Method and device for disassembling decommissioned photovoltaic modules based on area detection and feedback control

Abstract

The application discloses a kind of based on area detection and feedback control's decommissioned photovoltaic module disassembly method and device, respectively calculate the adhesion resistance value and damage risk value of detection area, and two types of area state are collectively determined to start peeling position, processing area and disassembly path;Multi-sensing feedback is implemented in the process of local heating-ultrasonic auxiliary-mechanical peeling;When early warning or shutdown event occurs, execute deterministic degradation, stop, rollback and recheck;Only within independent hard constraint, the result of completed area is used to update subsequent area parameters.The application solves the problem that when the adhesion state and damage state of the component interface are not uniform in the plane, uniform heating of the whole plate, fixed speed and fixed path can easily cause local overheating, insufficient separation, new glass breakage or instability during disassembly process, so that the area detection result can directly participate in the control of peeling, zoning, path and equipment parameters, and the scope of artificial intelligence parameter optimization is limited by independent safety interlocking.

Description

Technical Field

[0001] This invention relates to the field of recycling and processing of decommissioned crystalline silicon photovoltaic modules and separation of laminated structures, and particularly to a method and apparatus for dismantling decommissioned photovoltaic modules based on regional detection and feedback control. Background Technology

[0002] Crystalline silicon photovoltaic modules typically consist of a cover glass, a first encapsulating film, solar cells and solder ribbons, a second encapsulating film, a backsheet or rear glass, an aluminum frame, and a junction box. The modules form a continuous bonding interface during the lamination process. After the modules are decommissioned, to improve the recycling value of the glass, silicon wafers, solder ribbons, and backsheet materials, the aluminum frame and junction box are usually removed first, followed by crushing, heat treatment, solvent treatment, or mechanical stripping of the laminated components.

[0003] Mechanical crushing easily forms a mixture of glass, silicon, and metal with similar particle sizes, increasing the difficulty of subsequent sorting; whole-panel heat treatment has high energy consumption and requires handling of gases generated by the encapsulation film and backsheet when heated; solvent treatment requires solvent recovery and waste liquid treatment. In order to obtain more intact glass, existing technologies have proposed heat-assisted mechanical dismantling. However, after long-term outdoor service, retired modules often exhibit differences such as edge water ingress, local delamination, yellowing or changes in the degree of cross-linking of the encapsulation film, hot spots, microcracks in the cells, glass edge notches, and embrittlement of the backsheet. The interfacial bonding strength and peel load resistance of different areas in the same module are not consistent. When using a uniform temperature, fixed peeling speed, and fixed straight path for the whole panel, areas with weaker bonding may be overheated, while areas with stronger bonding may have residual adhesive or a sudden increase in peeling force. When the peeling edge passes through existing cracks or glass notches, it may also induce crack propagation.

[0004] In summary, existing heat-assisted mechanical dismantling methods have the following drawbacks: (1) Heating the whole board or fixed segments usually uses a uniform temperature, time and speed, which makes it difficult to handle the significant differences in the same retired component caused by water ingress, yellowing of the film, local delamination, hot spots, glass gaps and microcracks.

[0005] (2) Local heating, ultrasonic treatment, peeling block and stable traction have been disclosed. If they work independently, the test results cannot be directly converted into peeling position, path and local parameters, and may still induce new damage at the crack tip or suspended support area.

[0006] (3) Relying solely on temperature or average peel force for feedback makes it difficult to distinguish between the two situations in a timely manner: "the adhesive layer has not yet softened, resulting in increased load" and "the glass or battery cell is cracking".

[0007] (4) If parameter optimization is not subject to independent safety constraints, excessively high temperatures, power or traction loads may be selected in order to reduce residual glue or shorten time; if fixed parameters are used completely, the measured results of the completed area cannot be utilized. Summary of the Invention

[0008] To address the problems mentioned in the background art, this invention provides a method and apparatus for dismantling decommissioned photovoltaic modules based on regional detection and feedback control. It transforms the differences in interfacial adhesion and damage risk in different regions of the decommissioned photovoltaic module into executable peeling locations, processing areas, dismantling paths, and local processing parameters. During the dismantling process, multi-source feedback is used to limit local overheating, sudden load increases, and crack propagation. Simultaneously, artificial intelligence parameter optimization participates in control only within verifiable physical safety boundaries, thereby forming a closed loop of detection, planning, execution, and safety interlocking. This method is suitable for low-damage separation of laminated structures in decommissioned crystalline silicon photovoltaic modules.

[0009] A first aspect of this invention provides a method for dismantling decommissioned photovoltaic modules based on area detection and feedback control, comprising the following steps: S1, Establish the coordinates of the laminated components to be disassembled and divide them into multiple detection areas; S2, apply short-term thermal and ultrasonic excitation within the same coordinate detection area, and collect thermal response, ultrasonic response, damage characteristics and support status; S3. Input the detection feature vector composed of thermal response and ultrasonic response into the adhesion calibration model with the peel force per unit width or the interface fracture energy of the same structural component sample as the output calibration value to obtain the adhesion resistance value. S4, based on acoustic anomalies, damage characteristics, and support status, yields a damage risk value; S5. Based on the difference in adhesion resistance, the difference in damage risk, and the spatial continuity of the adjacent detection areas, merge the processing areas, select the peeling position at the edge of the component that is lower than the peeling threshold and the risk threshold and meets the support conditions, and select the path with the lowest cost from the candidate paths that do not pass through the prohibited area as the disassembly path. S6, select the peeling parameters for the current processing area within independent temperature, peeling force, power and stroke constraints. The peeling parameters can be conservative peeling parameters or peeling parameters predicted by a prediction model. S7, disassemble according to the disassembly path and the peeling parameters, and collect feedback signals of temperature, peeling force, displacement and acoustic emission in real time; S8, when any feedback signal meets the first warning value, the stripping parameter value of the current area is downgraded and then returned to step S7. When any feedback signal meets the second warning value or a stripping abnormality occurs, the disassembly is stopped and the current processing area is re-inspected and then returned to step S5. The warning level of the second warning value is higher than that of the first warning value. S9: After the current processing area is completed, perform quality checks on the stripping parameters and stripping result data and record the valid parameter-result data as the dataset for updating the prediction model. If there are still unprocessed areas, return to step S5 until there are no unprocessed areas.

[0010] In one embodiment of the present invention, in step S1, the outer edge of the laminated assembly to be disassembled, the direction of the busbar, glass cracks, edge notches, backplate damage, existing delamination and support boundaries are identified, and the detection area density is increased at the glass crack tip, edge notch, existing delamination and support transition position.

[0011] In one embodiment of the present invention, step S2, collecting damage features includes: Visible light imaging was used to capture glass cracks, edge notches, and backplate damage. Electroluminescent or photoluminescent imaging was used to obtain cell cracks; Glass stress anomalies are obtained using polarization imaging; Ultrasonic response is used to obtain stratification or ultrasonic anomalies.

[0012] In one embodiment of the present invention, step S3 specifically includes: Select a module sample that is identical to the laminated module to be disassembled in terms of glass structure, encapsulation film and backsheet type, and conduct short-time thermal excitation, ultrasonic excitation and local peeling tests in the corresponding test areas of the module sample in sequence. The detection feature vector consisting of maximum temperature rise, peak time, cooling slope, ultrasonic amplitude and ultrasonic propagation time is used as input, and one of the measured unit width peel force or interface fracture energy is used as output. The regression model is trained to obtain the adhesion calibration model. The detection feature vector of the laminated component to be disassembled is input into the adhesion calibration model for prediction, and the output value is used as the adhesion resistance value. When any detection feature exceeds the allowable range of the corresponding feature in the calibration sample, or the prediction standard deviation exceeds a preset uncertainty threshold, the corresponding detection area is marked as a high uncertainty area; the adhesion resistance value of the i-th detection area... for: in, For adhesion calibration model, Let be the detection feature vector composed of the temperature response and ultrasonic response of the i-th detection region. , The maximum temperature rise in the i-th detection area of ​​the laminated component to be disassembled. The time it takes for the i-th detection area of ​​the laminated component to be disassembled to reach its temperature peak. Let be the cooling slope of the i-th detection area of ​​the laminated component to be disassembled. Let be the ultrasonic amplitude of the i-th detection area of ​​the laminated component to be disassembled. The ultrasonic propagation time is the i-th detection area of ​​the laminated component to be disassembled.

[0013] In one embodiment of the present invention, in step S4, at least two of the following characteristics of the laminated assembly to be disassembled are normalized into risk features: glass cracks or edge notches, crack tip distance, cell cracks, backsheet damage, existing delamination, ultrasonic anomalies, glass stress anomalies, solder strip or busbar location, support boundary, and local suspension state. Calculate the damage risk value of the i-th detection region. : in, The cutoff function is used to truncate the weighted sum to the interval [0,1], where K≥2 represents the number of risk features selected. The weights are those verified by component samples or set by preset risk rules. Let k be the normalized risk feature of the i-th detection region.

[0014] In one embodiment of the present invention, step S5 specifically includes: Adjacent detection areas sharing a common edge are considered as adjacent areas. When the difference in adhesion resistance values ​​between adjacent areas is not greater than the adhesion difference threshold, the difference in damage risk values ​​is not greater than the risk difference threshold, and the two detection areas are not prohibited from being separated by the boundary, the adjacent areas are merged into the same processing area. For areas with through-crack ends, glass chipping, or unsupported regions, prohibition of peeling or prohibition of passage is set. Multiple detection areas are constructed into a graph with adjacent relationships as edges. Multiple candidate paths without prohibited passage areas are generated from the peeling initiation position to the current target boundary. The path cost of each candidate path is calculated, and the candidate path with the minimum path cost is determined as the disassembly path. The path cost of a candidate path is determined by a weighted sum of the normalized adhesion resistance value, damage risk value, travel distance, repeated heating cost, and number of turning times along the path. The path cost of candidate path π is... for: in, Let be the normalized adhesion resistance value of the i-th detection region. Let be the normalized moving distance of the i-th detection region. Let i be the normalized damage risk value for the i-th detection region. To the cost of repeated heating, For the number of turns, , , , , The weights are non-negative and their sum is 1.

