An AI vision-guided environmental protection recycling process for waste photovoltaic panels

Abstract

The application discloses an AI vision-guided environment-friendly recycling process for waste photovoltaic panels, and relates to the technical field of environmental protection, comprising the following steps: removing the aluminum frame; analyzing the waste photovoltaic panel after the aluminum frame is removed; removing the PET layer and the glass layer; crushing, screening, AI vision, and electromagnetic induction composite detection, metal micro-particle separation, silicon powder air separation, and silicon powder plasma treatment are performed on the photovoltaic panel after sand blasting treatment. The waste photovoltaic panel is comprehensively analyzed through a three-modal AI recognition model, the ultrasonic pre-peeling and gradient sand blasting combined process are linked, the surface peeling treatment is performed on the degradable sand blasting medium, the environment-friendly recycling mode without high-temperature cracking is realized, the environmental protection treatment cost can be greatly reduced, and the unified treatment of single-glass and double-glass photovoltaic panels is compatible, so that the problems that harmful gas is generated in the existing high-temperature cracking process to pollute the environment and the application range of the physical crushing process is limited can be solved.

Description

Technical Field

[0001] This invention relates to the field of environmental protection technology, specifically to an environmentally friendly recycling process for waste photovoltaic panels based on AI vision guidance. Background Technology

[0002] The environmentally friendly recycling process for waste photovoltaic panels can efficiently recover valuable metals such as silicon, silver, and aluminum, as well as reusable materials such as glass and frames, reducing the consumption of primary resources. It can also prevent the leakage of harmful substances such as lead and cadmium from the panels, thus preventing pollution of soil, water sources, and the atmosphere. At the same time, it complies with environmental regulations, helps the photovoltaic industry develop in a green and closed-loop manner, and reduces the environmental impact throughout its entire life cycle.

[0003] In existing technologies, environmentally friendly recycling processes for waste photovoltaic panels mostly employ high-temperature pyrolysis or single physical crushing processes. High-temperature pyrolysis generates harmful gases such as VOCs and fluorides, which seriously pollute the atmospheric environment and have high processing costs. Furthermore, single physical crushing processes lack adaptability to photovoltaic panel structures and are difficult to integrate with the unified processing of single-glass and double-glass photovoltaic panels, thus limiting their applicability. Based on this, the present invention provides an AI vision-guided environmentally friendly recycling process for waste photovoltaic panels. Summary of the Invention

[0004] The purpose of this invention is to provide an AI-guided environmentally friendly recycling process for waste photovoltaic panels. This invention uses a three-modal AI recognition model to accurately analyze the composition, layer thickness, and interlayer stress distribution of waste photovoltaic panels. It links a graded recycling and reuse system for biodegradable sandblasting media, combines an ultrasonic pre-stripping and gradient sandblasting process triggered by interlayer stress, and integrates AI vision and electromagnetic induction composite detection, electromagnetic adsorption and ultrasonic oscillation synergistic separation, and plasma purification technology to build a full-process AI-guided environmentally friendly recycling system. This system achieves high-temperature-free pyrolysis recycling of waste photovoltaic panels, efficient material circulation, and high-value-added purification.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] An AI-guided, vision-based environmentally friendly recycling process for waste photovoltaic panels includes:

[0007] S1. Pre-removal of aluminum frame: The aluminum frame of the waste photovoltaic panel is removed by mechanical clamping and hydraulic separation. The separated aluminum frame is collected and recycled separately. Subsequent processes are only for the photovoltaic panel body without aluminum frame.

[0008] The mechanical clamping pressure is 0.3-0.5 MPa, the hydraulic separation stroke speed is 5-8 mm / s, and the aluminum frame separation error is ≤ ±1 mm.

[0009] S2. A three-modal AI recognition model integrating visual imaging, ultrasonic thickness detection, and interlayer stress sensing is used to analyze the composition, layer thickness, and interlayer stress distribution of waste photovoltaic panels after the aluminum frame has been removed.

[0010] The composition consists of glass, PET, silicon wafer and EVA, the layer thickness is 0.1 to 3 mm, and the interlayer stress distribution is 0.1 to 0.8 MPa;

[0011] S3. Remove the PET and glass layers from the surface of the waste photovoltaic panel. Use biodegradable ceramic microspheres as the sandblasting medium. During the sandblasting process, AI identifies the material of the sandblasting area in real time through visual imaging. When the EVA layer is detected, the nozzle position is automatically switched to the untreated PET and glass layer area, while adjusting the sandblasting intensity and sandblasting time.

[0012] After sandblasting, the biodegradable ceramic microspheres are graded, recycled, and dried by hot air. The dried biodegradable ceramic microspheres are then reused in the sandblasting process. The grading, recycling, hot air drying, and reuse processes are controlled by AI. The AI ​​provides feedback on process parameters at a cycle of 0.18 seconds.

[0013] S4. When the interlayer stress value of the waste photovoltaic panel is ≥0.48MPa, AI triggers a combination of ultrasonic pre-peeling and gradient sandblasting process. The ultrasonic vibration weakens the bonding force between the PET and glass layers, and then gradient sandblasting removes the PET and glass layers.

[0014] The ultrasonic pre-peeling frequency is 42-58kHz and the duration is 6-9 seconds. The gradient sandblasting pressure is adjusted in steps in the order of 0.45MPa, 0.75MPa and 0.42MPa. The distance between the nozzle and the photovoltaic panel is 45-85mm. For every 10mm change in distance, a pressure of ±0.09MPa is compensated. The peeling area error is ≤±0.08mm.

[0015] S5. After being sandblasted, the photovoltaic panels are conveyed to a shredder to be crushed to a particle size of 2cm. After crushing, the materials are first separated by a vibrating screen to separate metal sand, PET fragments, glass fragments and 2cm photovoltaic panel body fragments.

[0016] The separated 2cm photovoltaic panel fragments are fed into a grinding mill to be processed to a particle size of 0.012-0.09mm, and then subjected to AI vision and electromagnetic induction composite detection, metal microparticle separation, silicon powder air classification and silicon powder plasma treatment in sequence.

[0017] Preferably, the trimodal AI recognition model includes 6 sets of distributed cameras, 2 ultrasonic thickness detection probes and 4 interlayer stress sensors. Among the 6 sets of distributed cameras, 4 sets are arranged symmetrically at the top and 2 sets are arranged symmetrically on the sides. The detection frequency of the ultrasonic thickness detection probes is 5.5 to 9.5 MHz and the detection range of the interlayer stress sensors is 0 to 1.8 MPa.

[0018] Preferably, the data analysis of the trimodal AI recognition model adopts a convolutional neural network algorithm to fuse visual imaging data, ultrasonic thickness data and interlayer stress data. The data fusion cycle and analysis response time are consistent, both ≤0.28 seconds / frame.

