A heat-conducting adhesive disassembly optimization method based on sensor data and intelligent algorithms

Abstract

The application provides a heat-conducting adhesive disassembly optimization method based on sensor data and intelligent algorithms, comprising: S1, real-time acquisition of the pulling force and angle of the heat-conducting adhesive disassembly, and determination of the nonlinear resistance change trend; S2, if the nonlinear resistance exceeds the resistance threshold, the pulling angle and speed of the feedback adjustment mechanical arm are activated, the real-time resistance feedback signal is obtained, and the damage degree of the heat-conducting adhesive structure is judged; S3, according to the damage degree, the action parameter sequence is iteratively optimized, the control instruction of gradual separation is generated, the surface stress of the battery assembly is synchronously monitored, and the damage risk is quantitatively evaluated; S4, if the damage risk is lower than the risk threshold, the adjusted pulling force is applied, the separation progress is monitored, the remaining bonding area information is extracted, it is determined whether the complete separation is achieved, and the final disassembly result is output. The application can accurately control the disassembly process, significantly reduce the damage risk of the battery assembly, and improve the disassembly efficiency and safety.

Description

Technical Field

[0001] This invention relates to the field of intelligent disassembly technology for lithium batteries, and in particular to an optimized method for disassembling thermally conductive adhesive based on sensor data and intelligent algorithms. Background Technology

[0002] Intelligent dismantling of lithium batteries is a crucial link in the new energy field, playing an indispensable role in promoting resource recycling and environmental protection. With the widespread application of electric vehicles and energy storage devices, the demand for recycling and dismantling used lithium batteries is increasing daily, which not only concerns economic benefits but also directly impacts the goal of sustainable ecological development. However, how to ensure efficiency while avoiding damage to battery components during the dismantling process has become a bottleneck that the industry urgently needs to overcome.

[0003] Currently, lithium battery disassembly largely relies on traditional mechanical operations or manual intervention, methods that often fall short when dealing with complex adhesive materials. Especially when separating the battery assembly from the bottom thermally conductive adhesive, existing technologies focus more on simply applying tensile force, neglecting the complex changing characteristics of the adhesive material under stress. These changing characteristics can lead to unpredictable resistance fluctuations during disassembly, increasing the difficulty of the operation and potentially causing structural damage to the battery assembly.

[0004] Against this backdrop, the core technological challenge in the field of intelligent lithium battery disassembly is increasingly focusing on how to address the adhesive resistance posed by the thermally conductive adhesive at the bottom. As a high-strength adhesive material, thermally conductive adhesive exhibits non-linear resistance changes under stress, making it difficult to effectively break its physical structure using simple linear pulling. Depending on the pulling angle and force, the adhesive force of the thermally conductive adhesive may suddenly increase or decrease at certain key points. If the direction and speed of the disassembly action cannot be adjusted in time, it may lead to pulling failure or irreversible damage to the battery components. For example, when disassembling a certain power battery, if the mechanical device cannot flexibly adjust the pulling angle according to changes in resistance, it may tear or deform the battery casing before the thermally conductive adhesive is completely separated.

[0005] Therefore, how to dynamically adapt to the nonlinear resistance changes of the thermally conductive adhesive during disassembly, and how to gradually destroy its bonding structure by flexibly adjusting mechanical actions, has become a key issue in the field of intelligent lithium battery disassembly. Solving this problem not only affects the improvement of disassembly efficiency, but also directly impacts the integrity protection of battery components and their subsequent recycling value. Summary of the Invention

[0006] To address the technical problems mentioned in the background section, this invention provides an optimized method for disassembling thermal conductive adhesive based on sensor data and intelligent algorithms, the method comprising: S1, real-time acquisition of the pulling force and angle of the thermal conductive adhesive disassembly to determine the trend of nonlinear resistance change; S2, if the nonlinear resistance exceeds the resistance threshold, then activate feedback to adjust the pulling angle and speed of the robotic arm, obtain real-time resistance feedback signals, and determine the degree of damage to the thermally conductive adhesive bonding structure. S3, based on the degree of damage, iteratively optimizes the sequence of action parameters, generates progressively separated control commands, synchronously monitors the surface stress of the battery module, and quantitatively assesses the risk of damage; S4. If the risk of damage is lower than the risk threshold, apply an adjusted pulling force, monitor the separation progress and extract information on the remaining adhesive area, determine whether the separation is complete, and output the final disassembly result.

