A multi-modal detection method and system for electrolyte leakage of a power battery pack

Abstract

The present application relates to a kind of power battery pack electrolyte leakage multi-modal detection method and system, multi-modal detection method includes, the image in the process of electric batch is close to the target screw of power battery pack, and the gray histogram of image is determined, based on gray histogram and normal surface histogram determine first leakage type and visual anomaly confidence;Real-time acquisition target screw is unwound in the process of electric batch torque data and motor current data, based on torque data and normal reference torque determine second leakage type and force sense anomaly confidence;Based on estimated torque, torque data and motor current data determine third leakage type and current anomaly confidence;Based on dynamic weight, visual anomaly confidence, force sense anomaly confidence, current anomaly confidence fusion confidence is calculated, based on fusion confidence to power battery pack electrolyte leakage is early warned, improve the accuracy of power battery pack electrolyte leakage detection.

Description

Technical Field

[0001] This invention relates to the field of battery recycling technology, and in particular to a multimodal detection method and system for electrolyte leakage in power battery packs. Background Technology

[0002] In the current field of automated dismantling of retired power battery packs, mainstream technical solutions mainly rely on single visual inspection or single force detection to monitor the dismantling process. Specifically, visual inspection solutions use industrial cameras to acquire images of the battery pack surface and use deep learning models to identify surface anomalies; force detection solutions use six-dimensional force sensors to monitor the interaction force between the dismantling tool and the workpiece to determine whether abnormal contact has occurred.

[0003] However, existing technical solutions have the following significant drawbacks: First, single visual inspection is unreliable in the scenario of disassembling retired battery packs. The surface of retired battery packs generally has problems such as electrolyte residue, dust coverage, metal oxide layer and reflective interference. These factors lead to a false alarm rate of over 18% and a false alarm rate of over 9% for visual inspection. Especially in the stage of trace electrolyte leakage, the recognition rate of visual inspection is only 35%, which cannot detect potential safety hazards in time. Second, single force detection cannot distinguish the type of abnormality. Force sensors can detect sudden torque changes, but cannot determine whether the change is caused by screw corrosion, electrolyte crystallization or mechanical jamming, which leads to frequent false shutdowns of the system and seriously affects the efficiency of the production line. Actual test data shows that the false alarm rate of single force detection reaches 22.5%. Therefore, the single detection mode has the problems of low reliability and high false alarm rate. Summary of the Invention

[0004] In view of this, it is necessary to provide a multimodal detection method and system for electrolyte leakage in power battery packs to solve the technical problems of low reliability and high false alarm rate of single detection mode.

[0005] To address the aforementioned problems, in a first aspect, the present invention provides a multimodal detection method for electrolyte leakage in a power battery pack, comprising: The image of the electric screwdriver approaching the target screw of the power battery pack is acquired, and the grayscale histogram of the image is determined. Based on the grayscale histogram and the normal surface histogram, the first leakage type and the confidence level of visual anomaly are determined. The torque data of the electric screwdriver and the motor current data are collected in real time during the process of unscrewing the target screw. Based on the torque data and the normal reference torque, the second leakage type and the confidence level of force abnormality are determined. Based on the calibration relationship between motor current and output torque, the estimated torque corresponding to the motor current data is determined, and based on the estimated torque, torque data, and motor current data, the third leakage type and the confidence level of current anomaly are determined. The final leakage type of the power battery pack electrolyte is determined based on the first leakage type, the second leakage type, and the third leakage type. A dynamic weight is determined based on the final leakage type. A fusion confidence level is calculated based on the dynamic weight, the confidence level of visual anomalies, the confidence level of force anomalies, and the confidence level of current anomalies. An early warning is issued for power battery pack electrolyte leakage based on the fusion confidence level and a preset grading threshold.

[0006] In one possible implementation, the first leak type includes liquid leak, crystallization leak, and mixed leak. Determining the first leak type and the confidence level of visual anomaly based on the grayscale histogram and the normal surface histogram includes: The height of abnormal peaks and the proportion of abnormal area are determined based on the grayscale histogram and the normal surface histogram. When the height of the abnormal peak is greater than the first peak value threshold, the leakage of electrolyte in the power battery pack is determined to be a liquid leakage. When the abnormal area ratio is greater than a preset area threshold and the abnormal peak height is greater than a second peak threshold and less than a third peak threshold, the electrolyte leakage of the power battery pack is determined to be a crystal leakage, wherein the first peak threshold and the second peak threshold are less than the third peak threshold, and the first peak threshold is greater than the second peak threshold. When the height of the abnormal peak is greater than the first peak value threshold and the proportion of the abnormal area is greater than the preset area threshold, the electrolyte leakage of the power battery pack is determined to be a mixed leakage. The peak anomaly factor, scaling factor, and area anomaly factor are determined based on the grayscale histogram and the normal surface histogram, and the visual anomaly confidence level is determined based on the peak anomaly factor, scaling factor, and area anomaly factor.

[0007] In one possible implementation, the confidence level for the visual anomaly is: , , , , in, For the confidence level of visual abnormalities, As a peak anomaly factor, This represents the maximum height of the main peak in the histogram of a normal surface. This represents the maximum height of the outlier peak in the grayscale histogram. As an area anomaly factor, The area of ​​the abnormal pixel region. The total pixel area of ​​the image. As a uniformity factor, The standard deviation of gray levels in the abnormal region. , , These are the weighting coefficients. This is the scaling factor for abnormal peak values. This is the area scaling factor.

