Unmanned crown block accurate positioning system based on multi-source vision fusion

Abstract

The invention relates to the technical field of electrolytic aluminum crown block positioning, and discloses an unmanned crown block accurate positioning system based on multi-source visual fusion. The system executes the following operations through the central control module: establishing an electrolytic aluminum workshop three-dimensional coordinate system, and setting a first type of calibration points and a second type of calibration points according to a preset distance in the longitudinal and horizontal coordinate directions respectively; an encoder is used for collecting moving track data of the crown block in the three-dimensional coordinate system, and a crown block moving track set is formed; performing error accumulation analysis based on the track set, and identifying a plurality of error accumulation regions; a plurality of groups of laser distance measuring sensors are mounted on the anode twisting device; and when the anode twisting and pulling device descends to a preset height, the multiple sets of laser distance measuring sensors are started to collect real-time distance measuring data. According to the system, through multi-source data fusion and error analysis, the positioning reliability of the unmanned crown block in a complex workshop environment is improved, and the high-precision requirement of electrolytic aluminum production for crown block operation is met.

Description

Technical Field

[0001] This invention relates to the field of positioning technology for electrolytic aluminum cranes, specifically to a precise positioning system for unmanned cranes based on multi-source visual fusion. Background Technology

[0002] In the electrolytic aluminum production process, unmanned overhead cranes undertake critical tasks such as anode handling and replacement, and their positioning accuracy directly affects the operational stability and production efficiency of the electrolytic cells. Currently, the electrolytic aluminum workshop environment is complex, with various adverse factors such as high temperature, dust, and electromagnetic interference, placing extremely high demands on the reliability and accuracy of the unmanned overhead crane positioning system.

[0003] Current unmanned overhead crane positioning technologies mostly rely on single sensors, such as encoder positioning, which calculates position information by collecting displacement data during crane operation. However, encoders are susceptible to vibrations from crane operation and track wear over long-term use, leading to a continuous accumulation of positioning errors. This is especially problematic in scenarios like aluminum electrolysis workshops where continuous operation is required for extended periods. As the operation time increases, the error accumulation problem becomes more pronounced. When the error exceeds the allowable range, it can cause collisions between the crane and equipment such as electrolytic cells and anode components, resulting in equipment damage and disrupting normal aluminum electrolysis production.

[0004] Some positioning systems employ visual sensors for assistance, but the high temperatures in aluminum smelting workshops generate significant heat radiation, and dust adheres to the sensor lenses. These factors interfere with the normal operation of the visual sensors, leading to a decline in the quality of the acquired image information and an inability to accurately extract the feature information required for positioning, thus affecting positioning accuracy. Furthermore, traditional positioning systems lack effective identification and targeted processing mechanisms for error accumulation areas, making it difficult to adjust positioning strategies based on the actual conditions of the overhead crane's trajectory. They can only rely on periodic manual calibration to correct errors, increasing labor costs and impacting production efficiency due to downtime during calibration. In actual production, the tight production pace in aluminum smelting workshops makes it difficult to strictly control the intervals for manual calibration, further exacerbating the problem of unstable positioning accuracy and failing to meet the high-precision, high-reliability operational requirements of unmanned overhead cranes. Summary of the Invention

[0005] The purpose of this invention is to provide a precise positioning system for unmanned overhead cranes based on multi-source visual fusion, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a precise positioning system for unmanned overhead cranes based on multi-source visual fusion, the system comprising: Establish a three-dimensional coordinate system for the electrolytic aluminum workshop, set first-type calibration points at preset intervals in the vertical axis direction, and set second-type calibration points at preset intervals in the horizontal axis direction; The crane's trajectory data in the three-dimensional coordinate system is collected by the encoder to form a set of crane trajectories; Error accumulation analysis was performed based on the set of overhead crane operating trajectories to identify multiple error accumulation areas; Multiple sets of laser rangefinders are installed on the anode torsion pulling device, each set containing at least two laser rangefinders; When the anode torsion device descends to a preset height, the multiple sets of laser ranging sensors are activated to collect real-time ranging data.

[0007] Preferably, establishing the three-dimensional coordinate system of the electrolytic aluminum workshop includes: A first-class calibration point is set every 8 electrolytic cells in the vertical axis direction, and the spacing between adjacent first-class calibration points is a preset value; Two second-type calibration points are set at preset intervals along the horizontal axis; The mechanical spring-loaded structure ensures that the positioning error of the first type of calibration point and the second type of calibration point is less than a preset threshold.

[0008] Preferably, the step of collecting the crane's trajectory data in the three-dimensional coordinate system via the encoder includes: The position coordinates of the overhead crane in the three-dimensional coordinate system are recorded in real time, and the position coordinate changes of the overhead crane at different times are stored. The operating error between adjacent calibration points is calculated based on the overhead crane's operating trajectory data.

[0009] Preferably, the error accumulation analysis based on the set of overhead crane operating trajectories includes: The overhead crane running trajectory set is segmented to obtain multiple running trajectory segments; Calculate the cumulative error for each trajectory segment and divide the trajectory into multiple error accumulation regions based on the magnitude of the cumulative error.

[0010] Preferably, the installation of the multiple sets of laser ranging sensors includes: At least two laser rangefinders are installed on each anode torsion device; For the double-torsion crane, a total of four sets of laser rangefinders are installed; Fix the installation position and measurement direction of each laser rangefinder.

[0011] Preferably, the step of activating the multiple sets of laser ranging sensors to collect real-time ranging data includes: When the anode torsion device descends to a preset height from the residual electrode guide rod, the distance measurement is activated. Continuously collect real-time measurement values ​​from each laser rangefinder and record the deviation data between each measurement point and the standard position.