[0019] Preferably, the particle size of the biodegradable ceramic microspheres is 55–145 μm;

[0020] The grading and recycling is achieved using grading sieves with specifications of 55μm, 105μm, and 145μm, with an allowable deviation of ±0.8μm in sieve aperture size, and a vibration frequency of 52 to 78Hz during grading and recycling.

[0021] The hot air drying process is carried out at a temperature of 42–58°C and a wind speed of 2.1–2.9 m / s. The relative humidity of the microspheres after drying is 6%–9% RH.

[0022] Preferably, the ultrasonic pre-stripping power is 160-240W, the vibration amplitude is 0.06-0.09mm, and the vibration direction is perpendicular to the photovoltaic panel surface;

[0023] The gradient sandblasting nozzle adopts a 0-360° omnidirectional rotating structure, the nozzle outlet diameter is 2.2-4.8mm, the jet velocity of the sandblasting medium is 9-24m / s, the real-time detection accuracy of the distance between the nozzle and the plate surface is ±0.8mm, and the pressure compensation accuracy is ±0.018MPa.

[0024] Preferably, the AI ​​vision and electromagnetic induction composite detection includes one high-speed industrial camera and two electromagnetic induction probes;

[0025] The high-speed industrial camera has a frame rate of 210 frames per second, and the image acquisition cycle is synchronized with the electromagnetic induction detection cycle, both being 4.8 ms / time.

[0026] The electromagnetic induction probe has a detection frequency of 52kHz and a detection signal sampling rate of 1.05MHz.

[0027] Preferably, in the AI ​​vision and electromagnetic induction composite detection, the high-speed industrial camera first acquires images to identify the preliminary location of metal microparticles, and then the electromagnetic induction probe detects the preliminary location. The identified metal microparticles are copper, silver and tin microparticles.

[0028] The particle size of the metal microparticles is 0.012–0.09 mm.

[0029] Preferably, the separation of metal microparticles adopts a combined method of gravity separation, electromagnetic adsorption and ultrasonic vibration. First, copper particles are separated by gravity separation, and then silver and tin microparticles are separated by electromagnetic adsorption and ultrasonic vibration.

[0030] The magnetic field strength of the electromagnetic adsorption is 0.85–1.15T, the magnetic field gradient is 0.12T / mm, and the frequency of the ultrasonic oscillation is 82kHz with a power of 110–190W.

[0031] Preferably, the silicon powder plasma treatment is used to remove residual EVA adhesive and carbon and oxygen impurities from the surface of the silicon powder;

[0032] The processing chamber temperature is 62-78℃, the vacuum degree is ≤4.8Pa, the vacuum degree maintenance accuracy is ±0.08Pa, the processing chamber has two plasma generating electrodes with an electrode spacing of 11-14mm, and the vacuum degree is controlled by a two-stage vacuum pump linkage.

[0033] Preferably, the plasma treatment power is 310-490W, the duration is 11-14s, and a mixture of argon and hydrogen is introduced during the treatment with a volume ratio of 8.5:1.5, the argon purity is ≥99.992%, the hydrogen purity is ≥99.992%, and the gas flow rate is 0.6-1.1L / min.

[0034] Gas is pulsed into the processing chamber through a quartz conduit with an inner diameter of 4.2–5.8 mm at a pulse frequency of 50 Hz. The conduit outlet is 5.5–7.5 mm away from the silicon powder surface.

[0035] Compared with the prior art, the beneficial effects of the present invention are:

[0036] 1. This invention uses a three-modal AI recognition model to comprehensively analyze the composition, layer thickness, and interlayer stress distribution of waste photovoltaic panels. It combines ultrasonic pre-stripping with gradient sandblasting processes and uses a biodegradable sandblasting medium for surface stripping treatment, achieving an environmentally friendly recycling mode without high-temperature pyrolysis. Compared with existing technologies, it can significantly reduce environmental treatment costs and is compatible with the unified treatment of single-glass and double-glass photovoltaic panels. Therefore, it can solve the problems of environmental pollution caused by harmful gases generated by existing high-temperature pyrolysis processes and the limited applicability of physical crushing processes.

[0037] 2. This invention uses AI-linked control to manage the graded recycling, hot air drying, and reuse of biodegradable sandblasting media. Combined with a dynamic compensation mechanism for sandblasting process parameters, it achieves the recycling of sandblasting media and precise control of process parameters. Compared with existing technologies, it can improve the utilization rate of sandblasting media and the stability of stripping operations. Therefore, it can solve the problems of serious media waste and damage to recycled materials caused by insufficient stripping precision in traditional sandblasting processes.

[0038] 3. This invention identifies the location of metal microparticles using AI vision and electromagnetic induction composite detection technology, separates the metal using electromagnetic adsorption and ultrasonic vibration, and then purifies the silicon powder through plasma treatment. This achieves precise sorting and high-purity purification of the recovered materials. Compared with existing technologies, it can improve the metal recovery rate and silicon powder purity, thus solving the problems of low sorting accuracy and low added value of recovered products in existing recycling processes. Detailed Implementation

[0039] The technical solutions in the embodiments of the present invention have been clearly and completely described. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] This embodiment provides an AI-guided, vision-based environmentally friendly recycling process for waste photovoltaic panels, including:

[0041] S1. Pre-removal of aluminum frame: The aluminum frame of the waste photovoltaic panel is removed by mechanical clamping and hydraulic separation. The separated aluminum frame is collected and recycled separately. Subsequent processes are only for the photovoltaic panel body without aluminum frame.

[0042] The mechanical clamping pressure is 0.3~0.5MPa, the hydraulic separation stroke speed is 5~8mm / s, and the aluminum frame separation error is ≤±1mm;

[0043] S2. A three-modal AI recognition model integrating visual imaging, ultrasonic thickness detection, and interlayer stress sensing is used to analyze the composition, layer thickness, and interlayer stress distribution of waste photovoltaic panels after the aluminum frame has been removed.

[0044] The structure consists of glass, PET, silicon wafers and EVA, with a layer thickness of 0.1 to 3 mm and an interlayer stress distribution of 0.1 to 0.8 MPa;

[0045] S3. Remove the PET and glass layers from the surface of the waste photovoltaic panel. Use biodegradable ceramic microspheres as the sandblasting medium. During the sandblasting process, AI identifies the material of the sandblasting area in real time through visual imaging. When the EVA layer is detected, the nozzle position is automatically switched to the untreated PET and glass layer area, while adjusting the sandblasting intensity and sandblasting time.

