, with sample with 9% moisture, 330 pF LC- tanks on both channel A and B. 3.5 Moisture Results The influence of sample moisture content on capacitance in continuous analysis is detailed in Tables 9 to 12 in Appendix G. Across all experimental runs with both chosen LC-tank configurations, the capacitance range for the moistures studied is between 0.5 and 1.2 pF. The corresponding box plots for the stationary tests are shown in Figure 14. An interesting phenomenon appears at 9.41 % with the 33 pF LC-tank, this is due to the measurement never stabilising, even when stationary. This phenomenon has been encountered in other tests, but the reasons are unclear. 25 3. Results 166.4 166.6 166.8 167.0 167.2 167.4 Ca pa cit an ce (p F) Capacitance given Moisture, Ch A 8.19 8.78 8.94 9.0 9.03 9.11 9.41 10.55 Moisture (%) 157.1 157.2 157.3 157.4 157.5 157.6 Ca pa cit an ce (p F) Capacitance given Moisture, Ch B 180.6 180.8 181.0 181.2 181.4 Ca pa cit an ce (p F) Capacitance given Moisture, Ch A 8.19 8.78 8.94 9.0 9.03 9.11 9.41 10.55 Moisture (%) 161.5 161.6 161.7 161.8 161.9 162.0 162.1 162.2 Ca pa cit an ce (p F) Capacitance given Moisture, Ch B Figure 14: Capacitance given sample moisture content, with 33 pF LC-tank on channel A and 150 pF LC-tank on channel B, aswel as 330 pF LC-tanks on both channel A and B. 26 3. Results 3.6 Nugget-To-Sill Ratio The nugget effect explained in Section 2.3.2 occurred when being tested. Figure 15 below indicate that small errors in moisture content can be observed. The difference in moisture content within the tests are between 0.08 % to 0.12 %. All the tests are done on lot 5 in Table 1. Figure 15: Box-plot illustrating the effect of Nugget-to-Sill ratio from one of the lots. 27 3. Results 3.7 Repeatability of Results Repeatability can be observed in every test, and results have proven to be quite consistent regarding measurements of the same sample at different times. Figure 16 illustrates the relative consistency of measuring the same samples five times without physically disturbing them. More results from similar tests can be viewed in Appendix G, Table 13 to 16. 180.50 180.75 181.00 181.25 181.50 Ca pa cit an ce Capacitance Measurements, Channel A, 330pF LC-tank 1 2 3 4 5 Cycle 161.6 161.8 162.0 162.2 Ca pa cit an ce Capacitance Measurements, Channel B, 330pF LC-tank 8.5 9.0 9.5 10.0 10.5 M oi st ur e Le ve l ( % ) 166.5 167.0 167.5 Ca pa cit an ce Capacitance Measurements, Channel A, 33pF LC-tank 1 2 3 4 5 Cycle 157.0 157.2 157.4 157.6 Ca pa cit an ce Capacitance Measurements, Channel B, 150pF LC-tank 8.5 9.0 9.5 10.0 10.5 M oi st ur e Le ve l ( % ) Figure 16: Eight samples measured at five separate times, both with 330 pF LC- tanks on both channels and 33 pF LC-tank on channel A and 150 pF LC-tank on channel B. 28 3. Results 3.8 Single Lot Averaging Tests To confirm that only one rotation is needed for these tests, an initial test was performed on lot 4 with two full rotations. The results of this test are shown in Figure 17 and confirm that the average is similar across rotations. 0 50 100 150 200 250 300 350 Angle (°) 177.5 178.0 Ca pa cit an ce (p F) Channel A Rotation 1 Rotation 2 Avg 1: 177.63 pF Avg 2: 177.63 pF 0 50 100 150 200 250 300 350 Angle (°) 161.6 161.8 162.0 Ca pa cit an ce (p F) Channel B Rotation 1 Rotation 2 Avg 1: 161.73 pF Avg 2: 161.74 pF Comparison of rotations Figure 17: Repeatability test for single lot averaging Figure 18 shows the average capacitance of lots 1-8 measured with the method described in Section 2.5.6. The tests for lots 1-3 were repeated after lot 8 due to an error in the measurement process. 1 2 3 4 5 6 7 8 Lot 177.60 178.20 178.80 179.40 180.00 180.60 181.20 Ca pa cit an ce (p F) Channel A Capacitance Boxplot 1 2 3 4 5 6 7 8 Lot 161.80 161.95 162.10 162.25 162.40 Ca pa cit an ce (p F) Channel B Capacitance Boxplot Figure 18: Full conveyor single moisture content tests for lots 1-8 As the tests in Figure 18 indicate inconsistencies in channel A and a sensor drift in channel B, another set of tests was performed. This time, focusing only on channel B and with a constant time between the tests in hopes of getting a constant drift that is possible to filter out. This time, only three lots were tested due to time 29 3. Results limitations, which were lots 7, 5 and 3 in that order. The results for these tests can be seen in Figure 19 and 20. 