A perception fusion adaptive compliant control method for non-ferrous metal smelting robot slag cleaning

Abstract

This invention discloses a perception-fusion adaptive compliant control method for slag cleaning in non-ferrous metal smelting robots, comprising: collecting joint operating parameters and joint friction forces during the slag cleaning process to construct joint spatial dynamic equations; collecting the temperature near the joint to correct the joint friction forces; calculating the joint external torque; mapping the joint external torque to obtain the end contact force; calculating the contact force error based on the desired contact force and the end contact force; using the contact force error and its rate of change as input, judging the current slag resistance change state through a fuzzy inference model, and adjusting the impedance stiffness parameter and damping coefficient in real time; calculating the impedance control force based on the impedance stiffness parameter and damping coefficient; combining the dynamic mechanics feedforward compensation torque to calculate the robot joint torque control command; after performing safety limiting processing on the joint torque control command, outputting limiting control or yielding control according to the safe state of the end contact force, and generating a slag cleaning trajectory.

Description

Technical Field

[0001] This invention relates to the field of compliant control technology, and in particular to a perception fusion adaptive compliant control method for slag cleaning in non-ferrous metal smelting robots. Background Technology

[0002] In non-ferrous metal production processes such as aluminum alloy casting, copper refining, and zinc smelting, slag is continuously generated on the surface of the molten metal in the furnace during the high-temperature melting process. Taking aluminum alloy casting as an example, the molten aluminum reacts with air and flux to generate oxide inclusions such as Al2O3, MgO, and AlN. The slag layer thickness is typically between 50 and 200 mm, and the amount of slag accounts for approximately 1.5% to 4.5% of the weight of the metal material. If not removed in time, the slag not only reduces the thermal efficiency of the furnace but also mixes into the castings, seriously affecting metallurgical quality and yield. It is a critical and frequently occurring necessary process in non-ferrous metal smelting production lines.

[0003] For a long time, slag cleaning operations in non-ferrous metals have relied mainly on manual labor. Workers must endure prolonged exposure to high-temperature radiation (700–900°C), toxic fumes such as fluorides and chlorides, and strenuous physical labor near the furnace opening, resulting in extremely high occupational safety risks. Furthermore, manual slag cleaning is inefficient, produces poor batch-to-batch consistency, and fails to meet the stringent requirements of modern precision casting for melt purity. Introducing industrial robots into slag cleaning operations to achieve inherent safety and standardization has become an urgent need for the intelligent manufacturing upgrade in the non-ferrous metals industry.

[0004] Robotic slag cleaning is a complex contact operation. The physical properties (viscosity, plastic shear strength) of non-ferrous metal slag vary in real time with the furnace cycle, melt temperature, and flux ratio. The plastic shear strength of aluminum alloy slag typically fluctuates significantly between 0.8 and 15 kPa, exceeding tenfold. The molten liquid level changes dynamically with the amount of feed, and the slag layer thickness is unevenly distributed. The contact force between the slag cleaning tool and the slag exhibits strong nonlinear variation, placing extremely high demands on the adaptability and reliability of the robot torque control system.

[0005] However, existing robotic slag removal systems have the following technical problems: High-temperature environments cause joint friction characteristics to drift, leading to a continuous deterioration in the accuracy of current-based torque estimation. Existing slag-cleaning robots primarily use the current of the servo motors driving the joints to infer the joint output torque as the main method of contact force sensing, eliminating the need for independent torque sensors. However, the high temperatures of 700–900°C near the furnace mouth cause the viscosity of the joint lubricating grease to decrease significantly with temperature, resulting in the actual joint friction torque deviating from the room-temperature calibration model value. This introduces a non-negligible systematic error into the current-based torque estimation, resulting in consistently low contact force feedback. This is the primary technical bottleneck restricting the application of flexible robot control in non-ferrous metal smelting scenarios. Some slag removal robots still use pure position control. The robot moves along a preset trajectory and does not sense contact force. Once it encounters a hard slag layer, it is very easy for joint overload damage to occur. Fixed-parameter impedance control is difficult to simultaneously handle both soft and hard slag conditions, resulting in poor robustness; moreover, it lacks a systematic safety protection mechanism, making it difficult to simultaneously guarantee thorough slag removal and operational safety.

[0006] The purpose of this invention is to design a perception fusion adaptive compliant control method for slag cleaning in non-ferrous metal smelting robots, which addresses the problems existing in the prior art. Summary of the Invention

[0007] In view of this, the purpose of this invention is to propose a perception fusion adaptive compliant control method for slag cleaning in non-ferrous metal smelting robots, which can solve the above-mentioned problems.

[0008] This invention provides a perception fusion adaptive compliant control method for slag cleaning in non-ferrous metal smelting robots, comprising: S1 divides the slag removal process into a probing and sensing stage, an active slag removal stage, and a leveling and confirmation stage, generating corresponding expected joint angles, expected joint angular velocities, expected contact forces, and initial impedance parameters respectively. The S2 robot collects joint operating parameters and joint friction during the slag removal process to construct the joint space dynamics equation, collects the temperature near the joint to correct the joint friction, calculates the joint external torque, and maps the joint external torque to obtain the end contact force; S3 calculates the contact force error based on the expected contact force and the end contact force. Using the contact force error and its rate of change as input, it judges the current state of slag resistance change through a fuzzy inference model and adjusts the impedance stiffness parameter and damping coefficient in real time. The impedance control force is calculated using the S4 root impedance stiffness parameters and damping coefficients. Combined with the dynamics feedforward compensation torque, the robot joint torque control command is calculated. After performing safety limiting processing on the joint torque control command, S5 outputs limiting control or yielding control based on the safety status of the end contact force, and generates a slag removal trajectory.