[0046] After sandblasting, the biodegradable ceramic microspheres are graded, recycled, and dried by hot air. The dried biodegradable ceramic microspheres are then reused in the sandblasting process. The grading, recycling, hot air drying, and reuse processes are controlled by AI. The AI ​​provides feedback on process parameters at a cycle of 0.18 seconds.

[0047] S4. When the interlayer stress value of the waste photovoltaic panel is ≥0.48MPa, AI triggers a combination of ultrasonic pre-peeling and gradient sandblasting process. The ultrasonic vibration weakens the bonding force between the PET and glass layers, and then gradient sandblasting removes the PET and glass layers.

[0048] The ultrasonic pre-peeling frequency is 42–58 kHz and the duration is 6–9 seconds. The gradient sandblasting pressure is adjusted in steps in the order of 0.45 MPa, 0.75 MPa and 0.42 MPa. The distance between the nozzle and the photovoltaic panel is 45–85 mm. For every 10 mm change in distance, a pressure of ±0.09 MPa is compensated. The peeling area error is ≤ ±0.08 mm.

[0049] S5. After being sandblasted, the photovoltaic panels are conveyed to a shredder to be crushed to a particle size of 2cm. After crushing, the materials are first separated by a vibrating screen to separate metal sand, PET fragments, glass fragments and 2cm photovoltaic panel body fragments.

[0050] The separated 2cm photovoltaic panel fragments are fed into a grinding mill to be processed to a particle size of 0.012-0.09mm, and then subjected to AI vision and electromagnetic induction composite detection, metal microparticle separation, silicon powder air classification and silicon powder plasma treatment in sequence.

[0051] In some embodiments, the trimodal AI recognition model includes 6 sets of distributed cameras, 2 ultrasonic thickness detection probes and 4 interlayer stress sensors. Among the 6 sets of distributed cameras, 4 sets are arranged symmetrically at the top and 2 sets are arranged symmetrically on the sides. The detection frequency of the ultrasonic thickness detection probes is 5.5 to 9.5 MHz, and the detection range of the interlayer stress sensors is 0 to 1.8 MPa.

[0052] In some embodiments, the data analysis of the trimodal AI recognition model adopts a convolutional neural network algorithm to fuse visual imaging data, ultrasonic thickness data and interlayer stress data. The data fusion cycle is consistent with the analysis response time, both ≤0.28 seconds / frame.

[0053] In some embodiments, the particle size of the biodegradable ceramic microspheres is 55–145 μm;

[0054] The grading and recycling are achieved using grading sieves with specifications of 55μm, 105μm, and 145μm. The allowable deviation of the sieve aperture is ±0.8μm, and the vibration frequency during grading and recycling is 52~78Hz.

[0055] The temperature for hot air drying is 42–58℃, the wind speed is 2.1–2.9 m / s, and the relative humidity of the microspheres after drying is 6%–9% RH.

[0056] In some embodiments, the power of ultrasonic pre-peeling is 160–240 W, the vibration amplitude is 0.06–0.09 mm, and the vibration direction is perpendicular to the surface of the photovoltaic panel.

[0057] The gradient blasting nozzle adopts a 0-360° omnidirectional rotating structure, with a nozzle outlet diameter of 2.2-4.8mm, a blasting medium jet velocity of 9-24m / s, a real-time detection accuracy of the distance between the nozzle and the plate surface of ±0.8mm, and a pressure compensation accuracy of ±0.018MPa.

[0058] In some embodiments, the AI ​​vision and electromagnetic induction composite detection includes one high-speed industrial camera and two electromagnetic induction probes.

[0059] The high-speed industrial camera has a frame rate of 210 frames per second, and the image acquisition cycle is synchronized with the electromagnetic induction detection cycle, both being 4.8ms / time.

[0060] The electromagnetic induction probe has a detection frequency of 52kHz and a detection signal sampling rate of 1.05MHz.

[0061] In some embodiments, in the AI ​​vision and electromagnetic induction combined detection, a high-speed industrial camera first acquires images to identify the preliminary location of metal microparticles, and then an electromagnetic induction probe detects the preliminary location. The identified metal microparticles are copper, silver and tin microparticles.

[0062] The particle size of the metal microparticles is 0.012–0.09 mm.

[0063] In some embodiments, the separation of metal microparticles adopts a combined approach of gravity separation, electromagnetic adsorption and ultrasonic vibration. Copper particles are first separated by gravity separation, and then silver and tin microparticles are separated by a combined approach of electromagnetic adsorption and ultrasonic vibration.

[0064] The magnetic field strength of electromagnetic adsorption is 0.85–1.15T, the magnetic field gradient is 0.12T / mm, the frequency of ultrasonic oscillation is 82kHz, and the power is 110–190W.

[0065] In some embodiments, silicon powder plasma treatment is used to remove residual EVA adhesive and carbon and oxygen impurities from the surface of silicon powder;

[0066] The processing chamber temperature is 62–78℃, the vacuum degree is ≤4.8Pa, the vacuum degree maintenance accuracy is ±0.08Pa, the processing chamber has two built-in plasma generating electrodes with an electrode spacing of 11–14mm, and the vacuum degree is controlled by a two-stage vacuum pump linkage.

[0067] In some embodiments, the plasma treatment power is 310-490W, the duration is 11-14s, and a mixture of argon and hydrogen gas is introduced during the treatment with a volume ratio of 8.5:1.5, the argon purity is ≥99.992%, the hydrogen purity is ≥99.992%, and the gas flow rate is 0.6-1.1L / min.

[0068] Gas is pulsed into the processing chamber through a quartz conduit with an inner diameter of 4.2–5.8 mm at a pulse frequency of 50 Hz. The conduit outlet is 5.5–7.5 mm away from the silicon powder surface.

[0069] Based on the aforementioned embodiments, the following sets of experiments were conducted.

[0070] It should be noted that the raw materials used in the following embodiments are all commercially available.

[0071] The process of the present invention will be described in detail below with reference to specific embodiments. Embodiment 1, S1. Pre-removal treatment of aluminum frame: The aluminum frame of the waste photovoltaic panel is first removed by mechanical clamping and hydraulic separation. The separated aluminum frame is collected and recycled separately. The subsequent process is only for the photovoltaic panel body without aluminum frame.

[0072] The mechanical clamping pressure is 0.4MPa, the hydraulic separation stroke speed is 6.5mm / s, and the aluminum frame separation error is ≤±0.8mm;

[0073] S2. A three-modal AI recognition model integrating visual imaging, ultrasonic thickness detection, and interlayer stress sensing is used to analyze the composition, layer thickness, and interlayer stress distribution of waste photovoltaic panels after the aluminum frame has been removed.