[0007] Furthermore, step S1 includes: Step S11: The pulling force and angle of the thermal conductive adhesive during the disassembly process are collected in real time by the sensor. After pre-processing, cleaning and standardization, a standardized resistance value distribution dataset is constructed. Step S12: Apply convolutional neural network to extract features and analyze the trend of nonlinear resistance change. If the feature threshold is exceeded, the classification is refined in layers and combined with the update frequency to obtain the dynamically adjusted resistance change prediction result. Step S13: If the predicted resistance change after dynamic adjustment deviates significantly from the actual value, calibrate the sensor to determine the final accurate resistance change trend.

[0008] Furthermore, step S11 includes: acquiring the pulling force and angle value of the thermally conductive adhesive in real time at a sampling frequency of 50 Hz using a six-axis torque sensor and a high-precision tilt sensor configured at the end of the robotic arm.

[0009] Furthermore, step S12 includes: constructing a convolutional neural network model containing three convolutional layers and two pooling layers for the resistance value distribution dataset.

[0010] Furthermore, by taking time-series data from 200 consecutive sampling points as input, key feature values ​​of local tensile force abrupt changes and angle shifts are extracted through a one-dimensional convolution kernel, and the core mode characterizing colloidal viscous fracture is output.

[0011] Furthermore, step S2 includes: Step S21: When the nonlinear resistance exceeds the resistance threshold, analyze the fluctuation characteristics to obtain preliminary abnormal results, activate the feedback mechanism, send instructions to the robotic arm to adjust the pulling angle and speed, and optimize the motion parameters. Step S22: Perform the adjusted pulling operation according to the optimized parameters, collect real-time feedback data to obtain the latest resistance signal, identify abnormal patterns through SVM classification, and determine the degree of damage to the thermally conductive adhesive bonding structure. Step S23: If the degree of damage exceeds the safe range, generate a new adjustment command to recalculate the pulling angle and speed, update the motion parameters, continuously monitor the resistance signal and record the feedback data; Step S24: When the monitoring shows that the resistance signal has returned to the normal range, save the current action parameters as a reference to complete the entire adjustment process.

[0012] Furthermore, step S21 includes: extracting the first derivative of the nonlinear resistance data as an abnormal fluctuation feature.

[0013] Furthermore, step S3 includes: Step S31: Generate initial control commands based on the degree of damage, synchronously monitor the stress on the battery surface, and record the stress fluctuations corresponding to the commands; Step S32: Mark high-risk areas of stress anomalies, adjust motion parameters accordingly, and generate new control commands to reduce abnormal fluctuations; Step S33: If there is still a high risk after the new instruction is executed, the action parameters are iteratively updated until the stress stabilizes, and finally the damage risk is determined and a safety control instruction is output.

[0014] Furthermore, step S32 includes: extracting the peak value and variance characteristics of the battery surface stress data within a specific time window; if the stress peak value is detected to exceed the stress threshold, the corresponding operation area is marked as a high-risk area.

[0015] Furthermore, step S4 includes: Step S41: Obtain initial damage risk data, generate control commands after assessment meets the standards, dynamically adjust the pulling force and continue to apply force, and collect separation progress indicators in real time. Step S42: Extract the information of the remaining adhesive area. If the proportion meets the standard, it is judged as a complete separation; otherwise, the force parameters are readjusted. Step S43: Combine the full-process risk data and progress indicators to generate the final breakdown result, confirm the task completion status, and iteratively optimize the control instructions if the target is not met.

[0016] The technical solution provided by this invention has the following beneficial effects: This invention discloses an optimized method for thermally conductive adhesive disassembly based on sensor data and intelligent algorithms. Addressing the unique business scenario of thermally conductive adhesive separation during battery assembly disassembly, it integrates logically related issues such as real-time monitoring of pulling force and angle, analysis of nonlinear resistance change trends, optimization of robotic arm motion parameters, and damage risk control. This invention collects pulling force and angle data using sensors, analyzes resistance distribution using a convolutional neural network model, determines the nonlinear change trend, and activates a feedback mechanism to adjust robotic arm parameters if the trend exceeds a threshold. Subsequently, iteratively updates the motion sequence using reinforcement learning algorithms, assesses damage risk using surface stress data, and continuously applies optimized pulling force when the risk remains below the threshold, ultimately achieving complete separation of the thermally conductive adhesive. The core innovation of this invention lies in its precise control of the disassembly process through multi-algorithm collaboration and a real-time feedback mechanism, significantly reducing the risk of battery assembly damage and improving disassembly efficiency and safety. Attached Figure Description