[0015] In one embodiment of the present invention, in step S6, the peeling parameters of the processing area include one or more of the following: target temperature, heating time, heating power, ultrasonic power, ultrasonic frequency, peeling speed, and peeling angle. The peeling parameters of the processing area are predicted using a prediction model. When the effective parameter-result data volume in the first processing area, high uncertainty area, or dataset does not reach the number of enabled samples of the prediction model, conservative peeling parameters verified by the same structure component sample are used. The dataset used to update the prediction model consists of qualified and valid parameter-result data from the completed treatment areas. The parameter-result data includes area coordinates, adhesion resistance values, damage risk values, peeling parameters, effective separation area ratio, film residue area ratio, energy consumption per unit area, peak peel force, peak temperature, new damage, and interlocking events. The input to the j-th set of data in the prediction model and output Represented as: in, Let J be the adhesion resistance value of the j-th data set. Let j be the damage risk value of the j-th data group. The stripping parameter for the j-th data group. The effective separation area ratio of the j-th data set. The percentage of residual film area in the j-th data set. Let j be the normalized value of energy consumption per unit area for the j-th data set. The peak peeling force of the j-th data set. Let be the peak temperature of the j-th data set. This represents the newly added damage normalized value for the j-th data group; When the effective parameter-result data volume in the dataset of the prediction model reaches the number of samples to be used in the prediction model, the constrained acquisition function of the candidate stripping parameters is calculated. The constrained acquisition function is composed of the product of the improved expectation of the minimized comprehensive objective function, the predicted probability of temperature and stripping force satisfying the constraints, and the maximum value of the constrained acquisition function within the candidate set that satisfies the lower probability limit and the equipment boundary. The candidate stripping parameter must simultaneously satisfy the following conditions: the probability that the predicted peak temperature does not exceed the upper limit of the allowable temperature is not less than 1-α, the probability that the predicted peak stripping force does not exceed the upper limit of the allowable stripping force is not less than 1-α, and the deterministic boundary of the power and movement stroke of the disassembly equipment. The maximum value of the constrained acquisition function is solved in the candidate stripping parameter set that satisfies the constraints, and the stripping parameter corresponding to the maximum value is taken as the stripping parameter of the current processing area. Here, α is the constraint tail probability, 0 < α < 0.5. The comprehensive objective function for the candidate stripping parameters is: in, Let be the comprehensive objective function to be minimized. To strip the parameter vector, , , , , The weights are non-negative and their sum is 1. To effectively separate the area ratio, This represents the percentage of residual adhesive film area. This is the normalized value of energy consumption per unit area. This is the normalized value of the peak peel force. This is the newly added damage normalization value.

[0016] In one embodiment of the present invention, in step S8, the first warning value and the second warning value are set for the feedback signals of temperature, peeling force and acoustic emission, respectively; the peeling abnormality includes the peeling leading edge displacement being less than the preset minimum increment and the peeling force continuously increasing, or the peeling force suddenly dropping and accompanied by continuous high-energy acoustic emission events; when the first warning value is met, at least one of reducing the heating power, reducing the ultrasonic power or reducing the peeling speed is executed; when the second warning value is met or a peeling abnormality occurs, the energy input and peeling movement are stopped, and the peeling mechanism is retracted by a preset distance and the current area is re-detected.

[0017] In one embodiment of the present invention, in step S9, the quality check includes at least the correspondence of regional coordinates, the absence of failure markers in sensor signals, the complete recording of actual stripping parameters, and the ability to verify the results. Only the valid parameter-result data that pass the quality check are added to the training dataset of the prediction model. When returning to step S5, the subsequent processing region and disassembly path are regenerated based on the updated processed region markers and the re-inspection results. Then, in step S6, the training dataset of the prediction model is updated, and the stripping parameters of the subsequent processing region are determined.

[0018] A second aspect of the present invention provides a decommissioned photovoltaic module dismantling device based on area detection and feedback control, used to execute the decommissioned photovoltaic module dismantling method based on area detection and feedback control described in the above embodiments, comprising: The support platform includes a positioning component, a flexible support component, and a partitioned negative pressure adsorption unit. The positioning component is used to determine the reference edge of the component, the flexible support component is used to support the laminated component to be disassembled to compensate for component warping, and the partitioned negative pressure adsorption unit is used to adsorb the laminated component to be disassembled to restrict component movement. The region detection component includes a short-time thermal exciter, an infrared camera, an ultrasonic detection device, and an image detection unit. The short-time thermal exciter applies short-time thermal excitation to each detection region and then uses the infrared camera to collect the thermal response of each detection region. The ultrasonic detection device applies pulse or frequency sweep excitation to the detection region and collects the ultrasonic response. The image detection unit collects one or more of the following: visible light image, electroluminescent image, photoluminescent image, or polarization image. The controller includes a region state calculation module, a path planning module, a parameter selection module, and a feedback control module. The region state calculation module outputs the adhesion resistance value and damage risk value of each detection region; the path planning module determines the processing area, peeling position, and dismantling path; the parameter selection module calls a conservative parameter library or a prediction model subject to hard constraints; and the feedback control module outputs power reduction, speed reduction, stop, or retraction commands to the local processing components and peeling mechanism based on the feedback signal. The local processing component is located at or in front of the peeling front edge and can move with the peeling front edge. The local processing component includes a local heater and an ultrasonic transducer. The local heater acts on the strip-shaped area at or in front of the peeling front edge, and the ultrasonic transducer applies vibration to the strip-shaped area or the peeled layer. The peeling mechanism includes a wedge-shaped peeling component, a clamping and traction component, and a drive mechanism, which are used to disassemble according to the peeling speed and peeling angle in the peeling parameters of the disassembly path; The feedback detection component includes a temperature detection unit, a force sensor, a displacement sensor, and an acoustic emission sensor. The force sensor is set on the load transmission path of the clamping traction component or the drive mechanism to collect feedback signals during the disassembly process. The current collection shunt assembly collects glass, backsheet, and encapsulation layer-cell composite material respectively.

[0019] The method and apparatus for dismantling decommissioned photovoltaic modules based on regional detection and feedback control according to embodiments of the present invention have the following beneficial effects: 1. The adhesion resistance value obtained by peel test calibration and the damage risk value obtained by crack, notch and support conditions are solved separately according to different evidence and different calculation relationships, respectively characterizing the expected disassembly resistance and the new damage risk, avoiding the use of a single comprehensive score with ambiguous meaning to simultaneously characterize the degree of peeling difficulty and the damage risk.

[0020] 2. The peeling location, processing area, and dismantling path are jointly determined by the two types of area states, which can bypass prohibited areas and adjust the temperature, speed, or allowable load in the restricted processing area that is manually confirmed, so that the test results and mechanical peeling actions form a clear technical link.

[0021] 3. The training data, input, output, and update time of the prediction model are all limited, and parameter optimization can only be carried out within independent temperature, peeling force, power, and stroke constraints. Peeling parameter prediction is only performed after the effective dataset meets the model activation conditions, and is subject to independent hard constraints and deterministic interlocking. When the effective dataset is insufficient, the prediction model is unreliable, or the sensor fails, conservative peeling parameters are restored.

[0022] 4. The first and second warning values ​​correspond to downgrade and shutdown respectively. The displacement-peeling force relationship and acoustic emission events are used to identify peeling stagnation or brittle cracking signs, so that warning, stop, rollback and re-inspection have clear triggering conditions, and can limit the current processing when abnormal load or cracking signs occur.

[0023] 5. Regional detection features, actual stripping parameters, stripping results, and interlocking events are bound to regional coordinates, facilitating traceable parameter updates for subsequent processing areas of the current component and components with the same structure.

[0024] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0025] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 A schematic diagram of a method for dismantling decommissioned photovoltaic modules based on regional detection and feedback control provided by the present invention; Figure 2 A schematic diagram of the adhesion resistance distribution (left) and damage risk distribution (right) provided for this invention; Figure 3 A schematic diagram illustrating the merging of processing areas, selection of peeling starting position, and obstacle bypass dismantling path provided by the present invention; Figure 4 A schematic diagram of the structure of the decommissioned photovoltaic module dismantling device based on regional detection and feedback control provided by the present invention.

[0026] Reference numerals: Support platform 100, positioning component 110, flexible support component 120, zoned negative pressure adsorption unit 130; area detection component 200, short-time thermal exciter 210, infrared camera 220, ultrasonic detection device 230, image detection unit 240; controller 300, area status calculation module 310, path planning module 320, parameter selection module 330, feedback control module 340; local processing component 400, local heater 410, ultrasonic transducer 420; peeling mechanism 500, wedge-shaped peeling component 510, clamping traction component 520, drive mechanism 530; feedback detection component 600, temperature detection unit 610, force sensor 620, displacement sensor 630, acoustic emission sensor 640; collection and diversion component 700. Detailed Implementation

[0027] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0028] Figure 1 This invention provides a schematic flowchart of a method for dismantling decommissioned photovoltaic modules based on regional detection and feedback control.

[0029] like Figure 1 As shown, the method for dismantling decommissioned photovoltaic modules based on regional detection and feedback control includes the following steps: S1, Establish the coordinates of the laminated components to be disassembled and divide them into multiple detection areas.

[0030] In an embodiment of the present invention, step S1 involves component positioning and detection area division. The aluminum frame and junction box are removed or avoided, and the laminated component to be disassembled is fixed to the support surface. The outer edge of the laminated component to be disassembled, the direction of the busbar, glass cracks, edge notches, backplate damage, existing delamination, and support boundaries are identified. A two-dimensional coordinate system consistent with the motion coordinates of the peeling mechanism is established, and the detection area density is increased at the glass crack tip, edge notches, existing delamination, and support transition positions.

[0031] S2, apply short-term thermal and ultrasonic excitation within the same coordinate detection area, and collect thermal response, ultrasonic response, damage characteristics and support status.

