INTRODUCTION

The present application concerns a method and an apparatus for automatic post- cleaning of carbon anodes in aluminium production.

BACKGROUND

In aluminium production, large carbon anodes are lowered into molten electrolyte in reduction cells. The carbon anodes react with the oxygen from dissolved alumina in the electrolyte and aluminium and CO ₂ are produced. The carbon anodes are large rectangular blocks of carbon attached to an iron structure. Each carbon anode is fixed onto an anode hanger. The anode hanger consists of a steel yoke with four studs sticking into the top of the anode and an aluminium rod connected to the superstructure. The carbon anode is changed regularly due to wear.

A used anode is called a butt. It consists mainly of leftover carbon together with a substantial amount of frozen electrolyte and alumina on top. This frozen electrolyte needs to be removed in order to recycle the leftover carbon. The removal process is a partly automated process. An anode butts which has been through the cleaning process is called a clean butts. Used carbon anodes are in some cases, not sufficiently cleaned and must be manually inspected and cleaned in a post- cleaning process. After inspection and post-cleaning the anode butt is separated from the anode hanger and sent to recycling.

NO19853325 describes cleaning of anodes by the use of a robot. An optic- electronic device (camera) is used for controlling the robot. A sensor detects the profile of the leftover carbon during the cleaning and controls the robot. An impact tool, scraping tool or milling tool, water beam or pressurized air is used.

SUMMARY OF THE INVENTION

The present invention provides a system and a method for automation of post- cleaning of carbon anodes in aluminium production. The automated method and system provides electrolyte material detection on the used carbon anode and an advanced robotic system for removing this material from the used carbon anode.

In a first aspect the invention provides a system for post-cleaning of electrolyte from a used carbon anode in aluminium production, the system comprising:

- a robot with manipulator arm provided with a tool for removal of the electrolyte;

- a vision system comprising:

at least one 3D sensor for detection of a 3D shape of a surface of said carbon anode and

an analysis system for identifying the electrolyte on said carbon anode from the 3D shape of the surface;

- a sensor for detection of tool breakthrough in the electrolyte on said carbon anode; and

- a control system for controlling the robot manipulator arm and the tool based on information from the vision system and the sensor device for detection of tool breakthrough.

In an embodiment, the sensor for detection of tool breakthrough may comprise a force/torque (FT) sensor. Alternatively, the sensor for detection of tool

breakthrough may comprise using at least one of an audio sensor and an impedance measurement device.

The control system may comprise a force torque (FT) sensor for force-feedback control, wherein the force torque (FT) sensor detecting force interaction between the tool and the carbon anode. The control system may further comprise a motion planner for path optimalization of the trajectory of the robot manipulator arm and tool with respect to the surface of the carbon anode to be post-cleaned.

The at least one 3D sensor may comprise at least one of structural light cameras, stereo cameras, time of flight cameras and 3D laser scanners. The analyses system may provide colour differentiation and positioning of the electrolyte in a coordinate system related to the 3D shape of the surface of the carbon anode to be cleaned. The tool may be a mechanical tool selected from at least one of a chiseling tool, a grinding tool and a sandblasting tool. The tool holder may comprise at least one vibration damper.

In a second aspect the invention provides a method for post-cleaning of electrolyte from a used carbon anode in aluminium production, the system comprising a robot with manipulator arm provided with a tool for removal of the electrolyte, the method comprising:

- detecting a 3D shape of a surface of said carbon anode;

- analysing the 3D shape and identifying positions of the electrolyte on said carbon anode,

- controlling the robot manipulator arm with tool based on the 3D shape and the identified positions of the electrolyte, and

- detecting tool breakthrough in the electrolyte on said carbon anode for further control of the robot manipulator arm with tool.

In an embodiment, the method may further comprise detecting tool breakthrough in the electrolyte on said carbon anode by using a force-torque sensor (FT) arranged on an end-effector of the robot manipulator arm. Alternatively, detecting tool breakthrough in the electrolyte on said carbon anode may be performed by using an audio sensor or by using impedance measurements between the tool and the cleaned carbon anode.

The method may further comprise identifying the positions of the electrolyte on said carbon anode by use of color differentiation. Selecting at least one tool to be used in the post-cleaning of the carbon anode may be performed based on the identified positions of the electrolyte. The method may further comprise generating a point to point path for cleaning of the carbon anode optimized with respect to time. Generating the point to point path may be based on the method Rapidly- exploring Random Trees. A collision free path for the tool in view of the surface of the carbon anode to be cleaned may be generated. A number of paths may be simulated before initiating the post-cleaning process and deciding an actual path based on cleaning efficiency in view of time. The path to be followed by the tool during the cleaning process may be updated based on real time detection of the 3D shape of the surface of the carbon anode.