[0012] Preferably, the central control module further includes the ability to execute: The average deviation value is calculated based on the measurement data from multiple sets of laser rangefinders. Adjust the position of the overhead crane in the vertical axis direction based on the average deviation value; After the vertical coordinate adjustment is completed, the crane angle is adjusted according to the differences in the measurement values ​​of each sensor.

[0013] Preferably, adjusting the position of the overhead crane in the vertical axis direction includes: Calculate the average value of the measurements from the first and second groups of laser rangefinder sensors; The adjustment distance is determined based on the difference between the average value and the standard value, and the crane is controlled to move the corresponding distance along the vertical axis.

[0014] Preferably, adjusting the crane angle includes: Compare the differences in measurement values ​​between two laser rangefinders on the same torsion pull device; Calculate the angle that needs to be adjusted based on the difference in measured values, and control the crane to rotate by the corresponding angle.

[0015] Preferably, the central control module further includes the ability to execute: Adjust the horizontal coordinate position based on the difference in measurement values ​​between two laser rangefinders on the same torsion device; For a double-torsion crane, adjust the horizontal coordinate position of the two torsion devices respectively; Repeat the measurement and adjustment until all measurements reach the preset accuracy range.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This unmanned overhead crane precise positioning system based on multi-source vision fusion establishes a three-dimensional coordinate system for the electrolytic aluminum workshop and sets first-type and second-type calibration points in the vertical and horizontal directions, respectively. This provides a unified and accurate spatial reference framework for crane positioning, enabling the crane's position information during operation to be calculated based on a clear coordinate benchmark. This avoids positioning deviations caused by inconsistent reference benchmarks and lays the foundation for the accuracy of subsequent positioning data.

[0017] In terms of positioning data acquisition, the system collects overhead crane trajectory data through encoders and forms a trajectory set, which comprehensively records the crane's motion process in a three-dimensional coordinate system, providing rich data support for subsequent error analysis. Based on the trajectory set, error accumulation analysis is performed and multiple error accumulation areas are identified. This allows the system to proactively discover areas where positioning errors tend to concentrate during crane operation, rather than passively waiting for errors to exceed allowable limits before taking action. This proactive identification method allows the system to understand error distribution patterns in advance, providing a basis for subsequent targeted optimization of positioning strategies and helping to reduce the impact of errors on positioning accuracy at the source.

[0018] In terms of sensor configuration, multiple sets of laser ranging sensors are installed on the anode torsion pulling device, with each set containing at least two laser ranging sensors. This multi-sensor collaborative operation effectively improves the reliability and accuracy of the sensor-acquired data. The multiple sensor sets enable redundant data acquisition; if one set of sensors experiences data anomalies due to workshop environmental factors, the data collected by other sets can still ensure normal positioning operation, avoiding system failure due to a single sensor malfunction. Furthermore, the configuration of at least two sensors per set allows for cross-verification and supplementation of multiple measurements at the same location, reducing errors caused by measurement deviations of a single sensor and further improving the accuracy of real-time ranging data.

[0019] When the anode torsion device descends to a preset height, multiple sets of laser rangefinders are activated to collect real-time ranging data. This on-demand activation method allows the high-precision characteristics of the laser rangefinders to acquire real-time position information at critical moments during key operations of the overhead crane. Laser rangefinders are highly resistant to interference and operate more stably than traditional vision sensors in the high-temperature, dusty environment of the aluminum electrolysis workshop, reducing environmental interference with data acquisition and ensuring accurate positioning data during critical operation phases. By combining the trajectory data collected by the encoder with the real-time ranging data collected by the laser rangefinders, multi-source data fusion is achieved. This allows for the comprehensive utilization of the advantages of different sensors. The trajectory data provided by the encoder reflects the overall movement trend of the overhead crane, while the real-time data provided by the laser rangefinders corrects the positioning accuracy at key locations. The two complement each other, effectively reducing the impact of single-sensor data deviations on overall positioning accuracy.

[0020] This system eliminates the need for frequent manual calibration. Through its own error accumulation analysis and multi-source sensor data fusion mechanism, it can autonomously maintain high positioning accuracy, reducing the need for manual intervention and lowering labor costs. Because the system can proactively identify and address error accumulation areas, it reduces the risk of equipment collisions caused by positioning errors, ensuring the safety of overhead crane operations and preventing production losses due to equipment damage and downtime. This helps maintain the continuity and stability of electrolytic aluminum production. In practical applications, the system can better adapt to the complex operating environment of electrolytic aluminum workshops, meeting the positioning needs of unmanned overhead cranes in different operating scenarios, improving the overall efficiency of unmanned overhead crane operations, and ensuring the efficient operation of electrolytic aluminum production. Attached Figure Description

[0021] Figure 1 This is a schematic diagram illustrating the working principle of the unmanned overhead crane precise positioning system based on multi-source vision fusion as described in this invention. Figure 2 A working principle diagram for establishing a three-dimensional coordinate system in an electrolytic aluminum workshop; Figure 3 This is a schematic diagram illustrating the working principle of installing multiple sets of laser rangefinder sensors. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figure 1 This invention provides a precise positioning system for unmanned overhead cranes based on multi-source vision fusion. The system includes an integrated central control module, encoders, laser ranging sensors, and other components to achieve precise positioning of overhead cranes within an aluminum electrolysis workshop. The central control module coordinates the various hardware units, performing tasks such as coordinate establishment, data acquisition, error analysis, and position adjustment. The system first establishes a three-dimensional coordinate system for the aluminum electrolysis workshop, setting first-type calibration points at preset intervals along the vertical axis and second-type calibration points at preset intervals along the horizontal axis. The encoder collects real-time trajectory data of the overhead crane in the three-dimensional coordinate system, forming a set of crane trajectory data. Based on this set, error accumulation analysis is performed to identify multiple error accumulation areas. Multiple sets of laser ranging sensors are installed on the anode torsion pulling device, each set containing at least two sensors. When the anode torsion pulling device descends to a preset height, the sensors are activated to collect real-time ranging data, thereby providing precise positioning feedback for the overhead crane.