[0032] In an embodiment of the present invention, step S2, acquiring damage features, includes: Visible light imaging was used to capture glass cracks, edge notches, and backplate damage. Electroluminescent or photoluminescent imaging was used to obtain cell cracks; Glass stress anomalies are obtained using polarization imaging; Ultrasonic response is used to obtain stratification or ultrasonic anomalies.

[0033] Specifically, step S2 performs multi-physics detection in the same coordinate system. A short-duration thermal excitation, insufficient to induce the encapsulating film into a formal softening and peeling state, is applied to each detection area, and temperature sequences during the heating and cooling phases are acquired. Pulsed or swept-frequency ultrasonic excitation is applied to the same coordinate region, and ultrasonic amplitude, phase, propagation time, attenuation, or local resonance response are acquired. Visible light images are used to obtain glass cracks, edge notches, and backplane damage; when the component circuitry is usable, electroluminescent or photoluminescent images are used to obtain cell cracks; polarization images are used to obtain glass stress anomalies. The above image information can be mapped to the detection area according to two-dimensional coordinates.

[0034] During the testing phase, the maximum temperature rise of the short-term thermal excitation is controlled below the softening initiation temperature of the encapsulating film to prevent premature softening and localized pre-peeling at the interface, thus ensuring that the thermal response signal accurately reflects the interface adhesion characteristics. During the peeling operation, the heating temperature is higher than the film's softening initiation temperature. Specifically, for commonly used ethylene-vinyl acetate copolymer (EVA) encapsulating films, the softening temperature is approximately 70–90℃, and the thermal decomposition initiation temperature is approximately 200–250℃; for polyolefin elastomer (POE) encapsulating films, the softening temperature is approximately 80–100℃. The heat flux density of the short-term thermal excitation during the testing phase is taken as 0.5–3.0 kW / m³. 2 The action time is 1 to 15 seconds, and the maximum surface temperature rise is controlled between 5 and 30°C. The reason for this selection is that this parameter range is sufficient to generate a temperature gradient and cooling curve difference on the component surface that can be distinguished by an infrared camera. However, the action time is short and the total heat input is low. The heat has not been fully conducted to the film-glass interface to make the film reach the softening temperature, so it will not cause a change in the interface bonding state.

[0035] S3 inputs the detection feature vector composed of thermal response and ultrasonic response into the adhesion calibration model with the peel force per unit width or the interface fracture energy of the same structural component sample as the output calibration value to obtain the adhesion resistance value.

[0036] In an embodiment of the present invention, step S3 specifically includes: Select a module sample that is the same as or similar to the laminated module to be disassembled in terms of glass structure, encapsulation film and backsheet type, and conduct short-time thermal excitation, ultrasonic excitation and local peeling tests in the corresponding test area of ​​the module sample in sequence. The detection feature vector consisting of maximum temperature rise, peak time, cooling slope, ultrasonic amplitude and ultrasonic propagation time is used as input, and one of the measured unit width peel force or interface fracture energy is used as output. The regression model is trained to obtain the adhesion calibration model. The detection feature vector of the laminated component to be disassembled is input into the adhesion calibration model for prediction, and the output value is used as the adhesion resistance value. When any detection feature exceeds the allowable range of the corresponding feature in the calibration sample, or the prediction standard deviation exceeds a preset uncertainty threshold, the corresponding detection area is marked as a high uncertainty area; the adhesion resistance value of the i-th detection area... for: in, For adhesion calibration model, Let be the detection feature vector composed of the temperature response and ultrasonic response of the i-th detection region. , The maximum temperature rise in the i-th detection area of ​​the laminated component to be disassembled. The time required to reach the peak temperature For cooling slope, This refers to the ultrasonic amplitude. This represents the ultrasound propagation time.

[0037] The adhesion calibration model takes the detection feature vector of a sample of the same structural component as input and outputs either the peel force per unit width or the interfacial fracture energy measured in the same area. The same model does not mix N / mm and J / m. 2The predicted value uses the unit of the selected physical quantity, or it can be normalized to 0 to 1 within the same model. If the model is normalized, the model output needs to be reverse normalized to restore the original physical quantity of adhesion resistance to obtain the adhesion resistance value. When the training sample is insufficient, a graded threshold table is used and the data is gradually corrected with new data. The specific operation method is as follows: (1) Threshold table construction: In the component sample calibration stage, even if the sample size is insufficient to train the regression model, several sets of data pairs (detection feature vector, measured unit width peeling force) can still be obtained through limited local peeling tests; the dimensions of the detection feature vector (maximum temperature rise, peak time, cooling slope, ultrasonic amplitude, ultrasonic propagation time) are divided into 3 to 5 levels according to the measured value range (e.g., maximum temperature rise <10℃ is low, 10 to 20℃ is medium, >20℃ is high), and each level combination corresponds to an adhesion resistance value range determined by test or experience, forming a graded lookup table; (2) Threshold table use: Extract the detection area of ​​the laminated component to be disassembled. After measuring the feature vector, find the corresponding adhesion resistance value range in the table according to the level of each dimension, and take the midpoint of the range as the estimated value of adhesion resistance in the region; if the feature combination exceeds the coverage of the table, mark it as a high uncertainty region and adopt conservative peeling parameters; (3) Stepwise correction: after each processing region is completed and passes the quality check, add the (detection feature, measured peeling force) data pair of the region to the sample library. When the measured data of a certain level combination reaches the preset number (such as no less than 5 groups), update the adhesion resistance value range corresponding to the level with the measured mean; when the total sample size reaches the number of samples for model activation, switch to regression model prediction, stop updating the grading threshold table and use it as a backup plan when the model is unavailable.

[0038] When establishing the adhesion calibration model, the component sample data is first divided into training data and validation data. Each detection feature is standardized according to the statistics of the training data, and then one of the following models—linear regression, ridge regression, Gaussian process regression, or tree regression—is trained. The preferred model is the Gaussian process regression model, which can output the predicted mean and predicted standard deviation. If any standardized feature exceeds the allowable range for the corresponding feature in the calibration sample, or if the predicted standard deviation exceeds a preset uncertainty threshold, the input is considered to be outside the calibration's applicable range. This region is marked as a high-uncertainty region, and conservative stripping parameters are used; the model's output for increasing temperature or load is not utilized. If a regression model that does not support outputting the predicted standard deviation is selected, the probability constraint mode is no longer enabled, and the conservative stripping parameters are directly switched to. Probability constraints only take effect under models that can output predicted uncertainty.

[0039] The allowable range and preset uncertainty threshold of the corresponding features of the calibration sample are obtained by using the training data statistics (mean, standard deviation and allowable extrapolation of each feature) and the prediction error distribution of the validation set through the component sample calibration test.

[0040] S4 is the damage risk value obtained based on acoustic anomalies, damage characteristics, and support conditions.

[0041] In step S4, at least two of the following characteristics of the laminated module to be disassembled are normalized as risk features: glass cracks or edge notches, crack tip distance, cell cracks, backsheet damage, existing delamination, ultrasonic anomalies, glass stress anomalies, solder strip or busbar location, support boundary, and local suspension state. Calculate the damage risk value of the i-th detection region. : in, The cutoff function is used to truncate the weighted sum to the interval [0,1], where K≥2 represents the number of risk features selected. The weights are those verified by component samples or set by preset risk rules. Let k be the normalized risk feature of the i-th detection region. Normalization can be achieved using minmax normalization, mapping each risk feature to the 0-1 range. Backplate damage is detected by visible light imaging and mapped to the corresponding detection region, normalized to a backplate damage area ratio feature. Its weight is set according to the impact of backplate damage on the integrity of the back support during the peeling process. Existing delamination is confirmed by ultrasonic C-scanning or time-domain reflectometry and directly mapped to a delamination area ratio feature of the detection region. Delamination areas typically have lower adhesion resistance but higher damage risk (delamination edges are prone to crack propagation or localized stress concentration during peeling). Therefore, the two types of region states are calculated separately and do not substitute for each other.

[0042] In embodiments of the invention, areas with through-crack ends, glass chipping, or areas without reliable support are designated as prohibited peeling zones or prohibited passage zones. The adhesion resistance value represents the expected dismantling resistance, and the damage risk value represents the risk of additional breakage under current support and defect conditions; both are calculated separately, and the same score is not simultaneously interpreted as dismantling difficulty and breakage risk. A validated classification model can also be used to output the damage risk value, but a single score identical to the adhesion resistance value cannot be simultaneously interpreted as dismantling difficulty and breakage risk.

[0043] S5. Based on the difference in adhesion resistance, the difference in damage risk, and the spatial continuity of the adjacent detection areas, the processing area is merged. At the edge of the component, a peeling position that is lower than the peeling threshold and the risk threshold and meets the support conditions is selected. The path with the lowest cost is selected from the candidate paths that do not pass through the prohibited area as the disassembly path.

[0044] In an embodiment of the present invention, step S5 specifically includes: Adjacent detection regions sharing a boundary are considered adjacent regions. When the difference in adhesion resistance values ​​between adjacent regions is no greater than the adhesion difference threshold, the difference in damage risk values ​​is no greater than the risk difference threshold, and the two detection regions are not prohibited from being separated by a boundary, the adjacent regions are merged into the same processing region. Specifically, for adjacent detection regions i and j sharing a boundary, the following formula is used to determine whether they can be merged: in, The adhesion difference threshold is determined based on the measured distribution of adhesion resistance in adjacent areas during the peeling test of the module sample and the process requirements for the uniformity of the treated area. The risk difference threshold is obtained based on the measured distribution of adhesion resistance in adjacent areas during the component sample peeling test and the process requirements for the uniformity of the treated area. The adhesion difference threshold can be taken as 75% of the difference in adhesion resistance between adjacent areas of the sample, and the risk difference threshold can be taken as 75% of the difference in damage risk between adjacent areas. Alternatively, an upper limit can be set manually according to the process requirements for the uniformity of parameters in the treated area. The threshold should be determined during the sample test stage and written into the controller configuration.

