A photosensitive resin configuration and verification system and method based on a large language model

Abstract

The application discloses a photosensitive resin configuration and verification system and method based on a large language model. The system comprises a liquid material supply unit, a solid material supply unit, a stirring and mixing unit, a ultraviolet light curing unit, a tensile test unit, and a host computer control system comprising a large language model processing module and a formula optimization module. The large language model processing module is used to extract formula-process-performance triplet information from literature data, construct a vector database, and generate an initial formula based on target performance requirements. The formula optimization module is used to receive measured performance data fed back by the tensile test unit, intelligently search and optimize based on a graph-structured formula search space and a search algorithm, and generate a next round of candidate formula. The application is fully automated and integrated, does not require manual intervention, greatly improves efficiency, and efficiently locates the optimal formula meeting the target performance requirements through the large language model combined with the search algorithm.

Description

Technical Field

[0001] This invention relates to the field of photocurable material preparation and testing technology, and in particular to a system and method for the preparation and verification of photosensitive resins based on a large language model. Background Technology

[0002] With the continuous maturation of additive manufacturing technology, photopolymer 3D printing, with its high-precision molding and excellent surface quality, has shown broad application prospects in aerospace, biomedicine, dental orthodontics, precision manufacturing, and cultural and creative industries. From high-temperature resistant engineering structural components to implantable human tissue scaffolds, the increasingly diversified application scenarios have placed more stringent and specific requirements on the performance of photosensitive resin materials, such as high strength, high toughness, biocompatibility, or special optical properties.

[0003] Photosensitive resins are complex multi-component systems, typically composed of oligomers (prepolymers), monomers (reactive diluents), photoinitiators, and related functional additives (such as leveling agents and defoamers). Currently, the development of resin formulations for specific performance requirements mainly relies on researchers manually mixing these components in different proportions based on experience. Since the final mechanical properties of the material (such as tensile strength, elongation at break, and flexural modulus) are unknown at the initial formulation stage, sample fabrication and secondary curing are necessary before tensile or flexural tests can be conducted using a mechanical testing machine to evaluate the formulation. This traditional "trial and error" development model suffers from significant efficiency bottlenecks: firstly, manual weighing and mixing not only consumes a large amount of manpower but also makes it difficult to avoid operational errors that affect the repeatability of experimental data; secondly, the entire process, from formulation and sample preparation to final performance verification, is lengthy, severely restricting the development speed and iteration efficiency of new materials.

[0004] In existing technologies, some general-purpose automated liquid handling workstations, while capable of dispensing liquids at microliter levels, often lack optimized designs for the rheological properties of high-viscosity photosensitive resins and do not include closed-loop linkage functions for subsequent performance testing. Although existing technologies exist for the preparation and performance verification of photosensitive resins, these technologies do not begin with the preparation of photosensitive resins from basic components. Instead, they first manually identify the oligomers, monomers, and initiators, prepare multiple initial formulations with different proportions, and then verify the performance by mixing these initial formulations.

[0005] The current technology for developing photosensitive resin formulations relies on manual trial and error, which is inefficient. It also lacks automated configuration schemes for multiple physical components and a closed-loop iterative mechanism from formulation generation to performance verification, which are technical problems that urgently need to be solved. Summary of the Invention

[0006] In view of the above problems, the present invention is proposed to provide a system and method for the preparation and verification of photosensitive resin based on a large language model to overcome or at least partially solve the above problems.

[0007] An embodiment of the present invention provides a photosensitive resin preparation and verification system based on a large language model, the system comprising: A liquid material supply unit includes a first extrusion mechanism for conveying low-viscosity monomer materials and a second extrusion mechanism for conveying high-viscosity oligomer materials, as well as a first guide rail structure for carrying containers and transporting them. A solid material supply unit includes a motor-driven screw feeding mechanism, a powder filling container, and a container transport mechanism equipped with a weighing sensor. The mixing unit includes a heatable magnetic stirring control device and a water bath ultrasonic degassing device. The ultraviolet curing unit includes a second guide rail structure on which a curing mold is mounted and an ultraviolet curing device. The tensile testing unit includes a sample transfer mechanism, a sample thickness measurement module, and a motor-driven guide rail for tensile testing. The robotic arm module is disposed between the liquid material supply unit, the solid material supply unit, the mixing unit, the ultraviolet curing unit, and the tensile testing unit, and is used to clamp and transfer the container; it also includes a host computer control system, which is communicatively connected to the liquid material supply unit, the solid material supply unit, the mixing unit, the ultraviolet curing unit, the tensile testing unit, and the robotic arm module. The host computer control system includes a large language model processing module and a recipe optimization module; The large language model processing module is used to extract formula-process-performance triplet information from literature data, construct a vector database, and generate an initial formula based on the target performance requirements. The formulation optimization module receives the measured performance data fed back by the tensile testing unit, performs intelligent search optimization based on the graph-structured formulation search space and search algorithm, and generates the next round of candidate formulations.

[0008] Optionally, the first extrusion mechanism includes a glass syringe, a floating connector, a right-angle adapter, and a linear electrically controlled displacement stage. The linear electrically controlled displacement stage drives the piston rod of the glass syringe to achieve quantitative extrusion of low-viscosity monomer material. The glass syringe is connected to the linear electrically controlled displacement stage through the floating connector and the right-angle adapter.

