A battery grabbing compensation method based on multi-sensor data fusion

Abstract

The application relates to the technical field of mine automation, and discloses a battery grabbing compensation method based on multi-sensor data fusion, which comprises the following steps: collecting multi-dimensional original data; monitoring the working condition in real time; when the vibration amplitude exceeds a threshold value, increasing the posture and deformation data weight and reducing the spacing data weight; when the dust concentration exceeds a threshold value, increasing the obstacle and grabbing force data weight and reducing the three-dimensional image data weight; fusing the data to generate comprehensive information; constructing a deviation prediction model based on mine car motion parameters and environment parameters, and outputting pre-compensation parameters; identifying the mine car type, calling a preset grabbing mode or generating a temporary grabbing scheme; adjusting the grabbing point in combination with the pre-compensation parameters; continuously collecting dynamic data in the grabbing process, evaluating the actual grabbing deviation, dynamically adjusting the weight and parameters, and stopping until stable grabbing is realized. The application can improve the grabbing stability and success rate to meet the efficient and reliable operation requirements of a mine.

Description

Technical Field

[0001] This application relates to the technical field of mine automation, and in particular to a battery grabbing and compensation method based on multi-sensor data fusion. Background Technology

[0002] Multi-vehicle battery grabbing in underground mines is crucial for ensuring the continuous operation of mining trucks. However, this scenario is affected by factors such as truck bumps, high dust concentrations, mixed vehicle types, and limited working space, placing stringent demands on grabbing accuracy and stability. Existing technologies mostly rely on a single sensor for detection and compensation, resulting in weak anti-interference capabilities—bumps can easily cause fluctuations in displacement sensor data, and dust can obstruct the field of view of the vision sensor, leading to decreased accuracy in battery position / angle detection and frequent grabbing misalignments, severely impacting operational efficiency and safety.

[0003] Existing technologies suffer from the following problems: First, they lack a dynamic sensor data fusion mechanism under multiple interference factors, making it impossible to adjust data weights according to real-time operating conditions, resulting in insufficient accuracy of fused information. Second, they have poor vehicle adaptability, making it difficult to adapt to known / unknown vehicle models, and they do not combine historical mine data to build a multi-factor coupled pre-compensation model, making it impossible to predict battery offset in advance. Third, most of them lack grasping closed-loop control systems, which cannot evaluate deviations and correct strategies in real time, resulting in low grasping stability and success rate when switching vehicle models in limited spaces, making it difficult to meet the needs of efficient and reliable operation in mines.

[0004] As can be seen from the above, how to improve the stability and success rate of grasping to meet the needs of efficient and reliable operation in mines still needs to be solved. Summary of the Invention

[0005] To improve the stability and success rate of battery gripping and meet the needs of efficient and reliable operation in mines, this application provides a battery gripping compensation method based on multi-sensor data fusion.

[0006] Firstly, this application provides a battery grasping compensation method based on multi-sensor data fusion, employing the following technical solution:

[0007] A battery grabbing compensation method based on multi-sensor data fusion includes:

[0008] The multi-dimensional data acquisition module is used to collect multi-dimensional raw data, including the mine truck's speed, steering angle, road slope data, obstacle information in the working environment, three-dimensional image displacement sensing of the battery interface, force take-off data and deformation data of the elastic damping component, distance information between the gripper and the battery, and real-time posture data of the robotic arm.

[0009] Based on real-time monitoring and judgment of multi-dimensional raw data, if the vibration amplitude of the mining truck exceeds the preset vibration threshold, the weight of the attitude data collected by the attitude monitoring sensor and the deformation data collected by the elastic damping component is increased, while the weight of the spacing data collected by the laser displacement sensor is decreased; if the dust concentration in the working environment exceeds the preset dust threshold, the weight of the obstacle data collected by the millimeter-wave radar and the gripping force data collected by the force sensing sensor is increased, while the weight of the three-dimensional image data collected by the binocular vision sensor is decreased; after adjusting the weights, the multi-dimensional raw data is fused to generate comprehensive information including working condition information and battery target information;

[0010] Based on the motion parameters of the mining truck and the environmental physical field parameters, a displacement prediction model corresponding to the battery position and angle is constructed. The motion parameters of the mining truck include the driving speed, steering angle and road slope data, and the environmental physical field parameters include the battery temperature and dust concentration. The displacement prediction model is used to calculate and predict the position displacement and angle displacement of the battery under the current working conditions to obtain the corresponding displacement prediction results. Based on the displacement prediction results, the corresponding pre-compensation parameters are output.

[0011] During the vehicle battery swapping phase, the model of the mining truck to which the battery to be grabbed belongs is determined. If the model is known, a pre-stored preset grabbing mode matching the known model is directly called. If the model is unknown, the probe claw set at the end of the robotic arm is activated to scan the external contour and surface hardness of the battery. A temporary grabbing plan is generated based on the scan results. The grabbing point position of the robotic arm's gripper is adjusted based on the preset grabbing mode or the temporary grabbing plan, combined with pre-compensation parameters.

[0012] During the battery gripping process, dynamic data from each sensor is continuously collected in real time. Based on the collected dynamic data, the actual gripping deviation between the robotic arm gripper and the battery is evaluated. If the actual gripping deviation exceeds the preset deviation range, the data fusion weights of each sensor are adjusted based on the actual gripping deviation, and the corresponding pre-compensation parameters are corrected. The gripping point of the robotic arm gripper is adjusted synchronously until the robotic arm stably grips the battery.

[0013] Optionally, when assessing and adjusting actual grasping deviations, the deviations are divided into three levels and corresponding to different adjustment strategies. The method also includes:

[0014] If the actual grasping deviation is a first-level deviation, fine-tune the gripper displacement compensation amount in the pre-compensation parameters, without adjusting the sensor fusion weights.

[0015] If the actual grasping deviation is a level two deviation, the sensor fusion weight and pre-compensation parameters are adjusted synchronously. Under vibration conditions, the weight of attitude data is increased, and under dust conditions, the weight of force sensing data is increased.

[0016] If the actual capture deviation is a level three deviation, immediately pause the capture action, re-execute the multi-dimensional data collection and dynamic weight fusion steps, locate the accurate position of the battery, and then restart the capture. At the same time, mark the deviation data as an abnormal sample and store it in the historical database.

[0017] Optionally, in evaluating the actual gripping deviation between the robotic arm gripper and the battery based on the collected dynamic data, the method further includes:

[0018] Retrieve the corresponding historical deviation data, obtain the corresponding current operating parameters, and automatically generate deviation correction coefficients based on the deviation correlation between historical deviation data and current operating parameters.

[0019] The deviation correction coefficient is combined with the current actual grasping deviation value to calculate the corrected deviation value; the gripper displacement compensation amount in the pre-compensation parameters is dynamically adjusted based on the corrected deviation value, wherein the adjustment range is proportional to the deviation correction coefficient and the adjustment rate does not exceed the preset dynamic adjustment upper limit.

[0020] Optionally, the method further includes:

[0021] A distributed temperature sensor array is set on the contact surface of the robotic arm gripper to monitor the temperature distribution on the battery surface in real time.

[0022] When the local temperature of the battery is detected to exceed a preset threshold based on the battery surface temperature distribution, the local cooling channel of the gripper contact surface is automatically activated to perform directional cooling of the high-temperature area;

[0023] Based on the correlation between battery temperature distribution and the migration prediction model, the weighting coefficients of temperature parameters in the migration prediction model are dynamically adjusted.

[0024] Optionally, the method further includes:

[0025] By analyzing the causal relationship between historical deviation data and current operating parameters using machine learning algorithms, a knowledge graph of operating conditions and deviation causes is established based on the causal relationship, and the causes of deviations are classified into vibration disturbance type, dust interference type, temperature influence type and multi-factor coupling type.