The automated system for post-cleaning of used carbon anodes (butt) provides a number of advantages in view of the manual prior art chiseling.

• Safer post-cleaning of used carbon node and increased HSE (Health,

Safety, Environment)

• More efficient post-cleaning process

• Faster response time and higher quality of used carbon node, as less electrolyte is left on the butts

• Fewer workers on the actual production line, reduced costs

• Increased productivity in the anode cleaning line

• Increased accuracy in the post-cleaning of butts

BRIEF DESCRIPTION OF DRAWINGS

Example embodiments of the invention will now be described with reference to the followings drawings, where:

Figure 1 a illustrates an automated robotic system for chiselling of used carbon node according to an embodiment of the present invention;

Figure 1 b and 1 c are flow charts of a cleaning operation according to an

embodiment of the present invention;

Figure 2a illustrates a tool holder and tool, a vibration damper and force sensor for the robot according to an embodiment of the present invention;

Figure 2b is a picture of a chiselling tool mounted on the tool holder with vibration absorbers, FT sensor and an adaptor plate connecting the tool to the robot arm according to an embodiment of the present invention;

Figure 3 schematically illustrates a flow chart for path planning of the chiselling operation according to an embodiment of the present invention;

Figure 4 shows a robot manipulator with a pneumatic hammer performing a chiselling operation according to an example embodiment of the present invention; Figure 5 shows a picture of an experimental result of chiselling performed by the robot manipulator from Figure 4 on carbon anode with electrolyte; Figure 6 shows a graph of force (N) as a function of time (s) as detected by a force sensor of the robot shown in Figure 4, showing breakthrough detection to the carbon butts according to an example embodiment of the present invention; and Figure 7 shows a pneumatic grinder mounted on a robot manipulator for testing out the concept of the present invention.

DETAILED DESCRIPTION

The overall concept of the present invention is illustrated in Figure 1 a. The robotic system in Figure 1 a provides an integrated system with subsystems for detection and control. Detection of a 3D shape on the clean carbon butt, detection of electrolyte on the clean carbon butt, as well as detection of tool breakthrough in the electrolyte deposited on the butt are provided. Information from the detection systems are input to the control systems controlling a manipulator arm on the robotic system.

In Figure 1 a the robot manipulator arm of the robotic system is provided with an appropriate tool for removal of electrolyte from a used clean carbon butt. A vision system is provided for data collection and analyses. A tool system provides calculation of position of electrolyte pits in the surface of the used clean butt and enables detection of bath materials and choice of tool. A motion planner generates a trajectory for the robot manipulator. Further the robotic system is provided with a robot controller. The robot controller provides functions for force and position control for the robot manipulator as well as tool mounting. Electrolyte breakthrough is also detected by the tool system. A user interface enables control of the robotic system by an operator. However, when the clean carbon butt is positioned for post cleaning, and the process initiated, the post cleaning procedure is fully automated and performed by the robotic system.

The robotic system comprises a robot manipulator with an appropriate tool. The robot controller includes a motion planner, path follower and force controller (hybrid controller). The robot manipulator may be a commercial available industrial manipulator. Industrial manipulators are supplied with commercial controllers for joint and Cartesian control of the manipulator. The motion planner will operate independently of the commercial controller. The hybrid controller will generate joint references to the commercial controller of the manipulator.

The motion planner will generate the path for the manipulator end-effector based on a detection of the unclean parts of the butts. The detection system may consist of different types of sensors. The path follower will ensure optimal motion between the unclean areas of the butts and the force controller will control the manipulation.

Vision system and identification of uncleaned areas

The vision system is provided by one or more 3D sensors. Examples of such sensors are structural light cameras, stereo cameras, time of flight cameras or 3D laser scanners. The measurements are distributed over a standard interface, such as CameraLink or Gigabit Ethernet to a PC for further analysis where commercial available software like Scorpion Vision from Tordivel can be used. The result is a 3D view of the surface of the butt to be post-cleaned. The positions of the unclean areas are identified by use of color differentiation and positioning in a Cartesian coordinate system related to the 3D shape of the butt to be cleaned. The 3D shape of the butt and the identified unclean areas are input into a motion planner system for planning a movement of the robot manipulator. The movement of the robot manipulator is planned in order to achieve a fast and reliable post-cleaning of the butt.