[0008] In one possible implementation, the second leakage type includes liquid leakage, crystallization leakage, and mixed leakage. Determining the second leakage type and the confidence level of force anomaly based on the torque data and a normal reference torque includes: Based on the torque data and the normal reference torque, determine the peak torque, torque fluctuation, and torque change rate; When the peak torque is greater than the torque threshold and the torque fluctuation is less than the first fluctuation threshold, the electrolyte leakage of the power battery pack is determined to be a crystal leakage. When the torque fluctuation is greater than the second fluctuation threshold and the torque change rate is greater than the change rate threshold, the electrolyte leakage of the power battery pack is determined to be a liquid leakage, wherein the second fluctuation threshold is greater than the first fluctuation threshold; When the peak torque is greater than the torque threshold, the torque fluctuation is greater than the second fluctuation threshold, and the torque change rate is greater than the change rate threshold, the electrolyte leakage of the power battery pack is determined to be a mixed leakage. Based on the torque data and the normal reference torque, determine the torque amplitude anomaly factor and the torque change rate anomaly factor, and based on the torque amplitude anomaly factor and the torque change rate anomaly factor, determine the confidence level of force perception anomaly.

[0009] In one possible implementation, the confidence level of the force perception anomaly is: , , , in, For the confidence level of force perception abnormality, , These are the weighting coefficients. This is an abnormal factor for torque amplitude. This is an abnormal factor in the rate of change of torque. This is the current measured torque value. This is the reference value for the rotation torque under normal operating conditions. This is the baseline value for the rate of change of torque under normal operating conditions. The threshold for abnormal torque change rate. This represents the current rate of change of torque.

[0010] In one possible implementation, the third leakage type includes liquid leakage and crystallization leakage, and the determination of the third leakage type and current anomaly confidence level based on the estimated torque, torque data, and motor current data includes: When the difference between the estimated torque and the torque data is greater than a preset threshold and the motor current data is greater than the current threshold, the electrolyte leakage of the power battery pack is determined to be either liquid leakage or crystallization leakage. The confidence level of the current anomaly is determined based on the estimated torque and the torque data.

[0011] In one possible implementation, the confidence level for the current anomaly is: , in, For current anomaly confidence level, To estimate torque, This is the measured torque value.

[0012] In one possible implementation, the fusion confidence level is: , in, To integrate confidence levels, , , For dynamic weights.

[0013] In one possible implementation, the early warning of electrolyte leakage in the power battery pack based on the fused confidence level and a preset grading threshold includes: When the fusion confidence level is greater than or equal to the first level threshold and less than the second level threshold, a yellow warning is triggered. When the fusion confidence level is greater than or equal to the second-level threshold and less than the third-level threshold, an orange alert is triggered. A red alert is triggered when the fusion confidence level is greater than or equal to the third-level threshold.

[0014] Secondly, the present invention also provides a multimodal detection system for electrolyte leakage in a power battery pack, comprising: The visual inspection module acquires an image of the electric screwdriver approaching the target screw of the power battery pack and determines the grayscale histogram of the image. Based on the grayscale histogram and the normal surface histogram, it determines the first leakage type and the confidence level of visual anomaly. The force detection module is used to collect torque data of the electric screwdriver and motor current data in real time during the process of unscrewing the target screw. Based on the torque data and the normal reference torque, the second leakage type and the confidence level of force abnormality are determined. The current detection module is used to determine the estimated torque corresponding to the motor current data based on the calibration relationship between motor current and output torque, and to determine the third leakage type and the current anomaly confidence level based on the estimated torque, torque data and motor current data. The early warning module is used to determine the final leakage type of the electrolyte in the power battery pack based on the first leakage type, the second leakage type, and the third leakage type; determine a dynamic weight based on the final leakage type; calculate a fusion confidence level based on the dynamic weight, the confidence level of visual anomalies, the confidence level of force anomalies, and the confidence level of current anomalies; and issue an early warning for electrolyte leakage in the power battery pack based on the fusion confidence level and a preset grading threshold.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: The multimodal detection method for electrolyte leakage in power battery packs provided by this invention integrates visual detection, force detection, and current detection by weighted fusion of visual anomaly confidence, force anomaly confidence, and current anomaly confidence, thereby improving the accuracy of electrolyte leakage detection in power battery packs and reducing the false alarm rate. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying 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.

[0017] Figure 1 A flowchart of an embodiment of the multimodal detection method for electrolyte leakage in a power battery pack provided by the present invention; Figure 2 This is a schematic diagram comparing the grayscale histogram of electrolyte leakage in a power battery pack with the histogram of a normal surface, as provided by the present invention. Figure 3 A weighted allocation table for querying the current operating condition of electrolyte leakage in a power battery pack, provided by the present invention. Figure 4 A schematic diagram comparing the performance of different detection strategies for electrolyte leakage in power battery packs provided by the present invention; Figure 5 This is a schematic diagram of an embodiment of the multimodal detection system for electrolyte leakage in a power battery pack provided by the present invention. Detailed Implementation

[0018] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0019] In this document, the term "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0020] A specific embodiment of the present invention discloses a multimodal detection method for electrolyte leakage in a power battery pack, such as... Figure 1 As shown, a multimodal detection method for electrolyte leakage in a power battery pack includes: S101. Acquire an image of the electric screwdriver approaching the target screw of the power battery pack, and determine the grayscale histogram of the image. Based on the grayscale histogram and the normal surface histogram, determine the first leakage type and the confidence level of visual anomaly.