7 5 3 Lot 161.80 161.95 162.10 162.25 162.40 Ca pa cit an ce (p F) Channel B Capacitance Boxplot Figure 19: Box plot of second set 7 5 3 Lot 162.03 162.04 162.06 162.07 162.09 162.10 Av er ag e ca pa cit an ce (p F) Channel B Average Capacitance with Moisture Average capacitance Moisture 7.2 7.4 7.6 7.8 8.0 8.2 8.4 M oi st ur e (% ) Figure 20: Average capacitance and verified moisture content of the second set In these tests, drift was not a problem and the data was usable. There also seems to be a clear proportional correlation between the capacitance and verified moisture. To get a formula between capacitance and moisture, linear regression was used on the data. The result of this is shown in Figure 21. 7.2 7.4 7.6 7.8 8.0 8.2 Moisture (%) 162.03 162.04 162.05 162.06 162.07 162.08 162.09 162.10 Ca pa cit an ce (p F) Linear Regression Between Capacitance and Moisture Data Points Regression Line: y = -0.06x + 162.57 Figure 21: Linear Regression on the Second Set of Tests 30 3. Results If this holds true, the moisture content in the iron ore concentrate can be found using the formula Moisture percentage = 162.5667 − C × 1012 0.0644 (12) where C is the capacitance in F. The R2 value for this line is 0.9957, which indicates a high accuracy. With only three points though, that does not mean much without further testing. 31 3. Results 32 4 Discussion Before continuing, it is worth discussing two notable qualities of the data. First being the varying capacitance ranges in the test results. The test evaluation did account for the LC-tank capacitance, by subtracting it and only reading the capacitance of the conveyor and the iron ore concentrate. Due to manufacturing error margins of the LC-tanks, some of the measured capacitances will vary by ap- proximately 12 pF, this will not affect analysis, since what is of interest is a relative measurement to a calibrated moisture content sample, rather than an absolute mea- surement. Secondly, there are flaws in the continuous tests, where at the maximal speed, each sample measured significantly fewer data points, leading to less reliable and some- times misleading data. The high speed is also hypothesised to slightly move the iron ore concentrate due to centrifugal forces and vibration, which compromises the validity of the results. 4.1 Analysis of Parameter Impact The measurements suggest that several parameters influence the capacitance of the measured iron ore concentrate and that it may be hard to isolate them in order to decide the moisture content. Below, tests aimed at isolating these parameters will be discussed. 4.1.1 Measuring Frequency Optimization The differences in measurement resolution certainly is related to the electrode sep- aration, as shown in Figure 3. Additionally, the variation in measurements, notably the minimal proportional deviation of 0.01 to 0.02 % observed in the 330 pF LC-tank configuration indicates high stability across the target intervals, likely due to the higher capacitance reducing sensitivity to external noise. To further narrow down the selection of measurement frequency, studies on the dielectric behaviour of concentrate could be performed. Given the theoretical un- derstanding that materials display different permittivity under different frequencies, it would be of interest to perform theoretical calculations and analytical measure- ments to find a more optimised frequency for a given quality of concentrate. 33 4. Discussion 4.1.2 Effect of Mass As shown in Table 3, the effects of mass seem to be significant. The large samples with a wider spread having a higher capacitance measurement than the baseline sam- ple indicates that there is some kind of electromagnetic effect of a large, continuous sample or the measured field extends further to in the x-y-plane than anticipated. The increase in capacitance from the other sample could be explained by the same hypothesised electromagnetic phenomenon but also a decreased distance to the sen- sor from deformation of the rotary dish or extension of the measurement field in the z-plane similar to that explained above. 4.1.3 Effect of Material Density The study found that the density of the material has a small impact on the results as presented in Figure 4. The observed decline in capacitance with higher compactness is contrary to the intuitive reasoning that less air within the material would lead to a higher total permittivity. It is speculated that the reduction could be explained either by more measurements being performed above the measured material or the dielectric properties of the material are altered by higher compactness. Further studies are needed to determine the cause. 