[0009] Optionally, the second extrusion mechanism includes a metal syringe, a heating sleeve fitted outside the metal syringe, and an air tube communicating with the inside of the metal syringe. The heating sleeve is used to heat the temperature inside the metal syringe to 35-50°C, and the air tube is used to introduce high-pressure gas to extrude the heated high-viscosity oligomer material from the metal syringe.

[0010] Optionally, the first guide rail structure of the liquid material supply unit is provided with a limit switch and a container holder. A weighing sensor is installed at the bottom of the container holder to monitor the weight of the material already put into the container in real time and feed the weight data back to the host computer control system to realize closed-loop control of the feeding amount.

[0011] Optionally, the screw pushing mechanism of the solid material supply unit includes a pusher motor, a screw connector, and a powder filling container. The pusher motor drives the screw to rotate to push the powdered material from the powder filling container into the container.

[0012] Optionally, the magnetic stirring control device of the stirring and mixing unit is configured to heat the container to 40-70°C and drive the stirring magnet inside the container to rotate, with a stirring time of 5-20 minutes; the degassing time of the water bath ultrasonic degassing device is 10-30 minutes.

[0013] Optionally, the sample thickness measurement module of the tensile testing unit includes a drive cylinder, a thickness sensor fixture, and a thickness sensor. The thickness sensor is used to measure the thickness of the sample to be tested to determine its cross-sectional area. The tensile testing unit also includes a tensile sensor and a high-torque motor. The tensile sensor is used to collect force-displacement data in real time during the tensile process, and the high-torque motor is used to drive the fixture to perform the tensile test.

[0014] Another embodiment of the present invention provides a method for configuring and verifying photosensitive resins based on a large language model, the method comprising: S1: Construct a primary knowledge database for the field of photocurable resins, perform layout analysis, table recognition, and structured parsing on the source text of the data, and use a strategy that combines entity recognition and relation extraction to extract information on formulation composition, synthesis process parameters, key performance indicators, and test conditions. Use a domain-pre-trained text embedding model to perform semantic vector encoding on the extracted information and construct a vector database that supports semantic similarity retrieval. S2: Based on the target performance requirements, design prompt words, retrieve matching recipe information from the vector database using a large language model, and generate an initial recipe; S3: Upload the initial formulation to the photosensitive resin preparation and verification system, which includes a host computer control system, and automatically prepare photosensitive resin samples according to the following sub-steps: S31: The liquid material supply unit, according to the proportion of each liquid component in the initial formula, quantitatively extrudes low-viscosity monomer material through the first extrusion mechanism and quantitatively extrudes high-viscosity oligomer material through the second extrusion mechanism into the container; S32: The solid material supply unit quantitatively pushes powdered photoinitiator and additives into the container according to the proportion of each solid component in the initial formula through a screw pushing mechanism; S33: The stirring and mixing unit heats and magnetically stirs the mixture in the container and performs ultrasonic degassing in a water bath to obtain a uniform photosensitive resin composition. S34: The ultraviolet curing unit injects the photosensitive resin composition into the photocuring mold, and cures it by ultraviolet light to obtain a cured sample; S4: The tensile testing unit performs thickness measurement and tensile testing on the cured sample, obtains tensile strength, elongation at break and elastic modulus data, and automatically saves the test data to the host computer control system; S5: Perform deviation analysis between measured performance data and expected performance. Based on the accumulated experimental data, discretize the multidimensional formulation space to construct a graph-structured search space. Use search algorithms to perform intelligent search optimization of the formulation space and generate the next round of candidate formulations. S6: Repeat steps S3 to S5 until a photosensitive resin formulation that meets the target performance requirements is obtained.

[0015] Optionally, the search algorithm in step S5 is the A* algorithm, which uses the currently verified formula as the starting node and the target performance region as the ending point, and guides the search direction through the following comprehensive evaluation function:

[0016] in For passing through the current node The overall estimated cost of reaching the target performance region From the initial node to the current node The actual cumulative experimental cost, For estimation from the current node based on machine learning agent model Heuristic distance to the target performance region.

[0017] Optionally, the formulation composition in step S1 includes monomer type and ratio, oligomer type and ratio, photoinitiator type and concentration, and auxiliary agent type and content; the synthesis process parameters include reaction temperature, reaction time, and photocuring conditions; the key performance indicators include tensile strength, elongation at break, elastic modulus, stress relaxation rate, water absorption rate, and light transmittance.

