Size measurement of iron ore pellets in industry is usually performed by manual sampling and sieving techniques. The manual sampling is performed infrequently and is inconsistent, invasive and time-consuming. Iron ore pellet's sizes are critical to the efficiency of the blast furnace process in the production of steel. Overly coarse pellets affect the blast furnace process negatively, however this affect can be minimized by operating the furnace with different parameters. An on-line system for measurement of pellet sizes would improve productivity through fast feedback and efficient control of the blast furnace. Also, fast feedback of pellet sizes would improve pellet quality in pellet production. Image analysis techniques promise a quick, inexpensive, consistent and non-contact solution to determining the size distribution of a pellet pile. Such techniques capture information of the surface of the pellet pile which is then used to infer the pile size distribution. However, there are a number of sources of error relevant to surface analysis techniques. The objective of this thesis is to address and overcome aspects of these sources of error relevant to surface analysis techniques. The research problem is stated as: How can the pellet pile size distribution be estimated with surface analysis techniques using image analysis? This problem is addressed by dividing the problem into sub-problems. The focus of the presented work is to develop techniques to overcome, or minimize, two of these sources of error; overlapped particle error and profile error. Overlapped particle error describes the fact that many pellets on the surface of a pile are only partially visible and a large bias results if they are sized as if they were smaller entirely visible pellets. No other researchers make this determination. Profile error describes the fact that only one side of an entirely visible pellet can be seen making it difficult to estimate pellets size. Statistical classification methods are used to overcome these sources of error. The thesis is divided into two parts. The first part contains an introduction to the research area together with a summary of the contributions, and the second part is a collection of four papers describing the research.

This thesis contributes to the field of machine vision and the theory of the sampling of particulate material on conveyor belts. The objective is to address sources of error relevant to surface-analysis techniques when estimating the sieve-size distribution of particulate material using machine vision. The relevant sources of error are segregation and grouping error, capturing error, profile error, overlapping-particle error and weight-transformation error. Segregation and grouping error describes the tendency of a pile to separate into groups of similarly sized particles, which may bias the results of surface-analysis techniques. Capturing error describes the varying probability, based on size, that a particle will appear on the surface of the pile, which may also bias the results of surface-analysis techniques. Profile error is related to the fact that only one side of an entirely visible particle can be seen, which may bias the estimation of particle size. Overlapping-particle error occurs when many particles are only partially visible, which may bias the estimation of particle size because large particles may be treated as smaller particles. Weight-transformation error arises because the weight of particles in a specific sieve-size class might significantly vary, resulting in incorrect estimates of particle weights. The focus of the thesis is mainly on solutions for minimizing profile error, overlapping-particle error and weight-transformation error.In the aggregates and mining industries, suppliers of particulate materials, such as crushed rock and pelletised iron ore, produce materials for which the particle size is a key differentiating factor in the quality of the material. Manual sampling and sieving techniques are the industry-standard methods for estimating the size distribution of these particles. However, as manual sampling is time consuming, there are long response times before an estimate of the sieve-size distributions is available. Machine-vision techniques promise a non-invasive, frequent and consistent solution for determining the size distribution of particles. Machine-vision techniques capture images of the surfaces of piles, which are analyzed by identifying each particle on the surface of the pile and estimating its size. Sampling particulate material being transported on conveyor belts using machine vision has been an area of active research for over 25 years. However, there are still a number of sources of error in this type of sampling that are not fully understood. To achieve a high accuracy and robustness in the analysis of captured surfaces, detailed experiments were performed in the course of this thesis work, towards the development and validation of techniques for minimizing overlapping-particle error, profile error and weight-transformation error. To minimise overlapping-particle error and profile error, classification algorithms based on logistic regression were proposed. Logistic regression is a statistical classification method that is used for visibility classification to minimize overlapping-particle error and in particle-size classification to minimize profile error. Commonly used size- and shape-measurement methods were evaluated using feature-selection techniques, to find sets of statistically significant features that should be used for the abovementioned classification tasks. Validation using data not used for training showed that these errors can be overcome.The existence of an effect that causes weight-transformation error was identified using statistical analysis of variance (ANOVA). Methods to minimize weight-transformation error are presented herein, and one implementation showed a good correlation between the results using the machine-vision system and manual sieving results.The results presented in this thesis show that by addressing the relevant sources of error, machine vision techniques allow for robust and accurate analysis of particulate material. An industrial prototype was developed that estimates the sieve-size distribution of iron-ore pellets in a pellet plant and crushed limestone in a quarry during ship loading. The industrial prototype also enables product identification of crushed limestone to prevent the loading of incorrectly sized products.

Size measurement of pellets in industry is usually performed by manual sampling and sieving techniques. Automatic on-line analysis of pellet size based on image analysis techniques would allow non-invasive, frequent and consistent measurement. We evaluate the statistical significance of the ability of commonly used size and shape measurement methods to discriminate among different sieve-size classes using multivariate techniques. Literature review indicates that earlier works did not perform this analysis and selected a sizing method without evaluating its statistical significance. Backward elimination and forward selection of features are used to select two feature sets that are statistically significant for discriminating among different sieve-size classes of pellets. The diameter of a circle of equivalent area is shown to be the most effective feature based on the forward selection strategy, but an unexpected five-feature classifier is the result using the backward elimination strategy. The discrepancy between the two selected feature sets can be explained by how the selection procedures calculate a feature's significance and that the property of the 3D data provides an orientational bias that favours combination of Feret-box measurements. Size estimates of the surface of a pellet pile using the two feature sets show that the estimated sieve-size distribution follows the known sieve-size distribution.