4.1.4 Effect of Speed One trend that became apparent is that the capacitance generally decreases as the speed of the material increases, as shown in Figures 12 and 13. There is, however, an exception at the highest speed shown in Figure 12, with the 33 pF LC-tank, where the capacitance unexpectedly increases. This is explained by the previously mentioned problems encountered when performing the continuous tests. On the other hand, Figure 12 suggests that the overall variation in measurements is not strongly affected by speed. This could mean that the impact of speed depends on variables like the type of sensor, properties of the material, or the specific capac- itance range being measured. Finally, the edges of the sensor are usually less accurate due to the way the electric field behaves in those areas. When the material moves quickly over these edge zones, the inaccuracies can become more noticeable. 4.1.5 Repeatability of Measurements The study found that measurements between rotations were essentially identical, indicating high repeatability. This is illustrated in Figures 16 and 17. However, between the different velocities as mentioned in 4.1.4, a similar connection could not be drawn. 34 4. Discussion In order to implement this system in a real scenario, it is a prerequisite to have repeatability in the results to be able to draw trustworthy conclusions. Therefore, it can be seen as one of the most important discoveries made in this project. During all tests temperature and humidity have been recorded. The relative hu- midity was found to be 25.5 to 49.2 % and the temperature 21.3 to 24.4 ◦C. Any analysis of their effect have not been made, however, the effect of them seems neg- ligible compared to other variations measured. 4.1.6 Effect of Moisture These readings did not consistently correlate with the moisture content of the con- centrate samples, suggesting other factors may influence capacitance measurements. A linear trend is still visible though, especially on channel A. The variations in moisture of the lot seem to be impacted by the nugget effect, which seems to have an impact on the results. Although steps are taken to ho- mogenise the material, it can not be ensured that the verification sample has the same moisture content as the sample on the test rig. As can be seen in Figure 15, the moisture content can differ more than some levels presented in Table 1, which makes it really hard to determine if the result with certainty can be trusted. However, the error of moisture variation in the tests may be affected by depurate from the ISO standard described in Section 2.3.2. If bigger samples and several sam- ples from each lot were tested as described in the standard, the validated moisture in the sample would more likely be representative for the whole lot. An important step in calculating moisture content is to compensate for the param- eters influencing on the reading, either with practical methods or by calculations. Given that the parameters have such a high variation and their non-linearity, it proved extremely difficult to create an equation or to model the effect of the param- eters accurately. The initial goal of the project was to test moisture content of 6 to 12 % as pre- sented in Table 1. This was abandoned after realising that the lot with the highest moisture content still had water accumulations present after homogenisation. 4.1.7 Single Lot Averaging Tests The averaging tests were done as a final effort to find a good correlation between capacitance and moisture by controlling as many variables as possible. The results were promising, but the method could not be explored further as these tests started only days before the project and report had to be finished. The initial averaging test showed two concerning sources of error that might be 35 4. Discussion present in other tests performed during this project. Channel A showed inconsis- tent results, with some tests having a much higher capacitance than the others. Channel B showed strong signs of sensor drift as the average capacitance increased with every test, even after going back to the first lots. Curiously, this sensor drift did not seem to be present in the second set of tests. The second set of tests showed promising results but would need more data points to confirm that it is not by chance. The resulting linear regression model seems to have a good accuracy but with only three points it is inconclusive. With further testing, including more data points and verifying the model, there is a chance this method could lead to a formula that can correlate capacitance and moisture closely. With the correlation seeming good in this test, it would seem that sifting the iron ore concentrate through a mesh, and thereby removing most of the lumps, helped give more accurate verification tests as well. If small-scale tests were to continue, this would be a good method to explore further, but it would be even better to go straight to large-scale testing, where many variances caused by the small scale would be avoided. 