[0074] The structure consists of glass, PET, silicon wafer and EVA, with a layer thickness of 1.5 mm and an interlayer stress distribution of 0.5 MPa;

[0075] The trimodal AI recognition model includes 6 sets of distributed cameras, 2 ultrasonic thickness detection probes, and 4 interlayer stress sensors. Among the 6 sets of distributed cameras, 4 sets are arranged symmetrically at the top and 2 sets are arranged symmetrically on the sides. The detection frequency of the ultrasonic thickness detection probes is 7.5MHz, and the detection range of the interlayer stress sensors is 0 to 1.8MPa.

[0076] The data analysis of the three-modal AI recognition model uses a convolutional neural network algorithm to fuse visual imaging data, ultrasonic thickness data, and interlayer stress data. The data fusion cycle and analysis response time are consistent, both being 0.25 seconds / frame.

[0077] S3. Remove the PET and glass layers from the surface of the waste photovoltaic panel. Use biodegradable ceramic microspheres as the sandblasting medium. During the sandblasting process, AI identifies the material of the sandblasting area in real time through visual imaging. When the EVA layer is detected, the nozzle position is automatically switched to the untreated PET and glass layer area, while adjusting the sandblasting intensity and sandblasting time.

[0078] After sandblasting, the biodegradable ceramic microspheres are graded, recycled, and dried by hot air. The dried biodegradable ceramic microspheres are then reused in the sandblasting process. The grading, recycling, hot air drying, and reuse processes are controlled by AI. The AI ​​provides feedback on process parameters at a cycle of 0.18 seconds.

[0079] The biodegradable ceramic microspheres have a particle size of 100 μm;

[0080] The grading and recycling are achieved using grading sieves with sizes of 55μm, 105μm, and 145μm. The allowable deviation of the sieve aperture is ±0.5μm, and the vibration frequency during grading and recycling is 65Hz.

[0081] The hot air drying process was carried out at a temperature of 50℃ and a wind speed of 2.5m / s, resulting in a relative humidity of 7%RH for the microspheres after drying.

[0082] S4. When the interlayer stress value of the waste photovoltaic panel is ≥0.48MPa, AI triggers a combination of ultrasonic pre-peeling and gradient sandblasting process. The ultrasonic vibration weakens the bonding force between the PET and glass layers, and then gradient sandblasting removes the PET and glass layers.

[0083] The ultrasonic pre-peeling frequency is 50kHz and the duration is 7.5 seconds. The gradient sandblasting pressure is adjusted in steps in the order of 0.45MPa, 0.75MPa and 0.42MPa. The distance between the nozzle and the photovoltaic panel is 65mm. For every 10mm change in distance, a pressure of ±0.09MPa is compensated. The peeling area error is ≤±0.05mm.

[0084] The ultrasonic pre-peeling power is 200W, the vibration amplitude is 0.075mm, and the vibration direction is perpendicular to the photovoltaic panel surface.

[0085] The gradient sandblasting nozzle adopts a 0-360° omnidirectional rotating structure, the nozzle outlet diameter is 3.5mm, the jet velocity of the sandblasting medium is 16m / s, the real-time detection accuracy of the distance between the nozzle and the plate surface is ±0.6mm, and the pressure compensation accuracy is ±0.018MPa.

[0086] S5. After being sandblasted, the photovoltaic panels are conveyed to a shredder to be crushed to a particle size of 2cm. After crushing, the materials are first separated by a vibrating screen to separate metal sand, PET fragments, glass fragments and 2cm photovoltaic panel body fragments.

[0087] The separated 2cm photovoltaic panel fragments are fed into a grinding mill to be processed to a particle size of 0.05mm, and then subjected to AI vision and electromagnetic induction composite detection, metal microparticle separation, silicon powder air classification and silicon powder plasma treatment in sequence.

[0088] AI vision and electromagnetic induction composite inspection includes one high-speed industrial camera and two electromagnetic induction probes.

[0089] The high-speed industrial camera has a frame rate of 210 frames per second, and the image acquisition cycle is synchronized with the electromagnetic induction detection cycle, both being 4.8ms / time.

[0090] The electromagnetic induction probe has a detection frequency of 52kHz and a detection signal sampling rate of 1.05MHz.

[0091] In the AI ​​vision and electromagnetic induction combined inspection, the high-speed industrial camera first acquires images to identify the preliminary location of metal microparticles, and then the electromagnetic induction probe detects the preliminary location. The identified metal microparticles are copper, silver and tin microparticles.

[0092] The particle size of the metal microparticles is 0.05 mm;

[0093] The separation of metal microparticles adopts a combined approach of gravity separation, electromagnetic adsorption and ultrasonic vibration. First, copper particles are separated by gravity separation, and then silver and tin microparticles are separated by electromagnetic adsorption and ultrasonic vibration.

[0094] The magnetic field strength of electromagnetic adsorption is 1.0T, the magnetic field gradient is 0.12T / mm, the frequency of ultrasonic oscillation is 82kHz, and the power is 150W.

[0095] Silicon powder plasma treatment is used to remove residual EVA adhesive and carbon and oxygen impurities from the surface of silicon powder;

[0096] The processing chamber temperature is 70℃, the vacuum degree is ≤4.0Pa, the vacuum degree maintenance accuracy is ±0.06Pa, the processing chamber has two built-in plasma generating electrodes with an electrode spacing of 12.5mm, and the vacuum degree is controlled by a two-stage vacuum pump.

[0097] The plasma treatment power was 400W, the duration was 12.5 seconds, and a mixture of argon and hydrogen gas was introduced during the treatment with a volume ratio of 8.5:1.5. The purity of argon gas was ≥99.992%, the purity of hydrogen gas was ≥99.992%, and the gas flow rate was 0.85L / min.

[0098] Gas is pulsed into the processing chamber through a quartz conduit with an inner diameter of 5.0 mm, at a pulse frequency of 50 Hz, and the conduit outlet is 6.5 mm away from the silicon powder surface.

[0099] Example 2: An AI-guided environmentally friendly recycling process for waste photovoltaic panels, comprising:

[0100] S1. Pre-removal of aluminum frame: The aluminum frame of the waste photovoltaic panel is removed by mechanical clamping and hydraulic separation. The separated aluminum frame is collected and recycled separately. Subsequent processes are only for the photovoltaic panel body without aluminum frame.

[0101] The mechanical clamping pressure is 0.3MPa, the hydraulic separation stroke speed is 5mm / s, and the aluminum frame separation error is ≤±0.5mm.

[0102] S2. A three-modal AI recognition model integrating visual imaging, ultrasonic thickness detection, and interlayer stress sensing is used to analyze the composition, layer thickness, and interlayer stress distribution of waste photovoltaic panels after the aluminum frame has been removed.