[0026] When determining the actual level of deviation, the system matches the current operating parameters with the operating condition-deviation cause knowledge graph to automatically identify the type of deviation cause.

[0027] The optimal compensation strategy is retrieved from the pre-stored compensation strategy library based on the type of deviation cause. The optimal compensation strategy includes the sensor weight adjustment ratio, the pre-compensation parameter correction coefficient, and the robot arm motion trajectory optimization parameters.

[0028] Optionally, the method further includes:

[0029] A battery interface alignment evaluation model is constructed based on the real-time comparison of displacement sensing data corresponding to the 3D image of the battery interface and the actual battery gripping position. The battery interface alignment evaluation model calculates the corresponding interface alignment index by analyzing the displacement vector of the feature points on the interface edge.

[0030] When the interface alignment index is lower than the preset alignment threshold, the gripping posture of the robotic arm is automatically adjusted, the rotation angle compensation is increased, and the gripper rotation angle compensation in the pre-compensation parameters is adjusted simultaneously.

[0031] Based on the correlation between the interface alignment index and the actual grasping deviation, the weight coefficients of the interface alignment parameters in the offset prediction model are dynamically adjusted to make the prediction results more consistent with the actual working conditions. After the robotic arm completes the grasping, the interface alignment index and deviation data are associated and stored.

[0032] Secondly, this application provides a battery grasping and compensation system based on multi-sensor data fusion, which adopts the following technical solution:

[0033] A battery grasping and compensation system based on multi-sensor data fusion includes:

[0034] The multi-dimensional data acquisition module is used to collect multi-dimensional raw data, including the mine truck's speed, steering angle, road slope data, obstacle information in the working environment, three-dimensional image displacement sensing of the battery interface, force take-off data and deformation data of the elastic damping component, distance information between the gripper and the battery, and real-time posture data of the robotic arm.

[0035] The adaptive weighted fusion module monitors and judges the current working condition in real time based on multi-dimensional raw data. If the vibration amplitude of the mining truck exceeds the preset vibration threshold, the weight of the attitude data collected by the attitude monitoring sensor and the deformation data collected by the elastic damping component is increased, while the weight of the spacing data collected by the laser displacement sensor is decreased. If the dust concentration in the working environment exceeds the preset dust threshold, the weight of the obstacle data collected by the millimeter-wave radar and the grasping force data collected by the force sensing sensor is increased, while the weight of the three-dimensional image data collected by the binocular vision sensor is decreased. After adjusting the weights, the multi-dimensional raw data is fused to generate comprehensive information that includes working condition information and battery target information.

[0036] The pre-compensation parameter output module constructs a displacement prediction model corresponding to the battery position and angle based on the mine car motion parameters and environmental physical field parameters. The mine car motion parameters include the mine car speed, steering angle, and road slope data, while the environmental physical field parameters include the battery temperature and dust concentration. The displacement prediction model calculates and predicts the battery's position and angle displacement under the current working conditions to obtain the corresponding displacement prediction results. Based on the displacement prediction results, the corresponding pre-compensation parameters are output.

[0037] The vehicle model adaptive grasping mode generation module determines the vehicle model of the mining truck to which the battery to be grasped belongs during the vehicle battery swapping stage. If the model is known, it directly calls the pre-stored preset grasping mode that matches the known model. If the model is unknown, it activates the probe claw set at the end of the robotic arm to scan the external contour and surface hardness of the battery, and generates a temporary grasping plan based on the scan results. The grasping point position of the robotic arm gripper is adjusted based on the preset grasping mode or the temporary grasping plan and pre-compensation parameters.

[0038] The real-time deviation dynamic compensation module continuously collects dynamic data from each sensor during the battery gripping process. Based on the collected dynamic data, it evaluates the actual gripping deviation between the robotic arm gripper and the battery. If the actual gripping deviation exceeds the preset deviation range, it adjusts the data fusion weights of each sensor and corrects the corresponding pre-compensation parameters based on the actual gripping deviation, and synchronously adjusts the gripping point of the robotic arm gripper until the robotic arm stably grips the battery.

[0039] Thirdly, this application provides a battery grasping and compensation system based on multi-sensor data fusion, which adopts the following technical solution:

[0040] A battery grabbing compensation system based on multi-sensor data fusion includes a processor, in which a program for a battery grabbing compensation method based on multi-sensor data fusion, any one of the above-mentioned methods, runs.

[0041] Fourthly, this application provides a storage medium, which adopts the following technical solution:

[0042] A storage medium storing a program for a battery grabbing compensation method based on multi-sensor data fusion, as described above.

[0043] In summary, this application includes at least one of the following beneficial technical effects:

[0044] By integrating multi-dimensional sensor data in real time and using a dynamic working condition adaptive mechanism, the stability of battery grasping is significantly improved: the sensor weights are adjusted in real time based on environmental parameters such as mine truck vibration and dust, a deviation prediction model is constructed to generate pre-compensation parameters in advance, and a three-level deviation classification strategy (first-level fine-tuning, second-level synchronous optimization, and third-level repositioning) is adopted during the grasping process to effectively suppress the cumulative deviation caused by environmental interference, ensuring the smoothness and accuracy of the robotic arm's grasping action under complex mine working conditions, and reducing the actual grasping deviation rate to below 5%.

[0045] Meanwhile, the innovative integration of the working condition-deviation cause knowledge graph and the interface alignment evaluation model automatically identifies deviation causes such as vibration disturbance and dust interference through historical data mining, dynamically retrieves the optimal compensation strategy (such as adjusting the sensor weight ratio and rotation angle compensation amount), and combines battery surface temperature monitoring and local cooling mechanism to prevent structural stress failure caused by thermal damage; the real-time calculation and feedback optimization of the interface alignment index further improves the battery interface matching accuracy, making the grab success rate stably reach more than 99.2%, meeting the needs of high-frequency and high-reliability battery swapping operations in mines. Attached Figure Description

[0046] Figure 1 This is a flowchart illustrating a battery capture compensation method based on multi-sensor data fusion according to an exemplary embodiment.

[0047] Figure 2 This is a structural block diagram of a battery grabbing and compensation system based on multi-sensor data fusion, according to an exemplary embodiment. Detailed Implementation

[0048] The embodiments of this application are described in detail below, and examples of the embodiments are shown in the accompanying drawings.

[0049] In the description of this specification, the references to "certain embodiments," "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples" indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer 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.

[0050] This application discloses a battery grasping compensation method based on multi-sensor data fusion, referring to... Figure 1 ,include:

[0051] The S100 collects multi-dimensional raw data, including the mine truck's speed, steering angle, road slope data, obstacle information in the working environment, three-dimensional image displacement sensing of the battery interface, force take-off data and deformation data of the elastic damping component, distance information between the gripper and the battery, and real-time posture data of the robotic arm.

[0052] The motion parameters of the mining truck are collected synchronously by on-board sensors such as inertial measurement units (IMU), GPS and gyroscopes. Specifically, these include the mining truck's speed (obtained in real time by wheel speed sensors or lidar, ranging from 0 to 10 km / h), steering angle (detected by steering angle sensors or steering motor encoders, with an accuracy of ±45 degrees), and road slope (measured by tilt sensors, with a dynamic range of ±15 degrees).

[0053] The operational environment perception data is achieved using millimeter-wave radar, binocular vision sensors, and LiDAR. It can detect the distance (e.g., 0.5 to 30 meters) and azimuth (±180 degrees) of surrounding obstacles in real time. At the same time, it uses laser dust sensors to monitor the concentration of particulate matter in the air (e.g., PM2.5 levels), providing quantitative basis for environmental interference assessment.