4.2 FMEA 4.2.1 Rotary Conveyor Some actions were highlighted as critical in the FMEA. As earlier mentioned, a PID controller was implemented to ensure a precise and constant speed during the tests. To be able to maintain the lowest speed and reach the highest torque required, a gearbox was constructed for the motor. To avoid overloading the motor, a current- limiting power supply was used to drive it. The frame was also reinforced after the first tests to ensure rigidity. All identified failure modes can be seen in Table 6 in Appendix E. 4.2.2 Sensor According to the FMEA analysis for the sensors, coaxial connectors designed to ensure reliable signal transmission and minimise electrical interference were chosen. This choice was made in order to minimize noise and other disturbances that could significantly affect the measurement results. The electrodes were secured using a 3D-printed casing that held them in place just below the acrylic sheet, maintaining a consistent distance and width to ensure uniform measurement conditions. All identified actions and potential issues have been addressed throughout the project to support accurate and stable measurements, see Table 7 in Appendix E. 36 4. Discussion 4.3 Ethical aspects This project has been performed in cooperation with LKAB, which has the vision to lead the industry into the future with innovative and competitive solutions. The system developed in this project could result in improvements across several fields. The system could lead to further improvements in pursuit of accomplishing climate goals, especially climate goals 9 and 12 set by the UN, by making the energy con- sumption more even [20]. 4.4 Social aspects In order to operate basic industry in Europe, especially in Sweden, adaptability and innovative capacity are crucial aspects to compete with industries in other parts of the world with lower production costs. This is essential to keep the industry, jobs and expertise in this part of the world, ensuring human rights and high general standard of living. The discoveries made in this project would, if implemented, lead to a safer environ- ment for the technicians at the plants. The manual work of taking out a sample for the gravimetric method will be reduced. Calibration to ensure the functionality of the sensors may, however, still be needed. 37 4. Discussion 38 5 Conclusion This project set out to answer two main questions: • What was the feasibility of implementing capacitive iron ore concentrate mois- ture content measurement in real-time? • How does the capacitive moisture measurement system compare to traditional methods? Even though the answer to these questions was not as straightforward as anticipated, the project has come to several conclusions which could provide a foundation for further studies. These conclusions, together with suggestions for areas of further studies, will be provided below. 5.1 Conclusion on Feasibility During the project, several parameters which has had as large, or larger, impact on the measurement results as change in moisture content have been identified. These include density, distance to sensor, amount of material, and some that are still to be identified. This makes it difficult to answer the question of feasibility accurately. However, the results have shown that there is indeed a measurable difference in capacitance between different moisture contents in the concentrate. It is hypothesised that several of the affecting parameters are the result of small samples and that by measuring larger volumes, many of the local variations will even out and show a more consistent result. Other studies have shown results that are less affected by these parameters, but there is limited research into the effects of relative speed between the sensor and the measured material. This project has shown that the relative speed does not influence the measurements more than what is acceptable. This makes it possible to conclude that it is indeed feasible to develop on-line measurement methods for iron ore concentrate using capacitance, but that the method needs further development using large-scale tests. 