Vision measurements are preferably performed by the vision system in intervals during the cleaning operation to check whether the anode is sufficiently clean. After a chiseling operation, the anode is scanned by the vision system to check whether all electrolyte is removed or if another tool should be used to clean the anode. The vision information of the 3D view of the surface is thus used as feedback to the tool system to ensure a successful removal of the electrolyte. Scanning of the anode by the vision system is normally not performed during the cleaning operation itself, as dust and debris from the cleaning operation destroys the quality of the vision measurement. The 3D shape of the butt is also input to the tool system for choice of tool/tools to be used for cleaning the butt. Figure 1 b shows a flow chart of the post-cleaning process of electrolyte from a used carbon anode. The anode is arranged in position in an anode holder. The anode is scanned by the vision system creating a 3D view of the anode. The 3D vision data from the vision system are analysed in an analysis system. The areas with electrolyte on the anode and the contour of the anode are detected, and a path created between the electrolyte covered areas. If no areas with electrolyte exists on the carbon anode, the cleaning is done. If areas with electrolyte exist on the anode, a tool is selected for the cleaning operation. Then, local waypoints are created based on the identified electrolyte area and the tool selected for the cleaning operation. The cleaning of the area is then performed. After cleaning of an area is performed by the tool, the anode is scanned again by the vision system obtaining a new 3D view of the anode. The new 3D view is analysed to detect remaining electrolyte and the contour of the anode. If more areas with electrolyte exist on the anode, a tool is again selected, locale waypoints created based on the identified remaining electrolyte areas and the area cleaned. The process is repeated until the carbon anode is cleaned.

Figure 1 c shows a flow chart of an example embodiment of the cleaning process of an electrolyte area on the carbon anode. If the area is cleaned, the area is done. If the area is not cleaned, the tool is moved to the waypoint for this area as previously created in the scanning and analysing steps. The cleaning with the tool is started when the tool is in position. The tool approaches the carbon anode surface.

If the tool is a grinding tool, the grinding tool is set to follow the contour of the anode to the next waypoint. Grinding is performed until the next waypoint is reached. When the next waypoint is reached, the tool is retreated. If the area is not cleaned, the tool is then again moved to the waypoint.

If the tool is a chisseling tool, chiselling is performed until break-through is detected. Then the chiselling tool is retreated. If break-through is not detected within a specified interval, the tool is retreated and the anode once again approached. An alternative tool may also be selected and the anode again approached with the alternative tool. If the area is identified as clean, the area is done. As explained above, the area to be cleaned is scanned with the 3D vision system to check whether the desired cleaning result was obtained or not. Cleaning is performed until the desired cleaning result is achieved.

Tool system

The robotic system comprises a robot manipulator with an appropriate tool for removal of electrolyte from the clean butts. The tool is mounted in/on a tool holder arranged on an end-effector of the robot manipulator. The manipulation may be performed by a mechanical tool. The mechanical tool may e.g. be a chiseling tool, a grinding tool or a sandblasting tool. The chiseling tool may e.g. be a pneumatic chiseling tool and the grinding tool may e.g. be a steel grinding tool (brush).

The choice of tool to be used for cleaning the butt is made by the tool system after an analysis of the 3D view of the surface of the butt input from the vision system. The 3D shape of the butt to be post-cleaned is identified and the positions of the unclean areas of the butt identified by use of colour differentiation. The choice of tool will depend on the amount of electrolyte on the butts and the complexity of the unclean areas of the butt. A predefined decision tree built on experience is used to identify the best tool and this will be part of the tool system. The decision tree applies a knowledge database which is built up continuously based on experience from previous cleaning operations. The knowledge database comprises sensor information from previous cleaning operations and linked with human knowledge. Operator based choice of tool will depend on the contaminated area and this experience will be linked to the images taken of the contaminated area. The decision tree may be a look up table comparing real time sensor information with content in the database.

The operation can be performed using a number of tools including a tool changing system, and can be performed using a single tool or a compounded multi-tool. The compounded multi-tool may be specially designed for the purpose and may comprise a number of tool elements as e.g. chisel, grinder and brush. Figure 4 shows an example where a pneumatic hammer is connected to a damper system to the end-effector of the robot manipulator. Figure 5 shows an example where a pneumatic grinder is mounted on the manipulator arm of the robot. The pneumatic grinder performs a brushing operation for post-cleaning of the butt. Tests were performed with these robots, where the manipulator arm followed a pre-programmed path. The tests showed that for the specific case the electrolyte was successfully removed within seconds. The robot was of the type industrial manipulator supplied ABB. Other types of industrial manipulators may also be applied.