[0054] The battery interface positioning data is completed collaboratively by a binocular vision sensor and a structured light scanner. A high-precision point cloud map of the battery interface is generated through stereo vision algorithms (positioning accuracy up to ±0.1 mm), and the offset of the interface in the X / Y / Z axis directions is calculated in real time to ensure accurate identification of the battery target position.

[0055] The acquisition of robotic arm dynamics data relies on joint torque sensors and strain gauges to record the output torque of each joint and the deformation of the gripper elastic damping component (measurement range 0-5 mm), providing real-time feedback for the dynamic response of the robotic arm.

[0056] Data on the interaction between the gripper and the battery is acquired synchronously through a laser displacement sensor and a force sensor, including the distance from the gripper tip to the battery surface (measurement range 0-100 mm) and the contact force applied by the gripper (range 0-200 N), accurately reflecting the physical interaction state during the gripping process.

[0057] The robotic arm's posture data is collected in real time by the end-effector IMU and encoder, and the output includes complete spatial information including end-effector pose coordinates (X / Y / Z) and attitude angles, ensuring real-time tracking and calibration of the robotic arm's motion trajectory.

[0058] By synchronously collecting and fusing high-precision, multi-source heterogeneous data, a complete dataset covering environmental conditions, mechanical motion, target positioning, and interactive physical characteristics was constructed. This provided real-time and reliable input for subsequent adaptive judgment of working conditions, construction of offset prediction models, and execution of dynamic compensation strategies, directly supporting a significant improvement in the stability and success rate of battery grabbing in complex mining environments.

[0059] The S200 monitors and judges the current working conditions in real time based on multi-dimensional raw data. If the vibration amplitude of the mining truck exceeds the preset vibration threshold, the weight of the attitude data collected by the attitude monitoring sensor and the deformation data collected by the elastic damping component is increased, while the weight of the spacing data collected by the laser displacement sensor is decreased. If the dust concentration in the working environment exceeds the preset dust threshold, the weight of the obstacle data collected by the millimeter-wave radar and the gripping force data collected by the force sensing sensor is increased, while the weight of the three-dimensional image data collected by the binocular vision sensor is decreased. After adjusting the weights, the multi-dimensional raw data is fused to generate comprehensive information that includes working condition information and battery target information.

[0060] First, the system continuously analyzes multi-dimensional raw data from the S100 to calculate the vibration amplitude of the mining truck (based on the root mean square value of acceleration data collected by the IMU sensor) and the dust concentration in the working environment (by obtaining real-time PM2.5 readings from a laser dust sensor). When the vibration amplitude exceeds a preset threshold of 0.5 m / s², the system automatically identifies it as a vibration condition; when the dust concentration exceeds a preset dust threshold, the system automatically identifies it as a dust condition.

[0061] Under vibration conditions, the system dynamically increases the weight of attitude data acquired by the attitude monitoring sensor and the weight of deformation data acquired by the elastic damping component, while decreasing the weight of spacing data acquired by the laser displacement sensor. Specifically, the weight of attitude data is increased by 10%, the weight of deformation data is increased by 10%, and the weight of spacing data is decreased by 10%, in order to avoid laser ranging distortion caused by vibration and ensure the reliability of attitude and deformation data in high vibration environments.

[0062] Under dusty conditions, the system dynamically increases the weight of obstacle data acquired by millimeter-wave radar and the weight of grasping force data acquired by force sensing sensor, while decreasing the weight of 3D image data acquired by binocular vision sensor. Specifically, the weight of obstacle data is increased by 15%, the weight of grasping force data is increased by 15%, and the weight of 3D image data is decreased by 15%, to overcome the blurring interference of dust on visual images and ensure the accuracy of obstacle recognition and grasping force measurement.

[0063] After the weights are adjusted, the system employs an adaptive weighted fusion mechanism to process the multi-dimensional raw data in real time, prioritizing the fusion of high-weight data sources. For example, under vibration conditions, it relies heavily on attitude and deformation data, while under dust conditions, it relies heavily on millimeter-wave radar and force sensing data. The fusion process is completed at a sampling frequency of 200Hz, ensuring that the data processing latency is less than 50 milliseconds, generating a comprehensive information matrix that includes the current operating condition level (such as vibration intensity and dust concentration) and precise information about the battery target (such as interface offset and attitude angle).

[0064] By employing real-time operational condition sensing and dynamic weight adjustment mechanisms, the impact of mine environmental interference on data quality was effectively suppressed: under vibration conditions, it avoided grasping offsets caused by spacing data distortion; under dust conditions, it overcame positioning errors caused by visual data blurring, thus improving the reliability of comprehensive information. This provides high-precision input for subsequent offset prediction models and dynamic compensation strategies, directly increasing the success rate of battery grasping in complex mine environments.

[0065] Through real-time operational condition perception and dynamic weight adjustment mechanisms, the impact of mine environmental interference on data quality was effectively suppressed: under vibration conditions, it avoided grasping offset caused by spacing data distortion; under dust conditions, it overcame positioning errors caused by visual data blurring, improving the reliability of comprehensive information to over 98.5%. This provides high-precision input for subsequent offset prediction models and dynamic compensation strategies, directly increasing the battery grasping success rate in complex mine environments.

[0066] The S300 constructs a displacement prediction model for the battery position and angle based on the mine car motion parameters and environmental physical field parameters. The mine car motion parameters include the mine car speed, steering angle, and road slope data, while the environmental physical field parameters include the battery temperature and dust concentration. The displacement prediction model calculates and predicts the battery position and angle displacement under the current working conditions to obtain the corresponding displacement prediction results, and outputs the corresponding pre-compensation parameters based on the displacement prediction results.

[0067] Firstly, during the system initialization phase, a deviation prediction model was constructed using historical battery swapping data from the mine. By analyzing detailed records of 12,000 battery swapping operations over the past three months, the system extracted the correlation patterns between mine truck speed, steering angle, road slope, and actual battery deviation, while also integrating environmental data such as battery temperature and dust concentration. For example, when the mine truck turns at a speed of 4–6 km / h, the battery often exhibits a lateral deviation of 0.3–0.5 mm along the X-axis; when the dust concentration exceeds a certain value, the deviation increases by an additional 0.1–0.2 mm. These patterns are encoded into a model knowledge base, and through deep learning algorithms, the nonlinear relationships between parameters are automatically learned, forming a predictive capability that can be invoked in real time.

[0068] In actual operation, the system continuously receives comprehensive information generated by the S200, and reads the current movement parameters of the mining truck (such as driving speed of 5.8 km / h, steering angle of 28 degrees, and road slope of 9 degrees) and environmental physical field parameters in real time. This data is input into the trained offset prediction model, which automatically matches historical working condition patterns and accurately calculates the specific offset of the battery in the current environment. For example, the system outputs an X-axis position offset of 0.45 mm, a Y-axis offset of 0.22 mm, and a Yaw angle offset of 1.6 degrees, ensuring that the prediction results are highly consistent with the actual working conditions.

[0069] Based on the predicted offset, the system automatically generates executable pre-compensation parameters. For example, based on an X-axis offset of 0.45mm, the system calculates that the robotic arm gripper needs to move 0.45mm in advance in the horizontal direction; for a yaw angle offset of 1.6 degrees, a rotation angle compensation command is generated. These parameters are converted into robotic arm motion commands and directly embedded into the initial stage of the gripping process, enabling the gripper to actively adjust its position and attitude before contacting the battery, thus avoiding gripping failure caused by dynamic offset.

[0070] By pre-quantifying the impact of environmental disturbances on battery position and generating precise compensation commands, the problem of grasping accuracy under complex mining conditions is fundamentally solved. This allows the robotic arm to proactively adapt to disturbances such as vibration and dust before contacting the battery, avoiding the lag of relying on post-processing adjustments in traditional methods. This mechanism increases the battery grasping success rate from 82% to 99.2%, reduces the single battery swapping time to within 12 seconds, and decreases the number of repeated adjustments by the robotic arm by 70%, significantly improving the stability and efficiency of high-frequency battery swapping operations in mines and providing key technical support for unmanned mining operations.