5.2 Conclusion Compared to Existing Methods Due to the issues mentioned above, no real comparative analysis was performed against other, more established methods. The project ran into several problems present in other methods, including sample variability. Tests showed sensor drift, indicating that the method may require periodic calibration, however, further studies may find ways to counteract this. This project is unable to draw any conclusion 39 5. Conclusion regarding the performance of capacitative methods compared to other methods of measuring moisture content of iron ore concentrate. 5.3 Areas for Further Study To enhance measurement accuracy, sensors should ideally be embedded directly within the iron ore concentrate. Given the abrasive nature of iron ore concentrate, this would most likely not work for a sensor that is supposed to perform measure- ments in a moving stream of material, but it would still be ideal to try to keep the air gap between sensor and measured material as small as possible. It is therefore suggested that further studies into the subject try designs that would not measure through the conveyor belt. Suggested methods for achieving this are to either mount the sensor to a plough riding on the material, thereby providing a fixed, small dis- tance; a sensor measuring from the side or measuring during free fall in a loading chute. This project has been exclusively small-scale and even though it has enabled tests with varying moisture contents in a way that would have been difficult to perform in a real production environment, many of the problems encountered most likely stem from the material not behaving the same in small scales. It is therefore recom- mended that further studies be performed on larger samples, ideally the same size as in a production environment or on-site. If tests are to be performed on a smaller scale, more emphasis should be put on proper sampling and homogenisation of the tested material since this has been iden- tified as a large source of error, mainly due to the nugget effect caused by uneven moisture distribution in the samples. If a new test rig is to be constructed, it is recommended to leave the circular design for a linear design. Even though the cir- cular conveyor has helped in the ease of repeated tests and material handling, it is not as representative of a real-world implementation. A linear belt design with two opposing directions would require transferring material from one belt to the other, which could provide mixing and thereby certain homogenisation of the material, which could improve results. To provide a full insight into the methods, areas of further studies include: • Effects of different minerals, i.e. hematite • Effects of additives • Effects of particle size • Impact from the environment, e.g. temperature and humidity 40 Bibliography [1] X.-h. Fan, G.-m. Yang, and X.-l. Chen, “Evaluation model on the ballability of iron ore concentrates,” Powder Technology, vol. 280, pp. 219–226, 2015. doi: 10.1016/j.powtec.2015.04.068. [2] A. Timofeeva, T. Nikitchenko, V. Fedina, and V. Kazartsev, “Effect of the composition and characteristics of the charge materials on the quality of iron- ore pellets,” Metallurgist, vol. 59, no. 3-4, pp. 212–215, 2015. doi: 10.1007/ s11015-015-0086-5. [3] SS-ISO 3087:2011, “Iron ores — determination of the moisture content of a lot,” International Organization for Standardization, Standard, 2011. [4] P. Cancilla, P. Barrette, and F. Rosenblum, “On-line moisture determination of ore concentrates ‘a review of traditional methods and introduction of a novel solution’,” Minerals Engineering, vol. 16, no. 2, pp. 151–163, 2003, Solid- liquid separation, issn: 0892-6875. doi: https : //doi . org/10 . 1016/S0892 - 6875(03)00046-3. [5] Engström, K.-O. Mickelsson, S. Töyrä, and K. Esbensen, “Improvement of sampling for moisture content analysis in iron ore pellet feed - variographic characterisation and on-line ir-analysis,” 2019, pp. 3523–3533. [6] D. G. Miljak, D. Bennett, T. Kazzaz, and N. G. Cutmore, “On-line microwave moisture measurement of iron ore and mineral concentrates in conveyor appli- cations,” in 2006 IEEE Instrumentation and Measurement Technology Con- ference Proceedings, (Sorrento, Italy), Apr. 2006, pp. 183–186. doi: 10.1109/ IMTC.2006.328366. [7] F. Harbert, B. Sjöberg, and J. Bak, “Moisture content measurement in iron ore,” Mining Magazine, pp. 339–341, Oct. 1973. [8] G. G. Archer and P. Wang, “The dielectric constant of water and debye-hückel limiting law slopes,” Journal of Physical and Chemical Reference Data, vol. 19, no. 2, pp. 371–411, 1990. doi: 10.1063/1.555853. [9] A. K. Rêgo Segundo, M. J. Da Silva, G. M. Freitas, P. M. De Barros Monteiro, and J. H. Martins, “Reply to comments: A novel low-cost instrumentation system for measuring the water content and apparent electrical conductivity of soils, sensors, 15, 25546–25563,” Sensors, vol. 18, no. 6, 2018, issn: 1424- 8220. doi: 10.3390/s18061742. [10] M. E. H. Silva, A. K. R. Segundo, S. A. L. d. Silva, and P. M. d. B. Mon- teiro, “Capacitive, non-invasive and coplanar-electrode transducer for measur- ing iron ore moisture,” Proceedings, vol. 42, no. 1, 2020, issn: 2504-3900. doi: 10.3390/ecsa-6-06579. 