[0018] The beneficial effects of this invention are that it replaces all manual operations in the traditional "trial and error-verification" process with automated operations. From formulation preparation, mixing, ultrasonic degassing, UV curing to mechanical tensile testing, the entire process is fully automated and integrated without human intervention. This reduces the formulation preparation, sample preparation and performance verification cycle, which originally took several days or even weeks, to several hours, significantly improving the efficiency of new material development. It extracts formulation knowledge from literature through a large language model to generate an initial formulation, and combines a search algorithm to perform intelligent optimization in a graph-structured formulation search space. This efficiently locates the optimal formulation that meets the target performance requirements with far fewer trials than a full-factor experiment. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the structure of a photosensitive resin preparation and verification system based on a large language model according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a liquid material supply unit according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a solid material supply unit according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a stirring and mixing unit according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of an ultraviolet curing unit according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of a tensile testing unit according to an embodiment of the present invention; Figure 7 This is a detailed structural schematic diagram of a tensile testing unit according to an embodiment of the present invention; Figure 8 This is a flowchart illustrating a method for preparing and verifying photosensitive resin based on a large language model, according to an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0021] Figure 1 This is a schematic diagram of the structure of a photosensitive resin preparation and verification system based on a large language model, according to an embodiment of the present invention. Figure 1 As shown, the system of this embodiment includes: The liquid material supply unit 1 includes a first extrusion mechanism for conveying low-viscosity monomer materials and a second extrusion mechanism for conveying high-viscosity oligomer materials, as well as a first guide rail structure for carrying containers and transporting them. Solid material supply unit 2 includes a motor-driven screw pushing mechanism, a powder filling container, and a container transport mechanism equipped with a weighing sensor. The mixing unit 3 includes a heatable magnetic stirring control device and a water bath ultrasonic degassing device. The ultraviolet curing unit 4 includes a second guide rail structure on which a curing mold is mounted and an ultraviolet curing device. Tensile testing unit 5 includes a sample transfer mechanism, a sample thickness measurement module, and a motor-driven guide rail for tensile testing; The robotic arm module is located between the liquid material supply unit 1, the solid material supply unit 2, the mixing and blending unit 3, the ultraviolet curing unit 4, and the tensile testing unit 5, and is used to clamp and transfer containers; it also includes a host computer control system (not shown in the figure), which is communicatively connected to the liquid material supply unit 1, the solid material supply unit 2, the mixing and blending unit 3, the ultraviolet curing unit 4, the tensile testing unit 5, and the robotic arm module. The host computer control system includes a large language model processing module and a recipe optimization module; The large language model processing module is used to extract formula-process-performance triplet information from literature data, construct a vector database, and generate an initial formula based on the target performance requirements. The formulation optimization module receives the measured performance data fed back by the tensile testing unit, performs intelligent search optimization based on the graph-structured formulation search space and search algorithm, and generates the next round of candidate formulations.

[0022] In practical applications, the large language model processing module of this invention pre-constructs a raw knowledge database containing information related to photocurable resins and dental polymer materials, and performs preprocessing operations on the data. Specific operations include layout analysis, table recognition, and full-text structured parsing of the source text to extract key information. Based on this, a prompt-guided model is designed to automatically extract multi-dimensional key information from the structured text, including formulation composition (monomer type and ratio, isocyanate index, chain extender type and content, photoinitiator type and concentration, etc.), synthesis process parameters (reaction temperature, reaction time, catalyst dosage, photocuring conditions, etc.), key performance indicators (tensile strength, elongation at break, elastic modulus, stress relaxation rate, water absorption, light transmittance, etc.), and testing conditions (temperature, humidity, water bath time, testing standards, etc.). A strategy combining entity recognition and relation extraction is adopted to ensure the completeness and accuracy of the formulation-process-performance ternary information, and the reliability of the extraction results is verified through manual sampling. Furthermore, a domain-pre-trained text embedding model is used to encode the extracted formula information, performance descriptions, and experimental conditions into semantic vectors, mapping discrete literature knowledge to a continuous high-dimensional semantic space, and constructing a scalable vector database that supports semantic similarity retrieval.

[0023] This invention utilizes a large language model to obtain the initial formulation of the photosensitive resin with the desired performance, namely the proportions of different oligomers, monomers, photoinitiators, and functional additives. An intelligent resin formulation system is then used to automate the formulation and post-processing of the resin under these component proportions. The mechanical properties of the material are characterized in a tensile testing unit, obtaining performance parameters such as tensile strength and elongation at break, thereby constructing a performance-component relationship database for the photosensitive resin. An optimization algorithm is then used to generate the next set of potentially optimized material proportions based on the database, initiating the next round of formulation performance verification. This forms a closed-loop iterative process of "intelligent recommendation → experimental verification → data feedback → re-recommendation."

[0024] The photosensitive resin configuration and verification system based on a large language model in this invention replaces all manual operations in the traditional "trial and error-verification" process with automated operations. From formulation preparation, mixing, ultrasonic degassing, UV curing to mechanical tensile testing, the entire process is fully automated and integrated without human intervention. The cycle of formulation preparation, sample preparation and performance verification, which originally took several days or even weeks, is compressed to several hours, greatly improving the efficiency of new material development. The system extracts formulation knowledge from literature through a large language model to generate an initial formulation, and combines a search algorithm to intelligently optimize in a graph-structured formulation search space, efficiently locating the optimal formulation that meets the target performance requirements with far fewer experiments than all-factor experiments.

[0025] In an optional embodiment of the present invention, the first extrusion mechanism includes a glass syringe, a floating joint, a right-angle adapter, and a linear electrically controlled displacement stage. The linear electrically controlled displacement stage drives the piston rod of the glass syringe to achieve quantitative extrusion of low-viscosity monomer material. The glass syringe is connected to the linear electrically controlled displacement stage through the floating joint and the right-angle adapter.

[0026] Furthermore, the second extrusion mechanism includes a metal syringe, a heating sleeve fitted outside the metal syringe, and an air pipe communicating with the inside of the metal syringe. The heating sleeve is used to heat the temperature inside the metal syringe to 35-50°C, and the air pipe is used to introduce high-pressure gas to extrude the heated high-viscosity oligomer material from the metal syringe.

[0027] Furthermore, the first guide rail structure of the liquid material supply unit is equipped with a limit switch and a container holder. A weighing sensor is installed at the bottom of the container holder to monitor the weight of the material already added into the container in real time and to feed the weight data back to the host computer control system to achieve closed-loop control of the feeding amount.