[0103] The structure consists of glass, PET, silicon wafer and EVA, with a layer thickness of 0.8 mm and an interlayer stress distribution of 0.3 MPa;

[0104] The trimodal AI recognition model includes 6 sets of distributed cameras, 2 ultrasonic thickness detection probes, and 4 interlayer stress sensors. Among the 6 sets of distributed cameras, 4 sets are arranged symmetrically at the top and 2 sets are arranged symmetrically on the sides. The detection frequency of the ultrasonic thickness detection probes is 5.5MHz, and the detection range of the interlayer stress sensors is 0 to 1.8MPa.

[0105] The data analysis of the three-modal AI recognition model uses a convolutional neural network algorithm to fuse visual imaging data, ultrasonic thickness data, and interlayer stress data. The data fusion cycle and analysis response time are consistent, both being 0.22 seconds / frame.

[0106] S3. Remove the PET and glass layers from the surface of the waste photovoltaic panel. Use biodegradable ceramic microspheres as the sandblasting medium. During the sandblasting process, AI identifies the material of the sandblasting area in real time through visual imaging. When the EVA layer is detected, the nozzle position is automatically switched to the untreated PET and glass layer area, while adjusting the sandblasting intensity and sandblasting time.

[0107] After sandblasting, the biodegradable ceramic microspheres are graded, recycled, and dried by hot air. The dried biodegradable ceramic microspheres are then reused in the sandblasting process. The grading, recycling, hot air drying, and reuse processes are controlled by AI. The AI ​​provides feedback on process parameters at a cycle of 0.18 seconds.

[0108] The biodegradable ceramic microspheres have a particle size of 55 μm;

[0109] The grading and recycling are achieved using grading sieves with specifications of 55μm, 105μm, and 145μm. The allowable deviation of the sieve aperture is ±0.3μm, and the vibration frequency during grading and recycling is 52Hz.

[0110] The hot air drying temperature was 42℃, the wind speed was 2.1m / s, and the relative humidity of the microspheres after drying was 6%RH.

[0111] S4. Because the interlayer stress value of the waste photovoltaic panel is <0.48MPa, the ultrasonic pre-peeling process is not triggered. Only the gradient sandblasting process is used to remove the PET layer and the glass layer.

[0112] The pressure of gradient sandblasting is adjusted in steps in the order of 0.45MPa, 0.75MPa and 0.42MPa. The distance between the nozzle and the photovoltaic panel is 45mm. For every 10mm change in distance, a pressure of ±0.09MPa is compensated. The peeling area error is ≤±0.04mm.

[0113] The gradient sandblasting nozzle adopts a 0-360° omnidirectional rotating structure, with a nozzle outlet diameter of 2.2mm, a sandblasting medium jet velocity of 9m / s, a real-time detection accuracy of the distance between the nozzle and the plate surface of ±0.4mm, and a pressure compensation accuracy of ±0.018MPa.

[0114] S5. After being sandblasted, the photovoltaic panels are conveyed to a shredder to be crushed to a particle size of 2cm. After crushing, the materials are first separated by a vibrating screen to separate metal sand, PET fragments, glass fragments and 2cm photovoltaic panel body fragments.

[0115] The separated 2cm photovoltaic panel fragments are fed into a grinding mill to be processed to a particle size of 0.012mm, and then subjected to AI vision and electromagnetic induction composite detection, metal microparticle separation, silicon powder air classification and silicon powder plasma treatment in sequence.

[0116] AI vision and electromagnetic induction composite inspection includes one high-speed industrial camera and two electromagnetic induction probes.

[0117] The high-speed industrial camera has a frame rate of 210 frames per second, and the image acquisition cycle is synchronized with the electromagnetic induction detection cycle, both being 4.8ms / time.

[0118] The electromagnetic induction probe has a detection frequency of 52kHz and a detection signal sampling rate of 1.05MHz.

[0119] In the AI ​​vision and electromagnetic induction combined inspection, the high-speed industrial camera first acquires images to identify the preliminary location of metal microparticles, and then the electromagnetic induction probe detects the preliminary location. The identified metal microparticles are copper, silver and tin microparticles.

[0120] The particle size of the metal microparticles is 0.012 mm;

[0121] The separation of metal microparticles adopts a combined approach of gravity separation, electromagnetic adsorption and ultrasonic vibration. First, copper particles are separated by gravity separation, and then silver and tin microparticles are separated by electromagnetic adsorption and ultrasonic vibration.

[0122] The magnetic field strength of electromagnetic adsorption is 0.85T, the magnetic field gradient is 0.12T / mm, the frequency of ultrasonic oscillation is 82kHz, and the power is 110W.

[0123] Silicon powder plasma treatment is used to remove residual EVA adhesive and carbon and oxygen impurities from the surface of silicon powder;

[0124] The processing chamber temperature is 62℃, the vacuum degree is ≤3.5Pa, the vacuum degree maintenance accuracy is ±0.04Pa, the processing chamber has two built-in plasma generating electrodes with an electrode spacing of 11mm, and the vacuum degree is controlled by a two-stage vacuum pump.

[0125] The plasma treatment power was 310W, the duration was 11 seconds, and a mixture of argon and hydrogen gas was introduced during the treatment with a volume ratio of 8.5:1.5. The purity of argon gas was ≥99.992%, the purity of hydrogen gas was ≥99.992%, and the gas flow rate was 0.6L / min.

[0126] Gas is pulsed into the processing chamber through a quartz conduit with an inner diameter of 4.2 mm, at a pulse frequency of 50 Hz, and the conduit outlet is 5.5 mm away from the silicon powder surface.

[0127] Example 3: An AI-guided, vision-based environmentally friendly recycling process for waste photovoltaic panels, comprising:

[0128] S1. Pre-removal of aluminum frame

[0129] The aluminum frame of the waste photovoltaic panel is first removed by a combination of mechanical clamping and hydraulic separation. The separated aluminum frame is collected and recycled separately. Subsequent processes are only for the photovoltaic panel body without aluminum frame.

[0130] The mechanical clamping pressure is 0.5MPa, the hydraulic separation stroke speed is 8mm / s, and the aluminum frame separation error is ≤±1mm.

[0131] S2. A three-modal AI recognition model integrating visual imaging, ultrasonic thickness detection, and interlayer stress sensing is used to analyze the composition, layer thickness, and interlayer stress distribution of waste photovoltaic panels after the aluminum frame has been removed.

