212 results about "Product defect" patented technology

The invention provides a product defect tracing method based on knowledge graph reasoning. The method comprises the following steps: constructing a knowledge graph based on a product hierarchical structure of a bill of materials and a manufacturing process hierarchical structure of a process route; when it is detected that data of a certain node in the knowledge graph is changed, calculating an update priority weight of a change node based on the occurrence frequency of the change node in a historical defect path and the hierarchy depth in the double-hierarchy structure; determining an updating range according to the updating priority weight, and updating the knowledge graph; and positioning defect nodes in the updated knowledge graph according to system response, executing path search from the defect nodes, identifying potential root cause nodes, and realizing product defect tracing. According to the method, accurate and real-time defect traceability of the product is realized based on multi-level feature correlation analysis of causal reasoning and dynamic knowledge graph construction and enhanced reasoning.

The invention relates to the technical field of industrial mechanical arm control, in particular to a mechanical arm execution control method for industrial production, which comprises the following steps: collecting parameters in real time; generating a geometric risk index; generating a mechanical risk index; extracting image features; abnormal mechanical arm judgment; determining an adjusting mechanical arm; and generating an adjustment instruction. According to the method, the inclination angle, the plane offset distance, the contact image features, the tightening torque and the force value of the tail end of the mechanical arm are monitored in real time in a multi-dimensional mode, a dual-risk judgment model based on geometric offset and mechanical loading is constructed, image texture fluctuation and posture change are further combined, the continuously abnormal mechanical arm is accurately positioned, and the safety of the mechanical arm is improved. And geometric and mechanical risk indexes are evaluated again after adjustment, so that the aluminum shell positioning and bolt assembling precision is guaranteed, and the problems that the production efficiency is reduced and the product defect rate is increased due to assembling quality fluctuation caused by mechanical arm positioning errors and inaccurate force control are effectively solved.

The invention provides a product structure safety evaluation method considering detection probability, which comprises the following steps of: automatically establishing a product finite element model in batches under different defect distribution by adopting a Monte Carlo method by constructing a joint probability model fusing defect existence probability and nondestructive testing detection probability indexes; calculating M integral and regularization damage parameter pi of the model under each defect distribution condition by combining material configuration mechanical energy intensity indexes, determining the influence rule of uncertainty on intensity index probability distribution, and establishing the incidence relation among the random defect distribution condition of the product, the mechanical performance and the structural reliability; and effective analysis and quality evaluation of product defects are realized. According to the method, a nondestructive testing detection probability POD evaluation system is creatively combined with the M integral and regularization parameters of the configuration mechanics, the limitation of a traditional reliability evaluation method is broken through, the influence of uncertainty on the structural strength is quantitatively detected, the precision and efficiency of product reliability evaluation are remarkably improved, and the method is suitable for popularization and application. And effective technical support is provided for structural safety design.

The invention relates to the technical field of intelligent manufacturing equipment, in particular to a mold cavity pressure homogenization precision control method and system based on a two-way closed loop, and the method comprises the following steps: based on an injection mold cavity, completing the sudden change recognition and compensation of pressure data, the dynamic switching of a master closed loop path and a slave closed loop path, and the discrimination of the synchronous adjustment state of a driver in sequence; and setting a gain adjustment action in stages, and implementing closed-loop correction according to a feedback trend to obtain a dynamic offset correction amount. According to the method, continuous correction of the pressure state is achieved by correlating multi-point pressure collection, judging the pressure sudden change position in the cavity in real time and dynamically compensating local abnormal data, the motion state of the propelling device is combined, pressure adjusting strategies are switched according to different process stages, a linkage gain adjusting and controlling mode is adopted, dynamic adjustment of pressure distribution keeps high collaboration, and the control effect is good. Periodic trend closed-loop correction is carried out on feedback signals in the driving response stage, and forming errors and product defects caused by abnormal local pressure are remarkably reduced.

According to the pole piece thickness control method, a dynamic regulation and control mechanism is constructed by detecting the thickness of a pole piece in real time and calculating the thickness deviation in real time, once the thickness deviation is larger than a preset threshold value, a system can give out an instruction in real time, a rolling mechanism is driven to adjust a roll gap, and it is ensured that the rolling thickness of the follow-up pole piece is stable and consistent; the requirements on the thickness consistency and stability of the pole piece are met, so that the product defects are reduced, and the product qualification rate is improved; reworking and scrapping are reduced, the production cost is reduced, meanwhile, the downtime is shortened, and the production efficiency is improved.