[0024] Example 1: See Figure 2The establishment of a three-dimensional coordinate system in the aluminum electrolysis workshop and the acquisition of overhead crane trajectory data via encoders first involves the precise placement of calibration points within the workshop. In the vertical direction, first-type calibration points are set up in units of eight electrolytic cells, with the spacing between adjacent points set to a fixed value, such as 10 meters, based on the actual dimensions of the workshop. This spacing design considers the standardized layout of the electrolytic cells and the operating range of the overhead crane, ensuring coverage of the entire working area. In the horizontal direction, two second-type calibration points are arranged at preset intervals, such as 5 meters. The selection of calibration points is based on the flatness and structural stability of the workshop floor. Initial positioning is performed using high-precision measuring instruments such as a total station. During installation, a constant pressure is applied through a mechanical spring-loaded structure. This structure, consisting of spring components and a base, can adapt to minor ground undulations during installation, controlling the positioning error within a millimeter-level threshold, for example, less than 1 millimeter, thereby ensuring the long-term reliability of the calibration points. The installation process of the mechanical spring-loaded structure includes pre-compression testing and repeated calibration. A torque wrench is used to adjust the spring force to ensure that each calibration point can remain stable when subjected to external impact. The calibration points themselves are made of corrosion-resistant alloy materials and coated with a reflective coating for easy visual identification. After installation, the coordinate data is entered through the central control module to form a digital workshop map.

[0025] The overhead crane's trajectory data in a three-dimensional coordinate system is collected by encoders. Absolute rotary encoders are used and are directly mounted on the crane's traveling wheel axles and lifting mechanism to monitor displacement changes in real time. The encoder output signal is transmitted to the central control module via a high-speed data acquisition card. Data acquisition is performed at a fixed sampling frequency, for example, recording the crane's X, Y, and Z coordinate values ​​every 100 milliseconds. The coordinate values ​​are calculated based on the origin of the three-dimensional coordinate system, which is usually set in a corner of the workshop for easy reference. The central control module stores the position coordinate changes at different times in a circular buffer. The buffer size is dynamically adjusted according to the crane's running time to prevent data loss. The stored data includes timestamps, coordinate values, and speed information, used for subsequent analysis of the crane's motion pattern. When calculating the running error between adjacent calibration points, the actual coordinates of the crane passing through the calibration point are compared with the theoretical coordinates. The theoretical coordinates are derived from the workshop design drawings, while the actual coordinates are fed back in real time by the encoder. The error value is calculated using vector difference and recorded as an error log for subsequent processing.

[0026] When setting the first type of calibration points, the interval in the vertical axis direction is based on the arrangement pattern of the electrolytic cells, with one point set for every eight cells. This takes into account the typical stopping positions of the overhead crane in the electrolysis process. A laser rangefinder is used to assist in positioning during installation to ensure accurate spacing. The second type of calibration points are symmetrically arranged in the horizontal axis direction, in pairs, to cross-verify the lateral position of the overhead crane. During installation, the level is adjusted using a level instrument. The spring force of the mechanical spring-loaded structure is calibrated to ensure it maintains the calibration point position even under overhead crane vibration. The error threshold is determined by averaging multiple measurements. During encoder data acquisition, real-time recording of position coordinates relies on the multi-threaded processing capability of the central control module. Data storage uses a compression algorithm to reduce space occupation. The calculation of operating errors includes not only absolute deviation but also cumulative error analysis.

[0027] The selection and installation details of the encoders have been further refined. Absolute encoders feature power-off memory to prevent position information loss. Installation locations are chosen at key movement joints of the overhead crane, such as the trolley traveling mechanism and the trolley traversing mechanism. Each encoder is connected to the central control module via a shielded cable to reduce electromagnetic interference. Binary encoding is used for data storage to improve efficiency. A sliding window algorithm is employed to smooth data fluctuations during error calculation, ensuring the reliability of error values. Maintenance of the mechanical spring-loaded structure is incorporated into daily inspections, with regular cleaning and recalibration to maintain accuracy. The reflective coating at calibration points is replaced periodically to ensure visibility by the vision system even in low-light conditions.

[0028] Example 2: See Figure 3 The central control module first preprocesses the raw trajectory data collected by the encoder, removing abnormal jump points caused by signal interference. The preprocessed continuous trajectory is segmented according to the natural turning points of the crane's operation. Segmentation points are usually selected at the moments when the crane starts, stops, accelerates, decelerates, or changes its direction of travel. Each trajectory segment contains a set of coordinate points that are continuous in time and have similar motion characteristics. The cumulative error of each trajectory segment is calculated using an integral method, accumulating the position deviation values ​​of each sampling point within the segment. The position deviation is defined as the Euclidean distance between the actual coordinates and the theoretical coordinates of that point. The cumulative error reflects the overall accumulation of errors in that segment. Based on the calculated cumulative error of each segment, a clustering algorithm is used to divide the entire operating area into different error level regions. Regions with significantly higher cumulative error than the average level are identified as high error accumulation regions. These regions often correspond to locations in the workshop with poor track flatness, strong electromagnetic interference, or frequent equipment vibration.

[0029] Multiple sets of laser ranging sensors are installed on the anode torsion pulling device. Before installation, each sensor must be individually calibrated to ensure that its range, accuracy, and response frequency meet the design requirements. At least two laser ranging sensors are symmetrically installed on the rigid support of each anode torsion pulling device. The sensor mounting base has elongated holes to allow for fine-tuning to precisely align the measurement direction. For overhead cranes equipped with dual torsion pulling mechanisms, two sets of sensors need to be installed on each of the front and rear torsion pulling devices, for a total of four laser ranging units. The installation position is selected on the lower plane of the torsion pulling device so that its laser beam can be projected vertically downward onto the upper surface of the residual electrode guide rod. When fixing the installation position and measurement direction, a high-precision angle ruler and laser pointer are used for calibration to ensure that the measurement axes of all sensors are parallel to each other and perpendicular to the overhead crane reference plane. Finally, the sensors are firmly locked to the support with anti-loosening nuts, and protective sleeves are installed at the wiring points to prevent dust and electrolyte corrosion in the workshop.

