Glass intelligent cutting method and system based on machine vision

Abstract

The present application relates to the technical field of glass processing, and particularly relates to a glass intelligent cutting method and system based on machine vision, which comprises the following steps: collecting multi-dimensional data of glass through a machine vision multi-dimensional fusion perception module; matching the glass type based on the multi-dimensional data and a preset characteristic database, dynamically activating calibration logic to optimize positioning information and marking a processing risk area, and establishing a mapping relationship of the multi-dimensional data, the calibrated positioning information and a preset layout; generating a customized layout scheme suitable for the glass through a multi-objective optimization function of a work condition and glass characteristic double-linkage optimization model; collecting cutting machine work condition data in real time, outputting a dynamic parameter set through a double-linkage optimization model based on the linkage relationship between the glass type and the work condition data; and integrating the multi-dimensional data, the customized layout scheme, the dynamic parameter set and processing process data to construct a full-cycle digital twin traceability system. The present application can significantly improve cutting precision and production efficiency.

Description

Technical Field

[0001] This invention relates to the field of glass processing technology, and in particular to a machine vision-based intelligent glass cutting method and system. Background Technology

[0002] As a crucial link supporting building curtain walls, electronic displays (such as UTG ultra-thin glass and sapphire substrates), and automotive windows, glass deep processing is experiencing a surge in downstream demand for product precision and variety, making traditional processing methods inadequate. Traditional methods rely on single sensing devices, failing to simultaneously acquire multi-dimensional data such as glass morphology, stress distribution, and constraint pressure. Furthermore, the data formats are inconsistent, requiring 10 minutes of manual processing per batch, easily introducing errors. This makes it difficult to predict the risk of cutting edge chipping due to stress concentration and lacks standardized data support for subsequent analysis.

[0003] In current processing, the misjudgment rate of glass type identification by manual means is 8%-12%. Furthermore, UTG glass is brittle and sapphire is hard, yet a uniform positioning and calibration logic is used without taking into account their characteristics, resulting in a scrap rate of over 15% for special glass. Layout relies on manual design without considering glass characteristics and working conditions, resulting in a cutting rate of only 80%-85%. Cutting parameters remain unchanged, and adjustments cannot be made in time when tools wear or spindle speed fluctuates. This not only shortens tool life by 20% but also requires 15-20 minutes per batch for production changeover and debugging of multiple product orders, significantly reducing production efficiency.

[0004] Furthermore, existing processing and quality data are isolated from each other, and tracing quality issues takes 2-4 hours with no real-time early warning mechanism, which can easily lead to batch scrapping. Waste is manually sorted, and the resource recycling rate is less than 60%. Existing processing solutions mostly focus on a single step and only support ordinary clear glass, resulting in poor adaptability and difficulty in meeting the needs of intelligent production. An integrated intelligent cutting solution is urgently needed to solve the above pain points. Summary of the Invention

[0005] This invention can effectively solve the pain points of traditional glass cutting and processing, such as fragmented data, poor adaptability, and difficulty in control and traceability. It enables precise cutting of various types of glass, significantly improves cutting accuracy and production efficiency, and reduces the scrap rate of special glass.

[0006] The technical solution proposed in this invention is: a machine vision-based intelligent glass cutting method, the method comprising: Multi-dimensional data of the glass is collected through a machine vision multi-dimensional fusion perception module; Glass type matching is performed based on multi-dimensional data and a preset characteristic database. The calibration logic is dynamically activated to optimize the positioning information and mark the processing risk area. A mapping relationship between multi-dimensional data, post-calibration positioning information and preset layout is established. Based on the mapping relationship and processing risk area, a customized layout scheme adapted to the glass is generated through the multi-objective optimization function of the dual-linkage optimization model of working conditions and glass properties. Real-time acquisition of cutting machine operating data; based on the linkage relationship between glass type and operating data, output dynamic parameter set through dual linkage optimization model; By integrating multi-dimensional data, customized sampling schemes, dynamic parameter sets, and processing data, a full-cycle digital twin traceability system is constructed.

[0007] Preferably, the specific process for obtaining the multi-dimensional data is as follows: The surface reflectivity, thickness, and texture features of the glass surface are acquired by the image acquisition unit to obtain surface characteristic data; The entire glass area is scanned by a laser morphology detection unit to reconstruct three-dimensional morphology data and capture information on glass micro-deformation and edge smoothness. The photoelastic stress detection unit identifies areas of stress concentration in the glass and outputs stress distribution data. The lateral constraint pressure of the glass is detected by an integrated pressure control unit, and the constraint pressure data is output. Event correlation and encapsulation are performed on surface characteristic data, three-dimensional morphology data, stress distribution data, and constraint pressure data to form multi-dimensional data.