[0045] Adjacent regions satisfying the above formula are incorporated into the same processing region through region growth or connected component marking. The peeling start location is selected from the component edge or a region at an existing local layer, and simultaneously satisfies... Below the peeling threshold The peeling threshold and risk threshold are below the risk threshold and the support is reliable. The peeling success rate and new breakage rate of the component samples under different adhesion resistance and damage risk levels are obtained through peeling success rate and new breakage rate tests; the peeling threshold is the upper limit of adhesion resistance corresponding to a peeling success rate of not less than 90%, and the risk threshold is the upper limit of damage risk corresponding to a new breakage rate of not more than 5%. The peeling location must also meet the condition of reliable support (normal negative pressure adsorption, no local suspension).

[0046] For areas with through-crack ends, glass breakage, or unsupported areas, prohibition of peeling or prohibition of passage is set. Multiple detection areas are constructed into a graph with adjacent relationships as edges. Multiple candidate paths without prohibited passage areas are generated from the peeling start position to the current target boundary. The path cost of each candidate path is calculated, and the candidate path with the minimum path cost is determined as the disassembly path. The path cost of the candidate path is determined by the weighted sum of the normalized adhesion resistance value, damage risk value, movement distance, repeated heating cost, and number of turning times on the path. Specifically, the detection region is constructed as an adjacency graph, and prohibited regions are not added to the set of passable nodes. Dijkstra's algorithm can be used. Algorithms such as the k-shortest path algorithm generate one or more candidate paths π, and the path cost of candidate path π is... for: in, Let be the normalized adhesion resistance value of the i-th detection region. Let be the normalized moving distance of the i-th detection region. This represents the normalized damage risk value for the i-th detection region. Normalization can be achieved by using minmax normalization to the 0-1 range. To the cost of repeated heating, The number of turns; the summation range is determined by i∈π, and the total number of detection regions does not require a separate variable t. , , , , The weights are non-negative and sum to 1; the cost of repeated heating. The cumulative energy consumption and material thermal damage risk resulting from reheating previously heated areas along the path π can be obtained by weighted summation based on the number of previously heated detection areas along the path, and the number of turning points. A change in the direction of travel along the path that exceeds a set angle threshold is recorded as a turn.

[0047] choose The smallest candidate path is used as the current disassembly path. When there is no connecting path that completely bypasses the high-risk area, the prohibited area is not changed to a passable area. Instead, automatic path generation is stopped and manual confirmation is requested. Areas that are manually confirmed to be allowed to be processed but not allowed to be used as the starting point for peeling are marked as restricted processing areas. The peeling speed limit, temperature limit, or maximum allowable peeling force is reduced separately for these restricted processing areas. Therefore, the path can pass through the restricted processing area, but this area cannot be selected as the starting point for peeling.

[0048] S6 allows you to select the peeling parameters for the current processing area within independent constraints of temperature, peeling force, power, and stroke. The peeling parameters can be conservative or predicted by a predictive model.

[0049] In an embodiment of the present invention, in step S6, the peeling parameters of the processing area include one or more of the following: target temperature, heating time, heating power, ultrasonic power, ultrasonic frequency, peeling speed, and peeling angle. The peeling parameters of the processing area are predicted using a prediction model. When the effective parameter-result data volume in the first processing area, high uncertainty area, or dataset does not reach the number of samples required for the prediction model to be enabled, conservative peeling parameters verified by samples of components with the same structure are used. The number of samples needs to be greater than the dimension of the peeling parameters to be optimized to avoid underfitting of the model. The dataset used to update the prediction model consists of qualified and valid parameter-result data from the completed treatment areas. The parameter-result data includes area coordinates, adhesion resistance values, damage risk values, peeling parameters, effective separation area ratio, film residue area ratio, energy consumption per unit area, peak peel force, peak temperature, new damage, and interlocking events. The dataset D for the prediction model consists of qualified data from already processed regions, and the j-th set of data for the prediction model is input. and output Represented as: in, Let J be the adhesion resistance value of the j-th data set. Let j be the damage risk value of the j-th data group. The stripping parameter for the j-th data group. The effective separation area ratio of the j-th data set. The percentage of residual film area in the j-th data set. Let j be the normalized value of energy consumption per unit area for the j-th data set. The peak peeling force of the j-th data set. Let be the peak temperature of the j-th data set. is the new damage normalization value for the j-th group of data; Gaussian process regression models can be trained separately for each output, or a validated multi-output Gaussian process regression model can be trained; the length scale, magnitude and noise parameters of the model kernel function are determined by maximizing the marginal likelihood, and the prediction error and prediction interval coverage are checked using hold-out data or cross-validation.

[0050] When the effective parameter-result data volume in the prediction model's dataset reaches the number of samples required for the prediction model to be enabled, a constrained acquisition function for candidate stripping parameters is calculated based on the mean and standard deviation of the predicted values ​​output by the prediction model. This constrained acquisition function is composed of the product of the improved expectation of the minimized comprehensive objective function, the predicted probability that temperature and stripping force satisfy the constraints, and the maximum value of the constrained acquisition function within the candidate set that satisfies the lower probability limit and the equipment boundary. This maximum value is taken as the stripping parameter for the current processing area. Candidate stripping parameters must simultaneously satisfy the following conditions: the probability that the predicted peak temperature does not exceed the upper limit of the allowable temperature is not less than 1-α; the probability that the predicted peak stripping force does not exceed the upper limit of the allowable stripping force is not less than 1-α; and the power of the dismantling equipment... For the deterministic boundary of the motion stroke, the maximum value of the constrained acquisition function is solved from the set of candidate peel parameters that meet the constraints. The peel parameter corresponding to the maximum value is taken as the peel parameter of the current processing area. When the set of candidate peel parameters that meet the constraints is empty, a conservative peel parameter is selected, where 0 < α < 0.5. The upper limit of allowable temperature, the upper limit of allowable peel force, the equipment power boundary, and the motion stroke boundary are not modified by the process proxy model. The upper limit of allowable temperature and the upper limit of allowable peel force are set by thermal stability test of the encapsulation film and backing material and glass flexural strength test. α is the constraint tail probability, which is set according to the acceptable safety constraint violation probability (usually 0.05 or 0.1). Upper limit of allowable temperature: The thermal decomposition initiation temperature of EVA film is approximately 200-250℃. Considering yellowing and crosslinking after long-term service, 120-150℃ is recommended. For fluorinated backsheets (TPT / TPE), a lower temperature (100-120℃) is recommended. Upper limit of allowable peel force: Based on the flexural strength of 3.2mm cover glass (approximately 40-60 MPa) and the peel contact width, 50-150 N / cm is recommended. The probability of constraint tail is set at 0.05 corresponding to a 95% confidence level, and 0.1 corresponding to a 90% confidence level. The above values ​​are for illustrative purposes only and should be determined through sample testing.

[0051] Specifically, the comprehensive objective function for the candidate stripping parameters is: in, Let be the comprehensive objective function to be minimized. To strip the parameter vector, , , , , The weights are non-negative and their sum is 1. To effectively separate the area ratio, This represents the percentage of residual adhesive film area. This is the normalized value of energy consumption per unit area. This is the normalized value of the peak peel force. This is the newly added damage normalization value.

[0052] Safety is not considered a soft cost that can be offset by separation effectiveness and energy consumption; candidate stripping parameters must also satisfy the safety formula: in, Let be the conditional probability symbol, and D be the training dataset for the prediction model. and These are independently configured upper limits for allowable temperature and upper limits for allowable peel force. and These represent the upper limits of the allowable heating power and ultrasonic power of the equipment, respectively. The applicable ranges for speed, angle, stroke, and materials are also subject to deterministic boundary limitations imposed by the equipment configuration. When using a constrained improved desired acquisition function, the stripping parameters for the current processing area can be selected using the following formula: in, This represents the optimal stripping parameter vector corresponding to the maximum value of the constrained acquisition function. To satisfy the deterministic boundaries of power, speed, angle, and stroke, a set of candidate stripping parameters is required. For the comprehensive objective function to be minimized The improvement expectation is a commonly used acquisition function in Bayesian optimization, used to balance the benefits of the objective function with the uncertainty of model prediction. If the uncertainty of the prediction model exceeds the threshold, the predicted value and the measured value continuously exceed the allowable deviation, or the relevant sensor of any feedback signal fails, the prediction model recommendation is deactivated and the conservative stripping parameters are restored. The conservative stripping parameters are the parameters obtained through stripping experiments. The prediction model is only updated between adjacent processing regions and does not replace the real-time safety interlock in step S8. Continuously exceeding the allowable deviation means that the deviation between the model prediction result and the measured result in two or more consecutive processing regions exceeds the allowable deviation threshold.

[0053] Let be the comprehensive objective function to be minimized, representing the weighted cost of inadequate separation, residual adhesive, energy consumption, load, and new damage; it differs from the acquisition function used to select the next set of parameters. Let Given the minimum objective value of the evaluated feasible candidate parameters under the current regional conditions, the expected improvement is EI(x) = E[max(J best -J(x),0)|D]. Here, D is the training dataset of the prediction model, and the conditional expectation depends simultaneously on the objective function's prediction mean and prediction uncertainty. The constrained acquisition function is A(x)=EI(x)·Pr[T]. p (x)≤T lim D]·Pr[F p (x)≤F lim[D], the product of the two probability terms adopts the modeling assumption that temperature and peel force prediction conditions are independent. Only the parameter with the largest A(x) is selected from the candidate parameters that respectively satisfy the above probability lower bound and equipment deterministic boundary. In independent Gaussian process prediction, the temperature constraint probability is Φ((T) lim -μ T (x)) / σ T (x)), the peeling force constraint probability is similar; Φ is the standard normal distribution function, μ T (x) and σ T (x) represents the corresponding predicted mean and standard deviation, respectively. When σ is zero, it is treated as a deterministic comparison.

[0054] S7 disassembles according to the disassembly path and peeling parameters, and collects feedback signals of temperature, peeling force, displacement and acoustic emission in real time.