[0017] Figure 1 This is a flowchart of an optimized method for disassembling thermal conductive adhesive based on sensor data and intelligent algorithms, according to the present invention. Detailed Implementation

[0018] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0019] like Figure 1 As shown in the figure, this embodiment presents an optimized method for disassembling thermal conductive adhesive based on sensor data and intelligent algorithms, which may specifically include: S1 collects the pulling force and angle of the thermally conductive adhesive during disassembly in real time to determine the trend of nonlinear resistance change.

[0020] Optionally, this step also includes: Step S11: The pulling force and angle of the thermal conductive adhesive during the disassembly process are collected in real time by the sensor. After pre-processing, cleaning and standardization, a standardized resistance value distribution dataset is constructed.

[0021] The pulling force and angle values ​​during the disassembly process of thermal conductive adhesive are collected in real time by sensors to obtain raw resistance data. Preprocessing technology is used to clean and standardize the raw resistance data to construct a standardized resistance value distribution dataset.

[0022] Step S12: Apply convolutional neural network to extract features and analyze the trend of nonlinear resistance change. If the feature exceeds the threshold, refine the classification by layer and combine it with the update frequency to obtain the dynamically adjusted resistance change prediction result.

[0023] For a standardized resistance value distribution dataset, a convolutional neural network model is applied to extract key feature values ​​and determine the core patterns of the resistance data. Based on the extracted key feature values, the correspondence between the feature values ​​and nonlinear changes is analyzed to determine the nonlinear resistance change trend. If the determined nonlinear resistance change trend exceeds a preset feature threshold range, the feature values ​​are further stratified to obtain a refined change trend classification. Combined with the update frequency of real-time data acquisition, a dynamically adjusted resistance change prediction result is obtained.

[0024] Step S13: If the predicted resistance change after dynamic adjustment deviates significantly from the actual value, calibrate the sensor to determine the final accurate resistance change trend.

[0025] If the predicted resistance change after dynamic adjustment deviates significantly from the actual collected data, the sensor acquisition parameters are recalibrated to determine the final resistance change trend and support the disassembly decision.

[0026] Specifically, in the disassembly of the thermally conductive adhesive, a six-axis torque sensor and a high-precision tilt sensor configured at the end of the robotic arm are used to acquire the pulling force and angle values ​​in real time at a sampling frequency of 50 Hz. The raw resistance data often contains high-frequency noise from mechanical vibrations. A moving average filtering algorithm is used to clean the raw data, removing abnormal glitch signals. Then, a standard score normalization method is used to map the pulling force to an interval with a mean of 0 and a standard deviation of 1, thereby obtaining a normalized resistance value distribution dataset. This effectively eliminates the differences in the dimensions of different sensors, significantly improving the convergence speed and stability of subsequent model processing.

[0027] For example, for the aforementioned standardized resistance value distribution dataset, a convolutional neural network model containing three convolutional layers and two pooling layers was constructed. Using time-series data from 200 consecutive sampling points as input, key feature values ​​of local tensile force abrupt changes and angular shifts were extracted using a one-dimensional convolutional kernel. The output represented the core patterns of viscous fracture in the adhesive (such as internal tearing, interfacial peeling, or substrate deformation). Based on the extracted key feature values, a support vector regression algorithm was used to analyze the correspondence between the feature values ​​and nonlinear changes. When the tensile force increases exponentially within 0.5 seconds and the angular shift exceeds 15 degrees, it is identified as a high-viscosity nonlinear resistance change trend. This deep learning-based feature extraction method can accurately capture the micromechanical changes of thermally conductive adhesives at different curing degrees, improving the accuracy of trend judgment.