The invention relates to the technical field of coating, and particularly discloses a digital twinning-based coating production line intelligent decision-making method, which is applied to a coating production line and comprises the following steps: establishing an optimized coating digital twinning model based on the coating production line; real-time coating production line change operation data are obtained, and the real-time coating production line change operation data comprise the production takt, the equipment operation state, the product defect information, the inventory time of a buffer area and the vehicle body sequence state; and inputting the real-time coating production line change operation data into an optimized coating digital twinborn model to obtain coating production scheduling optimization information. According to the invention, the digital twinning-based intelligent decision-making system for the coating production line is constructed, so that multiple beneficial effects are realized. By establishing the optimized coating digital twin model, the physical layout, the process logic and the equipment state of the production line are accurately mapped, and the problem that a traditional static model is separated from actual production is solved.

The invention relates to the technical field of electronic product defect detection, in particular to an electronic product defect detection method based on multi-dimensional data fusion, and the method comprises the steps: selecting a target region circuit board, and carrying out the unit segmentation based on a specific circuit board image to obtain a region three-dimensional image; analyzing the image abnormal region based on defect features to determine a fatal defect or a non-fatal defect; the actual defect density of the circuit board is obtained based on the number of non-fatal defect categories and the reference area of the circuit board, and defect influence values corresponding to the non-fatal defect categories are determined based on the actual defect density and the standard defect density interval; and determining the defect grade of the electronic product based on the defect influence value and the corresponding defect influence weight. The method has the advantages of high detection accuracy, high automation degree, strong adaptability and the like, has wide application prospects in the fields of electronic product manufacturing, quality control and the like, can effectively improve the product quality and the production efficiency, and promotes the development of the electronic manufacturing industry.

The utility model provides a kind of mistaken position prevention's tire mould, belong to tire production and processing technical field, this mistaken position prevention's tire mould, including mould;Sliding block, sliding block fixedly connected in the circumference inner wall of mould;Upper disc, upper disc is connected in the upper end of mould;Connecting seat, connecting seat is connected in mould, and the side end of connecting seat is equipped with sliding slot, and sliding slot is matched with sliding block, through the matching of sliding block and sliding slot, and the accurate cooperation of clamping block and socket, the alignment when mould is closed is ensured, product defect caused by mould misplacement is reduced, the height of connecting seat and mould is equal, and through bolt screw thread connection upper disc and clamping block, ensure the stability of mould in mould pressing forming process.

The present application relates to the technical field of industrial robot control, and more particularly to a mechanical arm execution control method for industrial production, which comprises the following steps: collecting parameters in real time; generating a geometric risk index; generating a mechanical risk index; image feature extraction; abnormal mechanical arm determination; adjusting mechanical arm determination; and generating an adjustment instruction. Through real-time monitoring of the tilt angle, plane offset distance, contact image features, and torque and force values of the mechanical arm end, a dual risk judgment model based on geometric offset and mechanical loading is constructed, and further combined with image texture fluctuation and posture change, the continuously abnormal mechanical arm is accurately positioned, and after adjustment, the geometric and mechanical risk indexes are re-evaluated, thereby ensuring the aluminum shell positioning and bolt assembly precision, effectively solving the problems of assembly quality fluctuation due to mechanical arm positioning error and inaccurate force control, and thus reducing production efficiency and increasing product defect rate.

The invention relates to the technical field of product defect detection, in particular to a circuit board finished product defect detection device which comprises an overturning part, and the overturning part comprises a climbing part, a rotating roller, a pushing part and symmetrically-arranged supporting assemblies which are sequentially arranged in the moving direction of a circuit board. The supporting assembly comprises a servo motor, a connecting structure is arranged at the output end of the servo motor, a supporting plate is rotationally arranged on the connecting structure, a gear is arranged on the supporting plate, and a clamping assembly is arranged on the connecting structure; the clamping assembly comprises a hydraulic rod, an elastic telescopic rod is arranged at the telescopic end of the hydraulic rod, a flexible clamping plate is arranged at the telescopic end of the elastic telescopic rod, and a rack is arranged on the elastic telescopic rod; by arranging the overturning part composed of the climbing structure, the rotating roller and the pushing structure, the circuit board can be naturally lifted and overturned around the fulcrum in the conveying process, and the defects caused by a traditional clamping or adsorption mode are overcome.