[0071] During the battery swapping phase, the S400 determines the model of the mining truck to which the battery to be grabbed belongs. If the model is known, it directly calls the pre-stored preset grabbing mode that matches the known model. If the model is unknown, it activates the probe claw set at the end of the robotic arm to scan the external contour and surface hardness of the battery. Based on the scanning results, a temporary grabbing plan is generated. The grabbing point position of the robotic arm's gripper is adjusted based on the preset grabbing mode or the temporary grabbing plan, combined with pre-compensation parameters.

[0072] During the battery swapping phase, the system immediately performs rapid vehicle identification using a high-definition camera at the end of the robotic arm and an onboard RFID reader. The camera scans the mine truck's body markings (such as the chassis number or logo), and the RFID reader reads the mine truck's electronic tag (such as an embedded chip). The system completes the matching within 0.2 seconds, for example, identifying the mine truck model as "Mine Truck S-2025" and confirming it as a known model. If a match cannot be found (such as the newly deployed "Mine Truck T-2026"), the unknown model processing procedure is automatically triggered.

[0073] For known vehicle models, the system directly calls the pre-stored dedicated gripping mode. This mode is deeply optimized based on battery swapping data from the past 800 similar vehicle models, including precise gripper opening and closing angles, gripping force thresholds (140~185N), and motion trajectory parameters. For example, for the "Mining Truck S-2025", the system presets the initial compensation values ​​for the gripping point (X-axis +0.2mm, Y-axis -0.1mm) and intelligently integrates the real-time pre-compensation parameters output by the S300 (such as an additional 0.45mm adjustment for the X-axis) to ensure that the gripper completes precise positioning before contacting the battery, with the error controlled within 0.1mm.

[0074] If an unknown vehicle model is detected, the system immediately activates the probe claw at the end of the robotic arm. Equipped with a high-precision laser scanner and a piezoelectric hardness sensor, the probe claw performs a full-range scan of the battery at 12 frames per second: the laser scan generates a millimeter-level point cloud map of the battery's external contour (accuracy ±0.05mm, covering details such as battery edges and interface grooves), while the hardness sensor obtains surface hardness values ​​(range 0-98 Shore A, reflecting battery material characteristics) through micro-pressure testing. The scanning process takes only 1.3 seconds. The system analyzes contour features (such as interface depth 2.5mm, edge curvature radius 10mm) and hardness data in real time, automatically generating a temporary gripping plan, such as determining the optimal gripping point coordinates (X: 151.8mm, Y: 87.4mm) and recommended gripping force (162N), ensuring the plan is perfectly adapted to the battery's physical characteristics.

[0075] The system dynamically integrates preset gripping modes or temporarily generated schemes with the pre-compensation parameters provided by the S300. For example, if a temporary scheme suggests an X-axis offset of 0.35mm, while the pre-compensation parameters indicate an X-axis increase of 0.45mm, the system calculates a total compensation value of 0.8mm and generates precise robotic arm motion commands. This controls the gripper to complete a 0.8mm coordinate adjustment before contacting the battery, while dynamically matching the gripping force to 162N to ensure a smooth fit between the gripper and the battery surface, avoiding battery damage or gripping failure due to hard contact.

[0076] By employing intelligent vehicle identification and dynamic grabbing scheme generation, the challenge of battery swapping in mining environments with diverse vehicle types has been completely solved. This system can automatically handle 99.5% of known vehicle types (without manual intervention) and all unknown vehicle types (generating a scheme within 1.3 seconds), increasing the battery grabbing success rate from 85% to 99.5%, reducing the single battery swapping time to within 8.5 seconds, and decreasing the number of repetitive adjustments to the robotic arm by 85%. This mechanism not only ensures the continuity of high-frequency, high-reliability automated battery swapping operations in mines but also provides scalability for seamless integration of new vehicle types in the future, becoming a key technological pillar for achieving unmanned mining operations.

[0077] During the battery gripping process, the S500 continuously collects dynamic data from each sensor in real time. Based on the collected dynamic data, it evaluates the actual gripping deviation between the robotic arm gripper and the battery. If the actual gripping deviation exceeds the preset deviation range, it adjusts the data fusion weights of each sensor and corrects the corresponding pre-compensation parameters based on the actual gripping deviation, and synchronously adjusts the gripping point of the robotic arm gripper until the robotic arm stably grips the battery.

[0078] During the battery gripping process, high-frequency dynamic data is continuously collected by a laser displacement sensor, force sensor, and IMU at the end of the robotic arm. The laser sensor scans the distance between the gripper and the battery surface 100 times per second (range 0-100mm), the force sensor records 50 changes in contact force per second (0-200N), and the IMU provides 200 attitude angle data per second, ensuring that the data stream covers the entire gripping process in real time and provides sufficient basis for deviation assessment.

[0079] The system analyzes this dynamic data in real time to accurately calculate the actual gripping deviation between the gripper and the battery. For example, when the laser rangefinder shows an X-axis offset of 0.6mm (with a pre-compensation parameter of 0.4mm) and the force sensing data shows a contact force fluctuation exceeding 10N, the system determines that the actual deviation has exceeded the preset safety threshold of 0.5mm and triggers the dynamic adjustment mechanism.

[0080] When the deviation exceeds the limit, the system automatically optimizes the weighting of sensor data fusion. Specifically, the system increases the weighting of the attitude data from the force sensor and IMU by 15% each, while decreasing the weighting of the laser displacement sensor by 10% to prioritize the use of high-reliability data sources. For example, when the deviation is too large, the system increases the weighting of the force sensor data from 40% to 55% to ensure more accurate grasping force measurement.

[0081] Based on the adjusted weights, the system corrects the pre-compensation parameters in real time. For example, if the original pre-compensation parameter for the X-axis is 0.4mm, the system calculates a new compensation value of 0.45mm based on the current deviation of 0.6mm, generates precise coordinate adjustment instructions, and dynamically optimizes the gripping force threshold to 155N to ensure that the parameters match the real-time operating conditions.

[0082] The robotic arm controller synchronously executes adjustment commands to fine-tune the gripper's grasping point position. For example, the system sends a command to move the X-axis by 0.45mm, causing the gripper to complete position correction within 0.3 seconds. At the same time, the gripping force is adjusted to 155N to achieve a smooth fit between the gripper and the battery surface, avoiding battery damage or detachment due to hard contact.

[0083] The system continuously monitors the adjusted status and reassesses the deviation every 50 milliseconds. When the laser ranging deviation drops below 0.3mm and the force sensing data stabilizes in the 150-160N range, the system confirms that a stable grasping state has been reached, automatically stops adjusting, and enters the battery fixing stage, ensuring a seamless grasping process.

[0084] By employing real-time deviation detection and dynamic adaptive adjustment, the lag problem in the grasping process is completely eliminated. It increases the battery grasping success rate from 99.2% to 99.8%, reduces the single-grab failure rate to below 0.2%, and shortens the grasping adjustment time to within 0.5 seconds, significantly improving the continuity and reliability of battery swapping operations in mines. This mechanism enables the robotic arm to maintain high-precision grasping even under complex vibrations and dust interference, providing a crucial guarantee for high-frequency battery swapping in unmanned mines.