[0028] It should be noted that if the predicted peak tensile force corresponding to the determined nonlinear resistance change trend exceeds the preset feature threshold range of 80 Newtons, a decision tree algorithm is used to further stratify the feature values, refining them into more detailed trend classifications such as internal tearing of the colloid, interface peeling, or substrate deformation. Combined with a real-time data acquisition and update frequency of 50 Hz, a Kalman filter algorithm is used to fuse the refined classification results, obtaining a dynamically adjusted resistance change prediction result. If the deviation between this prediction result and the actual acquired tensile force data exceeds 10 Newtons, a feedback mechanism is triggered to recalibrate the sensor's zero-point drift parameters and gain coefficient. This establishes a closed-loop data verification and hardware optimization mechanism, ensuring the absolute reliability of the final resistance change trend under complex disassembly conditions.

[0029] S2, if the nonlinear resistance exceeds the resistance threshold, then activate feedback to adjust the pulling angle and speed of the robotic arm, obtain real-time resistance feedback signals, and determine the degree of damage to the thermally conductive adhesive bonding structure.

[0030] Optionally, this step also includes: Step S21: When the nonlinear resistance exceeds the resistance threshold, analyze the fluctuation characteristics to obtain preliminary abnormal results, activate the feedback mechanism, send instructions to the robotic arm to adjust the pulling angle and speed, and optimize the motion parameters.

[0031] If the detected nonlinear resistance data exceeds the preset resistance threshold, the data processing module analyzes the resistance change trend, extracts abnormal fluctuation characteristics, and obtains a preliminary anomaly judgment result. Based on the preliminary anomaly judgment result, the feedback mechanism is activated, sending instructions to the robotic arm adjustment module to dynamically update the pulling angle and pulling speed, and determining the optimized motion parameters.

[0032] Step S22: Perform the adjusted pulling operation according to the optimized parameters, collect real-time feedback data to obtain the latest resistance signal, identify abnormal patterns through SVM classification, and determine the degree of damage to the thermally conductive adhesive bonding structure.

[0033] By optimizing the motion parameters, the robotic arm is driven to perform an adjusted pulling operation. Real-time feedback data is collected during the execution process to obtain the latest resistance signal. The acquired resistance signal is then classified using a support vector machine algorithm to analyze abnormal patterns and determine the degree of damage to the thermally conductive adhesive bonding structure.

[0034] Step S23: If the degree of damage exceeds the safe range, generate a new adjustment command to recalculate the pulling angle and speed, update the motion parameters, continuously monitor the resistance signal and record feedback data.

[0035] Based on the assessment of the degree of damage, if the damage exceeds the safe range, a new adjustment command is generated to recalculate the pulling angle and pulling speed, and update the motion parameters. Using the updated motion parameters, changes in the resistance signal are continuously monitored, and feedback data after each adjustment is recorded to determine the stability of the thermally conductive adhesive bonding structure.

[0036] Step S24: When the monitoring shows that the resistance signal has returned to the normal range, save the current action parameters as a reference to complete the entire adjustment process.

[0037] Based on the continuous monitoring results of the stability status, if the resistance signal is detected to have returned to the normal range, the current action parameters are saved as a reference, and the adjustment process is completed.

[0038] In one possible implementation, when the nonlinear resistance data is detected to exceed a preset resistance threshold of 50 Newtons, the data processing module performs time-domain analysis on the trend value of the nonlinear resistance change.

[0039] Specifically, the first derivative of the nonlinear resistance data is extracted as an abnormal fluctuation feature. If the rate of change of resistance exceeds 15 Newtons per second for 0.5 seconds, a preliminary anomaly judgment is obtained. Based on this result, the feedback mechanism is activated to send instructions to the robotic arm adjustment module.

[0040] For example, the initial vertical 90-degree pulling angle is dynamically updated to 75 degrees, and the initial pulling speed of 5 millimeters per second is reduced to 2 millimeters per second, thereby determining the optimized motion parameters and driving the robotic arm to perform the adjusted pulling operation.

[0041] In one possible implementation, as the robotic arm executes according to optimized parameters, sensors acquire real-time feedback data at a frequency of 100 Hz. The latest resistance data signal is then classified using a support vector machine algorithm.

[0042] Specifically, the extracted peak tensile force and angle deviation are used as input feature vectors, and a support vector machine model is constructed using a radial basis kernel function to output a classification label for the degree of damage to the thermally conductive adhesive bonding structure.

[0043] For example, the classification labels are set to three levels: minor peeling, local breakage, and structural tearing, in order to analyze the abnormal patterns in the signal and determine the current damage state of the thermal conductive adhesive.