[0021] The electric screwdriver refers to a customized modular electric screwdriver, a Maxon DXL 32L motor, and a planetary gearbox, with a rated torque of 2.0 Nm and a maximum speed of 525 rpm.

[0022] S102. Real-time acquisition of electric screwdriver torque data and motor current data during the unscrewing of the target screw, and determination of the second leakage type and force abnormality confidence level based on the torque data and normal reference torque.

[0023] S103. Determine the estimated torque corresponding to the motor current data based on the calibration relationship between motor current and output torque. Determine the third leakage type and the current anomaly confidence level based on the estimated torque, torque data and motor current data.

[0024] S104. Determine the final leakage type of the electrolyte in the power battery pack based on the first leakage type, the second leakage type, and the third leakage type. Determine the dynamic weight based on the final leakage type. Calculate the fusion confidence level based on the dynamic weight, visual anomaly confidence level, force anomaly confidence level, and current anomaly confidence level. Provide an early warning for electrolyte leakage in the power battery pack based on the fusion confidence level and the preset classification threshold.

[0025] In some embodiments, a vision sensor, a force sensor, and an electric screwdriver drive unit are integrated into the same robot end effector. The robot collects image, torque, and current data during the disassembly process of the retired power battery pack. Based on the image, torque, and current data, the type of electrolyte leakage in the retired power battery pack is determined, and the confidence levels of visual anomalies, force anomalies, and current anomalies are calculated. A fusion confidence level is calculated based on the confidence levels of visual anomalies, force anomalies, and current anomalies, and an early warning of electrolyte leakage in the retired power battery pack is given based on the fusion confidence level.

[0026] In some embodiments, in step S101, an image is acquired during the process of the electric screwdriver approaching the target screw of the power battery pack, and the grayscale histogram of the image is determined. An Intel RealSense D405 depth camera is used, integrated into the side of the electric screwdriver tool and rigidly connected to the tool tip. The camera provides... The system uses RGB images of pixels and depth information based on stereo vision. The field of view covers a working area with a diameter of 80 mm centered on the tool tip. The relative position of the camera and the tool tip is ensured by precision machining, with a calibration error of less than 0.05 mm, eliminating the systematic errors introduced by traditional hand-eye calibration. As the electric screwdriver approaches the target screw, the depth camera continuously acquires images at a rate of 30 frames per second. After Gaussian filtering and noise reduction of each frame, the grayscale histogram of the image is calculated. Based on the grayscale histogram and the normal surface histogram, the first leakage type and the confidence level of visual anomaly are determined. The grayscale histogram is compared with the normal surface histogram to detect whether there is a bimodal distribution or abnormal skewness. If there is an obvious high grayscale peak or low grayscale peak in the grayscale histogram, and the area of ​​the peak region exceeds a preset threshold, it is marked as a candidate region for visual anomaly. Finally, the detection results of 5 consecutive frames are verified for temporal consistency. Only when anomalies are detected in more than 3 frames is a visual anomaly signal output to avoid instantaneous noise interference. The first leakage type includes liquid leakage, crystallization leakage, and mixed leakage. The height of abnormal peaks and the percentage of abnormal area are determined based on grayscale histograms and normal surface histograms. When the height of an abnormal peak is greater than a first peak value threshold, the battery pack electrolyte leakage is determined to be a liquid leakage; for example, if the abnormal peak height is greater than 200, the battery pack electrolyte leakage is classified as a liquid leakage. When the percentage of abnormal area is greater than a preset area threshold and the height of the abnormal peak is greater than a second peak value threshold but less than a third peak value threshold, the battery pack electrolyte leakage is determined to be a crystallization leakage. The first and second peak value thresholds must be less than the third peak value threshold, and the first peak value threshold must be greater than the second peak value threshold. For example, if the height of the abnormal peak is greater than 150 but less than 220 and the percentage of abnormal area is relatively large, the battery pack electrolyte leakage is classified as a crystallization leakage. When the height of the abnormal peak is greater than the first peak value threshold and the percentage of abnormal area is greater than a preset area threshold, the battery pack electrolyte leakage is determined to be a mixed leakage. For a comparison diagram of the grayscale histogram and the normal surface histogram, please refer to [link to diagram]. Figure 2 The core of visual inspection is to identify changes in surface optical properties caused by electrolyte leakage. The main components of electrolyte have optical properties that are completely different from those of metal surfaces: EC and DMC liquids in electrolyte produce specular reflection in the visible light band, resulting in a significantly higher gray value in the leak area than the surrounding metal surface; the crystals formed after LiPF6 deliquescence appear as white powder, and their gray value is also significantly higher; the gray value of the electrolyte-wetted area is actually lower than that of the normal surface because the liquid absorbs light. Peak anomaly factor, scaling factor, and area anomaly factor are determined based on gray-level histograms and normal surface histograms. Visual anomaly confidence is then determined based on these factors, calculated using a three-factor weighted model based on the anomaly severity of the gray-level histograms. The visual anomaly confidence is as follows: , , , , in, For the confidence level of visual abnormalities, As a peak anomaly factor, This represents the maximum height of the main peak in the histogram of a normal surface. This represents the maximum height of the outlier peak in the grayscale histogram. As an area anomaly factor, This represents the area of ​​abnormal pixel regions with grayscale values ​​greater than 190 (or less than 80). This represents the total pixel area of ​​the image (ROI region). As a uniformity factor, The standard deviation of gray levels in the abnormal region. , , Weighting coefficients ( , , ),coefficient The scaling factor for abnormal peak values ​​maps the factor to the 0-1 interval within a reasonable range. hour, ,coefficient The area scaling factor (factor = 1 when the abnormal area accounts for 12.5%) maps the factor to the range of 0 to 1.