[0085] During routine battery swapping operations in the mine, a mine car encountered complex conditions of high vibration (0.6 m / s²) and high dust concentration when turning at a speed of 5.2 km / h. Traditional systems often fail to grasp the battery due to misalignment caused by the gripper and blurred visual image due to vibration and dust. After the system starts up S100, the IMU and dust sensor collect vibration amplitude and dust data in real time. S200 quickly determines that it is a dual-condition system of vibration and dust, dynamically increases the weight of attitude data and deformation data by 10%, and reduces the weight of laser spacing to ensure data reliability. Based on the mine car's turning angle of 28 degrees, the road slope of 9 degrees and the dust concentration, S300 accurately predicts the battery's X-axis offset of 0.45mm and Yaw angle offset of 1.6 degrees, and generates pre-compensation parameters. S400 identifies the mine car model as "Mine Car T-2026" (unknown model), activates the detection claw to complete the battery contour scan and hardness test within 1.3 seconds, automatically generates a temporary gripping plan, and dynamically integrates the pre-compensation parameters with the plan, so that the gripper completes a 0.8mm coordinate adjustment before contact.

[0086] During the grasping process, the S500 continuously monitors the system using a laser displacement sensor (scanning 100 times per second) and a force sensor (recording 50 times per second). Upon detecting an actual X-axis offset of 0.6mm (exceeding the 0.5mm threshold), the system immediately increases the weight of the force sensing and IMU data by 15%, correcting the pre-compensation parameter from 0.4mm to 0.45mm, and simultaneously adjusting the grasping force to 155N. Within 0.3 seconds, the robotic arm precisely fine-tunes the gripper position, reducing the laser ranging deviation to below 0.3mm, and stabilizing the force sensing data within the 150-160N range, achieving a smooth fit. The entire process is seamless, avoiding repeated adjustments due to vibration and dust, ensuring the battery enters a stable grasping state immediately upon contact.

[0087] Through this collaborative execution, the system increased the battery grabbing success rate from the traditional 82% to 99.8%, reduced the single battery swapping time to within 8.5 seconds, and decreased the number of repeated adjustments by 85%. In 24-hour continuous operation, the system successfully handled 120 battery swaps without a single failure, significantly improving the stability and efficiency of high-frequency battery swapping in mines. This provides key support for achieving the reliability goal of "unmanned operation and continuous operation" in unmanned mines, and completely solves the industry pain point of battery grabbing in complex mine environments.

[0088] In this embodiment, during the battery gripping process, the dynamic interaction between the gripper and the battery is continuously monitored in real time using a laser displacement sensor, a force sensor, and an IMU. The system collects 100 distance data points, 50 contact force data points, and 200 attitude angle data points per second to ensure comprehensive perception of the gripping process. Based on historical data and on-site conditions, the system intelligently classifies actual gripping deviations into three levels: Level 1 deviation is between 0.1 and 0.3 mm, Level 2 deviation is between 0.3 and 0.5 mm, and Level 3 deviation is greater than 0.5 mm. This grading mechanism enables the system to adopt precise response strategies for deviations of varying severity.

[0089] When the system detects a first-level deviation (such as an X-axis offset of 0.25mm), it only finely adjusts the gripper displacement compensation amount in the pre-compensation parameters, for example, adjusting the original pre-compensation value of 0.4mm to 0.42mm. There is no need to change the data fusion weights of any sensors, ensuring that the system maintains stable operation while completing minor corrections and avoiding unnecessary resource consumption.

[0090] When a secondary deviation is detected (e.g., X-axis offset of 0.4mm), the system synchronously adjusts the sensor fusion weights and pre-compensation parameters. Under vibration conditions, the weight of attitude data collected by the attitude monitoring sensor and deformation data of the elastic damping component are automatically increased by 15%, while the weight of the laser displacement sensor is decreased by 10%. Under dust conditions, the weight of obstacle data from the millimeter-wave radar and gripping force data from the force sensing sensor are increased by 15%, while the weight of the binocular vision sensor is decreased by 15%. Based on the adjusted weights, the system corrects the pre-compensation parameter from 0.4mm to 0.43mm and synchronously adjusts the gripping force to 152N to ensure rapid recovery of stable gripping under moderate deviations.

[0091] When a Level 3 deviation is detected (e.g., an X-axis offset of 0.6mm), the system immediately pauses the grasping action to avoid battery damage or grasping failure due to severe deviation. Subsequently, the system re-executes the multi-dimensional data acquisition process, re-acquiring raw information such as mine vehicle motion parameters and environmental perception data starting from S100. Dynamic weight fusion is then performed in S200 to generate new comprehensive information. After confirming the accurate battery location, the system restarts the grasping process, simultaneously marking the current Level 3 deviation data as an anomalous sample and automatically storing it in the historical database for model optimization. This allows the system to more accurately predict and respond to similar situations in the future.

[0092] Through this three-level deviation classification and processing mechanism, the system has achieved a precise response strategy from fine-tuning to complete reset, increasing the battery grabbing success rate from 99.2% to 99.8%, reducing the single grabbing failure rate to below 0.2%, and reducing the number of invalid adjustments by 75%. This significantly improves the stability and reliability of battery grabbing in complex mine environments, providing a solid guarantee for high-frequency and high-continuity battery swapping operations in unmanned mines.

[0093] In this embodiment of the application, the method further includes, during the process of evaluating the actual gripping deviation between the robotic arm gripper and the battery based on the collected dynamic data:

[0094] When assessing actual grasping deviations, the system first automatically retrieves deviation records from the historical database that are highly similar to the current operating conditions. The historical database stores detailed deviation data from the past 1500 battery swapping operations, including key parameters such as vibration intensity, dust concentration, and mine car speed, along with their corresponding actual deviation values. For example, under operating conditions with a vibration amplitude of 0.5–0.6 m / s² and a dust concentration of 80–90 μg / m³, historical data shows that the battery's average X-axis offset is 0.42 mm, providing a basis for subsequent correlation analysis.

[0095] The system acquires current operating parameters in real time, including mine car vibration amplitude, dust concentration, and mine car speed (5.3 km / h), and intelligently matches them with key features in historical data. Through machine learning algorithms, the system analyzes the similarity between historical data and current parameters, automatically selecting the 50 most relevant historical records to ensure the accuracy of deviation correlation analysis. For example, it found that the current operating conditions highly match 92% of similar historical records.

[0096] Based on the selected historical deviation data and current operating parameters, the system automatically generates a deviation correction coefficient. For example, historical data shows that under similar operating conditions, the actual deviation is on average 0.15 mm higher than the pre-compensation prediction value. Based on this, the system calculates a deviation correction coefficient of 0.15, making the prediction closer to the actual influence of environmental disturbances and avoiding the prediction deviation caused by ignoring historical experience in traditional methods.

[0097] The system combines the deviation correction factor with the current actual grasping deviation value to calculate the corrected deviation value. For example, when the laser rangefinder shows that the current actual deviation of the X-axis is 0.4mm, combined with a correction factor of 0.15, the system calculates a corrected deviation value of 0.55mm, ensuring that the deviation correction takes into account both historical patterns and the current specific working conditions.

[0098] Based on the correction deviation value, the system dynamically adjusts the gripper displacement compensation amount in the pre-compensation parameters. The adjustment range is proportional to the deviation correction coefficient; for example, a correction coefficient of 0.15 corresponds to a compensation adjustment of 0.15 mm. At the same time, the system strictly controls the adjustment rate to not exceed the preset dynamic adjustment upper limit (0.2 mm per second) to avoid robot arm vibration or battery damage due to excessively rapid adjustment, ensuring a smooth and reliable adjustment process.

[0099] By intelligently utilizing historical deviation data and employing a dynamic correction mechanism, the system significantly improves the accuracy of pre-compensation parameters, reducing the number of invalid adjustments during battery grabbing by 65% ​​and further increasing the grabbing success rate from 99.2% to 99.7%. At the same time, it ensures that the robotic arm maintains high precision and high stability in complex working conditions, providing more reliable technical support for unmanned mining operations.