41 https://doi.org/10.1016/j.powtec.2015.04.068 https://doi.org/10.1007/s11015-015-0086-5 https://doi.org/10.1007/s11015-015-0086-5 https://doi.org/https://doi.org/10.1016/S0892-6875(03)00046-3 https://doi.org/https://doi.org/10.1016/S0892-6875(03)00046-3 https://doi.org/10.1109/IMTC.2006.328366 https://doi.org/10.1109/IMTC.2006.328366 https://doi.org/10.1063/1.555853 https://doi.org/10.3390/s18061742 https://doi.org/10.3390/ecsa-6-06579 Bibliography [11] R. P. Deshpande, “Introduction to capacitors,” en, in (Capacitors), 1st Edi- tion, Capacitors. New York: McGraw-Hill Education, 2015, isbn: 9780071848565. [12] G. Santos, É. Pinto, A. Silva, P. Monteiro, A. Mesquita, and A. K. Segundo, “Real-time capacitive sensor applied to a train wagon prototype for measuring iron ore moisture,” IEEE Latin America Transactions, vol. 21, no. 4, pp. 595– 601, 2023. doi: 10.1109/TLA.2023.10128932. [13] H. A. Radi and J. O. Rasmussen, “Inductance, oscillating circuits, and ac cir- cuits,” in Principles of Physics: For Scientists and Engineers. Berlin, Heidel- berg: Springer Berlin Heidelberg, 2013, pp. 961–998, isbn: 978-3-642-23026-4. doi: 10.1007/978-3-642-23026-4_28. [14] F. Reverter, X. Li, and G. C. M. Meijer, “Stability and accuracy of active shielding for grounded capacitive sensors,” Measurement Science and Technol- ogy, vol. 17, no. 11, pp. 2884–2890, 2006. doi: 10.1088/0957-0233/17/11/004. [15] FDC2x1x Multi-Channel, High Resolution Capacitance-to-Digital Converter for Capacitive Sensing Applications, 2024. [Online]. Available: https://www. ti.com/lit/gpn/FDC2214. [16] KERN EHA 500-2. [Online]. Available: https://www.kern-sohn.com/cosmoshop/ default/pix/a/media/TEHA%20500-2-A/TD_EHA+500-2_en.pdf. [17] Kern DE 60K10D. [Online]. Available: https://www.kern-sohn.com/cosmoshop/ default/pix/a/media/DE%2060K10D/TD_DE+60K10D_en.pdf. [18] B. Clarke, Microwave Measurements. The Institution of Engineering and Tech- nology, 2007, ch. 18, pp. 409–458. doi: 10.1049/PBEL012E_ch18. [19] Sauter FH 500. [Online]. Available: https://www.kern-sohn.com/cosmoshop/ default/pix/a/media/TFH%20500-B/TD_FH+500_en.pdf. [20] Luossavaara-Kiirunavaara AB (LKAB), Års- och hållbarhetsredovisning 2024, Luleå, Sverige, 2025. [Online]. Available: https : / / lkab . com / wp - content / uploads/2025/03/LKAB_2024_Ars-och-hallbarhetsredovisning.pdf. 42 https://doi.org/10.1109/TLA.2023.10128932 https://doi.org/10.1007/978-3-642-23026-4_28 https://doi.org/10.1088/0957-0233/17/11/004 https://www.ti.com/lit/gpn/FDC2214 https://www.ti.com/lit/gpn/FDC2214 https://www.kern-sohn.com/cosmoshop/default/pix/a/media/TEHA%20500-2-A/TD_EHA+500-2_en.pdf https://www.kern-sohn.com/cosmoshop/default/pix/a/media/TEHA%20500-2-A/TD_EHA+500-2_en.pdf https://www.kern-sohn.com/cosmoshop/default/pix/a/media/DE%2060K10D/TD_DE+60K10D_en.pdf https://www.kern-sohn.com/cosmoshop/default/pix/a/media/DE%2060K10D/TD_DE+60K10D_en.pdf https://doi.org/10.1049/PBEL012E_ch18 https://www.kern-sohn.com/cosmoshop/default/pix/a/media/TFH%20500-B/TD_FH+500_en.pdf https://www.kern-sohn.com/cosmoshop/default/pix/a/media/TFH%20500-B/TD_FH+500_en.pdf https://lkab.com/wp-content/uploads/2025/03/LKAB_2024_Ars-och-hallbarhetsredovisning.pdf https://lkab.com/wp-content/uploads/2025/03/LKAB_2024_Ars-och-hallbarhetsredovisning.pdf A FDC2214 Configuration Table 5: FDC2214 Configuration Property Address Value General DRDY_2INT 0x19[0:0] 1 WD_ERR2INT 0x19[5:5] 1 AL_WARN2OUT 0x19[11:11] 1 AH_WARN2OUT 0x19[12:12] 1 WD_ERR2OUT 0x19[13:13] 1 INTB_DIS 0x1A[7:7] 0 REF_CLK_SRC 0x1A[9:9] 1 RESERVED 0x1A[10:10] 1 RESERVED 0x1A[12:12] 1 SENSOR_ACTIVATE_SEL 0x1A[11:11] 0 DEGLITCH 0x1B[0:2] 4 HIGH_CURRENT_DRV 0x1B[6:6] 0 RR_SEQUENCE 0x1B[13:14] 0 AUTOSCAN_EN 0x1B[15:15] 1 Channel N RCOUNT 0x08+N 32000 SETTLECOUNT 0x10+N 4096 FREF_DIVIDER (0x14+N)[0:9] 2 FIN_SEL (0x14+N)[12:13] 1 IDRIVE (0x1E+N)[11:15] 17 I A. FDC2214 Configuration II B Sensor Mount Figure 22: Drawing of Sensor Mount III B. Sensor Mount IV C Gearbox Diagram Figure 23: Illustration of the gearbox driving the dish on the test rig. The encoder is connected to the