[0028] Understandably, the liquid material supply unit is the first working area in the system. Figure 2This is a schematic diagram of the structure of a liquid material supply unit according to an embodiment of the present invention. Figure 2 As shown, the first extrusion mechanism includes a glass syringe 13, a floating connector 14, a right-angle adapter 15, and a linear electrically controlled displacement stage 16. The support baffle 11 supports the fixed guide rail and provides protection; the fixed guide rail 12 fixes the syringe holder and syringe at the same height; the glass syringe 13 is used for loading and extruding the monomer material; the floating connector 14 reduces errors caused by syringe vibration and inaccurate positioning; the right-angle adapter 15 connects the motor displacement stage and the syringe; and the linear electrically controlled displacement stage 16 controls the linear movement of the syringe. The glass syringe 13 contains low-viscosity monomer material, and the linear electrically controlled displacement stage 16 drives the piston rod of the glass syringe 13 to achieve quantitative extrusion of the low-viscosity monomer material. The extrusion volume is calculated from the monomer ratio in the initial formulation.

[0029] The second extrusion mechanism includes a metal syringe 192, a heating sleeve 191 fitted over the metal syringe, and an air tube 193 communicating with the interior of the metal syringe 192. The heating sleeve 191 is used to raise the temperature of the metal syringe; the metal syringe 192 is used to load oligomer material; and the air tube 193 guides airflow to control the extrusion of the oligomer inside the syringe. The metal syringe 192 contains a high-viscosity oligomer material, which is continuously heated to 35-50°C by the surface heating sleeve module to reduce the viscosity of the internal oligomer. A host computer sends a signal to trigger airflow into the air tube, and under the action of air pressure, the internal oligomer material is extruded downwards. The extrusion volume is related to the timing of the trigger signal.

[0030] The liquid material supply unit also includes a first guide rail structure 17 for transporting the filling container, on which a first limit switch 171, a container holder, a weighing sensor 181, and a container cup 182 are installed. The first guide rail structure 17 is used to move the material; the first limit switch 171 is used to control the position and positioning of the material; the container holder and weighing sensor 181 are used to hold the material and weigh the filling material; the container cup 182 is the container for filling the material. The guide rail moves the container sequentially to below each syringe, completing the sequential feeding of each liquid component. A weighing sensor module is installed at the bottom of the container holder, which can monitor the weight of the material put into the container in real time, and is used to provide feedback on the real-time weight of the material that has been added, ensuring the accuracy of the feeding.

[0031] Furthermore, the screw pushing mechanism of the solid material supply unit includes a pusher motor, a screw connector, and a powder filling container. The pusher motor drives the screw to rotate to push the powdered material from the powder filling container into the container.

[0032] Figure 3 This is a schematic diagram of the structure of a solid material supply unit according to an embodiment of the present invention. Figure 3As shown, the optical axis 21 is used to improve the stability of the guide rail movement; the weighing sensor module 221 is used to weigh the powder; the container cup holder 222 is used to hold the material cup; the motor drive module 23 is used to control the movement of the material cup; the support rod 25 is used to fix the height of the pusher motor; the motor fixing part 261 is used to fix the position of the pusher motor; the pusher motor 262 is used to drive the screw connector to rotate and control the filling of powder; the screw connector 263 is used to connect the motor and control the filling of powder; the powder filling container 264 is used to hold the powder material; and the second limit switch 27 is used to determine the position and positioning of the material cup.

[0033] The screw feeding mechanism includes a motor mounting component 261, a feeding motor 262, a screw connector 263, and a powder filling container 264. The feeding motor drives the screw to rotate, moving the powder inside forward and dropping it into the container from the outlet. The amount of material fed can be accurately controlled by controlling the number of rotation steps of the motor. A weighing sensor module 221 is installed at the bottom of the container, which can monitor the change in the mass of the powder fed into the container in real time, thereby improving the accuracy of material feeding.

[0034] Figure 3 The first robotic arm module 24 is used to grip the container from the end of the guide rail of the liquid material supply unit and transfer it to the container holder of the solid material supply unit.

[0035] Furthermore, the magnetic stirring control device of the stirring and mixing unit is configured to heat the container to 40-70°C and drive the stirring magnet inside the container to rotate, with a stirring time of 5-20 minutes; the degassing time of the water bath ultrasonic degassing device is 10-30 minutes.

[0036] Figure 4 This is a schematic diagram of the structure of a stirring and mixing unit according to an embodiment of the present invention. Figure 4 As shown, the mixing unit includes a heatable magnetic stirring control device 32 and a water bath ultrasonic degassing device 33. The magnetic stirring control device 32 is used for heating and stirring the material; the water bath ultrasonic degassing device 33 is used for degassing the material. Figure 4 The second robotic arm module 31 is used to transfer the container from the solid material supply unit to the mixing unit. It is understood that the second robotic arm module 31 can be used in conjunction with... Figure 3 The first robotic arm module 24 represents different working positions of the same robotic arm, and can also be set as multiple independent robotic arm modules according to actual needs.

[0037] Figure 5 This is a schematic diagram of the structure of a UV curing unit according to an embodiment of the present invention. Figure 5As shown, the second guide rail structure 41 is used for the position movement of the mold, the UV curing mold 42 is a customized stretching sample mold, the UV irradiation lamp 43 is used for the curing treatment of the resin, the lamp holder adapter 44 is used to fix the UV curing lamp, and the linear drive module 45 is used for the movement of the UV curing lamp.