[0100] In this embodiment of the application, the method further includes: setting a distributed temperature sensor array on the contact surface of the robotic arm gripper to monitor the battery surface temperature distribution in real time; when the local temperature of the battery is detected to exceed a preset threshold based on the battery surface temperature distribution, automatically activating the local cooling channel on the gripper contact surface to perform directional cooling of the high-temperature area; and dynamically adjusting the weighting coefficient of the temperature parameter in the offset prediction model based on the correlation between the battery temperature distribution and the offset prediction model.

[0101] Specifically, the system integrates a high-precision distributed temperature sensor array on the contact surface of the robotic arm gripper, with 5 miniature temperature sensors arranged per square centimeter to form a 16×16 grid layout. This enables real-time monitoring of the temperature distribution on the battery surface with an accuracy of ±0.5 degrees Celsius, covering the entire area from the battery interface to the edge, providing a comprehensive sensing basis for temperature anomaly detection.

[0102] During battery grabbing, the system continuously analyzes real-time data from the temperature sensor array, updating the temperature distribution map 10 times per second to accurately identify hot spots on the battery surface. For example, when the temperature of the battery interface area reaches 55 degrees Celsius (exceeding the preset threshold of 50 degrees Celsius), the system immediately determines that there is a risk of local overheating, triggers the temperature anomaly handling process, and avoids battery performance degradation or safety hazards caused by high temperatures.

[0103] When the local temperature of the battery exceeds a preset threshold, the system automatically activates the local cooling channels on the gripper contact surface. The cooling channels consist of miniature liquid cooling pipes distributed at corresponding positions on the gripper contact surface. They can directionally cool the high-temperature area at a flow rate of 200 mL per second, reducing the local temperature to a safe range (below 45 degrees Celsius) within 3 seconds. This ensures the thermal stability of the gripper when in contact with the battery and prevents battery interface deformation or material softening caused by high temperatures.

[0104] Based on the correlation analysis between battery temperature distribution and historical operating condition data, the system dynamically adjusts the weighting coefficient of temperature parameters in the offset prediction model. For example, when the temperature distribution shows abnormally high temperatures in the battery interface area, the system increases the weight of the temperature parameter from the baseline value of 30% to 45%, making the offset prediction model more sensitive to the impact of temperature on battery thermal expansion and ensuring that the prediction results are highly consistent with actual operating conditions. For example, in high-temperature environments, the battery interface may experience a thermal expansion offset of 0.2 to 0.3 mm.

[0105] Through this temperature monitoring and adaptive adjustment mechanism, the system effectively solves the problem of decreased grasping accuracy caused by abnormal battery temperature, reducing the grasping failure rate due to temperature factors from 3.5% to 0.8%, while ensuring the thermal safety performance of the battery. This mechanism enables the system to maintain high-precision grasping even in high-temperature mining environments, increasing the battery grasping success rate from 99.2% to 99.6%, providing more comprehensive protection for high-frequency battery swapping operations in unmanned mines.

[0106] In this embodiment, during the system initialization phase, deep learning algorithms (such as causal inference networks) are used to perform in-depth analysis of historical data from 12,000 battery swapping operations, identifying the causal relationships between historical deviation data and operating parameters such as mine car vibration amplitude, dust concentration, and battery temperature. For example, the system found that when the vibration amplitude exceeds 0.5 m / s² and the dust concentration is below 80 μg / m³, the deviation is mostly due to mechanical structural disturbances caused by mine car vibration; when the dust concentration exceeds 90 μg / m³ and the vibration amplitude is low, the deviation is mainly caused by visual interference. These causal patterns are encoded into a structured knowledge graph, forming the core decision-making basis of the system.

[0107] The system intelligently categorizes the causes of deviations into four types: vibration disturbance type (mechanical structural offset caused by mine car vibration), dust interference type (visual recognition distortion caused by dust), temperature influence type (thermal expansion offset caused by battery temperature changes), and multi-factor coupled type (complex deviations caused by the combined effects of multiple environmental factors). For example, when a vibration amplitude of 0.6 m / s², a dust concentration of 85 μg / m³, and a battery temperature of 48°C are detected simultaneously, the system classifies it as a multi-factor coupled deviation, requiring the integration of multiple compensation strategies.

[0108] In actual operation, when the system detects that the actual grasping deviation exceeds a preset threshold, it immediately initiates the operating condition-deviation cause matching process. The system intelligently compares the current operating condition parameters (such as vibration amplitude of 0.58 m / s², dust concentration of 88 μg / m³, and battery temperature of 47°C) with the knowledge graph to automatically identify the type of deviation cause. For example, the system identifies the current operating condition as a dust interference type deviation because the dust concentration of 88 μg / m³ significantly exceeds the threshold, while the vibration amplitude and temperature are within the normal range.

[0109] Based on the identified cause of the deviation, the system accurately retrieves the optimal compensation strategy from the pre-stored compensation strategy library. For example, for dust interference-related deviations, the system retrieves the following pre-stored strategies: increase the weight of the force sensing sensor by 15%, reduce the weight of the binocular vision by 15%, adjust the pre-compensation parameter correction coefficient by 0.12, and optimize the robotic arm's motion trajectory parameters (such as increasing the pre-compensation buffer by 0.2mm), ensuring that the compensation strategy perfectly matches the cause of the deviation.

[0110] This mechanism, through causal-driven intelligent decision-making, upgrades the system's response to deviations from passive adjustment to precise prediction and proactive compensation. After identifying the type of deviation, the system can immediately apply the most suitable compensation strategy, increasing adjustment accuracy from 85% to 98.5%, reducing ineffective adjustments by 70%, and simultaneously improving battery retrieval success rate from 99.2% to 99.8%. This provides intelligent decision support for high-frequency, high-reliability battery swapping operations in complex mine environments.

[0111] In this embodiment of the application, the method further includes:

[0112] Before battery grasping, the system continuously acquires high-precision 3D image displacement sensing data of the battery interface using a binocular vision sensor and a structured light scanner at the end of the robotic arm. It generates 10 interface point cloud maps per second, accurately capturing the positions of 50 key feature points on the edge of the battery interface. The system intelligently compares this real-time image data with the preset target position before the robotic arm grasps the battery, automatically constructing a battery interface alignment evaluation model. This model analyzes the displacement vectors of the interface edge feature points (e.g., feature point A shifts 0.3mm towards the X-axis and 0.1mm towards the Y-axis) to calculate a comprehensive index reflecting the degree of interface alignment, ensuring the objectivity and accuracy of the alignment evaluation.

[0113] The system calculates the battery interface alignment index in real time. This index is derived from the displacement vectors of all feature points and ranges from 0 to 100 (a higher value indicates better alignment). For example, when the system detects that the average displacement of the feature points at the interface edge exceeds 0.3mm, the alignment index drops to 85 (below the preset threshold of 90), and the system immediately determines that the interface alignment is insufficient, triggering the on-site adjustment mechanism.

[0114] When the interface alignment index falls below a preset threshold, the system automatically adjusts the robotic arm's gripping posture. Specifically, the system increases the rotation angle compensation (e.g., from the original preset 1.5° to 2.1°) and simultaneously updates the gripper rotation angle compensation value in the pre-compensation parameters. This ensures the robotic arm completes precise rotational adjustments before contacting the battery, preventing gripping failures due to interface misalignment. For example, the system calculates an additional 0.6° rotation and completes the posture fine-tuning within 0.2 seconds to ensure precise matching between the interface and the gripper.

[0115] Based on the correlation analysis between the interface alignment index and the actual grasping deviation, the system dynamically adjusts the weight coefficient of the interface alignment parameter in the offset prediction model. For example, when the alignment index is consistently below 90, the system increases the weight of the interface alignment parameter from the baseline value of 25% to 35%, making the model more sensitive to the impact of interface alignment problems on the overall offset, thereby improving the accuracy of subsequent predictions. For example, under similar working conditions, the model can predict interface alignment offsets of 0.2 to 0.3 mm in advance.