[0009] The beneficial effects of this invention are: First, it has a strong adaptive capability to changes in the characteristics of non-ferrous metal slag. The fuzzy impedance adaptive mechanism does not require a precise mathematical model of the slag; it can automatically adjust the stiffness parameters based solely on the real-time contact force signal. It can adapt to large fluctuations of more than ten times the plastic shear strength of aluminum / copper / zinc slag, fundamentally solving the technical problem of fixed parameter controllers failing under varying slag conditions.

[0010] Secondly, the contact torque estimation remains consistently reliable under high-temperature conditions. A temperature-friction coupling compensation model is introduced to dynamically correct the friction torque deviation caused by changes in grease viscosity using measured temperature values. Compared with a fixed friction model, this reduces the current estimation error by more than 60%. The dual-path adaptive fusion and three-mode switching mechanism of the current-observer system ensures that the two estimation paths verify and back each other, keeping the contact torque estimation error within 5% of full scale under furnace opening conditions of 700–900°C. This provides a solid force feedback foundation for subsequent compliance control.

[0011] Third, the dual-layer safety protection significantly improves the reliability of the equipment. The joint torque saturation (first layer) and the end contact force over-limit retreat (second layer) complement each other: the former prevents the actuator from being damaged due to continuous overload, and the latter responds quickly to instantaneous impact within 2 ms. The two layers work together to greatly extend the fault-free operating life of the slag cleaning robot.

[0012] Fourth, complete dynamic feedforward compensation improves the accuracy of contact force control. With full compensation for inertia, Coriolis force, gravity and joint friction, the end contact force control error can be controlled within ±8% of the expected value at typical slag removal speeds (0.05~0.3 m / s).

[0013] Fifth, it organically integrates five functional modules: sensing and estimation, impedance control, adaptive adjustment, safety protection, and trajectory planning. It is applicable to various non-ferrous metal smelting equipment such as aluminum alloy casting furnaces, copper anode furnaces, and zinc refining furnaces, and supports parameter switching according to furnace type, exhibiting good engineering portability. It introduces a three-stage differentiated control strategy: P1 sensory detection → P2 active slag removal → P3 leveling confirmation. Based on the slag distribution heatmap, different speed and impedance parameters are applied to soft and hard slag zones respectively. Compared with the traditional single-parameter slag removal mode, it effectively reduces the impact risk in hard slag zones while ensuring thorough slag removal. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings required in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart of the method in this embodiment. Detailed Implementation

[0016] To facilitate understanding by those skilled in the art, the structure of the present invention will now be described in further detail with reference to the accompanying drawings. It should be understood that, unless otherwise specified, the order of the steps mentioned in this embodiment can be adjusted according to actual needs, and they can even be executed simultaneously or partially simultaneously.

[0017] like Figure 1 As shown, this embodiment of the invention provides a perception fusion adaptive compliant control method for slag cleaning in non-ferrous metal smelting robots, comprising: S1 divides the slag removal process into the probing and sensing stage, the active slag removal stage, and the leveling and confirmation stage, respectively generating the corresponding expected joint angle, expected joint angular velocity, and their estimated expected joint angular acceleration, expected contact force, and initial impedance parameters. When the S101 robot's movement speed is less than or equal to the first speed threshold, it is classified into the exploration and perception stage. When the first speed threshold of S102 is less than the robot's movement speed and less than the third speed threshold, it is divided into the active slag removal stage; When the S103 robot's movement speed is less than or equal to the second speed threshold, it is classified as the leveling confirmation stage.

[0018] In this step, the slag removal operation is divided into three control sub-stages (P1→P2→P3) according to the process stage. The parameters of each stage are configured independently to form a differentiated staged control strategy, avoiding the failure risk of a single parameter mode under different slag conditions.

[0019] Phase P1 (Touch and Sensing Phase): The robot slowly drives the tool to contact the slag surface at a low speed (≤0.05 m / s). The S3 fuzzy inference system starts with a low stiffness initial value (recommended 200~400 N / m) and identifies the hardness zones of the slag layer through the contact force response—soft slag zone (contact force <20 N), medium slag zone (20~50 N), and hard slag zone (>50 N), forming a heat map of the slag condition distribution on the entire furnace surface, providing prior input for the differentiated trajectory planning in Phase P2.

[0020] Phase P2 (Active Slag Removal Phase): Based on the slag condition heat map of Phase P1, a slag removal path covering the entire furnace surface is planned. In the soft slag zone, a low slag removal force (30-50 N for aluminum alloy furnaces) and a high speed (0.15-0.3 m / s) are used to quickly push and scrape the slag. In the hard slag zone, the speed is reduced (0.05-0.1 m / s) and the subsequent safety protection mechanism is prioritized. At the same time, the subsequent fuzzy adaptive adjustment layer continuously and dynamically updates the impedance stiffness and damping coefficient.