Size measurement of rocks is usually performed by manual sampling and sieving techniques. Automatic on-line analysis of rock size based on image analysis techniques would allow non-invasive, frequent and consistent measurement. In practical measurement systems based on image analysis techniques, the surface of rock piles will be sampled and therefore contain overlapping rock fragments. It is critical to identify partially visible rock fragments for accurate size measurements. In this research, statistical classification methods are used to discriminate rocks on the surface of a pile between entirely visible and partially visible rocks. The feature visibility ratio is combined with commonly used 2D shape features to evaluate whether 2D shape features can improve classification accuracies to minimize overlapped particle error.

The size distribution as a function of weight of particles is an important measure of product quality in the mining and aggregates industries. When using manual sampling and sieving, the weight of particles is readily available. However, when using a machine vision system, the particle size distributions are determined as a function of the number of particles. In this paper we first show that there can be a significant weight-transformation error when transforming from one type of size distribution to another. We also show how the problem can be overcome by training a classifier and scaling the results according to calibrated average weights of rocks. The performance of the algorithm is demonstrated with results of measurements of limestone particles on conveyor belts.

Evaluation of Spherical Fitting as a technique for sizing iron ore pellets is performed. Size measurement of pellet in industry is usually performed by manual sampling and sieving techniques. Automatic on-line analysis of pellet size would allow non-invasive, frequent and consistent measurement. Previous work has used an assumption that pellets are spherical to estimate pellet sizes. In this research we use a 3D laser camera system in a laboratory environment to capture 3D surface data of pellets and steel balls. Validation of the 3D data against a spherical model has been performed and demonstrates that pellets are not spherical and have physical structures that a spherical model cannot capture.

Size measurement of pellets in industry is usually performed by manual sampling and sieving techniques. Automatic on-line analysis of pellet size based on image analysis techniques would allow non-invasive, frequent and consistent measurement. We make a distinction between entirely visible and partially visible pellets. This is a significant distinction as the size of partially visible pellets cannot be correctly estimated with existing size measures and would bias any size estimate. Literature review indicates that other image analysis techniques fail to make this distinction. Statistical classification methods are used to discriminate pellets on the surface of a pile between entirely visible and partially visible pellets. Size estimates of the surface of a pellet pile show that the overlapped particle error is overcome by only estimating the surface size distribution with entirely visible pellets.

Methods for non-destructive inspection of layered materials are becoming more and more popular as a way of assuring product integrity and quality. In this paper, we present a model-based technique using ultrasonic measurements for classification of thin bonding layers within three-layered materials. This could be, for example, an adhesive bond between two thin plates, where the integrity of the bonding layer needs to be evaluated. The method is based on a model of the wave propagation of pulse-echo ultrasound that first reduces the measured data to a few parameters for each measured point. The model parameters are then fed into a statistical classifier that assigns the bonding layer to one of a set of predefined classes. In this paper, two glass plates are bonded together with construction silicone, and the classifiers are trained to determine if the bonding layer is intact or if it contains regions of air or water. Two different classification methods are evaluated: nominal logistic regression and discriminant analysis. The former is slightly more computationally demanding but, as the results show, it performs better when the model parameters cannot be assumed to belong to a multivariate Gaussian distribution. The performance of the classifiers is evaluated using both simulations and real measurements.

An industrial prototype 3D imaging and analysis system has been developed that measures the pellet sieve size distribution into 9 sieve size classes between 5mm and 16+mm. The system is installed and operational at a pellet production plant capturing and analysing 3D surface data of piled pellets on the conveyor belt. It provides fast, frequent, non-contact, consistent measurement of the pellet sieve size distribution and opens the door to autonomous closed loop control of the pellet balling disk or drum in the future. Segmentation methods based on mathematical morphology are applied to the 3D surface data to identify individual pellets. Determination of the entirely visible pellets is made using a new two feature classification, the advantage being that this system eliminates the resulting bias due to sizing partially visible (overlapped) particles based on their limited visible profile. Literature review highlights that in the area of size measurement of pellets and rocks, no other researchers make this distinction between entirely and partially visible particles. Sizing is performed based on best-fit-rectangle, classified into size classes based on one quarter of the measured sieving samples, and then compared against the remaining sieve samples.

This project developed an industrial prototype 3D imaging and analysis system for measuring the size distribution of iron ore green pellets. The system is installed and operational at a pellet production plant capturing and analysising 3D surface data of piled pellets on the conveyor belt. It provides fast, frequent, non-contact, consistent measurement of the pellet sieve size distribution and opens the door to autonomous closed loop control of the pellet balling disk or drum in the future. Segmentation methods based on morphological image processing are applied to the 3D surface data to identify individual pellets. Determination of entirely visible pellets (non-overlapped) is made using a two feature classification strategy, the advantage being that this system eliminates the bias the results from sizing overlapped particles based on their limited visible profile. The system achieves what a number of commercial 2D fragmentation measurement systems could not satisfactorily achieve for the pellet producer, that is, accurate sizing of the green pellets.