[0116] After the robotic arm completes the grasping process, the system associates the interface alignment index (e.g., 87) with the actual grasping deviation data (e.g., X-axis offset of 0.5mm) and stores it in the historical database, forming an "alignment-deviation" mapping relationship. For example, the system records the associated sample where "alignment index 87 corresponds to X-axis deviation of 0.5mm" for subsequent optimization of the interface alignment parameter weights in the offset prediction model, enabling the system to predict and compensate for interface alignment problems in advance under similar working conditions.

[0117] Through a battery interface alignment evaluation mechanism, the system reduced the failure rate of grasping caused by interface misalignment from 4.2% to 0.5%. Simultaneously, it enabled the robotic arm to achieve more precise interface matching under complex working conditions, increasing the battery grasping success rate from 99.2% to 99.8%, and shortening the adjustment time for a single grasping operation to within 0.4 seconds. This mechanism not only solved the interface alignment problem but also continuously optimized the prediction model through historical data accumulation, providing more comprehensive accuracy assurance for unmanned battery swapping operations in mines.

[0118] During routine battery swapping operations in a deep mine, a mine car travels at 5.5 km / h under complex conditions of 92 μg / m³ dust concentration and 0.58 m / s² vibration. Traditional systems suffer from blurred visual images due to dust and distorted laser ranging due to vibration, resulting in a 0.55 mm offset on the battery's X-axis and a gripper failure rate as high as 12%. After the system starts up S100, it detects that the temperature in the battery interface area has reached 54℃ (exceeding the 50℃ threshold) through the distributed temperature sensor array, and immediately activates the local cooling channel to reduce the temperature to 44℃; at the same time, S200 dynamically lifts the force sensing sensor weight by 15%, and S300 generates a deviation correction coefficient of 0.15 based on 1500 historical data, adjusting the pre-compensation parameter from 0.4mm to 0.55mm; S400 identifies the mining truck model as "Mining Truck V-2026" (unknown model), scans the battery outline within 1.3 seconds and generates a temporary grasping plan; S500 evaluates the interface alignment index in real time (82, below the 90 threshold), automatically increases the rotation compensation by 0.6°, and calls the knowledge graph to confirm that it is a dust interference type deviation, and executes the optimal compensation strategy.

[0119] The system completes all dynamic adjustments within 0.4 seconds: temperature anomaly handling ensures battery thermal stability, historical deviation data correction improves pre-compensation accuracy, interface alignment adjustment resolves misalignment issues, and a dust interference compensation strategy optimizes sensor weights. Before contacting the battery, the robotic arm completes 0.55mm displacement compensation on the X-axis and 0.25mm rotation compensation on the Y-axis, stabilizing the laser ranging deviation at 0.28mm and controlling force sensing data fluctuations within ±5N, achieving smooth gripping. The entire process avoids the 8-10 ineffective adjustments required in traditional systems; the gripper reaches a stable state immediately upon contact, and the battery interface alignment index is improved to 93.

[0120] Through this collaborative execution, the system increased the battery grabbing success rate from the traditional 85% to 99.8%, reduced the single battery swapping time to within 8.2 seconds, decreased the number of repeated adjustments by 82%, and reduced the grabbing failure rate caused by dust and temperature factors to 0.8% and 0.5%, respectively. In 24-hour continuous operation, the system successfully handled 150 battery swaps without a single failure, completely solving the accuracy and stability bottleneck in high-frequency battery swapping in mines. This provides a reliable guarantee for achieving "zero-error, high-efficiency" automated operation in unmanned mines, significantly reducing mine downtime and manual intervention costs.

[0121] This application discloses a battery grasping and compensation system based on multi-sensor data fusion, referring to... Figure 2 ,include:

[0122] The multi-dimensional data acquisition module 001 is used to collect multi-dimensional raw data, including the mine truck's driving speed, steering angle, road slope data, obstacle information in the working environment, three-dimensional image displacement sensing of the battery interface, force take-off data and deformation data of the elastic damping component, distance information between the gripper and the battery, and real-time posture data of the robotic arm.

[0123] The adaptive weighted fusion module 002 monitors and judges the current working condition in real time based on multi-dimensional raw data. If the vibration amplitude of the mining truck exceeds the preset vibration threshold, the weight of the attitude data collected by the attitude monitoring sensor and the deformation data collected by the elastic damping component is increased, while the weight of the spacing data collected by the laser displacement sensor is decreased. If the dust concentration in the working environment exceeds the preset dust threshold, the weight of the obstacle data collected by the millimeter-wave radar and the grasping force data collected by the force sensing sensor is increased, while the weight of the three-dimensional image data collected by the binocular vision sensor is decreased. After adjusting the weights, the multi-dimensional raw data is fused to generate comprehensive information that includes working condition information and battery target information.

[0124] The pre-compensation parameter output module 003 constructs a displacement prediction model corresponding to the battery position and angle based on the mine car motion parameters and environmental physical field parameters. The mine car motion parameters include the mine car speed, steering angle, and road slope data, while the environmental physical field parameters include the battery temperature and dust concentration. The displacement prediction model calculates and predicts the battery's position displacement and angle displacement under the current working conditions to obtain the corresponding displacement prediction results. Based on the displacement prediction results, the corresponding pre-compensation parameters are output.

[0125] The vehicle model adaptive grasping mode generation module 004 determines the vehicle model of the mining truck to which the battery to be grasped belongs during the vehicle battery swapping stage. If the model is known, it directly calls the pre-stored preset grasping mode that matches the known model. If the model is unknown, it activates the probe claw set at the end of the robotic arm to scan the external contour and surface hardness of the battery, and generates a temporary grasping scheme based on the scanning results. The grasping point position of the robotic arm gripper is adjusted based on the preset grasping mode or the temporary grasping scheme and the pre-compensation parameters.

[0126] The real-time deviation dynamic compensation module 005 continuously collects dynamic data from each sensor during the battery gripping process. Based on the collected dynamic data, it evaluates the actual gripping deviation between the robotic arm gripper and the battery. If the actual gripping deviation exceeds the preset deviation range, it adjusts the data fusion weights of each sensor and corrects the corresponding pre-compensation parameters based on the actual gripping deviation, and synchronously adjusts the gripping point of the robotic arm gripper until the robotic arm stably grips the battery.

[0127] This application also discloses a battery grabbing compensation method based on multi-sensor data fusion, including a processor, in which a program for the battery grabbing compensation method based on multi-sensor data fusion described above is running.

[0128] This application also discloses a storage medium storing a program of the battery capture compensation method based on multi-sensor data fusion, as described in any of the above embodiments.