[0044] In one possible implementation, if the degree of destruction output by the support vector machine is structural tearing, which exceeds the safe range, then an adjustment instruction is generated.

[0045] Specifically, the motion parameters were recalculated, further limiting the pulling speed to 1 millimeter per second and adjusting the pulling angle to 60 degrees. The variance of the resistance signal was continuously monitored using the updated motion parameters.

[0046] For example, the feedback data after each adjustment is recorded. If the variance of the resistance signal is found to be within 2.5 for 3 consecutive seconds, that is, it has returned to the normal range, the adhesive structure is determined to be in a stable state. Then the current action parameters are saved as a reference benchmark for subsequent disassembly, and the entire dynamic adjustment process is completed.

[0047] S3, based on the degree of damage, iteratively optimizes the sequence of action parameters, generates progressively separated control commands, synchronously monitors the surface stress of the battery components, and quantitatively assesses the risk of damage.

[0048] Optionally, this step also includes: Step S31: Generate initial control commands based on the degree of damage, synchronously monitor the stress on the battery surface, and record the stress fluctuations corresponding to the commands.

[0049] A reinforcement learning algorithm is used to iteratively update the action parameter sequence, generating an initial control command sequence in response to changes in the degree of damage. Real-time monitoring of the battery assembly is conducted using sensor devices to acquire surface stress data, recording stress fluctuations corresponding to the control command sequence.

[0050] Step S32: Mark high-risk areas of stress anomalies, adjust motion parameters accordingly, and generate new control commands to reduce abnormal fluctuations.

[0051] For the acquired surface stress data, the correspondence between it and the control command sequence is analyzed to determine the range of abnormal fluctuations in the stress data. If the fluctuation exceeds a preset fluctuation threshold, it is marked as a high-risk area. Based on the marked high-risk areas, the action parameters of the reinforcement learning algorithm are adjusted to generate an optimized new control command sequence, thereby reducing abnormal fluctuations in surface stress.

[0052] Step S33: If there is still a high risk after the new instruction is executed, the action parameters are iteratively updated until the stress stabilizes, and finally the damage risk is determined and a safety control instruction is output.

[0053] After executing the new control command sequence, acquire new surface stress data, analyze its fluctuation trend, and determine whether high-risk areas still exist. If high-risk areas still exist, continue iteratively updating the action parameter sequence to generate the latest control command sequence until the surface stress data fluctuations stabilize, obtaining the final damage risk level. By analyzing the final damage risk level, determine the safety status of the battery module, and output the corresponding control command sequence as the basis for subsequent operations.

[0054] In one possible implementation, a deep Q-network algorithm is used to iteratively update the sequence of motion parameters to account for changes in the degree of damage during the disassembly of the battery assembly. Taking the initial physical state characteristics of the current battery assembly as input, the deep Q-network algorithm outputs an initial control command sequence containing the robotic arm's movement trajectory and applied force.

[0055] For example, surface stress data is acquired in real time using fiber Bragg grating sensors deployed on the surface of the battery assembly. When the initial control command sequence is executed, the sensor records stress changes at a sampling rate of 50 Hz, forming a time-series stress data set to ensure a one-to-one correspondence between the stress data and the control commands in the time dimension.

[0056] It should be noted that a dynamic time warping algorithm is used to analyze the time alignment relationship between the acquired surface stress data and the control command sequence. The peak value and variance characteristics of the stress data within a specific time window are extracted. If the detected stress peak value exceeds a preset stress threshold of 45 MPa, the corresponding operating area is marked as a high-risk area.

[0057] In one possible implementation, the reward function weights of the deep Q-network algorithm are adjusted based on the marked high-risk areas, and a penalty term for stress exceeding limits is added. An optimized new control command sequence is regenerated based on the updated algorithm model, and new surface stress data is obtained after execution. If the peak stress of the new surface stress data drops below 30 MPa and the variance of the fluctuation is less than the variance threshold of 2.5, the surface stress data is considered stable. Based on this stable state, the final damage risk level assessment value is output, determining that the battery assembly is in a safe disassembly state, and the current control command sequence is used as the basis for subsequent operations.

[0058] S4. If the risk of damage is lower than the risk threshold, apply an adjusted pulling force, monitor the separation progress and extract information on the remaining adhesive area, determine whether the separation is complete, and output the final disassembly result.