[0021] Phase P3 (Leveling Confirmation Phase): After the active slag removal path is completed, the robot performs a final leveling scan on the furnace surface at a low speed (≤0.08 m / s) and low contact force (≤25 N).

[0022] The desired joint angle and desired joint angular velocity for slag removal are automatically generated by the online planning module based on furnace type parameters, and must simultaneously satisfy three types of process constraints: ① Slag removal angle constraint: The angle between the tool and the molten surface should be maintained between 15° and 45°; ② Slag removal line speed constraint: Recommended 0.05~0.3 m / s; ③ Tool-molten surface safety distance constraint: The bottom of the tool must be at least 5 mm away from the molten surface.

[0023] The S2 robot collects joint operating parameters and joint friction during the slag removal process to construct the joint space dynamics equation, collects the temperature near the joint to correct the joint friction, calculates the joint external torque, and maps the joint external torque to obtain the end contact force; S201 collects motor current, joint angle, joint angular velocity and estimated joint angular acceleration during robot operation to establish joint space dynamics equations, and collects temperature near the joint to calculate the external torque of the first joint using a temperature-compensated friction model. S2011 calculates the motor joint output torque using the phase current signal of the servo motor during robot operation. The calculation formula is as follows: , in, Indicates the output torque of the motor joint, Indicates torque coefficient, Indicates the effective value of the phase current; S2012 uses the Lagrange method to establish the joint space dynamics equations of the n-DOF cleaning robot manipulator, and calculates the robot's own dynamic torque. The joint space dynamics equations are as follows: , in, Indicates joint angle, Indicates angular velocity, Indicates angular acceleration, Represents the joint space inertia matrix, Represents the Coriolis force and centrifugal torque. Represents the gravitational moment vector, Represents theoretical joint friction torque, This indicates the driving torque command issued by the controller. Representing the geometric Jacobian matrix, This represents the six-dimensional generalized contact torque at the end point; In this step, the four terms on the left side of the equation describe the internal dynamic effects that the robot needs to overcome during its movement.

[0024] S2013 uses a Coulomb-viscosity-temperature coupled friction model to construct a temperature-compensated friction model, and calculates the joint friction torque affected by temperature. The calculation formula is as follows: ， ， in, Represents the parameters of the Coulomb friction term, Indicates the parameters of viscous friction. Indicates temperature correction factor, Indicates the friction-temperature sensitivity coefficient, Indicates real-time temperature measurement near the joint, Indicates room temperature reference temperature; S2014 subtracts the robot's own dynamic torque, calculated from the real-time collected joint angular velocity, joint angular velocity and estimated joint angular acceleration, and the joint friction torque affected by temperature from the motor joint output torque to obtain the first joint external torque.

[0025] In this step, the robot's own dynamic torque is the sum of the inertial torque term, Coriolis force and centrifugal torque term, and gravitational torque term on the left side of the joint space dynamic equation. It reflects the self-dynamic effects that the robot needs to overcome to complete the current motion and is independent of frictional torque and external contact torque. The calculation process is as follows: the joint encoder collects the joint angles in real time. q and joint angular velocity The joint angular velocity is then differentially calculated or estimated by an observer to obtain the joint angular acceleration. ;Will q , , Substitute the sum of the inertial moment term, Coriolis force and centrifugal moment term, and gravitational moment term into the joint space dynamics equation described in step S2012 for calculation. The calculation formula is as follows: τ dyn = M ( q ) + C ( q , q ) + G ( q ), in, τ dyn This represents the robot's own dynamic torque, which is the sum of the inertial torque, the Coriolis force and centrifugal torque, and the gravitational torque. The robot's own dynamic torque is then subtracted sequentially from the output torque of the motor joints. τ dyn The joint friction torque affected by temperature is used to obtain the first joint external torque.

[0026] In this step, the servo motor current signal driving the joint is used to infer the joint output torque as the main signal source for contact force sensing, eliminating the need to install a separate torque sensor.

[0027] However, in high-temperature environments of 700–900°C, the viscosity of joint lubricating grease decreases significantly with increasing temperature, causing the joint friction torque to deviate from the room-temperature calibration value. Directly applying a fixed friction model would introduce systematic torque estimation errors. Therefore, temperature-adaptive friction compensation is introduced, employing a Coulomb-viscosity-temperature coupled friction model, in which… β The friction-temperature sensitivity coefficient is determined by offline calibration experiments, with a recommended range of 0.005–0.02 °C. The compensated current torque is estimated by real-time correction of the friction term.

[0028] In terms of high-temperature friction compensation, a friction calibration method based on thermal imaging-assisted temperature field modeling is adopted, which can replace the Coulomb-viscosity-temperature coupled friction model based on actual temperature measurement. The joint temperature distribution is acquired in real time by infrared thermal imager, and a more refined two-dimensional (rotation speed-temperature) friction torque lookup table model is constructed to dynamically correct the current torque estimation error. Alternatively, a data-driven approach can be adopted, using machine learning (such as Gaussian process regression) to learn the friction-temperature-rotation speed three-dimensional mapping offline and query the alternative analytical model online, further expanding the applicable accuracy range of friction compensation in high-temperature scenarios of this invention.