The application discloses an industrial product defect open set detection method based on instance-level features and variation constraints, constructs a detection model composed of a multi-scale feature extraction network, a Transformer set prediction decoding module and a probability detection head, and discretizes dense features into instance-level query features; by introducing a variation Gaussian constraint and feature space interpolation regularization, the polarization divergence of known category features is limited, the inter-class decision boundary is softened, and the generalization ability of the model to unknown defects is improved. In the detection stage, the energy score is constructed by using the non-normalized category response value, and double-gated discrimination is carried out in combination with the maximum response probability, so that unknown defect identification and abnormal alarm are realized. The application can effectively reduce the false positive rate and the false negative rate in a complex industrial scene, and improve the accuracy and reliability of the open set defect detection of industrial products.

The invention discloses a plastic product defect online detection method based on machine vision, and relates to the technical field of plastic product defect detection.The method comprises the following steps that a unified shooting environment is configured, and a surface image of a plastic product is collected through an industrial camera; image processing is carried out on the surface image, and a surface effective image is obtained after processing is completed; performing feature analysis on the surface effective image, and extracting surface pixel characteristics of the plastic product in the surface effective image; analyzing normal pixel characteristics of the surface of the plastic product based on the normal sample, and analyzing a defect identification standard of the plastic product based on the normal pixel characteristics; performing defect detection on the plastic product based on the defect identification standard; the method is used for solving the problem that an existing plastic product defect detection technology is not comprehensive enough in consideration of the application range of color deviation, and consequently, the defect detection result of the plastic product is prone to errors.

The thin-wall plastic mold cavity structure comprises a male mold and a female mold which are matched with each other, a cavity is formed between the male mold and the female mold, the communicated exhaust grooves are formed in the periphery of the cavity, and each exhaust groove sequentially comprises a buffer section, a first gradually-changed exhaust groove unit and a second gradually-changed exhaust groove unit in the gas exhaust direction. The cross sectional area of the first gradually-changed exhaust groove unit is gradually increased to form a diffusion channel, the cross sectional area of the second gradually-changed exhaust groove unit is gradually decreased to form a convergence channel and generate back pressure, and the buffer section can prevent the melt front from rushing into the exhaust groove. Through cooperation of all the structures, all-region uniform exhaust is achieved, the problems of uneven exhaust, melt overflow and the like of an existing mold are effectively solved, product defects are reduced, the high-speed injection molding requirement of thin-wall parts is met, the structure is simple, machining is easy, maintenance is convenient and fast, universality is high, the production cost and the product reject ratio are reduced, and good practicability and popularization value are achieved.

The present disclosure relates to a training method and device of a defect detection model, and a defect detection method of an industrial product, and relates to the technical field of industrial product defect detection. The method comprises: performing defect detection and image reconstruction on historical defect detection images based on a network model to be trained, to obtain defect detection results and image reconstruction results; determining a first loss value according to the defect detection results and defect annotation results of the historical defect detection images; determining a second loss value according to the historical defect detection images and the image reconstruction results, and determining a total loss value according to the first loss value and the second loss value; training the network model to be trained based on the total loss value, to obtain a trained network model, and determining a product defect detection model based on the trained network model. The present disclosure improves the accuracy of the product defect detection model.

A method, apparatus, storage medium, and terminal for determining training datasets based on AIGC are disclosed. The method includes: inputting a first training dataset into an initial model for preliminary iterative training to obtain a pre-trained model; inputting a second training dataset into the pre-trained model for further iterative training to obtain a retrained model, wherein the second training dataset is obtained by expanding the first training dataset based on AIGC technology; inputting multiple images to be labeled into the retrained model for product defect prediction, and labeling the images to be labeled using the product defect prediction results to obtain a third training dataset; wherein the ratio of the number of images in the second training dataset to the first training dataset is greater than or equal to 10, and the ratio of the convergence threshold of the first loss function used in the preliminary iterative training to the second loss function used in the subsequent iterative training is greater than or equal to 2. This scheme can quickly obtain a large number of training datasets labeled with accurate product defect tags.