[0038] Furthermore, the sample thickness measurement module of the tensile testing unit includes a drive cylinder, a thickness sensor fixture, and a thickness sensor. The thickness sensor is used to measure the thickness of the sample to be tested to determine its cross-sectional area. The tensile testing unit also includes a tensile sensor and a high-torque motor. The tensile sensor is used to collect force-displacement data in real time during the tensile process, and the high-torque motor is used to drive the fixture to perform the tensile test.

[0039] Figure 6 This is a schematic diagram of the tensile testing unit according to an embodiment of the present invention. Figure 7 This is a detailed structural schematic diagram of a tensile testing unit according to an embodiment of the present invention. Figure 6 and Figure 7 As shown, suction cup 51 generates negative pressure to attract objects below it; suction cup adapter 52 connects the suction cup to the motor; sample transfer mechanism 53 places and moves samples; tensile testing module 54 performs tensile tests on samples; sample thickness measurement module 55 tests sample thickness changes; mechanical support structure 56 serves as a fixed bracket for the robotic arm; and drive motor 57 controls the movement of the robotic arm module. It can be understood that suction cup 51, suction cup adapter 52, and drive motor 57 together constitute the robotic arm module, which is used to remove the cured sample from the mold and transfer it to sample transfer mechanism 53.

[0040] The sample transfer mechanism 53 includes a long-distance drive cylinder 531, a miniature cylinder 532, and a sample fixing block 533. The long-distance drive cylinder 531 is used to stretch the sample horizontally; the miniature cylinder 532 is responsible for the vertical movement of the sample; and the sample fixing block 533 is responsible for placing the sample.

[0041] The tensile testing module 54 includes a high-thrust cylinder 541, a cylinder adapter clamp 542, a tensile sensor 543, a clamp adapter 544, a motor-driven guide rail 545, and a high-torque motor 546. The high-thrust cylinder 541 is responsible for clamping the sample and providing the holding force; the cylinder adapter clamp 542 connects the cylinder and the fixed end; the tensile sensor 543 collects mechanical data during the tensile process; the clamp adapter 544 fixes the clamp to the guide rail; the motor-driven guide rail 545 linearly drives the clamp movement; and the high-torque motor 546 provides high torque to achieve high tensile force.

[0042] The sample thickness measurement module 55 includes a drive cylinder 551, a thickness sensor fixture 552, and a thickness sensor 553. The drive cylinder 551 is used for vertical movement of the thickness sensor; the thickness sensor fixture 552 connects the thickness sensor and the cylinder; and the thickness sensor 553 measures the thickness change of the sample.

[0043] The following details the working process of each unit in the photosensitive resin preparation and verification system based on a large language model according to an embodiment of the present invention: (1) Operation of the liquid material supply unit The liquid material supply unit is the first area to start operation. Low-viscosity monomer material is contained in glass syringes, while high-viscosity oligomer material is contained in metal syringes. After the equipment starts operating, the first guide rail structure moves to its final position, moving the container holder to the leftmost area of ​​the guide rail. It stops after triggering a limit switch, and the host computer control system sends a command to start the feeding step. The first guide rail structure moves the container to below the first glass syringe, then stops. Subsequently, the electrically controlled displacement stage of the first set of syringes moves, causing the syringes to extrude downwards. The extrusion volume is calculated from the proportion of the first monomer in the initial formula. A weighing sensor module is installed at the bottom of the container holder to monitor the weight of the material being added in real time, providing feedback on the real-time weight of the added material and ensuring feeding accuracy. After the first displacement stage completes the material feeding, it stops for 2-4 seconds to allow the extruded syringes and the material in the containers to return to a stable state. Then, the first guide rail structure continues to operate, moving the container to below the second syringe, repeating the above steps until the first four sets of monomers have been extruded. Next, the first guide rail structure moves the container below the first metal syringe. The syringe contains a high-viscosity oligomer material, which is continuously heated to 35-50°C using a surface heating sleeve module, reducing the viscosity of the oligomer. The host computer control system sends a signal to trigger airflow into the endotracheal tube. Under air pressure, the oligomer material is extruded downwards. The extrusion volume is related to the timing of the trigger signal, and weight feedback is provided by a weighing sensor at the bottom of the container. After the first oligomer is extruded, operation stops for 2-4 seconds to allow the extruded syringe and the material in the container to return to a stable state. The first guide rail structure then continues to operate, moving the container below the second syringe. The same steps are repeated to extrude the second syringe. The first guide rail structure then moves, moving the container to the far right and triggering a limit switch to stop the movement, completing the first unit's operation.

[0044] (2) Solid material supply unit operation The solid material supply unit is the second working area. Three screws push containers filled with powdered materials such as initiators and additives. When the limit switch at the rightmost end of the first guide rail is triggered, the guide rail of the solid material supply unit begins to move, driving the container back to its original position. The robotic arm module then moves, gripping the container and transferring it to the container holder of the solid material supply unit. After a 2-5 second pause, the guide rail begins to move, moving the container below the first screw feeding module. Once the position is confirmed, the first screw extrusion motor starts to rotate, moving the powder forward and dropping it into the container. The amount of material fed can be accurately controlled by adjusting the number of rotation steps of the motor. Simultaneously, a weighing sensor module is installed at the bottom of the container holder, which can monitor the change in the mass of the powder fed into the container in real time, thereby improving the accuracy of material feeding. After feeding is complete, pause for 2-4 seconds to allow the screw feeding mechanism and the material inside the container to return to a stable state. Then, the guide rail starts running and moves the container below the second screw feeding module, repeating the above steps until all solid powder material has been fed. Once completed, the guide rail moves the container forward and triggers the limit switch at the end, stopping the movement and completing the work of the second unit.