second gear from the motor and the dish is connected to the centre of the last gear. V C. Gearbox Diagram VI D Motor Driver Circuit Figure 24: Description of the different parts of the motor driver box. The motor is controlled by a IBT2 motor driver. The ESP32 microcontroller sets the speed of the motor using pulse width modulation and takes inputs from the buttons. It gives outputs to the display and to the microcontroller logging data from the sensors. The encoder is used as feedback in a PID controller running on the ESP32 that makes sure the speed is correct while the load differs. The parameters for this algorithm are KP = 75000, KI = 50000 and KD = 1000. VII D. Motor Driver Circuit VIII E Failure Mode & Effects Analysis (FMEA) Table 6: FMEA for Rotary Conveyor Component Failure Mode Cause P E V RPN Action Motor Overloading Excessive load, incor- rect sizing 6 9 7 378 Implement PID controller for precise speed control Motor Misalignment Incorrect installation, wear 5 8 8 320 Use 3D-printed gearbox with precise alignment Conveyor Belt Belt Jamming Blockage due to ore lumps or debris 6 9 8 432 Ensure continuous clean- ing Bearings Wear and Failure load distribution, con- tamination 5 9 7 315 Inspect regularly Power Supply Power Loss Loose connection, overload 6 8 7 336 Use a stabilized power supply and ensure proper connections Frame/Structure Vibrations and Instability Inadequate design, in- correct materials 6 8 8 384 Reinforce the frame IX E. Failure Mode & Effects Analysis (FMEA) Table 7: FMEA for Sensor and Electronics Component Failure Mode Cause P E V RPN Action Sensor plates Signal deviation Poor grounding 6 8 8 384 Use shielded cables with grounded copper tape layer to improve grounding. Sensor plates Mechanical damage Wear/tear handling 7 7 9 441 Protective casing, more robust materials Electronics (PCB) Overheating Poor heat dissipation 5 9 7 315 Optimize placement to re- duce heat Electronics (Amp) Noise/Interference Poor shielding 6 8 8 384 Use Fakra cables Power Supply Power loss Faulty source/discon- nect 5 8 7 280 Improve stability Microcontroller Data error Software bug or faulty ADC reading 4 8 6 192 Debug and validate Ar- duino MKR Zero code Wiring Loose connection Poor grounding/vibra- tion 5 9 7 315 Secure connections, lock- ing connectors X F Test Moisture Content Table 8: Desired Moisture Content Calculation Function Expression Moisture (%) f(x) 1.93x - b(x) 3f(x) f(8) - x 9 − b(8) 6 x 9 − b(6) 8.19 x 9 − b(4) 8.78 x 9 − b(2) 8.94 x 9 9 x 9 + b(1) 9.03 x 9 + b(3) 9.11 x 9 + b(5) 9.41 x 9 + b(7) 10.55 x 9 + b(8) 12 The formula and calculation used to determine which moisture contents to use to be able to test an error margin down to 0.03% XI F. Test Moisture Content XII G Moisture Cycles Table 9: 33 pF, Channel A, Capacitance for each Cycle for different Moisture Contents, 0.008 rps Moisture (%) Capacitance (pF) Cycle 1 Cycle 2 Cycle 3 Cycle 4 Cycle 5 8.52 166.430 166.389 166.355 166.377 166.352 8.71 166.085 166.057 166.051 166.051 166.050 9.03 166.104 166.083 166.080 166.072 166.073 9.11 166.536 166.518 166.499 166.511 166.490 9.11 166.095 166.075 166.078 166.069 166.083 9.22 166.616 166.581 166.559 166.557 166.557 9.41 166.594 166.666 166.670 166.658 166.662 10.55 167.095 167.080 167.079 167.065 167.075 Table 10: 150pF, Channel B, Capacitance for each Cycle for different Moisture content Moisture (%) Capacitance (pF) Cycle 1 Cycle 2 Cycle 3 Cycle 4 Cycle 5 8.52 157.381 157.348 157.343 157.344 157.339 8.71 156.924 156.894 156.885 156.892 156.879 9.03 156.919 156.899 156.903 156.896 156.890 9.11 156.777 156.763 156.760 156.762 156.754 9.11 156.923 156.905 156.902 156.904 156.894 9.22 157.193 157.174 157.170 157.173 157.163 9.41 156.917 156.914 156.920 156.913 156.913 10.55 157.234 157.224 157.224 157.219 157.215 XIII G. Moisture Cycles Table 11: Channel A, 330pF, Capacitance for each Cycle for different Moisture content, 0.008 rps Moisture (%) Capacitance (pF) Cycle 1 Cycle 2 Cycle 3 Cycle 4 Cycle 5 8.52 180.576 180.515 180.531 180.512 180.523 8.71 180.323 180.303 180.291 180.296 180.285 9.03 180.280 180.264 180.260 180.260 180.245 9.11 180.676 180.664 180.659 180.648 180.645 9.11 180.282 180.275 180.264 180.265 180.255 9.22 180.707 180.675 180.661 180.659 180.644 9.41 180.669 180.731 180.720 180.720 180.714 10.55 181.105 181.112 181.097 181.097 181.081 Table 12: Channel B, 330pF, Capacitance for each Cycle for different Moisture content, 0.008 rps Moisture (%) Capacitance (pF) Cycle 1 Cycle 2 Cycle 3 Cycle 4 Cycle 5 8.52 162.434 162.437 162.426 162.429 162.421 8.71 161.974 161.979 161.966 161.972 161.957 9.03 161.977 161.978 161.973 161.969 161.963 9.11 161.878 161.986 161.869 161.983 161.868 9.22 162.242 162.230 162.234 162.224 162.224 9.41 162.016 162.008 162.010 162.000 162.001 10.55 162.300 162.295 162.295 162.286 162.294 XIV G. Moisture Cycles Table 13: Capacitance Measurements for different speeds, 2 cycles, 33 pF. Channel A. Sorted by Moisture (%) within each cycle. 