The invention discloses a product quality defect analysis method and device and a medium, and relates to the technical field of data processing. The method comprises the following steps: firstly, establishing a prediction model from a raw material performance index to a product quality index based on historical data by using a random forest algorithm; when product defects are detected, initial raw material data of the batch of products are obtained and subjected to multi-round random disturbance, a large number of virtual samples are generated, and the quality of the virtual samples is predicted through a model. Then, qualified samples are screened out from the virtual samples, and a sample subset with the highest model prediction reliability is further selected from the samples. And finally, performing statistics on the consistency of the disturbance direction of each raw material performance index in the subset, and judging the raw material index with the highest consistency as a key factor which is most likely to cause the current quality defect. Noise in industrial data can be effectively processed, product quality problems can be quickly and accurately attributed to specific raw material characteristics, and accurate guidance is provided for production control and process optimization.

The application discloses a kind of visual inspection methods based on inductance surface defect, first color image is carried out color channel decomposition, using the gray scale of product image carries out template positioning, obtains coordinate;Second, according to the relative area of positioning coordinate, then carry out morphological operation;Finally, through gray scale stretching, the contrast of defect and background is enhanced, so that it finds out product defect and carries out elimination, obtains quality qualified product.The detection result obtained by the above-mentioned detection method of the application reduces the misjudgment rate, improves the yield, at the same time meets the control of product quality of customer, improves the detection efficiency of product.As the gray scale linear variation of part of defects in different spatial domain is more obvious, the application decomposes the image of different spatial domain through different color component information, so that the defect is more easily obtained and detected.

This invention discloses a comprehensive thread production device, relating to the field of thread production technology. The invention includes a fixed base, a spindle box fixedly connected to the top of the fixed base, control buttons on the side of the spindle box, a three-jaw chuck rotatably connected to the side of the spindle box, a machine base fixedly connected to the side of the spindle box, a guide rail fixedly connected to the top of the machine base, a concentric mechanism on the top of the guide rail, a baffle fixedly connected to the side of the machine base, and a chip groove formed on the inner wall of the machine base. This invention uses a motor-driven drive gear and driven gear to precisely control the translational stroke of the moving seat, and with the double sliding guidance of the guide rail, the top cone sleeve and the three-jaw chuck maintain strict coaxiality, solving the problem of workpiece clamping eccentricity, improving the dimensional and positional accuracy of thread processing, and reducing the product defect rate.

The invention discloses a stamping process-oriented defect feedback and digital simulation closed-loop control system and method, and relates to the technical field of closed-loop control. The system comprises a working condition synchronous sensing and feature coding module, a digital simulation prediction module, a defect feedback and decision control module, a process parameter execution module and an online learning and model evolution module. According to the method, multi-source operation data are collected, a working condition feature packet is generated, material parameters are dynamically matched after a formed digital simulation model is input, initial disturbance is injected, and defect prediction parallel to physical stamping is achieved; calculating a prediction confidence coefficient by combining measured data, and generating a precise process adjustment decision by linking defect risk distribution and applying the decision to a next period; and based on actually measured and predicted errors, continuously updating a material parameter library and forming digital simulation model parameters through online learning. According to the method, the problems that traditional stamping simulation and reality are disjointed, and process adjustment is blind are effectively solved, the product defect rate is greatly reduced, and meanwhile the production efficiency and the quality stability are improved.

The utility model discloses a laser dadoing compound machine which comprises a supporting seat, a machining table is arranged at the top end of the supporting seat, and a plurality of fixing clamps which are distributed at equal intervals and used for clamping workpieces are arranged at the top end of the machining table. According to the utility model, the dadoing processing part and the laser processing part are slidably mounted on the vertical mounting plate through the driving sliding assembly, so that two different processing devices are combined into a whole according to respective characteristics, only one compound machine needs to be purchased for processing the same product, the purchase cost is reduced, and the production efficiency is improved. The same machining table and fixing clamps are used for laser machining and dadoing machining, the multiple fixing clamps distributed at equal intervals enable machining of two working procedures to be completed through one-time clamping, time is saved, efficiency is improved, the dadoing machining part moves firstly to machine products, the machining sequence is changed into the mode that dadoing machining is conducted firstly, then laser cutting is conducted, and the machining efficiency is improved. The loss of the planer tool and the reject ratio of products are reduced, and the situation that positioning is inconvenient due to the special shape is avoided.