[0045] (3) The mixing unit is working. The mixing and blending unit is the third working area. When the limit switch at the end of the solid material supply unit is triggered, the robotic arm module will move and grip the container, transporting it to the magnetic stirring control device in the mixing and blending unit. After a 2-4 second pause, the module rapidly heats up to 40-70°C, while the internal motor stirs the material inside the container (a stirring magnet is pre-placed inside). After 5-20 minutes of operation, heating and motor rotation cease, and the robotic arm moves the container to the water bath ultrasonic degassing device. This device performs ultrasonic degassing via a water bath, with a degassing time set to 10-30 minutes. After degassing, the robotic arm module removes the container and transports it to the UV curing unit, completing the work of the third unit.

[0046] (4) UV curing unit in operation The UV curing unit is the fourth working area. After the water bath ultrasonic degassing device stops operating, the guide rail of the UV curing unit moves the mold to the leftmost origin position. The robotic arm grips the magnetic recycle rod to collect the magnetic particles from the container, and then pours the mixed photosensitive resin into the mold. After this step, the robotic arm places the container in the recycling area to await cleaning and next use. The host computer control system sends a command to move the mold directly under the UV curing lamp, linearly controls the distance between the UV lamp and the mold, and then controls the UV lamp to light up, completing the resin curing operation. The irradiation time is set between 10-30 minutes. After this step, the guide rail moves the mold to the other end and stops, completing the work of the fourth unit.

[0047] (5) Operation of the tensile testing unit The tensile testing unit is the final working section. When the mold moves to the end of the guide rail, the motor-driven suction cup adsorbs the solidified sample from the mold and moves it to the sample transfer module. Subsequently, a cylinder moves the sample forward, stopping below the sample thickness measurement module. Simultaneously, the cylinder drives the thickness measurement sensor downward to monitor the sample's thickness change. After thickness measurement, the transfer module's cylinder continues to move the sample to the end, where a micro-cylinder moves upward, delivering the sample to the clamps of the tensile testing module. The high-thrust cylinders on both sides of the clamps operate simultaneously to clamp the sample. After this step, the transfer module's micro-cylinder returns to its initial position, detaching from the sample. The host computer control system controls the high-torque motor to rotate and uses the tensile sensor to collect real-time mechanical data. The acquired force-displacement data is saved to the host computer control system, and based on the previously collected thickness data, the cross-sectional area of ​​the material is determined, thereby obtaining data such as the Young's modulus and elongation at break of the material, completing the work of the tensile testing unit.

[0048] The embodiments of the present invention use closed-loop control with precise motor drive and real-time feedback from weighing sensors to avoid subjective errors caused by manual weighing and mixing, ensuring the consistency of each batch of resin formulation and providing a reliable and repeatable experimental basis for subsequent performance evaluation.

[0049] This invention employs a differentiated feeding design for the multi-phase properties of photosensitive resins. For low-viscosity liquids, a stepper motor drives a syringe for precise feeding; for high-viscosity liquids, heating and pneumatic assisted extrusion are used; and for powdered photoinitiators and additives, a screw pusher is used. This effectively solves the problem of automated feeding caused by the large differences in the physical properties of the various components of photosensitive resins.

[0050] Figure 8 This is a flowchart illustrating a method for preparing and verifying photosensitive resin based on a large language model, according to an embodiment of the present invention. Figure 8As shown, the method includes: S1: Construct a primary knowledge database for the field of photocurable resins, perform layout analysis, table recognition, and structured parsing on the source text of the data, and use a strategy that combines entity recognition and relation extraction to extract information on formulation composition, synthesis process parameters, key performance indicators, and test conditions. Use a domain-pre-trained text embedding model to perform semantic vector encoding on the extracted information and construct a vector database that supports semantic similarity retrieval. S2: Based on the target performance requirements, design prompt words, retrieve matching recipe information from the vector database using a large language model, and generate an initial recipe; S3: Upload the initial formulation to the photosensitive resin preparation and verification system, which includes a host computer control system, and automatically prepare photosensitive resin samples according to the following sub-steps: S31: The liquid material supply unit, according to the proportion of each liquid component in the initial formula, quantitatively extrudes low-viscosity monomer material through the first extrusion mechanism and quantitatively extrudes high-viscosity oligomer material through the second extrusion mechanism into the container; S32: The solid material supply unit quantitatively pushes powdered photoinitiator and additives into the container according to the proportion of each solid component in the initial formula through a screw pushing mechanism; S33: The stirring and mixing unit heats and magnetically stirs the mixture in the container and performs ultrasonic degassing in a water bath to obtain a uniform photosensitive resin composition. S34: The ultraviolet curing unit injects the photosensitive resin composition into the photocuring mold, and cures it by ultraviolet light to obtain a cured sample; S4: The tensile testing unit performs thickness measurement and tensile testing on the cured sample, obtains tensile strength, elongation at break and elastic modulus data, and automatically saves the test data to the host computer control system; S5: Perform deviation analysis between measured performance data and expected performance. Based on the accumulated experimental data, discretize the multidimensional formulation space to construct a graph-structured search space. Use search algorithms to perform intelligent search optimization of the formulation space and generate the next round of candidate formulations. S6: Repeat steps S3 to S5 until a photosensitive resin formulation that meets the target performance requirements is obtained.