[0030] The instantaneous error at each point within the calculation segment is obtained by comparing the encoder reading with the theoretical coordinates of the corresponding calibration point. The calculation of the cumulative error is not a simple arithmetic summation, but rather takes into account the time weight of the crane's operation in that segment, i.e., an approximation of the integral of the error over time, thus more accurately reflecting the error accumulation effect during dynamic operation.

[0031] After identifying high-error-accumulation areas (such as the track beam connection, coordinate range X: 15-20m, Y: 5-8m, Z: 0.5-1.2m), the central control module stores the coordinates of this area in the key monitoring database and configures a special working mode for high-error areas for the laser ranging sensor. The area trigger condition is: when the X / Y coordinates of the overhead crane fed back by the encoder in real time fall within the high-error area range stored in the key monitoring database, the special mode is triggered (response delay ≤ 100ms). Measurement parameter adjustment: In special mode, the sampling frequency of the laser ranging sensor is increased from the conventional 100Hz to 200Hz, and the number of measurements per sensor group is increased from 1 to 3 (the arithmetic mean of the 3 measurements is taken as valid data to reduce random errors). Data verification mechanism: If the standard deviation of the 3 measurements of a certain laser sensor group is > 0.1mm, visual sensor auxiliary verification is immediately activated. Through visual feature comparison (such as the deviation between the actual and theoretical positions of guide rod markers), it is confirmed whether the laser data is affected by environmental interference. If the verification confirms that the laser data is abnormal, it switches to vision + encoder fusion positioning.

[0032] The sensor bracket is securely connected to the torsion pull device body by welding. After welding, stress relief treatment is required to prevent deformation. After physical installation is completed, the unique identifier, installation position coordinates, installation angle, and corresponding torsion pull device number of each sensor are registered in the central control software to establish a complete sensor configuration mapping table.

[0033] Taking a specific application scenario of production line No. 3 in an electrolytic aluminum workshop as an example, the overhead crane needs to complete the material transfer task from the aluminum tapping station to the anode assembly station. The process of error accumulation analysis based on the overhead crane's running trajectory set begins with the central control module calling historical operating data. This module selects more than 200 trajectory records of the overhead crane running on this line in the past 30 shifts. Each record contains a complete three-dimensional coordinate sequence. The preprocessing stage first identifies and eliminates abnormal trajectory segments caused by signal interruption, such as trajectory jumps caused by the overhead crane's sudden stop or path replanning. The trajectory segmentation algorithm divides each complete trajectory into typical stages such as accelerated departure, constant speed travel, and deceleration entry, based on the characteristic points of the overhead crane's running state. The constant speed travel segment is further subdivided into multiple analysis units at fixed distance intervals. When calculating the error accumulation of each running trajectory segment, the system uses an integral method to accumulate the position deviation of each point within the segment. The position deviation is taken from the Euclidean distance between the encoder coordinates and the theoretical coordinates of the corresponding calibration point. The analysis found that the error accumulation of the third segment (near the ventilation equipment in the middle of the workshop) is consistently about 40% higher than that of other segments. The three error level regions, which are divided according to the magnitude of the accumulated error, are located in the area near the track beam connection. This area corresponds to the original building expansion joint in the workshop, where there are micro-fluctuations in the track flatness.

[0034] After identifying the high-error areas, the installation of laser rangefinder sensors commenced. Technicians determined the optimal sensor installation positions based on the error distribution map. Two laser rangefinder sensors were symmetrically installed on the support beam of each anode torsion pulling device. High-temperature resistant sensors were selected and mounted on specially designed stainless steel brackets, with shock-absorbing pads placed between the brackets and the support beams. For the dual-torsion pulling crane undertaking this task, a total of four sets of sensors were required to be installed on the front and rear torsion pulling devices. The sensor for the front torsion pulling device was installed 1.5 meters from the central axis, and the sensors for the rear torsion pulling device were installed in symmetrical positions. When fixing the installation positions, operators used a laser calibrator to adjust the sensor measurement direction so that the laser beam was projected vertically downwards. Then, a torque wrench was used to tighten the fixing bolts to the specified torque value. All sensor cables were run through metal flexible conduits for protection, bundled and secured along the torsion pulling device structure, and then connected to the top junction box. The junction box had an IP67 protection rating to prevent electrolyte dust intrusion.

[0035] The error accumulation analysis process is dynamic. The central control module automatically updates the trajectory database and recalculates the error distribution weekly. The most recent analysis showed that the high-error area has expanded compared to the initial stage, extending approximately 0.8 meters south of the track. Spatial interference issues were encountered during the installation of the laser rangefinder sensor. The distance between the sensor on the right side of the rear torsion pull device and the hydraulic pipeline was insufficient. Technicians designed a uniquely shaped bracket to offset the sensor's installation position by 5 centimeters, avoiding the pipeline while maintaining the perpendicularity of the measurement axis. Calibration after sensor installation lasted for two shifts. Operators repeatedly tested the sensor in the calibration area using the overhead crane, comparing sensor readings with total station measurements, and adjusting the sensor installation angle until the error was less than 0.2 millimeters. To address the characteristics of the high-error area, the control system optimized the sensor's operating strategy. During operation in this area, the sampling frequency was increased from 100 times per second to 200 times per second, while the overhead crane's operating speed was reduced to 70% of its rated value. This balance between speed and accuracy improved positioning reliability.