[0055] During disassembly, local heating is applied to the strip-shaped area in front of the peeling edge, ultrasonic vibration is applied to the strip-shaped area or the peeled layer, and mechanical peeling is carried out along the disassembly path determined in step S5 and according to the peeling speed and peeling angle in the peeling parameters determined in step S6. At the same time, multiple sensors are used to continuously collect temperature, peeling force, displacement and acoustic emission signals as feedback signals.

[0056] Specifically, a local heater and an ultrasonic transducer are positioned at or in front of the peeling edge and move with it. A temperature detection unit, force sensor, displacement sensor, and acoustic emission sensor continuously sample the data.

[0057] S8, when any feedback signal meets the first warning value, the stripping parameter value of the current area is downgraded and the process returns to step S7. When any feedback signal meets the second warning value or a stripping abnormality occurs, the disassembly is stopped and the current processing area is re-inspected and the process returns to step S5. The warning level of the second warning value is higher than that of the first warning value.

[0058] In an embodiment of the present invention, in step S8, the first warning value and the second warning value are set for feedback signals of temperature, peeling force, and acoustic emission, respectively; peeling anomalies include peeling leading edge displacement being less than a preset minimum increment while peeling force continues to increase, or peeling force suddenly dropping and accompanied by continuous high-energy acoustic emission events; when the first warning value is met, at least one of reducing heating power, reducing ultrasonic power, or reducing peeling speed is executed; when the second warning value is met or a peeling anomaly occurs, energy input and peeling movement are stopped, causing the peeling mechanism to retract a preset distance and re-detect the current area. The peeling mechanism retracts a preset distance, causing the wedge-shaped peeling member to exit the formed peeling gap, and then the current processing area is re-detected.

[0059] The retraction distance is determined based on the wedge-shaped peeler's insertion length, retraction gap, and available stroke of the drive mechanism. A distance sufficient for safe retraction is selected within the existing 1–30 mm range. If this range is insufficient for retraction or the retraction path is unsafe, the machine is stopped and manual intervention is initiated. After retraction, S2 to S4 are re-executed for the current area and adjacent affected areas to review cracks, delamination, support status, adhesion resistance, and damage risk. Then, S5 is re-planned. If the automatic processing conditions are not met during the re-inspection, the operator selects to terminate the area based on the re-inspection results, marks it as a restricted processing area after meeting support and safety conditions, or re-executes S5 and S6 after adjusting parameters within the existing safety boundaries. Manual confirmation cannot cancel the independent safety upper limit.

[0060] In embodiments of the present invention, a first warning value is used for warning downgrading, and a second warning value is used for shutdown protection. The first and second warning values ​​are set according to the material, equipment capacity, and sample testing, respectively. When the first warning value is met, at least one of the heating power, ultrasonic power, or peeling speed is reduced, and then the process returns to S7 to continue processing. When the second warning value is met, the peeling leading edge displacement is less than the preset minimum increment while the peeling force continues to increase, or the peeling force suddenly drops and is accompanied by continuous high-energy acoustic emission events, the energy input and peeling movement are stopped, the peeling mechanism is retracted, and the current processing area is re-inspected. After re-inspection, the process returns to S5. If the automatic processing conditions are still not met after re-inspection, manual confirmation or termination of processing in that area is initiated.

[0061] High-energy acoustic emission events refer to acoustic emission signal amplitudes exceeding a preset event threshold. Continuous high-energy acoustic emission events refer to the occurrence of more than a set number of high-energy acoustic emission events within a unit time window. This threshold is obtained by calibrating the component sample. The preset minimum increment is the minimum displacement that the peeling leading edge should reach within a set time window, used to determine whether peeling has stalled.

[0062] Furthermore, the displacement signal acquired in real time in step S7 is not used as an independent warning threshold in step S8, but is incorporated into the interlocking logic of step S8 as a peeling anomaly criterion. Specifically, peeling anomalies in step S8 include peeling leading edge displacement being less than a preset minimum increment while peeling force continues to increase. Here, the displacement signal and peeling force signal are jointly used to determine the peeling stagnation state: when the peeling leading edge displacement is less than the preset minimum increment (displacement signal criterion) within a set time window, and the peeling force continues to increase (force signal criterion), it is determined to be a peeling stagnation anomaly, triggering a second warning level shutdown and re-inspection. The reason why the displacement signal does not have an independent warning threshold is that the absolute value of the displacement itself is affected by component size, path planning, and mechanism stroke, and does not have independent safety physical meaning; however, its combined relationship with the peeling force, such as displacement stagnation and increased bonding force equaling interface not separated, and a sudden increase in displacement and a sudden decrease in bonding force equaling brittle cracking of the interface, can effectively distinguish between the two failure modes. A sudden decrease in peeling force accompanied by continuous high-energy acoustic emission events in step S8 is another peeling anomaly criterion, corresponding to the brittle cracking mode.

[0063] S9: After the current processing area is completed, perform quality checks on the stripping parameters and stripping result data and record the valid parameter-result data as the dataset for updating the prediction model. If there are still unprocessed areas, return to step S5 until there are no unprocessed areas.

[0064] After the current processing area is successfully completed, the area coordinates, adhesion resistance value, damage risk value, actual peeling parameters, effective separation area ratio, residual film area ratio, energy consumption per unit area, peak peeling force, peak temperature, new damage, and interlocking events are bound as parameter-result data. Interlocking events are those that meet the first and second warning values ​​in step S8. Quality checks include at least corresponding area coordinates, no failure markers in sensor signals, complete recording of actual peeling parameters, and verifiable results. Only valid parameter-result data that passes the quality check are added to the training dataset of the prediction model. When returning to step S5, the subsequent processing area and disassembly path are regenerated based on the updated processed area markings and re-inspection results. Subsequently, in step S6, the dataset D of the prediction model is updated, and the peeling parameters of the subsequent processing area are determined.

[0065] After updating the dataset for the prediction model, if there are still unprocessed areas, return to step S5 to regenerate the subsequent processing areas and paths based on the processed area markings and re-inspection results. Then, in step S6, read the updated dataset to determine the stripping parameters for the next processing area. The dataset is not used to recalculate the adhesion resistance value in step S3 or the damage risk value in step S4, thus avoiding confusing path replanning and prediction model training for the same purpose.

[0066] The method of the present invention will be described below through a specific embodiment.

[0067] (1) Sample calibration and model boundary determination Component samples with glass thickness and structure, encapsulation film type, and backsheet type that are the same as or similar to the laminated component to be disassembled are selected. The component samples are divided into test grids of the same size as the laminated component to be disassembled. Short-duration thermal and ultrasonic excitations are applied sequentially to each grid, and the maximum temperature rise, peak time, cooling slope, ultrasonic amplitude, and propagation time are extracted. Subsequently, independent local peel tests are conducted on the same coordinate test grid to measure the peel force per unit width. When interfacial fracture test conditions are available, a separate model with interfacial fracture energy as the output can be established, but the two output quantities should not be mixed in the same adhesion calibration model.

[0068] After thermal excitation, temperature sequences can be acquired by an infrared camera; the ultrasonic testing device acquires amplitude, phase, propagation time, attenuation, or local resonance response in the same coordinate region. Visible light, electroluminescence, photoluminescence, or polarization images are used to supplement crack, notch, or stress information, but do not replace adhesion calibration of thermal and ultrasonic responses.

[0069] The detection feature vectors of each detection grid are combined with the measured calibration data to form calibration data D0. The features are standardized according to the mean and standard deviation of the training data, and a Gaussian process regression model f is trained. cal The kernel function parameters are determined by maximizing the marginal likelihood, and the prediction error is checked using validation data. For each feature, the lower and upper bounds of the training samples are recorded, and an allowable expansion amount is set; if any input exceeds the corresponding allowable interval, or if the prediction standard deviation of the model output exceeds the uncertainty threshold, the detection region is considered to be outside the calibration range.

[0070] During the component sample calibration phase, conservative peeling parameters and the upper limit of allowable temperature T were also determined. lim Permissible peel force upper limit F lim The first warning value, the second warning value, and the fallback distance are specified. These values ​​are written into a separate security configuration and cannot be modified by the prediction model.

[0071] (2) Regional detection, merging and path determination Secure the laminated components, which have been removed or have been made away from the aluminum frame and junction box, to the support surface and establish a two-dimensional coordinate system. The side length of the basic inspection grid can be 10–50 mm, with smaller grids used for crack tips, edge notches, existing delamination, and support transition areas. Thermal excitation inspection can be performed using infrared radiation, hot air, or contact resistance heating, with a heat flux density of 0.5–3.0 kW / m² during the inspection stage. 2 The action time can be 1–15 s, and the maximum temperature rise during the detection phase can be controlled within 5–30℃; the ultrasonic center frequency can be 20–300 kHz, and the scanning step size can be 5–50 mm. The above ranges can be set according to actual conditions.

[0072] The adhesion resistance value and damage risk value are calculated, for example, Figure 2 Intersecting lines in the grid represent prohibited areas; these nodes are removed from adjacent graphs during path generation. Path planning first merges processed areas, then selects the starting point for stripping from areas that meet the following criteria: edge reachability, adhesion resistance below the stripping threshold, damage risk below the risk threshold, and reliable support. The path cost is calculated for each candidate path, and the path with the minimum cost is selected as the current stripping path. Figure 3 As shown. Figure 2 and Figure 3 The laminated components to be disassembled are divided into two-dimensional grid detection areas. The values ​​within the grid represent adhesion resistance and damage risk values, and the intersecting grid lines represent areas where passage is prohibited.

[0073] If a high-risk area completely isolates an untreated area, the automatic planning will stop and prompt for manual confirmation. Areas that are manually confirmed as treatable can only be converted to restricted treatment areas; their risk records cannot be deleted. Before entering a restricted treatment area, the maximum peeling speed, maximum temperature, or maximum permissible peeling force will be reduced, and the feedback sampling frequency will be increased.