[0027] In some embodiments, in step S102, the torque data of the electric screwdriver and the motor current data are collected in real time during the unscrewing process of the target screw. Based on the torque data and the normal reference torque, the second leakage type and the confidence level of force abnormality are determined. During force detection, the six-dimensional force / torque sensor built into the collaborative robot is used with a sampling frequency of 500Hz to monitor the force on the tool in the X, Y, and Z directions and the torque around the three axes in real time. The force in the Z-axis direction is used to detect the contact state, and the torque around the Z-axis is used to monitor the resistance change during the screw unscrewing process. The core of force detection is to identify the torque abnormality caused by electrolyte crystallization or liquid resistance. During the normal screw unscrewing process, the torque curve shows the following characteristics: the peak torque in the coupling stage is about 0.15 Nm, and the torque in the stable unscrewing stage is... The torque is maintained at 0.08±0.02 Nm, with a torque fluctuation range of less than 0.05 Nm. When electrolyte leakage occurs, the torque curve exhibits the following abnormal patterns: electrolyte crystallization adheres to the screw threads or the inner wall of the screw hole, causing a sharp increase in unscrewing resistance, with the peak torque reaching over 0.35 Nm, which is 2 to 4 times the normal value; liquid electrolyte enters the screw hole, generating viscous resistance, causing the torque fluctuation range to increase to over ±0.08 Nm; the electrolyte reacts with the metal to produce gas, causing periodic torque drops and rises during unscrewing; Z-axis torque data is collected at a frequency of 500 Hz, and after applying a moving average filter to the raw data, the deviation between the current torque value and the normal reference torque, as well as the torque change rate, are calculated; if the absolute value of the torque exceeds... or or the rate of change of torque exceeds If the torque value is / second, it is marked as an abnormal force signal. Based on torque data and normal reference torque, the peak torque, torque fluctuation, and torque change rate are determined. When the peak torque is greater than the torque threshold and the torque fluctuation is less than the first fluctuation threshold, the electrolyte leakage of the power battery pack is determined to be a crystal leakage. When the torque fluctuation is greater than the second fluctuation threshold and the torque change rate is greater than the change rate threshold, the electrolyte leakage of the power battery pack is determined to be a liquid leakage. If the second fluctuation threshold is greater than the first fluctuation threshold, and the peak torque is greater than the torque threshold, the torque fluctuation is greater than the second fluctuation threshold, and the torque change rate is greater than the change rate threshold, the electrolyte leakage of the power battery pack is determined to be a mixed leakage. The current spiral... Compared to the normal torque baseline (peak value 0.15, stable value 0.08, fluctuation less than 0.05), if the torque of the filament is consistently high (>0.30) with small fluctuations, it is judged as crystallization; if the torque fluctuates greatly (>0.08) with a rapid rate of change, it is judged as liquid; if both characteristics are present, it is judged as mixing. Based on the torque data and the normal baseline torque, torque amplitude anomaly factors and torque change rate anomaly factors are determined. Based on the torque amplitude anomaly factors and torque change rate anomaly factors, the force perception anomaly confidence level is determined. The force perception anomaly confidence level is calculated based on the combined degree of torque amplitude anomaly and torque change rate anomaly. The force perception anomaly confidence level is: , , , in, For the confidence level of force perception abnormality, , These are the weighting coefficients. This is an abnormal factor for torque amplitude. This is an abnormal factor in the rate of change of torque. This is the current measured torque value. This is the reference value for the rotation torque under normal operating conditions. This is the baseline value for the rate of change of torque under normal operating conditions. The threshold for abnormal torque change rate. This represents the current rate of change of torque.