The speed of the cleaning process depends on how much electrolyte that is to be cleaned, the efficiency of the tool, the desired quality of the cleaning result, but also on the type of sensor used and the measurement frequency. Today, the manual cleaning process requires several man-labour years. The use of the robotic cleaning system will require less than one man-labour year in an average aluminium work.

Today, post-cleaning is manual and highly dependent on the operator experience. Layers of dust cover the used carbon node, and frozen electrolyte lies under the dust.

The post-cleaning is today performed as a manual process by use of heavy manual chiseling tool where some unclean areas are detected by visual inspection, and the rest by tactile detection as frozen electrolyte feels softer than the carbon forming the butts. When the chiseling tool breaks through the frozen electrolyte, this is clearly felt by the operator and the operator stops the chiseling. This feeling experienced by the operator in manual post-cleaning is transferred to the robotic solution in the present invention by use of force-feedback sensors.

Figure 2a shows an example embodiment of a tool holder. The tool holder is in this embodiment provided with a force/torque sensor (FT sensor) providing force- feedback for detection of electrolyte breakthrough. The FT sensor is of an industrial type and robust with regards to harsh environments. The sensor is mounted on the end-effector of the tool holder such that the tool forces are aligned with the sensor elements. Adapter plates connect the FT sensor to the robot end- effector. The adapter plates are provided between the FT sensor, tool holder and vibration damper. The adapter plates are securely fixed on the robot end effector. The FT sensor is applied in feedback force control of the robot. This will be explained later. Figure 2b is a picture of a chiselling tool mounted on the tool as in Figure 2a holder with vibration absorbers, FT sensor and an adaptor plate connecting the tool to the robot arm.

Heavy vibrations from the tools, and in particular from pneumatic tools, may cause frequent errors to occur on the robot controller resulting in a poor post cleaning result. Vibration dampers are provided on the robot end-effector to remove these errors and enable precise control of the robot. In the embodiment shown in Figures 2a and 2b, the vibration dampers are in the form of parallel aluminum plates with rubber based vibration absorbers joined by machine screws. Other fastening means than machine screws may also be used. The parallel aluminum plates, , are attached between the adapter plates and the tool end-effector such that the parallel aluminum plates are aligned perpendicular to the tool forces. The plates may be attached by the use of machine screws. The parallel aluminum plates in the example embodiment of Figure 2b was 88 cm ² and 1 .5 cm thick. Other customized vibration dampers may also be envisaged.

Experiments have been performed with and without vibration dampers. Vibrations from the pneumatic tools caused the robot to stop due to overload. Non- observance of these errors resulted after the vibration dampers were added.

Motion planner

The 3D information of the surface of the butt is input into the motion planner for planning the trajectory of the robot arm to which the tool is attached. The motion planner is provided by a path planner for motion of the tool along the surface of the butts. Concepts span from manual to automatic path planning with obstacle avoidance. The manual path planning will be performed in case the system fails to detect unclean areas. There will be a continuous update of the path to be followed by the tool during the cleaning process based on real time information from the vision system. The vision system identifies the progress of the cleaning process. The path planning considered will be in two parts:

- a first part for generation of paths for cleaning of unclean areas and

- a second part for optimal path planning of robot motion between the

unclean areas.

Generation of paths for cleaning of unclean areas

The 3D information of the surface of the butt provided by the vision system identifies the unclean areas of the butts. Path planning can then be performed by analysis of the 3D information and generation of point to point paths. The cleaning process will thus be a point-to-point motion based on continuous update from the vision system, FT sensor and tool system module. This will be explained further below.

Optimal path-planning for motion between the unclean areas:

The motion of the robot arm with the tool between the unclean areas can be considered as a point-to-point path planning task in an environment with or without obstacles.

In order to post-clean the butt, the tool is brought forward to the unclean areas of the butt. Trajectory optimization with respect to time is provided by the path planner in order to achieve an efficient and reliable post-cleaning process. The point-to-point path planning methodology may be based on Rapidly-exploring Random Trees (RRT). The method constructs a collision free path by using randomly sampled points from the robot configuration space, where each point and path segment is tested against the collision space. Open source code from Lavalle can be applied for testing path concepts. Different paths may be simulated by the motion planner system before the actual post-cleaning process is initiated for the robot manipulator. The actual path chosen for each post-cleaning process is decided based on cleaning efficiency in view of time.

