Estimation of Phases Levels in a Teniente Converter Using Machine Vision Max Schaaf ∗ Zacar´ıas G´ omez ∗∗ Aldo Cipriano ∗ ∗
Pontificia Universidad Cat´ olica de Chile, Santiago, CO 307 C22 Chile (e-mail:
[email protected],
[email protected]) ). ∗∗ Potrerillos Smelter, Salvador Division Codelco-Chile (e-mail:
[email protected])
Abstract: The Teniente converter (TC) is used for smelting and converting copper concentrates. The TC process is very complex given that it combines continuous input flows with intermittent product extraction in an autogenous operation. Liquid levels inside the converter are important variables for deciding the operation of the overall process, but in the high-temperature environment of the TC, their measurement presents a challenge. Although various measuring techniques have been employed for the task, the traditional approach using a test bar remains the established one. This article describes an instrument for estimating the phase levels in a TC based on machine vision. The method employs a digital camera to capture images of the test bar. A video image processing algorithm extracts several of the surface characteristics of the bar in order to make the estimate. The main contribution of this work is to estimate criteria free of human intervention and directly connect the image processing algorithm to the smelter’s control network. Keywords: digital vision; image processing; instrumentation; Teniente converter; smelter. 1. INTRODUCTION
The Teniente converter (TC) is one of the principal existing technologies for primary copper smelting. It combines the smelting and converting of copper concentrate in a single operation to produce white metal (75% Cu) and slag (8% Cu), which are separated using gravity methods. Material may be charged to the TC through tuyeres, a Garr gun (top feed) or the converter mouth. Dry concentrate with a humidity of 0.2% is injected through four tuyeres (Bergh et al., 2002). The TC process is autogenous because of the heat generated by the exothermic chemical reactions is exploited to melt the concentrate. To improve efficiency, oxygen-enriched air (34%) is injected through 43 tuyeres at a pressure of about 2 atmospheres. To prevent the buildup of material in the tuyeres that could hamper the injected air flow, they are cleaned by two tuyere punchers known as MAPUCOS. Siliceous flux, revert and wet concentrates are introduced into the top part of the TC through the Garr gun feeder and distributed inside the converter by air flow, while slag from downstream processes is charged intermittently via the converter mouth. As a result of chemical reactions inside the TC, the charged material separates into two liquid phases, white metal and slag. The former contains an average of 75% Cu, 3% Fe and 21% S and the latter is composed of oxides, fayalite, magnetite, free silicates and gangue components.
Since the density of white metal is greater than that of slag, the former settles at the bottom of the converter. The two products are extracted intermittently through tap holes at either end of the converter. To facilitate selective extraction these outlets are located at different heights. The white metal is extracted at a temperature of 1220 o C into a bleeder passage or channel and then to ladles that are conveyed by crane to Peirce-Smith converters (PSC), the next stage in the overall process. The slag is extracted at 1240 o C into bleeder channels and either flows directly or is poured into ladles and transferred by crane to slag cleaning furnaces (SCF) for further processing to recover the entrapped copper. Average copper content of the slag is 8%. The gases formed during the conversion process, mainly containing sulfurous anhydride, are continuously extracted through the converter mouth at a temperature of 1260o C. Together with dust particles they are collected by a watercooled converter hood and sent first to a gas treatment circuit and then an acid plant. Due to the autogenous nature of the TC process and the lack of sensors for measuring certain key variables such as chemical composition and liquid phase levels, control of the converter’s operation is highly complex. Several studies have employed expert control and continuous model predictive control in an attempt to reduce variability in the controlled variables (Bergh et al. 2005; Gajardo, 2008). Phase levels are important variables because they directly characterize the products contained in the converter. This
IFACMMM 2009. Viña del Mar, Chile, 14 -16 October 2009.