[0036] During actual operation, when the torsion pulling device carrying the anode carbon block descends to a height of 30 centimeters from the residual electrode guide rod, all four sets of laser ranging sensors simultaneously activate. The initial value recorded by the front left sensor is 301.5 mm, the front right sensor is 299.2 mm, the rear left sensor is 302.1 mm, and the rear right sensor is 298.9 mm. These data are transmitted to the central control module in real time. The system calculates the average height difference between the front and rear torsion pulling devices to be 2.4 mm. It also finds that the readings of the right-side sensor group are generally 1.8 mm lower than those of the left-side, indicating a slight clockwise tilt of the overhead crane. Based on this, the control system generates adjustment commands. The longitudinal movement mechanism first raises the overhead crane by 2.4 mm, then the rotation mechanism adjusts counterclockwise by 0.1 degree, and the transverse movement mechanism finely adjusts to the left by 0.7 mm. After the initial adjustment, the sensors immediately resample. The data shows that the height difference has been reduced to 0.3 mm and the horizontal deviation has decreased to 0.4 mm. The system then initiates a second round of fine adjustments, ultimately controlling the positioning error within the range of 0.1 mm. The entire adjustment process took four and a half seconds, and the overhead crane successfully completed the precise alignment and installation of the anode carbon block.

[0037] Example 3: When the anode torsion drawing device descends smoothly to a preset height from the residual electrode guide rod under the command of the control system, multiple sets of laser ranging sensors installed around the device are simultaneously activated. This preset height is set within a specific numerical range according to process requirements. During the descent, the height encoder continuously monitors the current position. When it enters the allowable error range of the preset height, the control unit immediately sends a synchronous acquisition command to all sensors. After the sensors are activated, they operate at a specific sampling frequency. The laser beam is precisely projected onto the upper surface of the residual electrode guide rod. The returned distance data is transmitted to the signal processing module through shielded twisted-pair cables. This module has a built-in adaptive digital filter, which can effectively suppress signal interference caused by workshop vibration and electrolyte dust. During continuous acquisition, the central processing unit records the timestamp, sensor number, and corresponding measurement value of each measurement point, and compares these real-time data with the theoretical values ​​of the standard position stored in the database to calculate the offset of each measurement point from the standard position in three-dimensional space. These deviation data are temporarily stored in a high-speed buffer register in vector form.

[0038] Based on real-time measurement data from multiple sensors, the control system executes a multi-step calculation process to determine the spatial pose deviation of the overhead crane. First, a weighted average is calculated for the measurements from all valid sensors. This calculation automatically excludes outlier data points exceeding the measurement range. The weighting coefficients are dynamically allocated based on the sensor calibration accuracy and the importance of the installation position. Based on the calculated average deviation, the control algorithm generates a position adjustment command in the longitudinal direction. The adjustment amount comprehensively considers the current deviation magnitude, the crane's motion inertia, and positioning accuracy requirements. The command is converted into a pulse signal and sent to the longitudinal drive servo motor, controlling the crane to move a corresponding distance along the track to eliminate systematic deviations. After completing the initial positioning in the longitudinal direction, the system immediately enters the angle fine-tuning stage. This stage analyzes the differences between the measurement values ​​of different sensors on the same torsion puller to determine the tilt degree of the crane platform. The difference in measurement values ​​directly reflects the deflection angle between the crane's current attitude and the ideal attitude. The control system calculates the rotation angle that needs to be compensated based on this difference and drives the rotation mechanism to perform fine angle correction, keeping the crane platform parallel to the residual pole guide rod.

[0039] Throughout the data acquisition and processing process, the system employs a multi-redundancy verification mechanism to ensure reliability. The laser rangefinder sensor's activation signal is simultaneously monitored by a hardware comparator and a software watchdog to prevent unexpected non-triggered situations. Data transmission uses a verification protocol, and digital filter parameters are dynamically adjusted based on real-time noise spectrum characteristics. Deviation calculation utilizes an improved weighted average algorithm. Position adjustment commands are generated using a predictive control algorithm. When the drive motor moves, the grating encoder provides real-time feedback of the actual displacement, which is compared in a closed-loop manner with the command value. The output torque is dynamically adjusted via a regulator to ensure a smooth and accurate positioning process. The angle adjustment phase also employs closed-loop control. Simultaneously with the rotation mechanism's movement, the sensor group continuously samples to verify the adjustment effect in real time. If the predetermined accuracy requirement is not met, iterative fine-tuning is performed until the deviation values ​​at all measurement points are less than the set tolerance, ultimately achieving precise spatial matching between the overhead crane and the residual pole guide rod.

[0040] A fusion algorithm is used in the data processing to improve measurement accuracy. This algorithm is expressed as:

[0041] in: This represents the final fused deviation measurement used for control decisions; symbol Indicates the number of effective sensor units currently participating in the fusion computation; symbol Indicates the first Raw deviation data collected by each sensor unit; symbol Is assigned to the first The confidence coefficient of each sensor unit is dynamically calculated based on the historical accuracy performance and current operating status of the sensor unit.

[0042] Each sensor unit undergoes self-diagnostic verification before participating in calculations, including laser intensity detection, received signal quality assessment, and ambient temperature compensation, to ensure the validity of the input data. The control algorithm considers the dynamic response characteristics of the crane system during adjustments, employing a gradual adjustment strategy: first, a larger coarse adjustment is made, followed by fine-tuning. Throughout the adjustment process, the central processing unit continuously monitors the data trends of each sensor, analyzing data consistency to determine the correctness of the adjustment direction and promptly identifying and correcting any potential anomalies. After all adjustments are completed, the system records the final sensor readings, adjustment parameters, and execution time, forming a complete position calibration log for subsequent system performance analysis and optimization.