[0074] (3) Construction, training and use of prediction models The first processing area uses conservative peeling parameters obtained from component sample verification. Peeling parameters can be selected from the target temperature, heating time, heating power, ultrasonic power, ultrasonic frequency, peeling speed, and peeling angle—these are the quantities that need optimization; the remaining quantities are fixed at the equipment's allowable values. Data that has been processed and passed quality checks forms dataset D.

[0075] Gaussian process regression models were established for the effective separation area ratio, residual film area ratio, energy consumption per unit area, peak peel force, peak temperature, and new damage, respectively. The input to each model was the region state [a]. i ,r i The model is combined with the actual stripping parameter x to avoid ignoring regional differences based solely on the parameters. During model training, standardized parameters bound to the dataset D are used, and the kernel function length scale, magnitude, and noise parameters are determined by maximizing the marginal likelihood. Model updates are only performed after the current region has been successfully completed and passed quality checks.

[0076] The number of valid data points is less than the number of enabled samples n for the prediction model. min Model recommendation is not enabled at this time. The number of samples n required to enable the prediction model is reached. minThen, among the discrete or continuous candidate parameters allowed by the equipment, first eliminate candidate parameters that do not meet the following conditions: the probability that the predicted peak temperature does not exceed the upper limit of the allowable temperature is not less than 1-α; the probability that the predicted peak peeling force does not exceed the upper limit of the allowable peeling force is not less than 1-α; and the deterministic boundaries of the disassembly equipment power and motion stroke. Then, select the parameter with the largest constrained acquisition function value according to the constrained improvement expectation formula. If the candidate parameter set is empty, use a conservative peeling parameter or stop the automatic processing of the current area without relaxing the safety constraints.

[0077] (4) Local processing and two-level feedback handling The localized heating area is located at or in front of the peeling edge, and ultrasonic vibration acts on the same strip-shaped area or the peeled layer. The peeling speed can be selected within the preliminary verification range of 0.2 to 10 mm / s, the peeling angle can be determined within the range of 20° to 90° according to the structure of the mechanism and components, and the retraction distance can be determined within the range of 1 to 30 mm according to the size of the peeling head and the scope of re-inspection.

[0078] Temperature, peeling force, displacement, and acoustic emission signals are continuously sampled. When any monitored quantity reaches the first warning value, the heating power, ultrasonic power, or peeling speed is reduced, and the process returns to step S7 to observe the feedback after the downgrade. When the second warning value is reached, the peeling leading edge displacement is less than the preset minimum increment while the peeling force continues to increase, or the peeling force suddenly drops and is accompanied by continuous high-energy acoustic emission events, the energy input and peeling movement are stopped, the process is reversed and the current area is re-examined, and then the process returns to step S5.

[0079] (5) Data quality check and closed-loop update After the current processing area is completed normally, record the area coordinates, detection characteristics, adhesion resistance value, damage risk value, actual peeling parameters, effective separation area ratio, residual film area ratio, energy consumption per unit area, peak peeling force, peak temperature, new damage, and interlocking events. If the area coordinates do not match, the sensor has a failure mark, the actual execution parameters are missing, or the result cannot be verified, this set of data will not be included in dataset D.

[0080] For data that passes the quality check, the processed area labels are updated to allow step S5 to regenerate subsequent processing areas and disassembly paths. Simultaneously, dataset D is added to step S6 for training or updating the prediction model. These two updates are applied to steps S5 and S6, respectively. After updating the prediction model's dataset, if unprocessed areas still exist, steps S5, S6, S7, S8, and S9 are executed sequentially. Once all areas have been processed, the disassembly operation ends.

[0081] In an embodiment of the present invention, the ultrasonic testing device is an ultrasonic instrument for testing, which collects interface echoes in a low-energy pulse or frequency sweep mode, without applying vibrational energy sufficient to cause interface peeling; during the peeling operation, the ultrasonic transducer outputs high-power vibration to assist interface separation; the two can be two independent hardware sets, or the same hardware can switch energy levels in a time-sharing manner.

[0082] The disassembly method in this embodiment is applicable to single-glass backsheet modules. For double-glass modules, thermal and ultrasonic responses can be obtained from both sides to form adhesion resistance and damage risk values ​​for the two interfaces. The side with lower adhesion resistance and corresponding lower glass damage risk is prioritized. When the adhesion resistance and damage risk assessment results of the two interfaces conflict, the damage risk value is used as the priority criterion, and the interface with lower damage risk is selected for disassembly. After the first interface is separated, the coordinates and support boundaries of the remaining composite layer are re-established, and steps S2 to S5 are repeated. The regional evaluation results and disassembly path of the interface before separation must not be used. The adhesion resistance and damage risk of the two interfaces of a double-glass module may differ significantly, requiring separate calibration models. After the first interface is separated, the support conditions of the remaining composite layer change, necessitating the re-establishment of coordinates and support boundaries; the original evaluation results cannot be used. The reason for prioritizing the damage risk value when the evaluation results of both sides conflict is that safety takes precedence over efficiency.

[0083] This invention proposes a method for dismantling decommissioned photovoltaic modules based on regional detection and feedback control. It obtains the thermal and ultrasonic responses of the same coordinate region through controllable thermal and ultrasonic excitation; it obtains the adhesion resistance value calibrated by a peel test and the damage risk value determined by cracks, notches, and support conditions; it determines the peeling start point, processing area, and obstacle avoidance path based on the two types of regional conditions; it implements multi-sensor feedback during the local heating-ultrasonic-assisted-mechanical peeling process; it performs deterministic degradation, stop, rollback, and re-inspection when a warning or shutdown event occurs; and it updates the parameters of subsequent regions only within independent hard constraints using the results of the completed regions.

[0084] like Figure 4 As shown, this embodiment of the invention proposes a dismantling device for decommissioned photovoltaic modules based on regional detection and feedback control, comprising: The support platform 100 includes a positioning component 110, a flexible support component 120, and a partitioned negative pressure adsorption unit 130. The positioning component 110 is used to determine the reference edge of the component, the flexible support component 120 is used to support the laminated component to be disassembled to compensate for component warping, and the partitioned negative pressure adsorption unit 130 is used to adsorb the laminated component to be disassembled to restrict component movement. The region detection component 200 includes a short-time thermal exciter 210, an infrared camera 220, an ultrasonic detection device 230, and an image detection unit 240. The short-time thermal exciter 210 applies thermal excitation to each detection region and then uses the infrared camera 220 to collect the thermal response of each detection region. The ultrasonic detection device 230 applies pulse or frequency sweep excitation to the detection region and collects the ultrasonic response. The image detection unit 240 collects one or more of the following: visible light image, electroluminescent image, photoluminescent image, or polarization image. The controller 300 includes a region state calculation module 310, a path planning module 320, a parameter selection module 330, and a feedback control module 340. The region state calculation module 310 outputs the adhesion resistance value and damage risk value of each detection region; the path planning module 320 determines the processing area, peeling position, and disassembly path; the parameter selection module 330 calls a conservative parameter library or a prediction model subject to hard constraints; and the feedback control module 340 outputs power reduction, speed reduction, stop, or retraction commands to the local processing component 400 and the peeling mechanism 500 based on the feedback signal. A local processing component 400 is disposed at or in front of the peeling front edge and is movable with the peeling front edge. The local processing component 400 includes a local heater 410 and an ultrasonic transducer 420. The local heater 410 acts on the strip-shaped area at or in front of the peeling front edge, and the ultrasonic transducer 420 applies vibration to the strip-shaped area or the peeled layer. The peeling mechanism 500 includes a wedge-shaped peeling member 510, a clamping and traction member 520, and a drive mechanism 530, which is used to disassemble according to the peeling speed and peeling angle in the peeling parameters of the disassembly path. The feedback detection component 600 includes a temperature detection unit 610, a force sensor 620, a displacement sensor 630, and an acoustic emission sensor 640. The force sensor 620 is located on the load transmission path of the clamping traction member 520 or the drive mechanism 530 and is used to collect feedback signals during the disassembly process. The current collection shunt module 700 collects glass, backsheet and encapsulation layer-cell composite respectively.

[0085] For fluorinated backplane assemblies, local exhaust hoods and detectors for fluorinated gases or volatile organic compounds may also be included.

[0086] The operation of the device of the present invention will be described through a specific embodiment.

[0087] Remove or avoid the aluminum frame and junction box of the laminated component to be disassembled, and place the laminated component to be disassembled on the support platform 100. The reference edge of the component is determined by the positioning component 110, the component warping is compensated by the flexible support component 120, and the component movement is restricted by the partitioned negative pressure adsorption unit 130. The controller 300 identifies the outer edge of the component, the direction of the busbar, glass cracks, edge notches, back panel damage, existing delamination, and support boundaries, and establishes a planar coordinate system consistent with the drive mechanism 530.

[0088] Before the task begins, the maximum temperature, maximum permissible peel force, maximum heating power, maximum ultrasonic power, speed range, angle range, stroke, and retraction distance are independently configured. These safety configurations are determined by equipment capabilities, packaging materials, and sample testing; the parameter selection module 330 has no authority to modify them. If the positioning or adsorption state does not meet the set requirements, the controller 300 will not execute step S2.

[0089] The laminated components to be disassembled are divided into a basic grid with sides of 10–50 mm. The density of the detection area is increased at glass crack tips, edge notches, existing delamination, and support transition locations. A short-time thermal exciter 210 applies thermal excitation to each grid, and an infrared camera 220 records the temperature sequence during the heating and cooling phases. An ultrasonic testing device 230 applies ultrasonic excitation to the same coordinate area and acquires the ultrasonic response. When the component circuitry is available, an electroluminescent or photoluminescent image can be obtained by the image detection unit 240; when the component is not powered, the damage risk is assessed using visible light, ultrasonic anomalies, and support boundary information.

[0090] During calibration, the thermal and ultrasonic responses of the component sample area are first obtained, and then independent peel tests are performed on the same coordinate area to record the peel force per unit width. When the conditions for interface fracture testing are available, a separate model with interface fracture energy as the output can be built. The area state calculation module 310 calculates the adhesion resistance value and damage risk value. When the input exceeds the calibration range, the sensor signal is missing, or the uncertainty of the adhesion calibration model is higher than the set value, the output of the adhesion calibration model is not directly used to increase the temperature or load, but the corresponding area is marked as a high uncertainty area.