[0029] S202 inputs the joint angle and joint angular velocity into the joint space dynamics equation to obtain the actual measured value of the generalized momentum. It inputs the joint angle, joint angular velocity, driving torque command, and temperature near the joint into the temperature-compensated friction model to obtain the predicted value of the generalized momentum. The difference between the actual measured value and the predicted value of the generalized momentum is used as the second joint external torque. The calculation formula is as follows: ， in, Indicates the external torque of the second joint, Represents the observer gain diagonal matrix, This represents the measured value of the generalized momentum of the joint. This represents the generalized momentum prediction value of the temperature-compensated friction model; In this step, a generalized momentum observer (GMO) is designed as an independent second-path torque estimator: based on the joint space dynamics equations, the actual measured value of the generalized momentum is compared with the predicted value based on the temperature-compensated friction model. The difference between the two reflects the unmodeled external contact torque. The observer gain diagonal matrix is ​​recommended to be set between 20 and 80 rad / s. The current estimation path and the GMO observer path mutually verify each other. The difference between the two paths can be used to detect friction compensation residuals online and trigger recalibration alarms, improving the overall robustness of contact torque sensing under high-temperature environments.

[0030] In terms of contact torque estimation, an extended Kalman filter (EKF) or an unscented Kalman filter (UKF) can be used to replace the generalized momentum observer (GMO) to jointly estimate the robot's dynamic state and external contact torque, thereby improving the estimation accuracy under low signal-to-noise ratio conditions. Furthermore, an adaptive robust Kalman filter is used to optimally fuse the sensor measurements and observer estimates, and the fusion accuracy is further improved by estimating the noise covariance matrix in real time.

[0031] S203 calculates the difference between the external torques of the first and second joints. Based on this difference and the temperature near the joint, a fusion weight is determined. Then, based on this weight, the external torques of the first and second joints are adaptively weighted and fused to obtain the final external torque. The calculation formula is as follows: ， in, Indicates the external torque of the joint, Indicates the external torque of the first joint, This indicates the fusion weight.

[0032] S2031 If the difference in external joint torque is less than or equal to the first difference threshold, then the consistency confidence index of the external joint torque is set as the first index value. S2032 If the difference in external joint torque is greater than the first difference threshold, then the consistency confidence index of the external joint torque is set as the second index value. S2033 calculates the fusion weights using the consistency confidence index and the temperature near the joint, as shown in the following formula: ， in, Representing logical functions, Indicates real-time temperature measurement near the joint, Indicates the upper limit of the effective temperature range of the friction model. Represents shape factor, Indicates the consistency confidence index; S2034 If the difference in external torque of the joint is greater than the second difference threshold, a recalibration alarm is triggered.

[0033] In this step, an adaptive weighted fusion and mode switching strategy is designed based on the results of current estimation and generalized momentum observer estimation. An estimation consistency confidence index (with a value of 0 to 1) is introduced to quantify the consistency between the two estimates in real time: when the difference in joint external torque is ≤ the difference threshold (it is recommended that the difference threshold be 8% of the rated torque), the consistency confidence index takes a high value, and the quality of both signals is good; when the deviation exceeds the threshold, the consistency confidence index decreases, triggering the mode switching judgment.

[0034] The fusion weight is jointly determined by temperature and consistency, and a shape factor of 0.5 is recommended. An adaptive weighted fusion strategy is designed: when the temperature is normal ( ≤ And when the two estimated heights are consistent, →1. Prioritize the use of the current estimate after temperature-friction compensation; when the joint temperature exceeds the effective range of the compensation model or the deviation between the two paths continuously exceeds the threshold, →0, switching to GMO observer-dominant estimation; if the deviation between the two paths is extremely large (exceeding 15% of the rated torque) and lasts for more than 500 ms, the system triggers a friction model recalibration alarm and records it to the digital archive. The automatic switching between the three modes (normal fusion / observer-dominant / alarm recording) ensures continuous and reliable contact torque estimation across the entire temperature range.

[0035] S3 calculates the contact force error based on the expected contact force and the end contact force. Using the contact force error and its rate of change as input, it judges the current state of slag resistance change through a fuzzy inference model and adjusts the impedance stiffness parameter and damping coefficient in real time. S301 acquires the end-effector contact force of the robot during the slag removal process, calculates the difference between the end-effector contact force and the expected contact force as the contact force tracking error and its rate of change, and the calculation formula is as follows: , , in, Indicates contact force tracking error, Indicates expected contact force, This represents the end contact force, where n represents the contact normal direction. Indicates the rate of change of error, The contact force tracking error at the current control sampling moment is represented by k, where k represents the current control sampling moment. Indicates the control sampling period; In this step, a positive value of the contact force tracking error indicates insufficient slag removal force (stiffness needs to be increased), while a negative value indicates that the contact force exceeds the expected value (stiffness needs to be reduced). The rate of change of the error reflects the trend of deviation and is used to predict the direction of change in slag characteristics. The control sampling period can be set to 1 ms, i.e., 1000 Hz real-time control.