[0059] Optionally, this step also includes: Step S41: Obtain initial damage risk data, generate control commands after assessment meets the standards, dynamically adjust the pulling force and continue to apply force, and collect separation progress indicators in real time.

[0060] Initial damage risk data is acquired through the damage risk assessment module and compared with a preset risk threshold to determine whether the conditions for applying pulling force are met, thus obtaining the risk assessment result. If the risk assessment result is lower than the preset risk threshold, the control command generation module generates corresponding control commands to dynamically adjust the pulling force and determine the adjusted pulling force parameters. Based on the adjusted pulling force parameters, the pulling force is continuously applied, and separation progress data is collected in real time. Separation progress indicators are extracted from this data to determine the current completion status of the separation process.

[0061] Step S42: Extract the information of the remaining adhesive area. If the proportion meets the standard, it is judged as a complete separation; otherwise, the force parameters are readjusted.

[0062] Information on the remaining bonding area is extracted from the separation progress index. Image processing technology is used to identify the boundaries of the bonding area and obtain the distribution data of the remaining bonding area. If the proportion of the remaining bonding area is lower than the preset standard, it is determined to be a complete separation state, and the separation state result is obtained.

[0063] Step S43: Combine the full-process risk data and progress indicators to generate the final breakdown result, confirm the task completion status, and iteratively optimize the control instructions if the target is not met.

[0064] Based on the separation status results, combined with damage risk data and separation progress indicators during the disassembly process, the final disassembly result data is generated to determine the completion status of the disassembly task. If the separation status results do not reach a complete separation state, the pulling force parameters are readjusted through the control command generation module, and separation progress data is continuously collected to obtain updated separation progress indicators.

[0065] In one possible implementation, an initial damage risk data during the battery assembly disassembly process is acquired through a damage risk assessment module. This data is presented as a comprehensive risk score, with a value range of 0 to 100. The acquired risk score is compared with a preset risk threshold, for example, a risk threshold set to 30. If the current risk assessment result is 15, which is lower than the risk threshold, it is determined that the condition for applying a pulling force is met. A control command generation module then generates corresponding control commands, using a proportional-integral-derivative (PID) control algorithm to dynamically adjust the pulling force. The current risk score and target separation speed are input, and the adjusted pulling force parameters are output, for example, smoothly increasing the initial pulling force from 50 Newtons to 65 Newtons, thereby improving disassembly efficiency while ensuring safety.

[0066] For example, a pulling force is continuously applied based on the adjusted 65 Newton pulling force parameter, while image data of the battery component separation interface is simultaneously acquired in real time via a visual sensor as separation progress data. Separation progress indicators are extracted from this data, specifically including the width of the currently separated gap and the visual features of the remaining adhesive area. For the extracted remaining adhesive area information, an edge detection algorithm is used to identify the boundaries of the thermally conductive adhesive bonding area in the image. The high-resolution image is input, and the bonding edges are located by calculating the local maxima of the image pixel grayscale gradient, outputting precise distribution data of the remaining adhesive area. The ratio of the pixel area of ​​the remaining adhesive area to the initial total adhesive area is further calculated; for example, the current remaining adhesive area percentage is calculated to be 12%.

[0067] It should be noted that the remaining adhesive area distribution data is evaluated in conjunction with preset separation status judgment rules. The preset standard for complete separation is that the remaining adhesive area percentage is less than 5%. Since the current percentage of 12% does not meet this standard, it is judged that the separation status has not been completed. Based on this separation status result, the control command generation module readjusts the pulling force parameters, for example, maintaining a pulling force of 65 Newtons and fine-tuning the pulling angle, continuously collecting new separation progress data and updating the separation progress index. If the percentage is subsequently detected to drop to 3%, it is judged as a complete separation status. Combined with the damage risk data of the entire process, the final disassembly result data is generated, confirming that the non-destructive disassembly of the battery module has been successfully completed, effectively avoiding physical breakage of the module caused by excessive pulling.

[0068] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements and additions without departing from the principle of the present invention, and these improvements and additions should also be considered within the scope of protection of the present invention.