[0132] The structure consists of glass, PET, silicon wafer and EVA, with a layer thickness of 3mm and an interlayer stress distribution of 0.8MPa;

[0133] The trimodal AI recognition model includes 6 sets of distributed cameras, 2 ultrasonic thickness detection probes, and 4 interlayer stress sensors. Among the 6 sets of distributed cameras, 4 sets are arranged symmetrically at the top and 2 sets are arranged symmetrically on the sides. The detection frequency of the ultrasonic thickness detection probes is 9.5MHz, and the detection range of the interlayer stress sensors is 0 to 1.8MPa.

[0134] The data analysis of the three-modal AI recognition model uses a convolutional neural network algorithm to fuse visual imaging data, ultrasonic thickness data, and interlayer stress data. The data fusion cycle and analysis response time are consistent, both being 0.28 seconds / frame.

[0135] S3. Remove the PET and glass layers from the surface of the waste photovoltaic panel. Use biodegradable ceramic microspheres as the sandblasting medium. During the sandblasting process, AI identifies the material of the sandblasting area in real time through visual imaging. When the EVA layer is detected, the nozzle position is automatically switched to the untreated PET and glass layer area, while adjusting the sandblasting intensity and sandblasting time.

[0136] After sandblasting, the biodegradable ceramic microspheres are graded, recycled, and dried by hot air. The dried biodegradable ceramic microspheres are then reused in the sandblasting process. The grading, recycling, hot air drying, and reuse processes are controlled by AI. The AI ​​provides feedback on process parameters at a cycle of 0.18 seconds.

[0137] The biodegradable ceramic microspheres have a particle size of 145 μm;

[0138] The grading and recycling are achieved using grading sieves with sizes of 55μm, 105μm, and 145μm. The allowable deviation of the sieve aperture is ±0.8μm, and the vibration frequency during grading and recycling is 78Hz.

[0139] The hot air drying process was carried out at a temperature of 58℃ and a wind speed of 2.9m / s, resulting in a relative humidity of 9%RH for the microspheres after drying.

[0140] S4. When the interlayer stress value of the waste photovoltaic panel is ≥0.48MPa, AI triggers a combination of ultrasonic pre-peeling and gradient sandblasting process. The ultrasonic vibration weakens the bonding force between the PET and glass layers, and then gradient sandblasting removes the PET and glass layers.

[0141] The ultrasonic pre-peeling frequency is 58kHz and the duration is 9 seconds. The gradient sandblasting pressure is adjusted in steps in the order of 0.45MPa, 0.75MPa and 0.42MPa. The distance between the nozzle and the photovoltaic panel is 85mm. For every 10mm change in distance, a pressure of ±0.09MPa is compensated. The peeling area error is ≤±0.08mm.

[0142] The ultrasonic pre-peeling power is 240W, the vibration amplitude is 0.09mm, and the vibration direction is perpendicular to the photovoltaic panel surface;

[0143] The gradient sandblasting nozzle adopts a 0-360° omnidirectional rotating structure, the nozzle outlet diameter is 4.8mm, the jet velocity of the sandblasting medium is 24m / s, the real-time detection accuracy of the distance between the nozzle and the plate surface is ±0.8mm, and the pressure compensation accuracy is ±0.018MPa.

[0144] S5. After being sandblasted, the photovoltaic panels are conveyed to a shredder to be crushed to a particle size of 2cm. After crushing, the materials are first separated by a vibrating screen to separate metal sand, PET fragments, glass fragments and 2cm photovoltaic panel body fragments.

[0145] The separated 2cm photovoltaic panel fragments are fed into a grinding mill to be processed to a particle size of 0.09mm, and then subjected to AI vision and electromagnetic induction composite detection, metal microparticle separation, silicon powder air classification and silicon powder plasma treatment in sequence.

[0146] AI vision and electromagnetic induction composite inspection includes one high-speed industrial camera and two electromagnetic induction probes.

[0147] The high-speed industrial camera has a frame rate of 210 frames per second, and the image acquisition cycle is synchronized with the electromagnetic induction detection cycle, both being 4.8ms / time.

[0148] The electromagnetic induction probe has a detection frequency of 52kHz and a detection signal sampling rate of 1.05MHz.

[0149] In the AI ​​vision and electromagnetic induction combined inspection, the high-speed industrial camera first acquires images to identify the preliminary location of metal microparticles, and then the electromagnetic induction probe detects the preliminary location. The identified metal microparticles are copper, silver and tin microparticles.

[0150] The particle size of the metal microparticles is 0.09 mm;

[0151] The separation of metal microparticles adopts a combined approach of gravity separation, electromagnetic adsorption and ultrasonic vibration. First, copper particles are separated by gravity separation, and then silver and tin microparticles are separated by electromagnetic adsorption and ultrasonic vibration.

[0152] The magnetic field strength of electromagnetic adsorption is 1.15T, the magnetic field gradient is 0.12T / mm, the frequency of ultrasonic oscillation is 82kHz, and the power is 190W.

[0153] Silicon powder plasma treatment is used to remove residual EVA adhesive and carbon and oxygen impurities from the surface of silicon powder;

[0154] The processing chamber temperature is 78℃, the vacuum degree is ≤4.8Pa, the vacuum degree maintenance accuracy is ±0.08Pa, the processing chamber has two built-in plasma generating electrodes with an electrode spacing of 14mm, and the vacuum degree is controlled by a two-stage vacuum pump.

[0155] The plasma treatment power was 490W, the duration was 14 seconds, and a mixture of argon and hydrogen gas was introduced during the treatment with a volume ratio of 8.5:1.5. The purity of argon gas was ≥99.992%, the purity of hydrogen gas was ≥99.992%, and the gas flow rate was 1.1L / min.

[0156] Gas is pulsed into the processing chamber through a quartz conduit with an inner diameter of 5.8 mm, at a pulse frequency of 50 Hz, and the conduit outlet is 7.5 mm from the silicon powder surface.

[0157] Comparative Example 1 differs from Example 1 in that it does not use a three-modal AI recognition model, but adopts a traditional mechanical detection method, which only detects the composition structure and layer thickness, and cannot obtain the interlayer stress distribution. The detection response time is 5 seconds / frame. The remaining steps are the same as in Example 1.

[0158] Comparative Example 2 differs from Example 1 in that it does not use biodegradable ceramic microspheres as the blasting medium, but uses ordinary quartz sand as the blasting medium, and does not perform graded recycling and reuse. The remaining steps are the same as in Example 1.

[0159] Comparative Example 3 differs from Example 1 in that it does not trigger the ultrasonic pre-peeling process, but only uses a gradient sandblasting process for surface peeling treatment, while the remaining steps are the same as in Example 1.

[0160] Comparative Example 4 differs from Example 1 in that it does not use plasma treatment for silicon powder purification, but uses a traditional acid washing process for silicon powder purification. The remaining steps are the same as in Example 1.