[0036] S302 uses the contact force tracking error and its rate of change as input to the fuzzy inference model, and divides it into several semantic levels; S303 establishes fuzzy expert rules based on several semantic levels, and uses Gaussian membership function and centroid method to solve fuzzy output impedance stiffness adjustment increment; S304 updates the impedance stiffness parameters and damping coefficient in real time based on the impedance stiffness adjustment increment.

[0037] S3041 updates the impedance stiffness parameter in real time based on the impedance stiffness adjustment increment. The calculation formula is as follows: ， in, This represents the impedance stiffness parameter updated at the next control sampling time. k Indicates the current control sampling time, The impedance stiffness parameter at the current control sampling moment, Represents the amplitude limiting function, Indicates the increment of impedance stiffness adjustment. and These represent the lower and upper limits of the impedance stiffness parameter, respectively; S3042 calculates the damping coefficient based on the updated impedance stiffness parameters and the critical damping relationship. The calculation formula is as follows: ， in, Indicates the damping coefficient updated at the next control sampling time. This represents the desired inertia parameter.

[0038] In this step, a Mamdani-type fuzzy inference system is designed, which will... and Each is divided into 7 semantic levels (NL / NM / NS / ZO / PS / PM / PL), corresponding to Gaussian membership functions, establishing 49 expert rules in a 7×7 format. The stiffness adjustment increment is output after defuzzification using the centroid method. 200 N / m is recommended (to ensure minimum slag removal force). A damping coefficient of 3000 N / m is recommended (to prevent damage from excessive force); the damping coefficient is updated in real time with the stiffness according to the critical damping principle to ensure that the impedance system always maintains a stable response without overshoot.

[0039] In terms of impedance parameter adaptation, an online parameter adaptation method based on reinforcement learning (such as Proximal Policy Optimization, PPO) can be used to replace the fuzzy logic reasoning system. Through repeated interaction with the smelting environment, the optimal impedance parameter adjustment strategy can be automatically learned without the need for manual pre-setting of the fuzzy rule base.

[0040] The impedance control force is calculated using the S4 root impedance stiffness parameters and damping coefficients. Combined with the dynamics feedforward compensation torque, the robot joint torque control command is calculated. S401 acquires the impedance stiffness parameters and damping coefficient obtained from real-time updates. Based on the end-effector pose deviation, impedance stiffness parameters, and damping coefficient, it establishes a target impedance model for the robot's end effector and calculates the impedance control force using the following formula: , , in, Indicates the actual pose of the end effector. This represents the desired end joint angle, desired joint angular velocity, Indicates terminal error acceleration, Indicates the terminal error rate, Indicates end-effector pose error, Indicates impedance control force; S402 calculates the robot end-effector impedance control force based on the robot end-effector target impedance model, and maps it into joint space impedance control torque through the robot Jacobian matrix. S403 substitutes the desired joint angle, desired joint angular velocity, and desired joint angular acceleration into the joint space dynamics equation, and calculates the dynamic feedforward compensation torque using the joint friction torque. S404 superimposes the joint space impedance control torque with the dynamic feedforward compensation torque to obtain the robot joint torque control command.

[0041] In this step, the formula for calculating the dynamic feedforward compensation torque is as follows: τ ff = M ( q ) ref + C ( q , q ) ref + G ( q )+ F fric ( ref , T ), in, τ ff This represents the dynamic feedforward compensation torque. ref , ref These represent the desired joint angular acceleration and the desired joint angular velocity, respectively, which are generated online by the trajectory planning module based on the desired joint angle and the desired joint angular velocity. F fric ( ref , TThe value represents the frictional torque calculated by the temperature-compensated friction model described in step S2013. The dynamic feedforward compensation of this invention provides full compensation for inertia, Coriolis force, gravity, and friction, corresponding to the complete dynamic feedforward compensation described in the beneficial effects section of the specification.

[0042] In this step, an impedance control method is used, and the robot end effector is equivalent to a spring-mass-damped system. The desired inertia matrix is ​​recommended to be 0.5–2 kg in the slag removal normal direction. The desired joint angle and desired joint angular velocity of the end effector are obtained through step S1.

[0043] By combining the Cartesian space control law in the formula with the joint dynamics equations and mapping it to the joint space through the generalized inverse of the Jacobian matrix, a joint torque control command containing complete dynamic feedforward compensation (full compensation for inertia, Coriolis force, gravity, and friction) is obtained. When the robot configuration approaches a singular configuration, the method of damped least norm pseudo-inverse is automatically switched to handle the Jacobian inverse to ensure control continuity and stability.

[0044] After performing safety limiting processing on the joint torque control command, S5 outputs limiting control or yielding control based on the safety status of the end contact force, and generates a slag removal trajectory.