Claims

1. A method for optimizing the disassembly of thermally conductive adhesive based on sensor data and intelligent algorithms, characterized in that, The method includes: S1, real-time acquisition of the pulling force and angle of the thermal conductive adhesive disassembly to determine the trend of nonlinear resistance change; S2, if the nonlinear resistance exceeds the resistance threshold, then activate feedback to adjust the pulling angle and speed of the robotic arm, obtain real-time resistance feedback signals, and determine the degree of damage to the thermally conductive adhesive bonding structure. S3, based on the degree of damage, iteratively optimizes the sequence of action parameters, generates progressively separated control commands, synchronously monitors the surface stress of the battery module, and quantitatively assesses the risk of damage; S4. If the risk of damage is lower than the risk threshold, apply an adjusted pulling force, monitor the separation progress and extract information on the remaining adhesive area, determine whether the separation is complete, and output the final disassembly result.

2. The method according to claim 1, characterized in that, Step S1 includes: Step S11: The pulling force and angle of the thermal conductive adhesive during the disassembly process are collected in real time by the sensor. After pre-processing, cleaning and standardization, a standardized resistance value distribution dataset is constructed. Step S12: Apply convolutional neural network to extract features and analyze the trend of nonlinear resistance change. If the feature threshold is exceeded, the classification is refined in layers and combined with the update frequency to obtain the dynamically adjusted resistance change prediction result. Step S13: If the predicted resistance change after dynamic adjustment deviates significantly from the actual result, calibrate the sensor to determine the final accurate resistance change trend.

3. The method according to claim 2, characterized in that, Step S11 includes: using a six-axis torque sensor and a high-precision tilt sensor configured at the end of the robotic arm to obtain the pulling force and angle value of the thermal conductive adhesive in real time at a sampling frequency of 50 Hz.

4. The method according to claim 3, characterized in that, Step S12 includes: constructing a convolutional neural network model containing three convolutional layers and two pooling layers for the resistance value distribution dataset.

5. The method according to claim 4, characterized in that, Using time-series data from 200 consecutive sampling points as input, key feature values ​​of local tensile force abrupt changes and angle shifts are extracted through a one-dimensional convolution kernel, and the core mode characterizing colloidal viscous fracture is output.

6. The method according to claim 1, characterized in that, Step S2 includes: Step S21: When the nonlinear resistance exceeds the resistance threshold, analyze the fluctuation characteristics to obtain preliminary abnormal results, activate the feedback mechanism, send instructions to the robotic arm to adjust the pulling angle and speed, and optimize the motion parameters. Step S22: Perform the adjusted pulling operation according to the optimized parameters, collect real-time feedback data to obtain the latest resistance signal, identify abnormal patterns through SVM classification, and determine the degree of damage to the thermally conductive adhesive bonding structure. Step S23: If the degree of damage exceeds the safe range, generate a new adjustment command to recalculate the pulling angle and speed, update the action parameters, continuously monitor the resistance signal and record the feedback data; Step S24: When the monitoring shows that the resistance signal has returned to the normal range, save the current action parameters as a reference to complete the entire adjustment process.

7. The method according to claim 6, characterized in that, Step S21 includes: extracting the first derivative of the nonlinear resistance data as an abnormal fluctuation feature.

8. The method according to claim 1, characterized in that, Step S3 includes: Step S31: Generate initial control commands based on the degree of damage, synchronously monitor the stress on the battery surface, and record the stress fluctuations corresponding to the commands. Step S32: Mark high-risk areas of stress anomalies, adjust motion parameters accordingly, and generate new control commands to reduce abnormal fluctuations; Step S33: If there is still a high risk after the new instruction is executed, the action parameters are iteratively updated until the stress stabilizes, and finally the damage risk is determined and a safety control instruction is output.

9. The method according to claim 8, characterized in that, Step S32 includes: extracting the peak value and variance characteristics of battery surface stress data within a specific time window; if the stress peak value is detected to exceed the stress threshold, the corresponding operation area is marked as a high-risk area.

10. The method according to claim 1, characterized in that, Step S4 includes: Step S41: Obtain initial damage risk data, generate control commands after assessment meets the standards, dynamically adjust the pulling force and continue to apply force, and collect separation progress indicators in real time. Step S42: Extract the information of the remaining adhesive area. If the proportion meets the standard, it is judged as a complete separation; otherwise, the force parameters are readjusted. Step S43: Combine the full-process risk data and progress indicators to generate the final breakdown result, confirm the task completion status, and iteratively optimize the control instructions if the target is not met.

Images