information is crucial for determining when and how much white metal and slag can be extracted and sent to the Peirce-Smith converter (PSC) and clean slag furnace (CSF), respectively. Furthermore, TC operation must avoid low white metal level in order to prevent foaming. Halting the process due to the occurrence of this phenomenon causes a decrease in productivity. Phase level measurement in a Teniente converter is very complex because of the high temperatures involved. Some researchers have proposed non-invasive techniques to measure the phase levels on-line. B¨ orger (1996) presents a non-invasive method for the estimation of the temperature and phase levels based on temperature measurement at the converter’s shell. Rojas and Garret´ on (2003) present a solution for liquid level measurement in furnaces and converters based on a mechanical wave. The wave is transmitted to the interior of the converter through one of the blowing tuyeres and transducers detect the amplitude of the signal. Phase levels are estimated from the propagation time. Recent developments in instrumentation include a sensor that performs on-line determination of liquid phase levels inside the converter based on electroresistive measurements taken by a set of electrodes mounted on the slag tap hole wall. The No 1 Teniente Converter at the Caletones smelter has been using this system since September 2006 (IM2, 2007). However, these non-invasive methods are not widely employed in the industry and the traditional test bar measurement method continues to be employed by smelter operators. This invasive technique consists in introducing a metallic test bar through the observation hole situated at the top of the converter. After some minutes, it is withdrawn. Due to its interaction with the two liquid phases, the surface of the bar will have changed. A worker then visually compares it with a standard ruler to estimate the phase levels. This information is communicated to the control room and manually inputted to the distributed control systems. The tuyere injection system produces different agitation zones that perturb the phase levels (Valencia et al. 2006). The observation hole is situated next to the converter mouth and far from the tuyere zone to reduce the influence of agitation inside the converter. The main disadvantage of the test bar method of estimating phase level is that each worker will have a different observation criterion. This paper describes the estimation of liquid phase levels based on machine vision to replace human observation. In this system, information is captured with a digital camera and an image processing algorithm then extracts the relevant characteristics for making the estimate. Section 2 describes this capture method, the principal test bar characteristics and the image processing algorithm. Section 3 presents an evaluation of the method based on video sequences of the bar after it is withdrawn from
IFACMMM 2009. Viña del Mar, Chile, 14 -16 October 2009.
Fig. 1. Schema for image acquisition. the smelter converter. Conclusions and future research are discussed in section 4 2. PHASES LEVELS ESTIMATION BASED IN DIGITAL VISION The traditional phase level measurement method is based on changes in a test bar’s surface due to interaction with the white metal and slag phases inside the converter. The zone of the bar in contact with the slag is coated with material, while the zone in contact with white metal is corroded. The new instrument proposed here uses a digital camera to capture a video sequence of the test bar after it is withdrawn from the converter. In this application, the camera is placed perpendicular to the test bar with an unobstructed field of view (see Fig. 1). Upon its removal from the converter, the bar emits high levels of light radiation as it cools to the surrounding temperature. The incandescent bar displays zones of different intensity. The zone in contact with white metal emits light of greater intensity than the zone in contact with slag and for a longer period. Figure 2 shows two images of a test bar at different moments together with their respective intensity profiles. The profiles describe how radiation from each zone decreases during the cooling process. These properties of the incandescent bar are exploited to develop an image processing algorithm for estimating each phase level. The algorithm has three stages: image integration, contrast adjustment and transition detection (see Fig. 3). In the first stage, the algorithm captures video sequences from the digital camera. The video frames are integrated right after the bar is taken out of the converter to increase the contrast between the different zones and reduce disturbances such as movement of the bar and momentary obstructions in the field of view. The sum of the images gives an image result in which the intensity is out of range. In the second stage, therefore, the contrast must be adjusted. In the transition detection stage the algorithm attempts to locate changes of intensity between the different zones. The detection of transition zones yields three positions ia , ib and ic in pixels:
Fig. 2. Test bar image at two different moments, with corresponding intensity profiles.
Fig. 4. Estimation of white metal level: close-up field of view (test No 2). 3. RESULTS The method described in the previous section was tested on the basis of seven video sequences recorded during TC operation at a smelter from different fields of view.
Fig. 3. Machine vision schema. ia : Lower position of the test bar. ib : Transition position between white metal and slag zone. ic : Position of maximum slag influence. Using positions ia , ib and ic the white metal level Lwm and slag level Lsl are calculated as: Lwm = (ib − ia ) · ∆L = kwm · ∆L
(1)
Lsl = (ic − ib ) · ∆L = ksl · ∆L
(2)
where kwm and ksl represent the two phase levels in pixels. ∆L is the instrumental resolution, a function of the digital camera’s field of view (F OV ) and CCD resolution. The latter is defined by the number of vertical nj and horizontal ni pixels. Since the number of horizontal pixels is greater than the vertical ones (ni ≥ nj ), the camera is rotated 90o in order to minimize estimation error. We then have ∆L =
F OV [m] ni
(3)
For example, using a field of view F OV = 3 [m] and a camera with ni = 640 [pixel], then the resolution is ∆L = 4.7 [mm]. When the phase level estimation is completed the results are sent to the Open Process Control (OPC) server, thus avoiding manual inputting of the results to the control net.