[0045] S501 performs amplitude limiting on the torque control commands for each robot joint to obtain the saturated joint torque. The calculation formula is as follows: ， in, i Indicates the robot joint number, Indicates the first i Torque control commands for each joint Indicates the first i The rated continuous output torque of each joint Indicates safety factor, This represents the joint torque after the i-th joint saturates. Represents the amplitude limiting function; The S502 acquires the robot's end-effector contact force in real time and calculates the magnitude of the end-effector contact force. If the magnitude of the end-effector contact force is less than the preset safe contact force threshold, a safety status flag of 1 is generated; if the magnitude of the end-effector contact force is greater than or equal to the preset safe contact force threshold, a safety status flag of 0 is generated. When the safety status flag is 1, S503 outputs the saturated joint torque. When the safety status flag is 0, it switches to emergency retreat mode and generates a retreat torque to drive the robot's end effector to retreat in the opposite direction of the contact force. The calculation formula is as follows: , in, Indicates the final output torque command. Indicating safety status signs It represents the yielding torque.

[0046] In this step, the first layer is joint torque saturation protection, which sets an independent upper limit for torque saturation for each joint to prevent the drive from overheating and being damaged due to continuous overload or causing the reducer gears to break. A safety factor of 0.80 is recommended, meaning a 20% safety margin is maintained based on the rated torque. The rated continuous output torque for the i-th joint is given by the joint hardware specifications.

[0047] The second layer is the end-effector contact force over-limit retreat protection. It continuously monitors the estimated end-effector contact force, and when its magnitude exceeds the safety threshold, it immediately switches to emergency retreat mode, driving the end to quickly and smoothly retreat in the opposite direction of the contact force. The recommended safety threshold is 150 N for aluminum alloy furnaces and 200 N for copper furnaces. The drive end retracts approximately 30 mm to a safe position, and the trigger-to-execution response delay is less than 2 ms.

[0048] After the active slag removal path is completed, the robot performs a final leveling scan of the furnace surface at a low speed (≤0.08 m / s) and low contact force (≤25 N) to generate the slag removal motion trajectory.

[0049] S504 calculates the comprehensive performance index of the slag removal motion trajectory based on a weighted quantification of contact force error, trajectory tracking error, and joint torque load. Early warning is then issued based on this comprehensive performance index. The calculation formula is as follows: J = λ 1 e F,norm + λ 2 e traj,norm + λ 3 e τ,norm , in, J Indicates overall performance indicators; e F,norm The normalized value of the contact force error is the ratio of the root mean square value of the contact force tracking error described in step S301 to the expected contact force. e traj,norm This represents the normalized value of the trajectory tracking error, which is the ratio of the root mean square value of the deviation between the actual end joint angle, the expected joint angular velocity, and the expected joint angle and the expected joint angular velocity to the upper limit of the allowable deviation. e τ,norm This represents the normalized value of the joint torque load, which is the maximum value of the ratio of the saturated torque of each joint to the rated torque. λ1. λ 2. λ 3 represents the corresponding weight coefficient, and the sum of the three is 1. It is recommended that the initial values ​​be 0.4, 0.3, and 0.3 respectively. The recommended threshold for the comprehensive performance index is 0.3, and the recommended threshold for the preset alarm is 0.6.

[0050] If the overall performance index is lower than the qualified threshold, the slag cleaning is confirmed to be completed and the furnace is safely exited. Otherwise, the local return stage P2 is used to supplement the slag cleaning. After a maximum of 3 cycles, the furnace is forcibly exited and an alarm is triggered.

[0051] When the overall performance indicators continuously exceed the preset alarm threshold, the system automatically performs speed reduction and records the data to the process monitoring system to form a traceable digital file, without stopping the automatic slag removal operation. The final control output torque is distributed to the servo drives of each joint via the EtherCAT real-time bus (1 kHz) for execution.

[0052] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0053] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0054] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0055] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0056] It should be noted that any reference signs placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The words first, second, and third, etc., do not indicate any order. These words can be interpreted as names.

[0057] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the invention.

[0058] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0059] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0060] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

Claims

1. A perception fusion adaptive compliant control method for slag cleaning in non-ferrous metal smelting robots, characterized in that, include: S1 divides the slag removal process into the probing and sensing stage, the active slag removal stage, and the leveling and confirmation stage, respectively generating the corresponding expected joint angle, expected joint angular velocity, and their estimated joint angular acceleration, expected contact force, and initial impedance parameters. The S2 robot collects joint operating parameters and joint friction during the slag removal process to construct the joint space dynamics equation, collects the temperature near the joint to correct the joint friction, calculates the joint external torque, and maps the joint external torque to obtain the end contact force; S3 calculates the contact force error based on the expected contact force and the end contact force. Using the contact force error and its rate of change as input, it judges the current state of slag resistance change through a fuzzy inference model and adjusts the impedance stiffness parameter and damping coefficient in real time. The impedance control force is calculated using the S4 root impedance stiffness parameters and damping coefficients. Combined with the dynamics feedforward compensation torque, the robot joint torque control command is calculated. After performing safety limiting processing on the joint torque control command, S5 outputs limiting control or yielding control based on the safety status of the end contact force, and generates a slag removal trajectory.

2. The perception fusion adaptive compliant control method for slag cleaning in non-ferrous metal smelting robots according to claim 1, characterized in that, The process of dividing the slag removal process into a sensory detection stage, an active slag removal stage, and a leveling confirmation stage includes: When the S101 robot's movement speed is less than or equal to the first speed threshold, it is classified into the exploration and perception stage. When the first speed threshold of S102 is less than the robot's movement speed and less than the third speed threshold, it is divided into the active slag removal stage; When the S103 robot's movement speed is less than or equal to the second speed threshold, it is classified as the leveling confirmation stage.