[0161] Performance testing: Performance tests were conducted on the recycled products and environmental indicators of the processes after treatment in Examples 1, 2, 3, Comparative Examples 1, 2, 3, and 4.

[0162] The test items include silicon powder purity, copper metal recovery rate, silver metal recovery rate, tin metal recovery rate, sandblasting media utilization rate, emissions of harmful gases (VOCs, fluorides), stripping area error, and the percentage of intact silicon powder particles recovered.

[0163] The purity of silicon powder was tested according to GB / T14849.1-2023 "Chemical Analysis Methods for Industrial Silicon - Part 1: Determination of Iron Content - 1,10-Diazaphenanthroline Spectrophotometric Method";

[0164] Metal recovery rate testing was performed according to GB / T3884.21-2018 "Chemical Analysis Methods for Copper Concentrates - Part 21: Determination of Copper, Sulfur, Lead, Zinc, Iron, Aluminum, Calcium, Magnesium and Manganese by Wavelength Dispersive X-ray Fluorescence Spectroscopy";

[0165] The emission of harmful gases was tested in accordance with GB16297-1996 "Integrated Emission Standard for Air Pollutants";

[0166] The peeling area error was directly measured using a laser rangefinder.

[0167] The utilization rate of the blasting media is calculated by the ratio of the recycled and reused mass to the initial input mass;

[0168] The percentage of intact silicon powder particles in the recycled silicon powder was determined by microscopic observation.

[0169] The obtained test data are recorded in Table 1 below:

[0170]

[0171] In the performance test, the purity level of the example was significantly higher than that of Comparative Example 1 and Comparative Example 4 in terms of silicon powder purity. This is because the three-modal AI recognition model accurately analyzes the structure of the photovoltaic panel and has the efficient impurity removal capability of plasma purification technology. In contrast, Comparative Example 1 relies on traditional mechanical detection and lacks accurate data support, while Comparative Example 4 uses traditional acid washing purification method, which has limited impurity removal effect.

[0172] Regarding metal recovery rates, the copper, silver, and aluminum recovery efficiencies of the examples were all at a high level. However, Comparative Example 1 lacked AI-guided precision, resulting in a lack of targeted metal separation and insufficient separation. Comparative Example 3 did not employ an ultrasonic pre-stripping process, making it difficult for metal particles to be completely encased in silicon powder. Therefore, the recovery efficiencies of both examples were significantly lower than those of the examples.

[0173] Regarding the utilization rate of the sandblasting media, the utilization rate of the embodiment is much higher than that of the comparative example 2. The embodiment achieves efficient circulation of the media through the graded recycling of biodegradable ceramic microspheres, hot air drying, and AI linkage reuse mechanism. In contrast, the comparative example 2 uses ordinary quartz sand and does not recycle and reuse it, resulting in serious waste of the media and a significant decrease in utilization rate.

[0174] Regarding harmful gas emissions, the emissions of the embodiment are significantly lower than those of Comparative Example 4, and fully comply with the requirements of GB16297-1996 "Integrated Emission Standard for Air Pollutants". The embodiment does not involve high-temperature pyrolysis and acid washing steps throughout the process, which reduces the generation of harmful gases from the source. Comparative Example 4 uses a traditional acid washing purification process, which is accompanied by the release of a large amount of harmful gases, and the emission indicators far exceed the environmental protection standards.

[0175] Regarding the peeling area error, the error control accuracy of the embodiment is far superior to that of the comparative example. The embodiment achieves precise control of the peeling process through the synergistic effect of AI dynamic pressure compensation and ultrasonic pre-peeling. The comparative example uses traditional mechanical detection, which cannot obtain photovoltaic panel parameters in real time and dynamically adjust the process, resulting in insufficient peeling accuracy and significantly increased error.

[0176] Regarding the percentage of intact silicon powder particles recovered, the proportion of intact particles in the embodiment was significantly higher than that in Comparative Example 1 and Comparative Example 3. Comparative Example 3 lacked an ultrasonic pre-exfoliation process, which made the silicon powder prone to secondary breakage during sandblasting. Comparative Example 1 had a detection lag, which resulted in a poor match between the process parameters and the actual state of the photovoltaic panel, leading to a high silicon powder breakage rate. In contrast, the embodiment effectively reduced silicon powder breakage through AI-precise parameter control of ultrasonic pre-exfoliation pretreatment.

[0177] Comparative analysis reveals that this invention achieves environmentally friendly and efficient recycling of waste photovoltaic panels through the synergistic effects of trimodal AI recognition, biodegradable sandblasting medium recycling, a combination of ultrasonic pre-stripping and gradient sandblasting processes, and AI vision and electromagnetic induction composite detection combined with plasma purification. The AI ​​vision-guided environmental recycling process for waste photovoltaic panels not only significantly improves the purity and metal recovery rate of the recycled products but also reduces environmental treatment costs and harmful gas emissions, while ensuring the stability of process operation and applicability to different types of photovoltaic panels. Therefore, this invention demonstrates that the AI ​​vision-guided environmental recycling process for waste photovoltaic panels has a broader market prospect and is more suitable for widespread adoption.

[0178] In the description of this specification, references to terms such as "an experiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that experiment or example is included in at least one experiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same experiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more experiments or examples.