[0043] Example 4: When adjusting the position of the overhead crane in the longitudinal direction, the central control module calls a preset data processing program. This program first reads the real-time measurement values ​​of multiple sets of laser rangefinders installed on the torsion puller. The sensors are logically divided into a first group (located at the front of the overhead crane) and a second group (located at the rear of the overhead crane) according to their physical location. The program preprocesses the valid measurement values ​​of all sensors in each group, removes abnormal readings caused by instantaneous interference, and calculates the arithmetic mean of the measurement values ​​of the first and second groups of sensors respectively. The average value calculation takes into account the measurement accuracy level of the sensors themselves, assigning different weight coefficients to sensors with different accuracies. The readings of high-accuracy sensors have a higher weight in the average value calculation. The calculated average value of the front and rear groups is compared with a preset standard value. The standard value comes from the theoretical height coordinate of the target residual pole guide rod in the three-dimensional coordinate system. The difference value is obtained by subtraction. This difference value directly reflects the overall deviation and direction of the overhead crane in the longitudinal direction (positive value is too high, negative value is too low). Determining the adjustment distance is not simply a matter of the difference value; rather, it requires the introduction of a control algorithm. This algorithm integrates the crane's current operating speed, load mass, and mechanism inertia to output an optimized adjustment distance command. This command is converted into the number of pulses required to drive the motor and sent to the longitudinal drive controller. The process of controlling the crane to move the corresponding distance along the longitudinal axis adopts a closed-loop control mode. After the motor starts, the encoder provides real-time feedback on the actual displacement, which is continuously compared with the command value. The PID controller dynamically adjusts the motor speed and torque to ensure a smooth and precise movement, ultimately positioning the crane near its theoretical position.

[0044] The process of adjusting the crane angle begins immediately after the longitudinal coordinate adjustment is completed. The central control module retrieves real-time measurements from two laser rangefinders located on the same torsion and pulling device. These two sensors are physically installed with a fixed lateral distance between them. Comparing the differences in the measurements from these two sensors is a simple subtraction operation. The result directly reflects the tilt of the crane platform relative to the horizontal plane. A positive result indicates that the left side of the device is too high, and a negative result indicates that the right side is too high. When calculating the angle to be adjusted based on the difference in measurements, the calculation principle is based on trigonometric functions. The installation distance between the two sensors is taken as the base of a right triangle, and the difference in measurements is taken as the opposite side. The arctangent function is used to solve for the angle value. The calculated angle value is a signed floating-point number, with the sign indicating the direction of rotation (positive for clockwise, negative for counterclockwise), and the absolute value representing the required angle adjustment range. The task of controlling the crane to rotate by the corresponding angle is performed by an independent rotating mechanism servo motor. The central control module converts the angle value into motor control commands, which include the target angle and rotation speed curve. The rotating mechanism typically uses a precision reducer to ensure rotational accuracy. During rotation, the encoder continuously feeds back the actual angle of the rotating shaft and performs a closed-loop comparison with the command value to ensure that the actual angle and the command angle are highly consistent after the rotation stops. See Table 1.

[0045] Table 1: Calculation Table for Adjustment of Vertical Axis Position

[0046] In a specific implementation example, after the overhead crane carrying the anode torsion device descends to the predetermined height, the laser ranging sensor group begins to operate. The central control module collects the data shown in the table. The data processing program first identifies that all readings are within the effective range and then calculates the average value. The average deviation of the first group of sensors (numbered F-01, F-02) is +0.74 mm after weighted calculation, and the average deviation of the second group of sensors (numbered R-01, R-02) is +1.50 mm. The control system further calculates the mean of these two groups of average deviations, obtaining a comprehensive average deviation of +1.12 mm for the overhead crane on the vertical axis. This positive value indicates that the current overall position of the overhead crane is higher than the theoretical standard position. Based on this comprehensive deviation value and combined with the overhead crane's current light load and low speed state parameters, the control algorithm calculates the theoretical adjustment distance as a downward movement of 1.12 mm. This value is converted into the pulse equivalent of the drive motor and sent out. The longitudinal drive motor starts and moves the overhead crane smoothly downwards in low-speed mode. The encoder monitors the falling distance in real time with micron-level resolution. When the cumulative movement reaches 1.12 mm, the motor stops and the brake is engaged, completing the initial adjustment of the longitudinal coordinate position.

[0047] Light load and low speed state parameter definition: Light load, the load weight of the crane is <5t (collected in real time by the weighing sensor (model: ZEMIC H3-C3-5t-3B) on the top of the crane, with a measurement accuracy of ±0.1t); Low speed, the longitudinal movement speed of the crane is <0.5m / s (feedback in real time by the encoder of the longitudinal movement mechanism, with a speed calculation accuracy of ±0.01m / s).

[0048] Theoretical adjustment distance calculation method: The theoretical adjustment distance needs to incorporate a working condition correction factor. To offset the effects of crane inertia and track clearance on adjustment accuracy, the formula is as follows:

[0049] in: Theoretical adjustment distance in the longitudinal direction of the overhead crane (unit: mm, positive value indicates upward adjustment, negative value indicates downward adjustment); The overall average deviation in the longitudinal direction of the overhead crane (here) (The positive value indicates that the overhead crane is above its theoretical position). : Working condition correction factor (dynamically determined based on crane load and speed: light load and low speed) When under heavy load (load ≥ 10t) and high speed (speed ≥ 1.0m / s) Medium load and medium speed The overhead crane was operating under light load (3.2t) and low speed (0.3m / s) conditions. Substituting into the formula, we get: ,because If it is positive, it needs to be adjusted downwards by 1.12mm.