[0091] The path planning module 320 merges spatially adjacent grids with similar adhesion resistance and damage risk values ​​into processing areas, and sets grids with through-crack ends, glass chipping, and unsupported areas as prohibited peeling zones or prohibited passage zones. Peeling locations are preferentially selected at component edges or existing local delamination sites, provided that the adhesion resistance and damage risk values ​​are below their respective thresholds and the support is reliable.

[0092] Candidate paths are ranked and their costs calculated using the cost calculation formula. Paths that pass through prohibited areas are eliminated. If the current area is stopped and re-inspected due to a second warning value event, S5 is re-executed, and paths that were not completed before the shutdown are not used.

[0093] The first processing region invokes conservative stripping parameters validated by the sample. The parameter selection module 330 reads the parameter-result dataset formed by S9; the number of samples enabled for the prediction model is determined before the experiment based on the number of parameters to be updated, repeatability, and data quality, and is used as a controller configuration item, not automatically reduced by the prediction model during operation.

[0094] Once the effective area data reaches the required sample size for the prediction model, the parameter selection module 330 establishes a prediction model based on regression models such as Gaussian process regression, evaluates candidate stripping parameters, and only selects parameters that satisfy the safety formula and equipment deterministic boundary conditions as safe candidate stripping parameters. Constrained improvement expectations can be used to select the next parameter. Model recommendations are not executed when data is insufficient, input exceeds the calibration range, model uncertainty is too high, or there are continuous prediction errors.

[0095] A local heater 410 and an ultrasonic transducer 420 act on the peeling front edge or in front of it, a wedge-shaped peeling member 510 establishes a peeling opening, and a clamping traction member 520 is advanced at a set speed and angle under the action of a drive mechanism 530. A temperature detection unit 610, a force sensor 620, a displacement sensor 630, and an acoustic emission sensor 640 continuously sample.

[0096] When the peeling force reaches a preset ratio of the maximum allowable peeling force, the temperature approaches the upper limit, or the acoustic emission signal is higher than the normal baseline, the feedback control module 340 reduces the peeling speed, heating power, or ultrasonic power, and then continues to execute S7. When the peeling force reaches the second warning value, the temperature exceeds the upper limit, the displacement-peeling force relationship indicates that the peeling leading edge is stalled, or high-energy acoustic emission events occur continuously, the local processing and peeling movement are stopped, the peeling mechanism 500 is retracted 1-30 mm, and the current area is re-inspected; after re-inspection, it returns to S5. If the processing conditions are still not met after re-inspection, manual confirmation or termination of processing in that area is initiated.

[0097] Once the current processing area completes normally, execute S9 to write the parameter-result data to the dataset. If there is an unprocessed area, return to S5, and execute S5-S9 sequentially; if there is no unprocessed area, the component decomposition ends.

[0098] Example 1: Regional inspection and disassembly of EVA single-glass backsheet assembly.

[0099] Place the crystalline silicon single-glass backsheet assembly, with the aluminum frame and junction box removed, on the support platform, ensuring the glass side faces the flexible support surface. Activate the negative pressure adsorption zone corresponding to the current processing area and establish two-dimensional coordinates at the edge of the assembly.

[0100] The components were divided into a 20mm × 20mm detection grid. The short-time thermal exciter operated at 1.2kW / m². 2The heat flux density was applied for 5 seconds, and an infrared camera recorded the surface temperature within 20 seconds after the start of heating at a frame rate of 25 Hz. An ultrasonic testing device completed the scan at a center frequency of 80 kHz and a step size of 20 mm. When the component circuit was available, electroluminescence images were acquired; when the circuit was unavailable, the location of cracks and notches was determined by visible light images and ultrasonic anomalies.

[0101] For each detection grid, the maximum temperature rise, peak time, cooling slope, ultrasonic amplitude, and ultrasonic propagation time are extracted. An estimated peel force per unit width is obtained using a pre-calibrated adhesion calibration model, and this estimated value is used as the adhesive resistance value. At least two of the following are converted into a breakage risk value: glass crack or edge notch, crack tip distance, cell crack, ultrasonic anomaly, glass stress anomaly, weld strip or busbar location, support boundary, and localized suspension state.

[0102] The first treatment area employs a conservative combination of peeling parameters: target temperature 95℃, action time 8s, ultrasonic frequency 40kHz, peeling speed 1mm / s, and peeling angle 45°. The local treatment component is located in front of the peeling leading edge, and the wedge-shaped peeling element advances along the obstacle-around path. The first warning value can be set to 0.80–0.90 of the maximum peeling force; when the first warning value is reached, the peeling speed is reduced to 50%–80% of the current value. When the maximum peeling force, maximum peeling temperature, or second acoustic emission warning value is reached, the process stops and retracts 5mm. Temperature and acoustic emission signals are also configured with first warning values; for example, the temperature warning is set below the upper limit of the film softening temperature, and the first acoustic emission warning is set below the crack characteristic event baseline.

[0103] 95℃ is a candidate process temperature in this embodiment. It is adopted when comparing adjacent candidate temperatures on samples with the same structure, ensuring stable peeling without triggering a preset safety boundary. This temperature is not the universal softening temperature of EVA material, nor does it represent the proven optimal temperature. Threshold configurations can be, for example, set the first and second temperature warning values ​​to 100℃ and 110℃ respectively, and the first and second peeling force warning values ​​to 80 N and 100 N respectively, provided sample verification allows. The corresponding independent safety upper limit must not be lower than the aforementioned second warning value. The acoustic emission threshold is determined based on the normal peeling baseline and crack event calibration under the same installation and acquisition conditions. The above values ​​are used to illustrate the control configuration; the specific values ​​used are verified by the corresponding component samples.

[0104] After completing the treatment area, record the parameter-result data. Once the preset number of samples is reached, update the peeling parameters for the next treatment area using the prediction model. If the predicted peak peeling force continuously exceeds the allowable deviation from the measured value, discontinue the model recommendation.

[0105] After the first treatment zone is completed, the separation area and residual adhesive area are measured, and energy consumption, peak peel force, and new cracks are recorded. The controller incorporates these results into the prediction model and selects the treatment parameters for the next treatment zone from candidate parameters that meet the temperature, peel force, and power limits. Before entering the crack-limited zone, the path preferentially avoids the crack ends; if avoidance is not possible, the peeling speed and maximum allowable peel force are reduced.

[0106] Example 2: Dual interface treatment of POE double glass module.

[0107] Thermal and ultrasonic responses were obtained from the front and rear glass sides, respectively, to determine the adhesion resistance and damage risk values ​​at the two interfaces. The side with lower adhesion resistance and corresponding lower glass damage risk was treated first. During the first interface separation, the untreated side was fixed by a flexible support and a zoned negative pressure adsorption unit. After the first interface glass separated, the zoned negative pressure adsorption unit dynamically adjusted the adsorption switch in the corresponding area to maintain reliable support and adsorption of the remaining composite layer.

[0108] After the first interface separates, coordinates and support boundaries are re-established for the remaining composite layer, and S2-S5 are re-executed, without directly using the area evaluation and path from the other interface before separation. The thinner or higher-risk glass side is used to determine a more conservative maximum permissible peel force.

[0109] Example 3: Safety control of fluorinated backsheet assembly.

[0110] When component material files or verified material identification results indicate that the backsheet contains fluorine, the controller invokes a separate backsheet-side temperature limit and prioritizes mechanical stripping and ultrasonic assistance. Localized treatment areas are equipped with exhaust hoods and gas detection units.

[0111] When the gas detection value reaches the first warning threshold, the heating power is reduced and the ventilation volume is increased. When the second warning threshold is reached, heating is stopped while ventilation is maintained. The automatic disassembly process can only resume after the fault is resolved and the gas detection signal returns to the safe threshold. The gas interlock is independent of the parameter selection module, and the prediction model cannot select parameters that exceed the upper limit of the temperature of the fluorinated backplane. The gas detection interlock corresponding to the fluorinated backplane is an independent safety interlock and does not share the same set of warning thresholds with the peel force, temperature, and acoustic emission warning system.

[0112] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0113] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A method for dismantling decommissioned photovoltaic modules based on regional detection and feedback control, characterized in that, Includes the following steps: S1, Establish the coordinates of the laminated components to be disassembled and divide them into multiple detection areas; S2, apply short-term thermal and ultrasonic excitation within the same coordinate detection area, and collect thermal response, ultrasonic response, damage characteristics and support status; S3. Input the detection feature vector composed of thermal response and ultrasonic response into the adhesion calibration model with the peel force per unit width or the interface fracture energy of the same structural component sample as the output calibration value to obtain the adhesion resistance value. S4, based on acoustic anomalies, damage characteristics, and support status, yields a damage risk value; S5. Based on the difference in adhesion resistance, the difference in damage risk, and the spatial continuity of the adjacent detection areas, merge the processing areas, select the peeling position at the edge of the component that is lower than the peeling threshold and the risk threshold and meets the support conditions, and select the path with the lowest cost from the candidate paths that do not pass through the prohibited area as the disassembly path. S6, select the peeling parameters for the current processing area within independent temperature, peeling force, power and stroke constraints. The peeling parameters can be conservative peeling parameters or peeling parameters predicted by a prediction model. S7, disassemble according to the disassembly path and the peeling parameters, and collect feedback signals of temperature, peeling force, displacement and acoustic emission in real time; S8, when any feedback signal meets the first warning value, the stripping parameter value of the current area is downgraded and then returned to step S7. When any feedback signal meets the second warning value or a stripping abnormality occurs, the disassembly is stopped and the current processing area is re-inspected and then returned to step S5. The warning level of the second warning value is higher than that of the first warning value. S9: After the current processing area is completed, perform quality checks on the stripping parameters and stripping result data and record the valid parameter-result data as the dataset for updating the prediction model. If there are still unprocessed areas, return to step S5 until there are no unprocessed areas.