[0028] In some embodiments, in step S103, the estimated torque corresponding to the motor current data is determined based on the calibration relationship between motor current and output torque. The third leakage type and current anomaly confidence level are determined based on the estimated torque, torque data, and motor current data. During force detection, the motor current data of the electric screwdriver is simultaneously measured and transmitted via an electric screwdriver servo driver (such as Maxon Escon). 50 / 5) Real-time monitoring of motor current at a sampling frequency of 1kHz. Based on the calibration relationship between motor current and output torque, the output torque of the electric screwdriver is indirectly estimated with a resolution of 0.02 Nm. The cost is only 15% of that of a dedicated torque sensor. Current detection serves as a redundant verification method for force sensing. By monitoring the motor current of the electric screwdriver, the load status is indirectly reflected. There is a stable linear relationship between motor current and output torque. This relationship is established through offline calibration experiments: using a standard torque sensor to measure the output torque of the electric screwdriver under different loads, and simultaneously recording the corresponding motor current, a current-torque calibration relationship is established. The motor current is collected at a frequency of 1kHz. Secondly, the current is converted into an estimated torque through the calibration relationship between motor current and output torque. The estimated torque is cross-validated with the torque directly measured by the force sensor. If the deviation between the two exceeds 15%, it is determined that the sensor is abnormal or there is an unexpected load change. When the difference between the estimated torque and the actual torque data exceeds a preset threshold, and the motor current data exceeds a current threshold, the electrolyte leakage in the power battery pack is determined to be either a liquid leak or a crystallization leak. The current is converted into an estimated torque, and compared with the actual torque measured by the force sensor. If the deviation is small, the force sensor is considered reliable, and no separate type is determined. If the deviation is large and the current is high, it tends to be a crystallization or liquid leak, used to assist in confirming the force sensor judgment. The current anomaly confidence level is calculated based on the degree of deviation between the estimated torque and the measured torque. The current anomaly confidence level is determined based on the estimated torque and the actual torque data. The current anomaly confidence level is: , in, For current anomaly confidence level, To estimate torque, The value is the measured torque. The coefficient 2.5 is the deviation scaling factor. When the deviation ratio reaches 40%, the confidence level is 1.0.

[0029] In some embodiments, in step S104, the final leakage type of the electrolyte in the power battery pack is determined based on the first leakage type, the second leakage type, and the third leakage type. Voting is performed based on the confidence levels of the first leakage type, the second leakage type, and the third leakage type. For example, if the first leakage type based on vision is crystallization leakage with a confidence level of 0.86, the second leakage type based on force is crystallization leakage with a confidence level of 0.95, and the third leakage type based on current is normal with a confidence level of 0.12, the force weight is the maximum in the crystallization leakage scenario, which is 0.45. The voting result is crystallization leakage, and the final leakage type of the electrolyte in the power battery pack is crystallization leakage. Dynamic weights are determined based on the final leak type. After determining the final leak type, the weights of each mode are automatically adjusted according to the current operating conditions and leak type. Under normal operating conditions: visual weight 0.33, force weight 0.34, and current weight 0.33, with mutual verification among the three. For EC / DMC liquid leaks: visual weight increased to 0.40, force weight 0.30, and current weight 0.30. For LiPF6 crystallization leaks: force weight increased to 0.45, visual weight 0.25, and current weight 0.30. For mixed leaks: the weights of the three are evenly distributed and comprehensively judged. The weight allocation table is consulted based on the current operating conditions. (For details on how to consult the weight allocation table for the current operating conditions, please refer to [link to relevant documentation]). Figure 3 The dynamic weights are determined according to the weight allocation table. Based on the dynamic weights, the confidence levels for visual anomalies, force anomalies, and current anomalies, the fusion confidence level is calculated. The fusion confidence level is: , in, To integrate confidence levels, , , Dynamic weights; A warning system for electrolyte leakage in power battery packs is implemented based on fusion confidence level and preset grading thresholds. A yellow warning is triggered when the fusion confidence level is greater than or equal to the first grading threshold and less than the second grading threshold; an orange warning is triggered when the fusion confidence level is greater than or equal to the second grading threshold and less than the third grading threshold; and a red warning is triggered when the fusion confidence level is greater than or equal to the third grading threshold. A yellow alert is triggered when the value is greater than or equal to 0.60 and less than 0.80. An orange alert is triggered when the value is greater than or equal to 0.80 and less than 0.90. A value greater than or equal to 0.90 triggers a red alert; based on the fusion confidence level and the severity of the leakage, a Level IV emergency response is initiated, while a green response indicates normal operation. ): Continue normal disassembly operations, maintain current parameters, yellow alert ( Reduce the electric screwdriver speed to 50% of the normal value (e.g., from 105 rpm to 52 rpm), while increasing the visual sampling frequency to 60 frames per second, strengthening monitoring, and issuing an orange alert. Immediately stop disassembling the screws, keep the tool in place, activate the local negative pressure collection device (negative pressure value -50kPa), and continue suction for 30 seconds while sending an early warning message to the monitoring terminal; Red Emergency ( ): Immediately execute an emergency stop. The robot stops moving at maximum deceleration, activates the full-area negative pressure collection system (negative pressure value -80kPa), triggers an audible and visual alarm, notifies the operator to evacuate and put on protective equipment, and records abnormal data packets (containing all sensor data from the preceding and following 10 seconds) for subsequent analysis.