[0050] After the vertical coordinate adjustment is completed, the system immediately initiates the angle adjustment program. The central control module reads the data from two sensors on the same torsion pull device at this moment. For example, the reading of the left sensor (L-X01) on the front torsion pull device is 301.5 mm, and the reading of the right sensor (R-X01) is 298.9 mm. The lateral installation distance between the two sensors is 1000 mm. The difference between the two measured values ​​is calculated as 301.5 - 298.9 = +2.6 mm. This positive difference indicates that the left side of the torsion pull device is higher than the right side, indicating a clockwise tilt. Based on the installation distance and height difference, the system calculates the angle value that needs to be corrected by counterclockwise rotation. This angle value is quickly obtained through a preset trigonometric function lookup table to avoid the delay caused by real-time floating-point calculations. After receiving the command to rotate counterclockwise by 0.149 degrees, the rotary servo motor starts to move, using a speed curve that is fast at first and then slows down to reduce mechanism sway. The high-precision encoder provides feedback on the rotation angle, and the motor decelerates before reaching the target position, finally stabilizing at the angle required by the command, completing the attitude fine-tuning. The entire adjustment process, from data collection to execution, is completed within seconds. During the process, various types of data are recorded in real time and timestamped for possible subsequent analysis and optimization.

[0051] Angle calculation formula: The tilt angle of the overhead crane platform can be calculated by using the opposite side of a right triangle (…). - Adjacent edge ( The arctangent function of ")" is calculated using the following formula:

[0052] in: The result is in radians and needs to be multiplied by... Convert to angle value; The difference in measurement values ​​between the left and right laser sensors on the same torsion device (left sensor reading) - Right side sensor reading (Unit: mm) The lateral mounting spacing between the two laser sensors (calibrated and fixed at the factory using a laser interferometer) , deviation ≤0.02mm); : The rotation angle that the overhead crane needs to be adjusted (unit: °, positive value indicates clockwise rotation, negative value indicates counterclockwise rotation).

[0053] Reading of sensor (L-X01) on the left side of the front torsion pull device Reading from the right-side sensor (R-X01) ,but ; Substituting into the formula, we get: ; because A positive reading (left side reading > right side reading) indicates that the left side of the torsion puller is too high and needs to be rotated counterclockwise. To correct the horizontal attitude.

[0054] Example 5: Adjusting the horizontal coordinate position based on the difference in measurements from two laser rangefinders on the same torsion pull device. The central control module first reads the real-time data from the two sensors located on the same horizontal plane. These two sensors are symmetrically installed on the torsion pull device, with their measurement axes parallel to the lateral movement direction of the crane. The control system calculates the instantaneous difference in the measurements from the two sensors. This difference directly reflects the amount and direction of the crane's offset relative to the target position in the horizontal coordinate direction. If the measurement value from the left sensor is greater than that from the right, it indicates that the crane has deviated to the right and needs to be corrected to the left. The calculation of the adjustment amount not only considers the instantaneous difference but also incorporates historical deviation data to form trend predictions. The influence of measurement noise is eliminated through a filtering algorithm, and finally, a horizontal coordinate adjustment command is generated and sent to the lateral movement drive mechanism.

[0055] For overhead cranes equipped with dual torsion pullers, the control system needs to process the positioning data of the front and rear torsion pullers separately. The sensor groups on the front and rear torsion pullers collect data independently, and the central processing unit processes these two sets of data in parallel. The adjustment process follows the principle of independent adjustment first, then coordinated adjustment. First, the front adjustment amount is calculated based on the difference in sensor data from the front torsion puller, and simultaneously, the rear adjustment amount is calculated based on the data from the rear torsion puller. The two adjustment commands are then sent to the front and rear traverse motors, respectively. During adjustment, the control system monitors the relative position changes of the two torsion pullers in real time, using a kinematic model to ensure that the dual torsion pullers remain synchronized during traverse, avoiding amplification of structural stress or positioning errors caused by asynchronous adjustments.

[0056] The system repeats measurements and adjustments until all measured values ​​reach the preset accuracy range. After each adjustment, the control system re-triggers the laser rangefinder sensor for a new round of data acquisition, comparing the new measured values ​​with the target values. If the measurement deviation of any sensor still exceeds the allowable tolerance range, a new round of adjustment calculation is initiated, with the adjustment amount dynamically corrected based on the current deviation and the effect of the previous adjustment. The iterative adjustment process employs a gradual convergence strategy, allowing for larger adjustments in the initial stage to quickly approach the target position. As the deviation decreases, the adjustment amplitude is gradually reduced to improve positioning accuracy. When all sensor data are within the allowable error range for three consecutive measurements, the system determines that positioning is complete, locks all drive mechanisms, and sends a ready signal.

[0057] In a specific execution scenario, the overhead crane, carrying a dual-torsion pulling mechanism, positions itself at the target workstation. The left sensor PL-01 of the front torsion pulling device reads 298.5 mm, and the right sensor PR-01 reads 297.2 mm, a difference of +1.3 mm, indicating that the front device is offset to the right by approximately 1.3 mm. The left sensor RL-01 of the rear torsion pulling device reads 299.1 mm, and the right sensor RR-01 reads 297.8 mm, also a difference of +1.3 mm, similarly indicating a rightward offset. The control system first calculates the adjustment amount for the front, determining, based on the difference and the mechanism's characteristics, that a lateral movement of 1.25 mm to the left is required. Simultaneously, the rear adjustment amount is also calculated to be a lateral movement of 1.25 mm to the left. Both lateral movement motors start synchronously, performing the leftward movement at the same speed. During the movement, a grating ruler provides real-time feedback on the actual displacement, stopping precisely when the 1.25 mm target is reached.