[0051] The photosensitive resin configuration and verification method based on a large language model in this invention replaces all manual operations in the traditional "trial and error-verification" process with automated operations. From formulation preparation, mixing, ultrasonic degassing, UV curing to mechanical tensile testing, the entire process is fully automated and integrated without human intervention. The cycle of formulation preparation, sample preparation and performance verification, which originally took several days or even weeks, is compressed to several hours, greatly improving the efficiency of new material development. The method uses a large language model to extract formulation knowledge from literature to generate an initial formulation, and combines a search algorithm to intelligently optimize in a graph-structured formulation search space, efficiently locating the optimal formulation that meets the target performance requirements with far fewer experiments than all-factor experiments.

[0052] In an optional embodiment of the present invention, the search algorithm in step S5 is the A* algorithm, which uses the currently verified formula as the starting node and the target performance region as the ending point, and guides the search direction through the following comprehensive evaluation function:

[0053] in For passing through the current node The overall estimated cost of reaching the target performance region. From the initial node to the current node The actual cumulative experimental cost, For estimation from the current node based on machine learning agent model Heuristic distance to the target performance region.

[0054] Specifically, the formulation composition in step S1 includes monomer type and ratio, oligomer type and ratio, photoinitiator type and concentration, and auxiliary agent type and content; the synthesis process parameters include reaction temperature, reaction time, and photocuring conditions; the key performance indicators include tensile strength, elongation at break, elastic modulus, stress relaxation rate, water absorption rate, and light transmittance.

[0055] After configuring and verifying a set of formulations in this embodiment of the invention, the actual experimental results (including the proportions of each component, the intensity and duration of UV curing, the tensile strength of the sample, and the data on the moving distance) are transmitted back to the host computer control system for deviation analysis compared with the expected performance. Based on the accumulated experimental data, the multidimensional formulation space is discretized into a graph-structured search space, where each node represents a specific combination of formulation parameters, the edges between nodes represent single-step adjustment operations of the formulation parameters, and the cost function of the edges is jointly defined by the degree of performance deviation from the target caused by the adjustment and the experimental cost.

[0056] Building upon this, the A* algorithm is employed for intelligent search and optimization of the recipe space. The A* algorithm is a graph search or path planning algorithm based on a heuristic evaluation function, incorporating actual path costs. and target estimated cost ,pass The algorithm selects nodes (i.e., combinations of experimental parameters) for priority expansion. It starts with currently validated formulations as the initial node and ends with the formulation region that meets the target performance requirements, through a comprehensive evaluation function. Guiding the search direction. During the search process, the algorithm calculates f(n) for all nodes to be searched and prioritizes the node with the smallest f(n) for further expansion. Therefore, the search direction tends to favor nodes that simultaneously satisfy "low cost of the already traversed path" and "closer to the estimated target," thereby reducing invalid searches and improving path search efficiency. Heuristic function The construction utilizes an alternative model trained with existing experimental data to predict the distance from the current recipe node to the target performance, ensuring the permissibility of the heuristic function.

[0057] In each iteration, the A* algorithm selects the recipe node with the smallest comprehensive evaluation function value from the open list as the priority experiment target for the next round. The experimental results of each round are used to update the actual performance data and neighborhood edge costs of the corresponding node in the graph structure, simultaneously optimizing the estimation accuracy of the heuristic function, forming a closed-loop iterative process of "intelligent recommendation → experimental verification → data feedback → graph structure and heuristic function update → re-recommendation". As the number of iterations increases, the verified nodes in the search graph gradually expand, the performance estimation of the unknown recipe region by the heuristic function becomes more accurate, and the algorithm's search path continuously converges towards the globally optimal recipe, thus efficiently locating the optimal photocuring recipe that meets the target performance requirements with far fewer trials than a full-factor experiment.

[0058] It should be noted that: Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0059] The above description is merely a specific embodiment of the present invention. Under the guidance of the above teachings of the present invention, those skilled in the art can make other improvements or modifications based on the above embodiments.

Claims

1. A system for configuring and verifying photosensitive resin based on a large language model, characterized in that, include: A liquid material supply unit includes a first extrusion mechanism for conveying low-viscosity monomer materials and a second extrusion mechanism for conveying high-viscosity oligomer materials, as well as a first guide rail structure for carrying containers and transporting them. A solid material supply unit includes a motor-driven screw feeding mechanism, a powder filling container, and a container transport mechanism equipped with a weighing sensor. The mixing unit includes a heatable magnetic stirring control device and a water bath ultrasonic degassing device. The ultraviolet curing unit includes a second guide rail structure on which a curing mold is mounted and an ultraviolet curing device. The tensile testing unit includes a sample transfer mechanism, a sample thickness measurement module, and a motor-driven guide rail for tensile testing. A robotic arm module is disposed between the liquid material supply unit, the solid material supply unit, the mixing unit, the ultraviolet curing unit, and the tensile testing unit, and is used to clamp and transfer containers. It also includes a host computer control system, which is communicatively connected to the liquid material supply unit, solid material supply unit, stirring and mixing unit, ultraviolet curing unit, tensile testing unit and robotic arm module; The host computer control system includes a large language model processing module and a recipe optimization module; The large language model processing module is used to extract formula-process-performance triplet information from literature data, construct a vector database, and generate an initial formula based on the target performance requirements. The formulation optimization module receives the measured performance data fed back by the tensile testing unit, performs intelligent search optimization based on the graph-structured formulation search space and search algorithm, and generates the next round of candidate formulations.