2. The method according to claim 1, characterized in that, In step S1, the outer edge of the laminated component to be disassembled, the direction of the busbar, glass cracks, edge notches, backplate damage, existing delamination and support boundaries are identified, and the detection area density is increased at the tip of the glass crack, edge notches, existing delamination and support transition positions.

3. The method according to claim 1, characterized in that, In step S2, the damage features collected include: Visible light imaging was used to capture glass cracks, edge notches, and backplate damage. Electroluminescent or photoluminescent imaging was used to obtain cell cracks; Glass stress anomalies are obtained using polarization imaging; Ultrasonic response is used to obtain stratification or ultrasonic anomalies.

4. The method according to claim 1, characterized in that, Step S3 specifically includes: Select a module sample that is identical to the laminated module to be disassembled in terms of glass structure, encapsulation film and backsheet type, and conduct short-time thermal excitation, ultrasonic excitation and local peeling tests in the corresponding test areas of the module sample in sequence. The detection feature vector consisting of maximum temperature rise, peak time, cooling slope, ultrasonic amplitude and ultrasonic propagation time is used as input, and one of the measured unit width peel force or interface fracture energy is used as output. The regression model is trained to obtain the adhesion calibration model. The detection feature vector of the laminated component to be disassembled is input into the adhesion calibration model for prediction, and the output value is used as the adhesion resistance value. When any detection feature exceeds the allowable range of the corresponding feature in the calibration sample, or the prediction standard deviation exceeds a preset uncertainty threshold, the corresponding detection area is marked as a high uncertainty area; the adhesion resistance value of the i-th detection area... for: in, For adhesion calibration model, Let be the detection feature vector composed of the temperature response and ultrasonic response of the i-th detection region. , The maximum temperature rise in the i-th detection area of ​​the laminated component to be disassembled. The time it takes for the i-th detection area of ​​the laminated component to be disassembled to reach its temperature peak. Let be the cooling slope of the i-th detection area of ​​the laminated component to be disassembled. Let be the ultrasonic amplitude of the i-th detection area of ​​the laminated component to be disassembled. The ultrasonic propagation time is the i-th detection area of ​​the laminated component to be disassembled.

5. The method according to claim 1, characterized in that, In step S4, at least two of the following characteristics of the laminated assembly to be disassembled are normalized into risk features: glass cracks or edge notches, crack tip distance, cell cracks, backsheet damage, existing delamination, ultrasonic anomalies, glass stress anomalies, solder strip or busbar location, support boundary, and local suspension state. Calculate the damage risk value of the i-th detection region. : in, The cutoff function is used to truncate the weighted sum to the interval [0,1], where K≥2 represents the number of risk features selected. The weights are those verified by component samples or set by preset risk rules. Let k be the normalized risk feature of the i-th detection region.

6. The method according to claim 1, characterized in that, Step S5 specifically includes: Adjacent detection areas sharing a common edge are considered as adjacent areas. When the difference in adhesion resistance values ​​between adjacent areas is not greater than the adhesion difference threshold, the difference in damage risk values ​​is not greater than the risk difference threshold, and the two detection areas are not prohibited from being separated by the boundary, the adjacent areas are merged into the same processing area. For areas with through-crack ends, glass chipping, or unsupported regions, prohibition of peeling or prohibition of passage is set. Multiple detection areas are constructed into a graph with adjacent relationships as edges. Multiple candidate paths without prohibited passage areas are generated from the peeling initiation position to the current target boundary. The path cost of each candidate path is calculated, and the candidate path with the minimum path cost is determined as the disassembly path. The path cost of a candidate path is determined by a weighted sum of the normalized adhesion resistance value, damage risk value, travel distance, repeated heating cost, and number of turning times along the path. The path cost of candidate path π is... for: in, Let be the normalized adhesion resistance value of the i-th detection region. Let be the normalized moving distance of the i-th detection region. Let i be the normalized damage risk value for the i-th detection region. To the cost of repeated heating, For the number of turns, , , , , The weights are non-negative and their sum is 1.

7. The method according to claim 1, characterized in that, In step S6, the peeling parameters of the processing area include one or more of the following: target temperature, heating time, heating power, ultrasonic power, ultrasonic frequency, peeling speed, and peeling angle. The peeling parameters of the processing area are predicted using a prediction model. When the effective parameter-result data volume in the first processing area, high uncertainty area, or dataset does not reach the number of samples enabled by the prediction model, conservative peeling parameters verified by samples of the same structure component are used. The dataset used to update the prediction model consists of qualified and valid parameter-result data from the completed treatment areas. The parameter-result data includes area coordinates, adhesion resistance values, damage risk values, peeling parameters, effective separation area ratio, film residue area ratio, energy consumption per unit area, peak peel force, peak temperature, new damage, and interlocking events. The input to the j-th set of data in the prediction model and output Represented as: in, Let J be the adhesion resistance value of the j-th data set. Let j be the damage risk value of the j-th data group. The stripping parameter for the j-th data group. The effective separation area ratio of the j-th data set. The percentage of residual film area in the j-th data set. Let j be the normalized value of energy consumption per unit area for the j-th data set. The peak peeling force of the j-th data set. Let be the peak temperature of the j-th data set. This represents the newly added damage normalized value for the j-th data group; When the effective parameter-result data volume in the dataset of the prediction model reaches the number of samples to be used in the prediction model, the constrained acquisition function of the candidate stripping parameters is calculated. The constrained acquisition function is composed of the product of the improved expectation of the minimized comprehensive objective function, the predicted probability of temperature and stripping force satisfying the constraints, and the maximum value of the constrained acquisition function within the candidate set that satisfies the lower probability limit and the equipment boundary. The candidate stripping parameter must simultaneously satisfy the following conditions: the probability that the predicted peak temperature does not exceed the upper limit of the allowable temperature is not less than 1-α, the probability that the predicted peak stripping force does not exceed the upper limit of the allowable stripping force is not less than 1-α, and the deterministic boundary of the power and movement stroke of the disassembly equipment. The maximum value of the constrained acquisition function is solved in the candidate stripping parameter set that satisfies the constraints, and the stripping parameter corresponding to the maximum value is taken as the stripping parameter of the current processing area. Here, α is the constraint tail probability, 0 < α < 0.

5. The comprehensive objective function for the candidate stripping parameters is: in, Let be the comprehensive objective function to be minimized. To strip the parameter vector, , , , , The weights are non-negative and their sum is 1. To effectively separate the area ratio, This represents the percentage of residual adhesive film area. This is the normalized value of energy consumption per unit area. This is the normalized value of the peak peel force. This is the newly added damage normalization value.

8. The method according to claim 1, characterized in that, In step S8, the first warning value and the second warning value are set for the feedback signals of temperature, peeling force and acoustic emission, respectively; the peeling abnormality includes the peeling leading edge displacement being less than the preset minimum increment and the peeling force continuously increasing, or the peeling force suddenly dropping and accompanied by continuous high-energy acoustic emission events; when the first warning value is met, at least one of reducing the heating power, reducing the ultrasonic power or reducing the peeling speed is executed; when the second warning value is met or a peeling abnormality occurs, the energy input and peeling movement are stopped, and the peeling mechanism is retracted by a preset distance and the current area is re-detected.

9. The method according to claim 1, characterized in that, In step S9, the quality check includes at least the correspondence of area coordinates, the absence of failure markers in sensor signals, the complete recording of actual stripping parameters, and the ability to verify the results. Only the valid parameters and result data that pass the quality check are added to the training dataset of the prediction model. When returning to step S5, the subsequent processing area and disassembly path are regenerated based on the updated processed area markers and re-inspection results. Then, in step S6, the training dataset of the prediction model is updated, and the stripping parameters of the subsequent processing area are determined.

10. A decommissioned photovoltaic module dismantling device based on area detection and feedback control, used to execute the decommissioned photovoltaic module dismantling method based on area detection and feedback control as described in any one of claims 1-9, characterized in that, include: The support platform includes positioning components, flexible support components, and zoned negative pressure adsorption units. The positioning component is used to determine the reference edge of the component, the flexible support component is used to support the laminated component to be disassembled to compensate for component warping, and the partitioned negative pressure adsorption unit is used to adsorb the laminated component to be disassembled to restrict component movement. The region detection component includes a short-time thermal exciter, an infrared camera, an ultrasonic detection device, and an image detection unit. The short-time thermal exciter applies short-time thermal excitation to each detection region and then uses the infrared camera to collect the thermal response of each detection region. The ultrasonic detection device applies pulse or frequency sweep excitation to the detection region and collects the ultrasonic response. The image detection unit collects one or more of the following: visible light image, electroluminescent image, photoluminescent image, or polarization image. The controller includes a region state calculation module, a path planning module, a parameter selection module, and a feedback control module. The region state calculation module outputs the adhesion resistance value and damage risk value of each detection region; the path planning module determines the processing area, peeling position, and dismantling path; the parameter selection module calls a conservative parameter library or a prediction model subject to hard constraints; and the feedback control module outputs power reduction, speed reduction, stop, or retraction commands to the local processing components and peeling mechanism based on the feedback signal. The local processing component is located at or in front of the peeling front edge and can move with the peeling front edge. The local processing component includes a local heater and an ultrasonic transducer. The local heater acts on the strip-shaped area at or in front of the peeling front edge, and the ultrasonic transducer applies vibration to the strip-shaped area or the peeled layer. The peeling mechanism includes a wedge-shaped peeling component, a clamping and traction component, and a drive mechanism, which are used to disassemble according to the peeling speed and peeling angle in the peeling parameters of the disassembly path; The feedback detection component includes a temperature detection unit, a force sensor, a displacement sensor, and an acoustic emission sensor. The force sensor is set on the load transmission path of the clamping traction component or the drive mechanism to collect feedback signals during the disassembly process. The current collection shunt assembly collects glass, backsheet, and encapsulation layer-cell composite material respectively.

Images