[0030] Please refer to the performance comparison diagram of different detection strategies. Figure 4 ,like Figure 4 As shown, the accuracy rate of single vision detection is 72.5%, the accuracy rate of single force detection is 68.3%, the accuracy rate of dual-modal fusion of vision and force is 91.8%, and the accuracy rate of tri-modal fusion of vision, force, and current reaches 97.2%. Even in electrolyte crystallization scenarios, the accuracy rate of tri-modal fusion detection remains at 96.8%, significantly better than single-modal detection, demonstrating a substantial improvement in detection accuracy. During 30 days of continuous operation testing, single vision detection had a cumulative false positive rate of 56.3, while tri-modal fusion detection had only 9.0, a reduction of 84.0%. This means that efficiency losses due to unintended production line shutdowns are reduced by more than 80%, and the false alarm rate is significantly lower. In severe leakage scenarios, tri-modal fusion detection... With a false negative rate of 0.7%, leakage events were essentially captured comprehensively, fundamentally eliminating potential safety hazards caused by missed detections. The false negative rate was close to zero, and the total response time was 0.90 seconds, far less than the time required for electrolyte leakage to spread to dangerous areas, providing ample time for the effective implementation of emergency measures. The introduction of yellow and orange alerts avoided the crude mode of shutting down the entire line as soon as an anomaly was detected. After adopting the tiered strategy, the downtime caused by leakage detection was reduced by 65%, and the overall production capacity was increased by about 18%. Current detection utilizes the existing servo drive of the electric screwdriver, eliminating the need to purchase additional torque sensors. The hardware cost increased by no more than 5%, but the detection reliability was improved by more than 40%.

[0031] For the automated dismantling of retired lithium iron phosphate battery packs, the dismantling objectives are to remove the module fixing screws, separate the module from the casing, and recover the battery cells. The configuration is as follows: Robot: Universal Robots UR10e, 10kg payload, 1300mm working radius; Vision sensor: Intel RealSense D405, RGB resolution. Depth resolution The frame rate is 30fps; the electric screwdriver is a custom modular electric screwdriver with a Maxon DXL 32L motor and planetary gearbox, rated torque 2.0Nm, and maximum speed 525rpm; the drive is a Maxon Escon 50 / 5, supporting real-time current monitoring; the negative pressure collection device is an industrial vacuum pump equipped with a HEPA filter and activated carbon adsorption layer, with an air volume of 120 cubic meters per hour; its operation process is as follows: Step 1: The robot moves with the end effector to about 300mm above the first module, and the vision system scans the module surface to identify and classify all visible screw positions; Step 2: The robot automatically selects the corresponding bit according to the screw type, and the vision system confirms that the bit color label matches the screw type; Step 3: The robot moves to 60mm directly above the target screw, and the vision system accurately locates the screw center and calculates the relative pose deviation between the tool and the screw; Step 4: The robot moves along the Z-axis at a speed of 5mm... The speed decreases by 1 / s, and the Z-axis force is monitored. When the force exceeds 5N, the spiral search mode is triggered, and the electric screwdriver rotates clockwise at 20rpm to assist the coupling of the screwdriver bit and the screw head. Step 5: After successful coupling, the electric screwdriver switches to 105rpm counterclockwise rotation, and the robot releases the Z-axis degree of freedom and automatically retracts as the screw is unscrewed, keeping the Z-axis force close to zero. Step 6: During the unscrewing process, the three-modal detection system continues to run. If an anomaly is detected, a graded emergency response is immediately executed. An example of an anomaly detection is as follows: During the disassembly of the 7th M6 screw in module 3, an anomaly was detected: the vision module found a bright area with a grayscale value of 205 around the screw, and the confidence level of the visual anomaly was determined. The force sensing module detected a sudden increase in torque from 0.08 Nm to 0.32 Nm, with a rate of change of 0.8 Nm / s, and determined the confidence level of the force sensing anomaly. The current module detected a sudden increase in motor current from 1.2A to 3.5A, estimating torque at 0.30Nm, a deviation of 6.7% from the force measurement. The confidence level for determining the current anomaly was [not specified]. Query the weight allocation table; the current operating condition is suspected to be LiPF6 crystallization. Assign weights accordingly. , , Calculate the fusion confidence score: ,because If an orange alert is triggered: immediately stop unscrewing, maintain the current position, start local negative pressure collection, and send an alert message to the monitoring terminal. After remote confirmation by the operator, replace the bit with a corrosion-resistant material and continue operation.

[0032] By fusing visual, force, and current detection modes, the detection accuracy is significantly improved: single visual detection accuracy is 72.5%, single force detection accuracy is 68.3%, visual + force dual-modal fusion accuracy is 91.8%, and the three-modal fusion detection accuracy reaches 97.2%. Even in electrolyte crystallization scenarios, the three-modal fusion detection accuracy remains at 96.8%, significantly better than single-modal detection, and the false alarm rate is significantly reduced: in 30 days of continuous operation testing, single visual detection had a cumulative false alarm of 56.3 times, while the three-modal fusion detection had only 9.0 times, a reduction of 84.0%. This means that the efficiency loss caused by unintended production line shutdowns is reduced by more than 80%; the false alarm rate is close to zero: in severe leakage scenarios, the false alarm rate of three-modal fusion detection is 0.7%, essentially achieving near-zero false alarm rate. It achieves comprehensive capture of leakage incidents, fundamentally eliminating potential safety hazards caused by missed detections; the response speed meets real-time requirements: the total response time is 0.90 seconds, far less than the time required for electrolyte leakage to spread to the dangerous area, providing ample time window for the effective implementation of emergency measures; the tiered emergency strategy improves production line continuity: by introducing yellow and orange alerts, the crude mode of shutting down the entire line as soon as an anomaly is detected is avoided. Actual test data shows that after adopting the tiered strategy, the downtime caused by leakage detection is reduced by 65%, and the overall capacity is increased by about 18%; high reliability is achieved at low cost: the current detection module utilizes the existing servo driver of the electric screwdriver, eliminating the need to purchase an additional torque sensor, increasing hardware costs by no more than 5%, but improving detection reliability by more than 40%.