2. The photosensitive resin preparation and verification system based on a large language model according to claim 1, characterized in that, The first extrusion mechanism includes a glass syringe, a floating connector, a right-angle adapter, and a linear electrically controlled displacement stage. The linear electrically controlled displacement stage drives the piston rod of the glass syringe to achieve quantitative extrusion of low-viscosity monomer material. The glass syringe is connected to the linear electrically controlled displacement stage through the floating connector and the right-angle adapter.

3. The photosensitive resin preparation and verification system based on a large language model according to claim 1, characterized in that, The second extrusion mechanism includes a metal syringe, a heating sleeve fitted outside the metal syringe, and an air tube communicating with the inside of the metal syringe. The heating sleeve is used to heat the temperature inside the metal syringe to 35-50°C, and the air tube is used to introduce high-pressure gas to extrude the heated high-viscosity oligomer material from the metal syringe.

4. The photosensitive resin preparation and verification system based on a large language model according to claim 1, characterized in that, The first guide rail structure of the liquid material supply unit is equipped with a limit switch and a container holder. A weighing sensor is installed at the bottom of the container holder to monitor the weight of the material already added in the container in real time and feed the weight data back to the host computer control system to realize closed-loop control of the feeding amount.

5. The photosensitive resin preparation and verification system based on a large language model according to claim 1, characterized in that, The screw pushing mechanism of the solid material supply unit includes a pusher motor, a screw connector, and a powder filling container. The pusher motor drives the screw to rotate to push the powdered material from the powder filling container into the container.

6. The photosensitive resin preparation and verification system based on a large language model according to claim 1, characterized in that, The magnetic stirring control device of the stirring and mixing unit is configured to heat the container to 40-70℃ and drive the stirring magnet inside the container to rotate, with a stirring time of 5-20 minutes; the degassing time of the water bath ultrasonic degassing device is 10-30 minutes.

7. The photosensitive resin preparation and verification system based on a large language model according to claim 1, characterized in that, The sample thickness measurement module of the tensile testing unit includes a drive cylinder, a thickness sensor fixture, and a thickness sensor. The thickness sensor is used to measure the thickness of the sample to be tested to determine its cross-sectional area. The tensile testing unit also includes a tensile sensor and a high-torque motor. The tensile sensor is used to collect force-displacement data in real time during the tensile process, and the high-torque motor is used to drive the fixture to perform the tensile test.

8. A method for preparing and verifying photosensitive resin based on a large language model, characterized in that, include: S1: Construct a primary knowledge database for the field of photocurable resins, perform layout analysis, table recognition, and structured parsing on the source text of the data, and use a strategy that combines entity recognition and relation extraction to extract information on formulation composition, synthesis process parameters, key performance indicators, and test conditions. Use a domain-pre-trained text embedding model to perform semantic vector encoding on the extracted information and construct a vector database that supports semantic similarity retrieval. S2: Based on the target performance requirements, design prompt words, retrieve matching recipe information from the vector database using a large language model, and generate an initial recipe; S3: Upload the initial formulation to the photosensitive resin preparation and verification system, which includes a host computer control system, and automatically prepare photosensitive resin samples according to the following sub-steps: S31: The liquid material supply unit, according to the proportion of each liquid component in the initial formula, quantitatively extrudes low-viscosity monomer material through the first extrusion mechanism and quantitatively extrudes high-viscosity oligomer material through the second extrusion mechanism into the container; S32: The solid material supply unit quantitatively pushes powdered photoinitiator and additives into the container according to the proportion of each solid component in the initial formula through a screw pushing mechanism; S33: The stirring and mixing unit heats and magnetically stirs the mixture in the container and performs ultrasonic degassing in a water bath to obtain a uniform photosensitive resin composition. S34: The ultraviolet curing unit injects the photosensitive resin composition into the photocuring mold, and cures it by ultraviolet light to obtain a cured sample; S4: The tensile testing unit performs thickness measurement and tensile testing on the cured sample, obtains tensile strength, elongation at break and elastic modulus data, and automatically saves the test data to the host computer control system; S5: Perform deviation analysis between measured performance data and expected performance. Based on the accumulated experimental data, discretize the multidimensional formulation space to construct a graph-structured search space. Use search algorithms to perform intelligent search optimization of the formulation space and generate the next round of candidate formulations. S6: Repeat steps S3 to S5 until a photosensitive resin formulation that meets the target performance requirements is obtained.

9. The method for preparing and verifying photosensitive resin based on a large language model according to claim 8, characterized in that, The search algorithm described in step S5 is the A* algorithm, which uses the currently verified formula as the starting node and the target performance region as the ending node, and guides the search direction through the following comprehensive evaluation function: in For passing through the current node The overall estimated cost of reaching the target performance region. From the initial node to the current node The actual cumulative experimental cost, For estimation from the current node based on machine learning agent model Heuristic distance to the target performance region.

10. The method for preparing and verifying photosensitive resin based on a large language model according to claim 8, characterized in that, The formulation composition in step S1 includes monomer type and ratio, oligomer type and ratio, photoinitiator type and concentration, and auxiliary agent type and content; the synthesis process parameters include reaction temperature, reaction time, and photocuring conditions; the key performance indicators include tensile strength, elongation at break, elastic modulus, stress relaxation rate, water absorption rate, and light transmittance.

Images