The application developed to implement the algorithm estimates the phase levels and displays them as video output (Figs. 4 and 5), shown here in close-up FOV (Fig 4) and extended FOV (Fig 5). Figure 4 presents the results using close up field of view, while figure 5 presents the result using extended field of view. Since the light radiation intensity decreases as the test bar cools, variations are produced in the phase level estimates. An error analysis is presented to evaluate the estimation quality on the basis of the maximum kwmmax and minimum kwmmin levels estimated during a period of 27 seconds. Thus, the error percentage is calculated as the ratio of the maximum variation of the estimates to their average: (kwmmax − kwmmin ) %error = 2 · · 100 (4) (kwmmax + kwmmin ) Table 1 summarizes the error percentages for the seven video sequence phase level estimation tests. The method thus yields good results in these preliminary tests, with an error of less than or equal to 3%. The variation in the white level estimation is caused by the diffuse transition zone. The comparison between close-up and extended fields of view show that the error is not associated with FOV. The absolute error e is smaller in close-up, however, because the resolution ∆L is also smaller. e = (kwmmax − kwmmin ) · ∆L (5)
IFACMMM 2009. Viña del Mar, Chile, 14 -16 October 2009.
tance of the Salvador Division of Codelco in carrying out this study. REFERENCES
Fig. 5. Estimation of white metal level: extended field of view (test No 5). Table 1. Error percentages for seven phase level estimation tests. Test No
kwmmax [pixels]
kwmmin [pixels]
variation [pixels]
% error
1 2 3 4 5 6 7
430 432 213 428 295 198 126
427 432 210 421 288 197 124
3 0 3 7 7 1 2
0.7 0.0 1.4 1.7 2.4 0.5 1.6
4. CONCLUSION Machine vision as a method for phase level estimation is a robust technique due to the fact that the video sequence integration reduces the error in the estimate. It should be noted that the digital camera’s exposure adjustment must be used to attenuate the high level of radiation and reduce the influence of ambient illumination. Briefly, the principal advantages of the artificial vision algorithm are: • Direct human observation of test bar is eliminated, thus avoiding risk of ocular damage due to high levels of light radiation. • Resolution for phase level estimation is higher. • White metal level estimate is delivered directly to smelter control network. The results presented here are currently under analysis with a view to their possible implementation at a smelter. ACKNOWLEDGEMENTS The first author is grateful for grants received from the Pontificia Universidad Cat´ olica de Chile and for the assis-
IFACMMM 2009. Viña del Mar, Chile, 14 -16 October 2009.
B¨ orger A. (1996). Estimaci´on de temperaturas y niveles de fases en el convertidor tipo Teniente, mediante mediciones sobre el manto del convertidor usando t´ecnicas no invasivas. Universidad T´ecnica Federico Santa Mar´ıa. Bergh, L., Chacana, P., Achurra, G., and Delgado, P. (2002). Improvements in the control of the injection system of copper concentrate in Teniente converters. Minerals Engineering, 15(5), 369 – 372. Bergh L.G. , Chacana P., and Carrasco C. (2005). Control strategy for a Teniente converter. Minerals Engineering, 18, 1123–1126. Gajardo M. (2008). Multivariable control application at Codelco norte’s Teniente converter No 2. In proceeding I International Congress on Automation in the Mining Industry, 335–342. Instituto de innovaci´on en miner´ıa y metalurgia (IM2) (2007). Sensor de niveles para Convertidor Teniente. (http://www.im2.cl/listado productos.htm). Rojas L. and Garret´on A. (2003). System for noninvasive online discrete measurement of phases levels in converters or pyrometallurgical furnaces. United State Patent 6594596. Valencia A., Rosales M., Paredes R., Leon C., and Moyano A. (2006). Numerical and experimental investigation of the fluid dynamics in a Teniente type copper converter. International Communications in Heat and Mass Transfer, 33(3), 302–310.