The present application relates to the technical field of textile production, in particular to a processing method for defect samples in light-weight product defect detection, aiming to solve the technical problems of knitted product defect sample scarcity, insufficient capture of micro-scale defects, easy confusion in similar defect classification, and mismatch between generated defect positions and industrial reality, and sequentially performing image acquisition preprocessing, defect recognition and classification, improved CycleGAN model construction, iterative training, sample generation and screening, and effect verification; the present application realizes unpaired batch generation and expansion of defect samples, improves the matching degree between generated defect sample positions and industrial reality, and adapts the obtained standardized sample set to HCF-YOLO light-weight defect detection model training, greatly improving the industrial practicality of samples and the industrial application effect of the defect detection model.

The application relates to the technical field of heater detection, and discloses an automatic air tightness and insulation voltage resistance detection method for a high-pressure heater product, which comprises the following steps: step S1 product association and parameter calling, step S2 composite air tightness detection, step S3 graded insulation voltage resistance detection, step S4 construction of a traceable data link, and step S5 product shunting and feedback data optimization process. The two-stage composite air tightness detection and the graded insulation voltage resistance detection are cooperatively designed, the data of a production management system and a detection system are synchronized, a complete traceable data link which is deeply bound by a product, detection and a production process is constructed, the detection precision and efficiency are cooperatively improved, the nature of product defects is accurately positioned and determined, the causes of defects are quickly traced, a precise data basis is built for production process optimization, and the problems that detection and production data are split and defect tracing is difficult in the prior art are effectively solved.

The application discloses a model parameter automatic adjustment method and device. The method comprises the following steps: when the model performance index of a product defect detection model is lower than a preset performance threshold, obtaining a plurality of groups of inferred data of the product defect detection model, wherein each group of inferred data comprises a defect detection result corresponding to an image identifier and a detection state, and the defect detection result comprises a defect category and a detection parameter group corresponding to each defect detection frame in a product image corresponding to the image identifier; for each defect category preset in the product defect detection model, determining a plurality of detection parameter groups corresponding to the defect category from the plurality of groups of inferred data, and determining a model performance index corresponding to each detection parameter group; and taking the detection parameter group with the highest model performance index as a target detection parameter group corresponding to the defect category. The application solves the technical problems of low efficiency and poor accuracy in manually adjusting the model parameters of the product defect detection model.

The invention provides a classification method and system based on category similarity fusion training, and relates to the technical field of computer vision and industrial automation, and the method comprises the steps: obtaining defect detection data sets of different products, and carrying out the preprocessing of the defect detection data sets; performing double-layer similarity cross-product sample fusion based on the preprocessed defect detection data set to obtain a fusion data set; performing general classification model training based on the fused data set to obtain a general defect detection model; and performing cross-product defect classification detection based on the general defect detection model to obtain a defect classification result. According to the method, cross-product sample fusion is carried out by adopting a mode based on double-layer similarity to form a fusion data set which is more compact in distribution and clearer and more uniform in boundary, and model training is carried out through the fusion data set, so that a general defect detection model can realize classified detection of the same defects of different products.

The utility model provides a self-reeling pipe forming jig which solves the technical problems that in the forming process of a self-reeling pipe, the inside and the outside are heated unevenly, deformation is prone to occurring, and products are poor. The self-coiling pipe forming jig comprises a shell and a forming pipe, the forming pipe is located in the shell, an annular gap is defined between the outer wall of the forming pipe and the inner wall of the shell, the annular gap is provided with a feeding port and a discharging port, and the inner diameter of the annular gap is gradually decreased in the direction from the feeding port to the discharging port; a first heating part wraps the shell, a second heating part is arranged in the forming pipe, and the second heating part is arranged in the axial direction of the forming pipe; the first heating part and the second heating part conduct heating on the outer side and the inner side of the annular gap, and it can be guaranteed that the temperatures of the inner layer and the outer layer are kept consistent when the self-coiling pipe is formed. When the temperature of the inner layer and the temperature of the outer layer are kept consistent, the inner layer and the outer layer of the product can be synchronously shrunk, so that the undesirable phenomena of inner folding, bulging and the like of the self-coiling pipe are avoided, and the finished product yield of the thickened self-coiling pipe is greatly improved.