[0033] In summary, the multimodal detection method for electrolyte leakage in power battery packs provided by this invention acquires images of an electric screwdriver approaching a target screw in the power battery pack and determines the grayscale histogram of the image. Based on the grayscale histogram and the normal surface histogram, a first leakage type and a visual anomaly confidence level are determined. Real-time acquisition of torque data from the electric screwdriver and motor current data during the screw unscrewing process is used to determine a second leakage type and a force-feel anomaly confidence level based on the torque data and the normal reference torque. Based on the calibration relationship between motor current and output torque, an estimated torque corresponding to the motor current data is determined. Based on the estimated torque, torque data, and motor current data, a third leakage type and a current anomaly confidence level are determined. Based on the first, second, and third leakage types, the final leakage type of the power battery pack electrolyte is determined. Dynamic weights are determined based on the final leakage type. A fusion confidence level is calculated based on the dynamic weights, visual anomaly confidence level, force-feel anomaly confidence level, and current anomaly confidence level. Based on the fusion confidence level and a preset grading threshold, an early warning is issued for power battery pack electrolyte leakage, thus improving the accuracy of power battery pack electrolyte leakage detection.

[0034] To better implement the multimodal detection method for electrolyte leakage in power battery packs in this embodiment of the invention, based on the multimodal detection method for electrolyte leakage in power battery packs, correspondingly, as follows: Figure 5 As shown, this embodiment of the invention also provides a multimodal detection system 500 for electrolyte leakage in a power battery pack. The multimodal detection system 500 for electrolyte leakage in a power battery pack includes: The visual inspection module 501 acquires images of the electric screwdriver approaching the target screw of the power battery pack and determines the grayscale histogram of the image. Based on the grayscale histogram and the normal surface histogram, it determines the first leakage type and the confidence level of visual anomaly. The force detection module 502 is used to collect the torque data of the electric screwdriver and the motor current data in real time during the process of unscrewing the target screw. Based on the torque data and the normal reference torque, the second leakage type and the confidence level of force abnormality are determined. The current detection module 503 is used to determine the estimated torque corresponding to the motor current data based on the calibration relationship between motor current and output torque, and to determine the third leakage type and the current anomaly confidence level based on the estimated torque, torque data and motor current data. The early warning module 504 is used to determine the final leakage type of the electrolyte in the power battery pack based on the first leakage type, the second leakage type, and the third leakage type; determine the dynamic weight based on the final leakage type; calculate the fusion confidence level based on the dynamic weight, the confidence level of visual anomalies, the confidence level of force anomalies, and the confidence level of current anomalies; and issue an early warning for electrolyte leakage in the power battery pack based on the fusion confidence level and the preset classification threshold.

[0035] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A multimodal detection method for electrolyte leakage in a power battery pack, characterized in that, include: The image of the electric screwdriver approaching the target screw of the power battery pack is acquired, and the grayscale histogram of the image is determined. Based on the grayscale histogram and the normal surface histogram, the first leakage type and the confidence level of visual anomaly are determined. The torque data of the electric screwdriver and the motor current data are collected in real time during the process of unscrewing the target screw. Based on the torque data and the normal reference torque, the second leakage type and the confidence level of force abnormality are determined. Based on the calibration relationship between motor current and output torque, the estimated torque corresponding to the motor current data is determined, and based on the estimated torque, torque data, and motor current data, the third leakage type and the confidence level of current anomaly are determined. The final leakage type of the battery pack electrolyte is determined based on the first leakage type, the second leakage type, and the third leakage type. A dynamic weight is determined based on the final leakage type. A fusion confidence level is calculated based on the dynamic weight, the confidence level of visual anomalies, the confidence level of force anomalies, and the confidence level of current anomalies. An early warning is issued for battery pack electrolyte leakage based on the fusion confidence level and a preset grading threshold.

2. The multi-modal detection method for electrolyte leakage in a power battery pack according to claim 1, characterized in that, The first leakage type includes liquid leakage, crystallization leakage, and mixed leakage. Determining the first leakage type and the confidence level of visual anomalies based on the grayscale histogram and the normal surface histogram includes: The height of abnormal peaks and the proportion of abnormal area are determined based on the grayscale histogram and the normal surface histogram. When the height of the abnormal peak is greater than the first peak value threshold, the leakage of electrolyte in the power battery pack is determined to be a liquid leakage. When the abnormal area ratio is greater than a preset area threshold and the abnormal peak height is greater than a second peak threshold and less than a third peak threshold, the electrolyte leakage of the power battery pack is determined to be a crystal leakage, wherein the first peak threshold and the second peak threshold are less than the third peak threshold, and the first peak threshold is greater than the second peak threshold. When the height of the abnormal peak is greater than the first peak value threshold and the proportion of the abnormal area is greater than the preset area threshold, the electrolyte leakage of the power battery pack is determined to be a mixed leakage. The peak anomaly factor, scaling factor, and area anomaly factor are determined based on the grayscale histogram and the normal surface histogram, and the visual anomaly confidence level is determined based on the peak anomaly factor, scaling factor, and area anomaly factor.