[0179] The preferred experiments disclosed above are merely illustrative of the invention. These preferred experiments do not exhaustively describe all details, nor do they limit the invention to the specific embodiments described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these experiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize it. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. An AI-guided, vision-based environmentally friendly recycling process for waste photovoltaic panels, characterized in that, include: S1. Pre-removal of aluminum frame: The aluminum frame of the waste photovoltaic panel is removed by mechanical clamping and hydraulic separation. The separated aluminum frame is collected and recycled separately. Subsequent processes are only for the photovoltaic panel body without aluminum frame. The mechanical clamping pressure is 0.3-0.5 MPa, the hydraulic separation stroke speed is 5-8 mm / s, and the aluminum frame separation error is ≤ ±1 mm. S2. A three-modal AI recognition model integrating visual imaging, ultrasonic thickness detection, and interlayer stress sensing is used to analyze the composition, layer thickness, and interlayer stress distribution of waste photovoltaic panels after the aluminum frame has been removed. The composition consists of glass, PET, silicon wafer and EVA, the layer thickness is 0.1 to 3 mm, and the interlayer stress distribution is 0.1 to 0.8 MPa; S3. Remove the PET and glass layers from the surface of the waste photovoltaic panel. Use biodegradable ceramic microspheres as the sandblasting medium. During the sandblasting process, AI identifies the material of the sandblasting area in real time through visual imaging. When the EVA layer is detected, the nozzle position is automatically switched to the untreated PET and glass layer area, while adjusting the sandblasting intensity and sandblasting time. After sandblasting, the biodegradable ceramic microspheres are graded, recycled, and dried by hot air. The dried biodegradable ceramic microspheres are then reused in the sandblasting process. The grading, recycling, hot air drying, and reuse processes are controlled by AI. The AI ​​provides feedback on process parameters at a cycle of 0.18 seconds. S4. When the interlayer stress value of the waste photovoltaic panel is ≥0.48MPa, AI triggers a combination of ultrasonic pre-peeling and gradient sandblasting process. The ultrasonic vibration weakens the bonding force between the PET and glass layers, and then gradient sandblasting removes the PET and glass layers. The ultrasonic pre-peeling frequency is 42-58kHz and the duration is 6-9 seconds. The gradient sandblasting pressure is adjusted in steps in the order of 0.45MPa, 0.75MPa and 0.42MPa. The distance between the nozzle and the photovoltaic panel is 45-85mm. For every 10mm change in distance, a pressure of ±0.09MPa is compensated. The peeling area error is ≤±0.08mm. The interlayer stress value of the waste photovoltaic panel is <0.48MPa, which does not trigger the ultrasonic pre-peeling process. Only the gradient sandblasting process is used to remove the PET layer and the glass layer. S5. After being sandblasted, the photovoltaic panels are conveyed to a shredder to be crushed to a particle size of 2cm. After crushing, the materials are first separated by a vibrating screen to separate metal sand, PET fragments, glass fragments and 2cm photovoltaic panel body fragments. The separated 2cm photovoltaic panel fragments are fed into a grinding mill to be processed to a particle size of 0.012-0.09mm, and then subjected to AI vision and electromagnetic induction composite detection, metal microparticle separation, silicon powder air classification and silicon powder plasma treatment in sequence.

2. The environmentally friendly recycling process for waste photovoltaic panels based on AI vision guidance according to claim 1, characterized in that: The trimodal AI recognition model includes 6 sets of distributed cameras, 2 ultrasonic thickness detection probes, and 4 interlayer stress sensors. Among the 6 sets of distributed cameras, 4 sets are arranged symmetrically at the top and 2 sets are arranged symmetrically on the sides. The detection frequency of the ultrasonic thickness detection probes is 5.5 to 9.5 MHz, and the detection range of the interlayer stress sensors is 0 to 1.8 MPa.

3. The environmentally friendly recycling process for waste photovoltaic panels based on AI vision guidance according to claim 2, characterized in that: The data analysis of the trimodal AI recognition model adopts a convolutional neural network algorithm to fuse visual imaging data, ultrasonic thickness data and interlayer stress data. The data fusion cycle and analysis response time are consistent, both ≤0.28 seconds / frame.

4. The environmentally friendly recycling process for waste photovoltaic panels based on AI vision guidance according to claim 1, characterized in that: The biodegradable ceramic microspheres have a particle size of 55–145 μm; The grading and recycling is achieved using grading sieves with specifications of 55μm, 105μm, and 145μm, with an allowable deviation of ±0.8μm in sieve aperture size, and a vibration frequency of 52 to 78Hz during grading and recycling. The hot air drying process is carried out at a temperature of 42–58°C and a wind speed of 2.1–2.9 m / s. The relative humidity of the microspheres after drying is 6%–9% RH.

5. The environmentally friendly recycling process for waste photovoltaic panels based on AI vision guidance according to claim 1, characterized in that: The ultrasonic pre-stripping power is 160-240W, the vibration amplitude is 0.06-0.09mm, and the vibration direction is perpendicular to the photovoltaic panel surface. The gradient sandblasting nozzle adopts a 0-360° omnidirectional rotating structure, the nozzle outlet diameter is 2.2-4.8mm, the jet velocity of the sandblasting medium is 9-24m / s, the real-time detection accuracy of the distance between the nozzle and the plate surface is ±0.8mm, and the pressure compensation accuracy is ±0.018MPa.

6. The environmentally friendly recycling process for waste photovoltaic panels based on AI vision guidance according to claim 1, characterized in that: The AI ​​vision and electromagnetic induction composite detection includes one high-speed industrial camera and two electromagnetic induction probes. The high-speed industrial camera has a frame rate of 210 frames per second, and the image acquisition cycle is synchronized with the electromagnetic induction detection cycle, both being 4.8 ms / time. The electromagnetic induction probe has a detection frequency of 52kHz and a detection signal sampling rate of 1.05MHz.

7. The environmentally friendly recycling process for waste photovoltaic panels based on AI vision guidance according to claim 6, characterized in that: In the AI ​​vision and electromagnetic induction composite detection, the high-speed industrial camera first acquires images to identify the preliminary location of metal microparticles, and then the electromagnetic induction probe detects the preliminary location. The identified metal microparticles are copper, silver and tin microparticles. The particle size of the metal microparticles is 0.012–0.09 mm.

8. The environmentally friendly recycling process for waste photovoltaic panels based on AI vision guidance according to claim 1, characterized in that: The separation of the metal microparticles adopts a combined approach of gravity separation, electromagnetic adsorption and ultrasonic vibration. First, copper particles are separated by gravity separation, and then silver and tin microparticles are separated by electromagnetic adsorption and ultrasonic vibration. The magnetic field strength of the electromagnetic adsorption is 0.85–1.15T, the magnetic field gradient is 0.12T / mm, and the frequency of the ultrasonic oscillation is 82kHz with a power of 110–190W.

9. The environmentally friendly recycling process for waste photovoltaic panels based on AI vision guidance according to claim 1, characterized in that: The silicon powder plasma treatment is used to remove residual EVA adhesive and carbon and oxygen impurities from the surface of the silicon powder; The processing chamber temperature is 62–78℃, the vacuum degree is ≤4.8Pa, the vacuum degree maintenance accuracy is ±0.08Pa, the processing chamber has two built-in plasma generating electrodes with an electrode spacing of 11–14mm, and the vacuum degree is controlled by a two-stage vacuum pump linkage.

10. The environmentally friendly recycling process for waste photovoltaic panels based on AI vision guidance according to claim 1, characterized in that: The silicon powder plasma treatment has a power of 310-490W and a duration of 11-14s. During the treatment, a mixture of argon and hydrogen is introduced with a volume ratio of 8.5:1.

5. The purity of argon is ≥99.992%, the purity of hydrogen is ≥99.992%, and the gas flow rate is 0.6-1.1L / min. Gas is pulsed into the processing chamber through a quartz conduit with an inner diameter of 4.2–5.8 mm at a pulse frequency of 50 Hz. The conduit outlet is 5.5–7.5 mm away from the silicon powder surface.

Images