The invention provides a stewed noodle automatic production line product defect analysis method and system based on multi-source detection, and relates to the technical field of product defect detection.The stewed noodle automatic production line product defect analysis method comprises the steps that a plurality of product defect detection nodes are inserted into a stewed noodle automatic production line, and a product image sequence is obtained; meanwhile, collecting real-time production data, extracting rheological characteristic parameters of the stewed noodle dough, performing multi-dimensional characteristic space mapping comparison with an intrinsic characteristic parameter set of a product quality sample to obtain multi-dimensional characteristic space mapping comparison information, and performing defect prediction and defect grading judgment in combination with a product image sequence, so as to obtain a multi-dimensional characteristic space mapping comparison result. And outputting the product defect grade and the risk coefficient, and performing product defect reminding. The technical problem that the dough defects cannot be comprehensively and accurately analyzed due to the lack of systematic modeling and dynamic sensing capabilities for the product quality evolution process in the prior art is solved, and the technical effects of improving the dough defect detection accuracy and enhancing the production line operation stability and the product quality consistency are achieved.

This invention relates to a wall switch assembly machine, comprising a base and a bracket mounted on the base. A first positioning mold and a spring positioning mechanism are mounted on the base. A feeding mechanism and a pressing mechanism are connected to the bracket. The feeding mechanism is positioned above the first positioning mold. Two spring positioning mechanisms are symmetrically arranged at the left and right ends of the first positioning mold. The pressing mechanism is suspended above the first positioning mold. This invention automates the assembly process of the switch base and button panel of the wall switch through the coordinated operation of these mechanisms, reducing manual operation and increasing assembly speed. The positioning of the switch base by the first positioning mold, the positioning and clamping of the spring by the spring positioning mechanism, the positioning of the button panel by the feeding mechanism, and the precise pressing of the pressing mechanism ensure accurate spring positioning and good engagement between the button panel and the switch base, reducing product defect rates and improving product reliability and stability.

The utility model discloses a double-CCD (charge coupled device) simultaneous shooting and double-head simultaneous pasting structure, which belongs to the technical field of product pasting equipment and is characterized in that a Y-axis moving platform and X-axis supports arranged on two sides of the Y-axis moving platform are arranged on a base, moving pasting head mechanisms and upper CCD camera units are arranged on the two X-axis supports, and feeders are arranged on two sides of the Y-axis moving platform. The reinforcing sheet laminating machine with the double CCD cameras and the double laminating heads has remarkable advantages. And in the aspect of laminating efficiency, the double CCDs take pictures at the same time, and the double heads are laminated at the same time, so that compared with a traditional single-CCD and single-laminating-head structure, the laminating time is greatly shortened, the production efficiency is improved, and the large-scale production requirement is met. In terms of fitting precision, the upper CCD camera and the lower CCD camera work in a matched mode, precise positioning and angle adjustment are conducted on the product and the reinforcing piece, multi-color annular light auxiliary recognition is added, it is guaranteed that the fitting position of the reinforcing piece is precise, and the defective rate of the product is reduced.

The invention discloses a consumer complaint information mining method, device and equipment based on a large language model, and relates to the field of product defect risk assessment and recall management. Constructing a standardized product security data set according to the multi-modal data, and performing semantic recognition and labeling on multi-modal complaint information in the data set to obtain a structured product security event information flow; in response to expiration of a preset update period, analyzing the product security event information stream and the current latest product security dynamic knowledge graph to obtain a product security event; the current latest product security dynamic knowledge graph is generated by fusing newly added product security event information flow on the basis of a historical knowledge graph; and carrying out situation awareness, risk early warning and defect correlation analysis on the product security event to finally obtain a product defect assessment result. The result is fed back to the current latest product safety dynamic knowledge graph to be optimized and updated; according to the method, semantic recognition and understanding capability can be realized, and dynamic association analysis capability is improved.

The invention relates to the technical field of defect positioning and machine learning, and discloses a large language model-based product defect positioning method, which comprises the following steps of: obtaining low-order structured features and high-order semantic features of original defect data corresponding to a target product through a large language model; wherein the low-order structured features and the high-order semantic features respectively comprise numerical features and text features; training a plurality of classifiers based on the low-order structured features and the high-order semantic features; and inputting target defect data into the plurality of classifiers, fusing obtained classification results, and determining a product defect type corresponding to the target defect data. According to the technical scheme, deep semantic analysis and regularization preprocessing are conducted on multi-source heterogeneous original defect data (such as logs, monitoring indexes and user feedback) through the large language model and the text encoder, a multi-dimensional feature expression system is constructed in combination with thinking chain design, and the defect recognition accuracy is remarkably improved.