3. The perception fusion adaptive compliant control method for slag cleaning in non-ferrous metal smelting robots according to claim 1, characterized in that, The joint operating parameters and joint friction forces of the collection robot during the slag removal process are used to construct the joint spatial dynamics equation. The temperature near the joint is collected to correct the joint friction force, and the joint external torque is calculated. The end contact force is obtained by mapping the joint external torque, including: The S201 collects motor current, joint angle, joint angular velocity and estimated joint angular acceleration during the robot's operation to establish joint space dynamics equations, and collects the temperature near the joint and calculates the external torque of the first joint using a temperature-compensated friction model. S202 inputs the joint angle and joint angular velocity into the joint space dynamics equation to obtain the actual measured value of the generalized momentum. It inputs the joint angle, joint angular velocity, driving torque command, and temperature near the joint into the temperature-compensated friction model to obtain the predicted value of the generalized momentum. The difference between the actual measured value and the predicted value of the generalized momentum is used as the second joint external torque. The calculation formula is as follows: ， in, Indicates the external torque of the second joint, Represents the observer gain diagonal matrix, This represents the measured value of the generalized momentum of the joint. This represents the generalized momentum prediction value of the temperature-compensated friction model; S203 calculates the difference between the external torques of the first and second joints. Based on this difference and the temperature near the joint, a fusion weight is determined. Then, based on this weight, the external torques of the first and second joints are adaptively weighted and fused to obtain the final external torque. The calculation formula is as follows: ， in, Indicates the external torque of the joint, Indicates the external torque of the first joint, This indicates the fusion weight.

4. The perception fusion adaptive compliant control method for slag cleaning in non-ferrous metal smelting robots according to claim 3, characterized in that, The acquisition of motor current, joint angles, joint angular velocities, and estimated joint angular acceleration during the robot's operation is used to establish joint space dynamics equations. The acquisition of temperature near the joint, combined with a temperature-compensated friction model, is used to calculate the external torque of the first joint, including: S2011 calculates the motor joint output torque using the phase current signal of the servo motor during robot operation. The calculation formula is as follows: ， in, Indicates the output torque of the motor joint, Indicates torque coefficient, Indicates the effective value of the phase current; S2012 uses the Lagrange method to establish the joint space dynamics equations of the n-DOF cleaning robot manipulator, and calculates the robot's own dynamic torque. The joint space dynamics equations are as follows: ， in, Indicates joint angle, Indicates joint angular velocity, Represents joint angular acceleration, Represents the joint space inertia matrix, Represents the Coriolis force and centrifugal torque. Represents the gravitational moment vector, Represents theoretical joint friction torque, This indicates the driving torque command issued by the controller. Representing the geometric Jacobian matrix, This represents the six-dimensional generalized contact torque at the end point; S2013 uses a Coulomb-viscosity-temperature coupled friction model to construct a temperature-compensated friction model, and calculates the joint friction torque affected by temperature. The calculation formula is as follows: ， ， in, Represents the parameters of the Coulomb friction term, Indicates the parameters of viscous friction. Indicates temperature correction factor, Indicates the friction-temperature sensitivity coefficient, Indicates real-time temperature measurement near the joint, Indicates room temperature reference temperature; S2014 subtracts the robot's own dynamic torque, calculated from the real-time collected joint angular velocity, joint angular velocity and estimated joint angular acceleration, and the joint friction torque affected by temperature from the motor joint output torque to obtain the first joint external torque.

5. The perception fusion adaptive compliant control method for slag cleaning in non-ferrous metal smelting robots according to claim 3, characterized in that, The calculation of the difference between the external torques of the first and second joints, and the determination of the fusion weight based on the difference in external torques and the temperature near the joint, includes: S2031 If the difference in external joint torque is less than or equal to the first difference threshold, then the consistency confidence index of the external joint torque is set as the first index value. S2032 If the difference in external joint torque is greater than the first difference threshold, then the consistency confidence index of the external joint torque is set as the second index value. S2033 calculates the fusion weights using the consistency confidence index and the temperature near the joint, as shown in the following formula: ， in, Representing logical functions, Indicates real-time temperature measurement near the joint, Indicates the upper limit of the effective temperature range of the friction model. Represents shape factor, Indicates the consistency confidence index; S2034 If the difference in external torque of the joint is greater than the second difference threshold, a recalibration alarm is triggered.

6. The perception fusion adaptive compliant control method for slag cleaning in non-ferrous metal smelting robots according to claim 1, characterized in that, The process of calculating the contact force error based on the expected contact force and the end contact force, using the contact force error and its rate of change as input, and determining the current state of slag resistance change through a fuzzy inference model to adjust the impedance stiffness parameters and damping coefficient in real time includes: S301 acquires the end-effector contact force of the robot during the slag removal process, calculates the difference between the end-effector contact force and the expected contact force as the contact force tracking error and its rate of change, and the calculation formula is as follows: ， ， in, Indicates contact force tracking error, Indicates expected contact force, This represents the end contact force, where n represents the contact normal direction. Indicates the rate of change of error, The contact force tracking error at the current control sampling moment is represented by k, where k represents the current control sampling moment. Indicates the control sampling period; S302 uses the contact force tracking error and its rate of change as input to the fuzzy inference model, and divides it into several semantic levels; S303 establishes fuzzy expert rules based on several semantic levels, and uses Gaussian membership function and centroid method to solve fuzzy output impedance stiffness adjustment increment; S304 updates the impedance stiffness parameters and damping coefficient in real time based on the impedance stiffness adjustment increment.

