Search

A machine vision sensor was developed for predicting deviations from the optimum amount of pitch in anode formulations using paste texture analysis. It could help operators mitigate the impact of the increasing variability of anode raw materials (coke and pitch). Paste samples were formulated in the laboratory using dry aggregate mixes obtained using two cokes having different properties and various amounts of pitch. These were imaged, formed into small cylindrical anodes, and baked to measure their density. A combination of image texture methods was used for extracting relevant paste textural features. The latter were then used as inputs of partial least squares regression models to predict deviations from the maximum baked density. Good prediction results were obtained. Furthermore, the sensor was able to detect when the paste was at the optimal amount of pitch for both cokes and to measure deviations from it.

A rapid and nondestructive machine vision sensor was developed for predicting the vibrated bulk density (VBD) of petroleum coke particles based on image texture analysis. It could be used for making corrective adjustments to a paste plant operation to reduce green anode variability (e.g., changes in binder demand). Wavelet texture analysis (WTA) and gray level co-occurrence matrix (GLCM) algorithms were used jointly for extracting the surface textural features of coke aggregates from images. These were correlated with the VBD using partial least-squares (PLS) regression. Coke samples of several sizes and from different sources were used to test the sensor. Variations in the coke surface texture introduced by coke size and source allowed for making good predictions of the VBD of individual coke samples and mixtures of them (blends involving two sources and different sizes). Promising results were also obtained for coke blends collected from an industrial-baked carbon anode manufacturer.

Reduction of pot draft in aluminum smelting cells is a promising approach for energy saving and waste heat recovery. The control of fugitive emissions (hazardous by-product gases escaping from cells to potroom) is investigated in a smelting cell with the reduced pot draft, down to half of the current level based on CFD simulations. Physical models with different length scales are created in order to directly resolve the very narrow escaping gaps on the pot shell in the physical model with smallest scale. A systematic analysis of the pot tightness is presented by considering various factors in the simulations, e.g., pot draft and hood placement. The results have shown that current pot structure, even within ideal operating conditions, fails to maintain 100% hooding efficiency under a 50% reduced pot draft. Two design modifications are proposed to enhance the pot tightness. An efficient sealing is observed when covering the lower half of the gaps between hoods. An estimation of the leaking hydrogen fluoride is made under different scenarios in order to quantitatively verify the modifications.

The new Sequential Multi-block PLS algorithm is applied for establishing multivariate specification regions jointly on multiple types of raw materials. SMB-PLS distinguishes between process variations associated with raw materials from other orthogonal sources of variations such as operating policies. Combinations of raw material properties not compensated by the control schemes are clearly identified. Multivariate specifications are required for these combinations to avoid negative impact on process performance and product quality. The method was applied to an industrial aluminum smelter. Bad combinations of raw material properties were identified and validated against process knowledge.

High quality baked carbon anodes contribute to the optimal performance of aluminum reduction cells. However, the currently decreasing quality and increasing variability of anode raw materials (coke and pitch) make it challenging to manufacture the anodes with consistent overall quality. Intercepting faulty anodes (e.g., presence of cracks and pores) before they are set in reduction cells and deteriorate their performance is therefore important. This is a difficult task, even in modern and well-instrumented anode plants, because lab testing using core samples can only characterize a small proportion of the anode production due to the costly, time-consuming, and destructive nature of the analytical methods. In addition, these results are not necessarily representative of the whole anode block. The objective of this work is to develop a rapid and non-destructive method for quality control of baked anodes using acousto-ultrasonic (AU) techniques. The acoustic responses of anode samples (sliced sections) were analyzed using a combination of temporal features computed from AU signals and principal component analysis (PCA). The AU signals were found sensitive to pores and cracks and were able to discriminate the two types of defects. The results were validated qualitatively by submitting the samples to X-ray Computed Tomography (CT scan).

The baked anode quality control strategy currently used by most carbon plants based on testing anode core samples in the laboratory is inadequate for facing increased raw material variability. The low core sampling rate limited by lab capacity and the common practice of reporting averaged properties based on some anode population mask a significant amount of individual anode variability. In addition, lab results are typically available a few weeks after production and the anodes are often already set in the reduction cells preventing early remedial actions when necessary. A database approach is proposed in this work to develop a soft-sensor for predicting individual baked anode properties at the end of baking cycle. A large historical database including raw material properties, process operating parameters and anode core data was collected from a modern Alcoa plant. A multivariate latent variable PLS regression method was used for analyzing the large database and building the soft-sensor model. It is shown that the general low frequency trends in most anode physical and mechanical properties driven by raw material changes are very well captured by the model. Improvements in the data infrastructure (instrumentation, sampling frequency and location) will be necessary for predicting higher frequency variations in individual baked anode properties. This paper also demonstrates how multivariate latent variable models can be interpreted against process knowledge and used for real-time process monitoring of carbon plants, and detection of faults and abnormal operation.

A metallurgical process generally processes some raw materials or ores into concentrates or primary metals using different technologies. However, the quality of the incoming raw materials has an effect on the process performance. Hence, engineers have to apply different set points in order to account or compensate for raw material quality variations. This is frequently performed using a set of univariate specifications. In this paper, a methodology to build multivariate specifications is presented and illustrated using a case study from an aluminum reduction smelter.

In this paper, the use of different image analysis techniques is investigated for predicting alumina content of anode cover materials used in primary aluminum smelters. This approach is proposed in to order to allow on-line estimation of alumina content for feedback control purposes, which is not currently possible due to the long time delays and limited number of samples that can be analyzed in the laboratory. Both color and textural features of various anode cover materials are extracted from digital RGB images, and partial least squares (PLS) regression models are developed for predicting alumina content from these features. Most alumina content prediction errors are within ± 2 σ of the X-ray fluorescence laboratory measurements, when larger variations are required by operators to make control decisions. Some challenges arising from the use of image analysis techniques for process control are also discussed.

Carbon anodes used in the production of primary aluminium have to be covered with a layer of particulate materials, a mixture of alumina and crushed electrolytic bath. Alumina composition analyses of this material are performed daily in laboratory on a very limited number of samples. However, analyses results from this small number of samples do not capture all the composition variations occurring on a daily basis. In this paper, alumina composition estimation of anode cover material, based on the use of image analysis and regression models is presented.