[0058] After the initial adjustment, the system immediately resampled. The new readings for the front sensor group were PL-01 = 299.8 mm and PR-01 = 300.1 mm, with a difference of -0.3 mm. The new readings for the rear sensor group were RL-01 = 299.9 mm and RR-01 = 300.2 mm, with a difference of -0.3 mm. The control system determined that the deviation had decreased but not completely eliminated, and that the deviation direction had changed to slightly lower to the left. Therefore, it generated a second adjustment command: a slight adjustment of 0.25 mm to the right for both the front and rear sections. The traverse motor restarted to perform fine adjustments. After the movement was completed, the third sampling data showed that all sensor readings were within the range of 300.0 ± 0.1 mm. The system determined that the positioning accuracy met the requirements, terminated the adjustment cycle, and locked the mechanism. The entire adjustment process was completed within 8 seconds, with two adjustment actions performed, ultimately controlling the gantry crane's lateral coordinate positioning accuracy to within 0.1 mm.

[0059] During the adjustment process, the control system continuously records changes in data from various sensors, the execution of adjustment commands, and the mechanism's response time. This data is used not only for real-time control decisions but is also stored in the operation log. For the double-torsion pulling mechanism, the control system pays special attention to the synchronization error between the front and rear devices. If the detected asynchrony between the front and rear adjustments exceeds a safety threshold, the adjustment will be immediately paused and a correction procedure will be initiated. Each adjustment command includes acceleration and deceleration parameter settings to ensure a smooth and shock-free lateral movement, avoiding unnecessary vibration to the crane structure and the workpiece being measured. The final positioning accuracy verification employs a multi-verification mechanism. In addition to the data from the laser rangefinder sensor itself, the actual displacement feedback from the encoder is also used for cross-verification to ensure the reliability and accuracy of the positioning results.

[0060] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A precise positioning system for unmanned overhead cranes based on multi-source visual fusion, characterized in that, Includes a central control module, which is used to perform: Establish a three-dimensional coordinate system for the electrolytic aluminum workshop, set first-type calibration points at preset intervals in the vertical axis direction, and set second-type calibration points at preset intervals in the horizontal axis direction; The crane's trajectory data in the three-dimensional coordinate system is collected by the encoder to form a set of crane trajectories; Error accumulation analysis was performed based on the set of overhead crane operating trajectories to identify multiple error accumulation areas; Multiple sets of laser rangefinders are installed on the anode torsion pulling device, each set containing at least two laser rangefinders; When the anode torsion device descends to a preset height, the multiple sets of laser ranging sensors are activated to collect real-time ranging data.

2. The unmanned overhead crane precise positioning system based on multi-source vision fusion as described in claim 1, characterized in that, The establishment of the three-dimensional coordinate system for the electrolytic aluminum workshop includes: A first-class calibration point is set every 8 electrolytic cells in the vertical axis direction, and the spacing between adjacent first-class calibration points is a preset value; Two second-type calibration points are set at preset intervals along the horizontal axis; The mechanical spring-loaded structure ensures that the positioning error of the first type of calibration point and the second type of calibration point is less than a preset threshold.

3. The unmanned overhead crane precise positioning system based on multi-source vision fusion as described in claim 1, characterized in that, The acquisition of the overhead crane's trajectory data in the three-dimensional coordinate system via the encoder includes: The position coordinates of the overhead crane in the three-dimensional coordinate system are recorded in real time, and the position coordinate changes of the overhead crane at different times are stored. The operating error between adjacent calibration points is calculated based on the overhead crane's operating trajectory data.

4. The unmanned overhead crane precise positioning system based on multi-source vision fusion as described in claim 1, characterized in that, The error accumulation analysis based on the set of overhead crane operating trajectories includes: The overhead crane running trajectory set is segmented to obtain multiple running trajectory segments; Calculate the cumulative error for each trajectory segment and divide the trajectory into multiple error accumulation regions based on the magnitude of the cumulative error.

5. The unmanned overhead crane precise positioning system based on multi-source vision fusion as described in claim 1, characterized in that, The installation of the multiple sets of laser ranging sensors includes: At least two laser rangefinders are installed on each anode torsion device; For the double-torsion crane, a total of four sets of laser rangefinders are installed; Fix the installation position and measurement direction of each laser rangefinder.

6. The unmanned overhead crane precise positioning system based on multi-source vision fusion as described in claim 1, characterized in that, The step of activating the multiple sets of laser ranging sensors to collect real-time ranging data includes: When the anode torsion device descends to a preset height from the residual electrode guide rod, the distance measurement is activated. Continuously collect real-time measurement values ​​from each laser rangefinder and record the deviation data between each measurement point and the standard position.

7. The unmanned overhead crane precise positioning system based on multi-source vision fusion as described in claim 1, characterized in that, The central control module also includes execution: The average deviation value is calculated based on the measurement data from multiple sets of laser rangefinders. Adjust the position of the overhead crane in the vertical axis direction based on the average deviation value; After the vertical coordinate adjustment is completed, the crane angle is adjusted according to the differences in the measurement values ​​of each sensor.

8. The unmanned overhead crane precise positioning system based on multi-source vision fusion as described in claim 7, characterized in that, The adjustment of the overhead crane's position in the vertical axis direction includes: Calculate the average value of the measurements from the first and second groups of laser rangefinder sensors; The adjustment distance is determined based on the difference between the average value and the standard value, and the crane is controlled to move the corresponding distance along the vertical axis.

9. The unmanned overhead crane precise positioning system based on multi-source vision fusion as described in claim 7, characterized in that, The adjustment of the overhead crane angle includes: Compare the differences in measurement values ​​between two laser rangefinders on the same torsion pull device; Calculate the angle that needs to be adjusted based on the difference in measured values, and control the crane to rotate by the corresponding angle.

10. The unmanned overhead crane precise positioning system based on multi-source vision fusion as described in claim 1, characterized in that, The central control module also includes execution: Adjust the horizontal coordinate position based on the difference in measurement values ​​between two laser rangefinders on the same torsion device; For a double-torsion crane, adjust the horizontal coordinate position of the two torsion devices respectively; Repeat the measurement and adjustment until all measurements reach the preset accuracy range.

Images