7. The perception fusion adaptive compliant control method for slag cleaning in non-ferrous metal smelting robots according to claim 6, characterized in that, The real-time updating of the impedance stiffness parameter and damping coefficient based on the impedance stiffness adjustment increment includes: S3041 updates the impedance stiffness parameter in real time based on the impedance stiffness adjustment increment. The calculation formula is as follows: ， in, This represents the impedance stiffness parameter updated at the next control sampling time. k Indicates the current control sampling time, The impedance stiffness parameter at the current control sampling moment, Represents the amplitude limiting function, Indicates the increment of impedance stiffness adjustment. and These represent the lower and upper limits of the impedance stiffness parameter, respectively; S3042 calculates the damping coefficient based on the updated impedance stiffness parameters and the critical damping relationship. The calculation formula is as follows: ， in, Indicates the damping coefficient updated at the next control sampling time. This represents the desired inertia parameter.

8. The perception fusion adaptive compliant control method for slag cleaning in non-ferrous metal smelting robots according to claim 1, characterized in that, The root impedance stiffness parameter and damping coefficient are used to calculate the impedance control force. Combined with the feedforward compensation torque in dynamics, the robot joint torque control commands are calculated, including: S401 acquires the impedance stiffness parameters and damping coefficient obtained from real-time updates. Based on the end-effector pose deviation, impedance stiffness parameters, and damping coefficient, it establishes a target impedance model for the robot's end effector and calculates the impedance control force using the following formula: ， ， in, Indicates the actual pose of the end effector. This represents the desired end joint angle, desired joint angular velocity, Indicates terminal error acceleration, Indicates the terminal error rate, Indicates end-effector pose error, Indicates impedance control force; S402 calculates the robot end-effector impedance control force based on the robot end-effector target impedance model, and maps it into joint space impedance control torque through the robot Jacobian matrix. S403 substitutes the desired joint angle, desired joint angular velocity, and desired joint angular acceleration into the joint space dynamics equation, and superimposes the joint friction torque to calculate the dynamic feedforward compensation torque. τ ff The calculation formula is as follows: τ ff = M ( q ) ref + C ( q , q ) ref + G ( q )+ F fric ( ref , T )， in, τ ff Indicates the dynamic feedforward compensation torque, ref , ref These represent the desired joint angular acceleration and the desired joint angular velocity, respectively. S404 superimposes the joint space impedance control torque with the dynamic feedforward compensation torque to obtain the robot joint torque control command.

9. The perception fusion adaptive compliant control method for slag cleaning in non-ferrous metal smelting robots according to claim 1, characterized in that, After performing safety limiting processing on the joint torque control command, the output of limiting control or yielding control is based on the safe state of the end contact force, and the generation of the slag removal trajectory includes: S501 performs amplitude limiting on the torque control commands for each robot joint to obtain the saturated joint torque. The calculation formula is as follows: ， in, i Indicates the robot joint number, Indicates the first i Torque control commands for each joint Indicates the first i The rated continuous output torque of each joint Indicates safety factor, This represents the joint torque after the i-th joint saturates. Represents the amplitude limiting function; The S502 acquires the robot's end-effector contact force in real time and calculates the magnitude of the end-effector contact force. If the magnitude of the end-effector contact force is less than the preset safe contact force threshold, the generated safety status flag is 1; if the magnitude of the end-effector contact force is greater than or equal to the preset safe contact force threshold, the generated safety status flag is 0. When the safety status flag is 1, S503 outputs the saturated joint torque. When the safety status flag is 0, it switches to emergency retreat mode and generates a retreat torque to drive the robot's end effector to retreat in the opposite direction of the contact force. The calculation formula is as follows: ， in, Indicates the final output torque command. Indicating safety status signs Indicates the yielding moment; S504 calculates the comprehensive performance index of the slag removal motion trajectory based on a weighted quantification of contact force error, trajectory tracking error, and joint torque load. Early warning is then issued based on this comprehensive performance index. The calculation formula is as follows: J = λ 1 e F,norm + λ 2 e traj,norm + λ 3 e τ,norm ， in, J Indicates comprehensive performance indicators, e F,norm This represents the normalized value of the contact force error, which is the ratio of the root mean square value of the contact force tracking error to the expected contact force. e traj,norm This represents the normalized value of the trajectory tracking error, calculated as the ratio of the root mean square value of the deviation between the actual end joint angle, the desired joint angular velocity, and the desired joint angle, to the upper limit of the allowable deviation. e τ,norm This represents the normalized value of the joint torque load, which is the maximum value of the ratio of the saturated torque of each joint to the rated torque. λ 1. λ 2. λ 3 represents the corresponding weighting coefficient.

Images