The robot configuration space is given by the robot structure and kinematics of the actual robot being used. Collision space is defined and detected by the vision system. Robot controller

Accurate control of the robot manipulator ensures good quality of the post-cleaning of the butts. The predefined path planned by the motion planner should be followed closely in order to enable fast and reliable removal of the electrolyte. Also, the identified unclean areas may be small. Industrial robots have very high accuracy (typically 0.1 mm) and since this method is based on use of such robots this accuracy will be achieved in this process. The accuracy is thus higher than a human operator can achieve.

During the post-cleaning process the robot manipulator is controlled to ensure the robot manipulator follows the predefined path. Path following is achieved by use of motion control based on feedback from the position and velocity measurements of the motion of each robot joint. This functionality is supplied by the supplier of the robot controller.

The manipulation part of the operation will be based on force interaction between the tool and the carbon anode and/or a combination of force and motion control. The force interaction is detected by the FT sensor and the control methods will depend on the choice of tool. A chisseling tool will apply position control of the robot and on/off force control. A grinding tool will apply continuous position and force control in cascade.

Learning strategies as reinforcement learning will be employed in the force feedback loop and in the generation of the decision tree for choice of tool.

Detection of breakthrough

Several methods can be applied to detect breakthrough of electrolyte, as when the tool tip has reached the carbon. Examples of methods include use of a FT sensor, audio sound or impedance measurements. These methods will be explained further below.

FT sensor

FT sensor measurements are applied to detect the variations in opposing forces which will occur depending on contact with carbon or electrolyte. This is due to the difference in density for the two materials. Analyzing of these force measurements will be used to detect when the cleaning tool has a breakthrough through the electrolyte. The information from the force measurement is input to the robot controller. The force measurement procedure is schematically presented in Figure 3. If the analyses of the force measurement result in the presence of electrolyte, the robot controller continues the chiseling operation. If the analyses of the force measurement result in the presence of carbon, the robot controller stops the chiseling operation. When the chiseling operation stops, the area of the butt which has been chiseled is inspected by the vision system. If the vision system identifies the butt as clean in this area, this information is input to the robot controller, which selects the appropriate tool and moves the tool end-effector to the next position of the pre-defined trajectory and starts the chiseling operation in this area.

An example of detection of electrolyte breakthrough by the use of a force/torque sensor is shown in Figure 7. The curve in Figure 7 was produced by using the robot manipulator with pneumatic hammer as shown in Figure 4. Figure 7 shows a graph of force (N) as a function of time (s) as detected by the force sensor.

Breakthrough detection to the carbon butts is shown with a dip in the curve at approximately 20 seconds, 39 seconds and 57 seconds when the force goes to zero and becomes negative. The dips in the curve thus show the presence of carbon.

Audio

A different sound pattern is detected when the tool tip breaks through the electrolyte and hits the carbon. The sound is detected by a microphone and input to the robot controller.

Impedance

Impedance measurements can be applied by measuring deviations in electrical resistance between the tool tip and the electrolyte/carbon materials of the butt. The electrolyte and the carbon materials have different electrical properties. The result of the impedance measurement is input to the robot controller. Combination of methods

The methods described above can be combined in different settings using Kalman filter or other estimation methods to improve the detection of breakthrough. Or the breakthrough detection methods can be applied in cascaded, parallel or serial set of control loops. The choice of methods depends on the tool. For example audio is only applied when using chiseling tool. FT sensor is applied in all solutions.

Cascaded control loops are applied in cases where contour following is required. An advantage with use of cascaded control loops is implementation of several control loops based on different measurements, for example velocity and force. Parallel control loops enables use of different measurements in parallel that both are able to change the system state. For example, audio and force detection where the first loop detecting breakthrough changes the state of the system. Serial control is when one control loops gives the reference to the next loop; e.g. velocity control and breakthrough detection.

Also in the embodiments using audio sound, impedance measurements and in the combination of methods, the vision system inspects the area when carbon is detected to ensure the area is clean before the tool is moved to the next point of the predefined trajectory, as explained for the FT sensor.

The invention is of course not in any way restricted to the embodiments described above. On the contrary, many possibilities to modifications thereof will be apparent to a person with ordinary skill in the art without departing from the basic idea of the invention such as defined in the appended claims.