3. The multimodal detection method for electrolyte leakage in a power battery pack according to claim 2, characterized in that, The confidence level for the visual anomaly is: ， ， ， ， in, For the confidence level of visual abnormalities, As a peak anomaly factor, This represents the maximum height of the main peak in the histogram of a normal surface. This represents the maximum height of the outlier peak in the grayscale histogram. As an area anomaly factor, The area of ​​the abnormal pixel region. The total pixel area of ​​the image. As a uniformity factor, The standard deviation of gray levels in the abnormal region. , , These are the weighting coefficients. This is the peak anomaly scaling factor. This is the area scaling factor.

4. The multimodal detection method for electrolyte leakage in a power battery pack according to claim 1, characterized in that, The second leakage type includes liquid leakage, crystallization leakage, and mixed leakage. Determining the second leakage type and the confidence level of force anomaly based on the torque data and the normal reference torque includes: Based on the torque data and the normal reference torque, determine the peak torque, torque fluctuation, and torque change rate; When the peak torque is greater than the torque threshold and the torque fluctuation is less than the first fluctuation threshold, the electrolyte leakage of the power battery pack is determined to be a crystal leakage. When the torque fluctuation is greater than the second fluctuation threshold and the torque change rate is greater than the change rate threshold, the electrolyte leakage of the power battery pack is a liquid leakage, wherein the second fluctuation threshold is greater than the first fluctuation threshold; When the peak torque is greater than the torque threshold, the torque fluctuation is greater than the second fluctuation threshold, and the torque change rate is greater than the change rate threshold, the electrolyte leakage of the power battery pack is determined to be a mixed leakage. Based on the torque data and the normal reference torque, determine the torque amplitude anomaly factor and the torque change rate anomaly factor, and based on the torque amplitude anomaly factor and the torque change rate anomaly factor, determine the confidence level of force perception anomaly.

5. The multimodal detection method for electrolyte leakage in a power battery pack according to claim 4, characterized in that, The confidence level for the abnormal force perception is: ， ， , in, For the confidence level of force perception abnormality, , These are the weighting coefficients. This is an abnormal factor for torque amplitude. This is an abnormal factor in the rate of change of torque. This is the current measured torque value. This is the reference value for the rotation torque under normal operating conditions. This is the baseline value for the rate of change of torque under normal operating conditions. The threshold for abnormal torque change rate. This represents the current rate of change of torque.

6. The multimodal detection method for electrolyte leakage in a power battery pack according to claim 1, characterized in that, The third leakage type includes liquid leakage and crystallization leakage. Determining the third leakage type and the confidence level of current anomaly based on the estimated torque, torque data, and motor current data includes: When the difference between the estimated torque and the torque data is greater than a preset threshold and the motor current data is greater than the current threshold, the electrolyte leakage of the power battery pack is determined to be either liquid leakage or crystallization leakage. The confidence level of the current anomaly is determined based on the estimated torque and the torque data.

7. The multimodal detection method for electrolyte leakage in a power battery pack according to claim 6, characterized in that, The confidence level for the current anomaly is: ， in, For current anomaly confidence level, To estimate torque, This is the measured torque value.

8. The multimodal detection method for electrolyte leakage in a power battery pack according to claim 7, characterized in that, The fusion confidence level is: ， in, To integrate confidence levels, , , For dynamic weights.

9. The multimodal detection method for electrolyte leakage in a power battery pack according to claim 8, characterized in that, The method of providing early warning of electrolyte leakage in power battery packs based on the fused confidence level and a preset grading threshold includes: When the fusion confidence level is greater than or equal to the first level threshold and less than the second level threshold, a yellow warning is triggered. When the fusion confidence level is greater than or equal to the second-level threshold and less than the third-level threshold, an orange alert is triggered. A red alert is triggered when the fusion confidence level is greater than or equal to the third-level threshold.

10. A multimodal detection system for electrolyte leakage in a power battery pack, characterized in that, include: The visual inspection module acquires an image of the electric screwdriver approaching the target screw of the power battery pack and determines the grayscale histogram of the image. Based on the grayscale histogram and the normal surface histogram, it determines the first leakage type and the confidence level of visual anomaly. The force detection module is used to collect torque data of the electric screwdriver and motor current data in real time during the process of unscrewing the target screw. Based on the torque data and the normal reference torque, the second leakage type and the confidence level of force abnormality are determined. The current detection module is used to determine the estimated torque corresponding to the motor current data based on the calibration relationship between motor current and output torque, and to determine the third leakage type and the current anomaly confidence level based on the estimated torque, torque data and motor current data. The early warning module is used to determine the final leakage type of the electrolyte in the power battery pack based on the first leakage type, the second leakage type, and the third leakage type; determine a dynamic weight based on the final leakage type; calculate a fusion confidence level based on the dynamic weight, the confidence level of visual anomalies, the confidence level of force anomalies, and the confidence level of current anomalies; and issue an early warning for electrolyte leakage in the power battery pack based on the fusion confidence level and a preset grading threshold.

Images