0.001 rps 0.016 rps Moisture (%) Capacitance (pF) Moisture (%) Capacitance (pF) Cycle 1 8.52 166.521 8.52 166.371 8.71 166.151 8.71 166.049 9.03 166.227 9.03 166.069 9.11 166.166 9.11 166.068 9.11 166.592 9.11 166.498 9.22 166.858 9.22 166.549 9.41 166.954 9.41 166.612 10.55 167.240 10.55 167.043 Cycle 2 8.52 166.474 8.52 166.320 8.71 166.130 8.71 166.061 9.03 166.161 9.03 166.054 9.11 166.146 9.11 166.077 9.11 166.600 9.11 166.540 9.22 166.697 9.22 166.495 9.41 166.736 9.41 166.657 10.55 167.158 10.55 167.055 XV G. Moisture Cycles Table 14: Capacitance Measurements for different speeds, 2 cycles, 150 pF. Chan- nel B. Sorted by Moisture (%) within each cycle. 0.001 rps 0.016 rps Moisture (%) Capacitance (pF) Moisture (%) Capacitance (pF) Cycle 1 8.52 157.523 8.52 157.334 8.71 156.955 8.71 156.882 9.03 156.947 9.03 156.884 9.11 157.002 9.11 156.891 9.11 156.805 9.11 156.754 9.22 157.340 9.22 157.157 9.41 156.950 9.41 156.902 10.55 157.348 10.55 157.213 Cycle 2 8.52 157.412 8.52 157.316 8.71 156.956 8.71 156.852 9.03 156.964 9.03 156.871 9.11 156.953 9.11 156.879 9.11 156.817 9.11 156.761 9.22 157.242 9.22 157.145 9.41 156.953 9.41 156.912 10.55 157.268 10.55 157.208 XVI G. Moisture Cycles Table 15: Capacitance Measurements for different speeds, 2 cycles, 330 pF. Chan- nel A. Sorted by Moisture (%) within each cycle. 0.001 rps 0.016 rps Moisture (%) Capacitance (pF) Moisture (%) Capacitance (pF) Cycle 1 8.52 180.696 8.52 180.495 8.71 180.410 8.71 180.290 9.03 180.400 9.03 180.253 9.11 180.352 9.11 180.260 9.11 180.732 9.11 180.641 9.22 180.894 9.22 180.645 9.41 180.866 9.41 180.716 10.55 181.191 10.55 181.077 Cycle 2 8.52 180.612 8.52 180.468 8.71 180.357 8.71 180.289 9.03 180.321 9.03 180.248 9.11 180.315 9.11 180.274 9.11 180.719 9.11 180.658 9.22 180.749 9.22 180.628 9.41 180.765 9.41 180.739 10.55 181.155 10.55 181.075 XVII G. Moisture Cycles Table 16: Capacitance Measurements for different speeds, 2 cycles, 330 pF. Chan- nel B. Sorted by Moisture (%) within each cycle. 0.001 rps 0.016 rps Moisture (%) Capacitance (pF) Moisture (%) Capacitance (pF) Cycle 1 8.52 162.512 8.52 162.416 8.71 162.051 8.71 161.945 9.03 162.034 9.03 161.951 9.11 162.034 9.11 161.966 9.11 161.925 9.11 161.865 9.22 162.299 9.22 162.216 9.41 162.045 9.41 162.018 10.55 162.329 10.55 162.295 Cycle 2 8.52 162.474 8.52 162.412 8.71 162.016 8.71 161.944 9.03 162.001 9.03 161.951 9.11 162.001 9.11 161.959 9.11 161.888 9.11 161.870 9.22 162.259 9.22 162.215 9.41 162.014 9.41 162.008 10.55 162.306 10.55 162.288 XVIII DEPARTMENT OF SOME SUBJECT OR TECHNOLOGY CHALMERS UNIVERSITY OF TECHNOLOGY Gothenburg, Sweden www.chalmers.se www.chalmers.se Glossary List of Figures List of Tables Introduction Background Problem Definition Delimitations Conflict of Interests Existing Methods Gravimetric Methods Microwave Methods Optical Methods Capacitive Methods Scope Methodology and Theoretical Framework Theoretical Basis Basics of Capacitors Dielectrics and Permittivity Resonator Circuits Sensor Construction Electrodes Construction Conversion Device Construction Controller And Interface Construction Data Logging Evaluation Module Usage Iron Ore Concentrate Moisture Verification Sample Quality Assurance Rotary Conveyor Rotary Dish Frame and Sensor Mount Sample Partitioning Drive System FMEA and Safety Testing Methodology Determination of Measurement Frequency Sample Mass Sample Density Test Speed Sample moisture Single Lot Averaging Tests Data Analysis Stationary Analysis Moving Analysis Results Choice of Measuring Frequency Influence From Sample Mass Influence From Sample Density Influence From Speed Moisture Results Nugget-To-Sill Ratio Repeatability of Results Single Lot Averaging Tests Discussion Analysis of Parameter Impact Measuring Frequency Optimization Effect of Mass Effect of Material Density Effect of Speed Repeatability of Measurements Effect of Moisture Single Lot Averaging Tests FMEA Rotary Conveyor Sensor Ethical aspects Social aspects Conclusion Conclusion on Feasibility Conclusion Compared to Existing Methods Areas for Further Study Bibliography FDC2214 Configuration Sensor Mount Gearbox Diagram Motor Driver Circuit Failure Mode & Effects Analysis (FMEA) Test Moisture Content Moisture Cycles