[0129] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A battery grasping compensation method based on multi-sensor data fusion, characterized in that, include: Multi-dimensional raw data is collected, including mining truck speed, steering angle, road slope data, obstacle information in the working environment, three-dimensional image displacement sensing data of battery interface, force take-off data and deformation data of elastic damping component, distance information between gripper and battery, real-time posture data of robotic arm, and real-time monitoring of battery surface temperature distribution by setting a distributed temperature sensor array on the contact surface of robotic arm gripper. Based on real-time monitoring and judgment of multi-dimensional raw data, if the vibration amplitude of the mining truck exceeds the preset vibration threshold, the weight of the attitude data collected by the attitude monitoring sensor and the deformation data collected by the elastic damping component is increased, while the weight of the spacing data collected by the laser displacement sensor is decreased; if the dust concentration in the working environment exceeds the preset dust threshold, the weight of the obstacle data collected by the millimeter-wave radar and the gripping force data collected by the force sensing sensor is increased, while the weight of the three-dimensional image data collected by the binocular vision sensor is decreased; after adjusting the weights, the multi-dimensional raw data is fused to generate comprehensive information including working condition information and battery target information; A displacement prediction model corresponding to the battery position and angle is constructed based on the mining truck's motion parameters and environmental physical field parameters. The mining truck's motion parameters include the mining truck's speed, steering angle, and road slope data, while the environmental physical field parameters include battery temperature and dust concentration. Based on the correlation between battery temperature distribution and the displacement prediction model, the weight coefficient of the temperature parameter in the displacement prediction model is dynamically adjusted. The displacement prediction model is used to calculate and predict the battery's position and angle displacement under the current working conditions to obtain the corresponding displacement prediction results. Based on the displacement prediction results, the corresponding pre-compensation parameters are output. When the local temperature of the battery is detected to exceed a preset threshold based on the battery surface temperature distribution, the local cooling channel of the gripper contact surface is automatically activated to provide directional cooling to the high-temperature area. During the vehicle battery swapping phase, the model of the mining truck to which the battery to be grabbed belongs is determined. If the model is known, a pre-stored preset grabbing mode matching the known model is directly invoked. If the model is unknown, the probe claw at the end of the robotic arm is activated to scan the external contour and surface hardness of the battery, and a temporary grabbing scheme is generated based on the scan results. A battery interface alignment evaluation model is constructed based on the real-time comparison between the displacement sensing data corresponding to the 3D image of the battery interface and the actual battery grabbing position. The battery interface alignment evaluation model calculates the corresponding interface alignment index by analyzing the displacement vector of the interface edge feature points. When the interface alignment index is lower than the preset alignment threshold, the grabbing posture of the robotic arm is automatically adjusted, the rotation angle compensation is increased, and the gripper rotation angle compensation in the pre-compensation parameters is adjusted simultaneously. Based on the correlation between the interface alignment index and the actual grabbing deviation, the weight coefficient of the interface alignment parameter in the offset prediction model is dynamically adjusted to make the prediction results more consistent with the actual working conditions. The grabbing point position of the robotic arm gripper is adjusted based on the preset grabbing mode or the temporary grabbing scheme combined with the pre-compensation parameters. During the battery gripping process, dynamic data from each sensor is continuously collected in real time. Based on the collected dynamic data, the actual gripping deviation between the robotic arm gripper and the battery is evaluated. The deviation is divided into three levels and corresponding to different adjustment strategies: if the actual gripping deviation is a level one deviation, the gripper displacement compensation amount in the pre-compensation parameters is finely adjusted without adjusting the sensor fusion weights; if the actual gripping deviation is a level two deviation, the sensor fusion weights and pre-compensation parameters are adjusted simultaneously, the weight of attitude data is increased under vibration conditions, and the weight of force sensing data is increased under dust conditions. If the actual grasping deviation is a level three deviation, the grasping action is immediately paused, and the multi-dimensional data acquisition and dynamic weight fusion steps are re-executed to locate the accurate position of the battery before the grasping is restarted. At the same time, the deviation data is marked as an abnormal sample and stored in the historical database. If the actual grasping deviation exceeds the preset deviation range, the data fusion weights of each sensor are adjusted based on the actual grasping deviation, and the corresponding pre-compensation parameters are corrected. The grasping point of the robotic arm gripper is adjusted synchronously until the robotic arm stably grasps the battery. After the robotic arm completes the grasping, the interface alignment index is associated with and stored with the deviation data.

2. The battery grasping compensation method based on multi-sensor data fusion according to claim 1, characterized in that, In evaluating the actual gripping deviation between the robotic arm gripper and the battery based on the collected dynamic data, the method also includes: Retrieve the corresponding historical deviation data, obtain the corresponding current operating parameters, and automatically generate deviation correction coefficients based on the deviation correlation between historical deviation data and current operating parameters. The deviation correction coefficient is combined with the current actual grasping deviation value to calculate the corrected deviation value; the gripper displacement compensation amount in the pre-compensation parameters is dynamically adjusted based on the corrected deviation value, wherein the adjustment range is proportional to the deviation correction coefficient and the adjustment rate does not exceed the preset dynamic adjustment upper limit.

3. The battery grasping compensation method based on multi-sensor data fusion according to claim 1, characterized in that, The method also includes: By analyzing the causal relationship between historical deviation data and current operating parameters using machine learning algorithms, a knowledge graph of operating conditions and deviation causes is established based on the causal relationship, and the causes of deviations are classified into vibration disturbance type, dust interference type, temperature influence type and multi-factor coupling type. When determining the actual level of deviation, the system matches the current operating parameters with the operating condition-deviation cause knowledge graph to automatically identify the type of deviation cause. The optimal compensation strategy is retrieved from the pre-stored compensation strategy library based on the type of deviation cause. The optimal compensation strategy includes the sensor weight adjustment ratio, the pre-compensation parameter correction coefficient, and the robot arm motion trajectory optimization parameters.

4. A battery gripping compensation system based on multi-sensor data fusion, which operates the battery gripping compensation method based on multi-sensor data fusion as described in any one of claims 1-3, characterized in that, include: The multi-dimensional data acquisition module is used to collect multi-dimensional raw data, including the mine truck's speed, steering angle, road slope data, obstacle information in the working environment, three-dimensional image displacement sensing of the battery interface, force take-off data and deformation data of the elastic damping component, distance information between the gripper and the battery, and real-time posture data of the robotic arm. The adaptive weighted fusion module monitors and judges the current working condition in real time based on multi-dimensional raw data. If the vibration amplitude of the mining truck exceeds the preset vibration threshold, the weight of the attitude data collected by the attitude monitoring sensor and the deformation data collected by the elastic damping component is increased, while the weight of the spacing data collected by the laser displacement sensor is decreased. If the dust concentration in the working environment exceeds the preset dust threshold, the weight of the obstacle data collected by the millimeter-wave radar and the grasping force data collected by the force sensing sensor is increased, while the weight of the three-dimensional image data collected by the binocular vision sensor is decreased. After adjusting the weights, the multi-dimensional raw data is fused to generate comprehensive information that includes working condition information and battery target information. The pre-compensation parameter output module constructs a displacement prediction model corresponding to the battery position and angle based on the mine car motion parameters and environmental physical field parameters. The mine car motion parameters include the mine car speed, steering angle, and road slope data, while the environmental physical field parameters include the battery temperature and dust concentration. The displacement prediction model calculates and predicts the battery's position and angle displacement under the current working conditions to obtain the corresponding displacement prediction results. Based on the displacement prediction results, the corresponding pre-compensation parameters are output. The vehicle model adaptive grasping mode generation module determines the vehicle model of the mining truck to which the battery to be grasped belongs during the vehicle battery swapping stage. If the model is known, it directly calls the pre-stored preset grasping mode that matches the known model. If the model is unknown, it activates the probe claw set at the end of the robotic arm to scan the external contour and surface hardness of the battery, and generates a temporary grasping plan based on the scan results. The grasping point position of the robotic arm gripper is adjusted based on the preset grasping mode or the temporary grasping plan and pre-compensation parameters. The real-time deviation dynamic compensation module continuously collects dynamic data from each sensor during the battery gripping process. Based on the collected dynamic data, it evaluates the actual gripping deviation between the robotic arm gripper and the battery. If the actual gripping deviation exceeds the preset deviation range, it adjusts the data fusion weights of each sensor and corrects the corresponding pre-compensation parameters based on the actual gripping deviation, and synchronously adjusts the gripping point of the robotic arm gripper until the robotic arm stably grips the battery.

5. A battery grasping and compensation system based on multi-sensor data fusion, characterized in that, Includes a processor, in which a program for a battery grabbing compensation method based on multi-sensor data fusion as described in any one of claims 1-3 is running.

6. A storage medium, characterized in that, The program stores a battery grabbing compensation method based on multi-sensor data fusion as described in any one of claims 1-3.

Images