[0008] Preferably, the specific process for obtaining the mapping relationship is as follows: Extract the core feature parameters from the multi-dimensional data, perform cosine similarity calculation with the preset feature database, and take the type with similarity exceeding the preset threshold as the glass recognition result; Based on the identified glass type, the calibration logic of the corresponding detection unit is dynamically activated, and the initial positioning information is iteratively corrected according to the preset accuracy threshold to obtain the calibrated positioning information. Calculate the coordinate difference between the calibrated positioning information and the preset map, and establish a mapping relationship between multi-dimensional data, calibrated positioning information and preset map; Based on stress distribution data and glass type, processing risk areas are marked and associated with mapping relationships.

[0009] Preferably, the specific process for obtaining the customized layout scheme is as follows: Based on the glass thickness, hardness data and processing risk areas in the mapping relationship, plan the toolpath spacing and layout sequence; By optimizing the distribution of finished products through a nesting algorithm, processing risk areas can be avoided, thereby improving the cutting rate and the lifespan of consumables.

[0010] Preferably, the specific process for obtaining the dynamic parameter set is as follows: Establish a correlation model between glass type, operating condition data, and parameter adjustment rules, and set differentiated parameter adjustment thresholds for different glass types; When the operating data is detected to be outside the preset range, the cutting speed, pressure and toolpath indentation are adjusted in real time based on the correlation model. The parameter adjustment logic of the correlation model is optimized based on feedback from actual processing results to improve parameter adaptation accuracy.

[0011] Preferably, the specific process for obtaining the full-cycle digital twin traceability system is as follows: A standardized interface library enables data integration between multi-dimensional data, customized layout schemes, dynamic parameter sets, and production management, warehousing, and recycling systems. Assign a unique traceability identifier to each cutting task, link the entire process data from original piece positioning and cutting execution to finished product warehousing and waste recycling, and build a digital twin model; Based on the digital twin model, closed-loop control of the cutting process, real-time early warning of processing anomalies, and precise collaborative utilization of resources are achieved.

[0012] Preferably, the full-cycle digital twin traceability system enables real-time early warning of processing anomalies and precise collaborative utilization of resources. The specific process is as follows: The actual toolpath trajectory during the cutting process is compared with the theoretical toolpath of the customized nesting scheme in real time. When the deviation exceeds the preset threshold, an abnormal warning is triggered and adjustment suggestions are pushed. Based on the waste data associated with traceability tags, and combined with the glass type, a customized recycling plan is pushed.

[0013] Preferably, the preset characteristic database contains core parameters of various types of glass; when there is no matching glass type, an initial adaptation scheme is generated based on glass types with similar parameters to adapt to multi-dimensional data, and after actual processing verification that it meets the quality requirements, it is updated to the characteristic database.

[0014] The present invention also provides a machine vision-based intelligent glass cutting system, the system being used to execute the aforementioned machine vision-based intelligent glass cutting method.

[0015] The present invention also provides a computer-readable storage medium storing a computer program that is executed by a processor to implement the aforementioned machine vision-based intelligent glass cutting method.

[0016] The beneficial effects of this invention are: 1. By simultaneously acquiring glass surface characteristics, three-dimensional morphology, stress distribution, and constraint pressure data through a machine vision multi-dimensional fusion perception module, and performing data deduplication, completion, noise reduction, and standardization based on process identifiers, the system completely solves the pain points of traditional single-dimensional data acquisition, such as fragmented data, inconsistent formats, and time-consuming manual processing. On the one hand, simultaneous acquisition of multi-dimensional data ensures that no glass features are missed, avoiding cutting risks caused by missing key data such as stress distribution and micro-deformation; on the other hand, the integrity of the standardized data is improved to over 99%, eliminating the need for secondary manual processing and reducing the processing time for a single batch of data by 80%. This provides high-quality data support for subsequent glass type identification and positioning calibration, ensuring the foundation of processing accuracy from the source.

[0017] 2. Intelligent glass type recognition is achieved through similarity matching between multi-dimensional data and a preset characteristic database. Combined with the glass's rigidity characteristics, the positioning calibration logic is dynamically activated. Furthermore, a customized nesting scheme and dynamic parameter set are output through a dual-linkage optimization model based on working conditions and glass characteristics. This solves the problems of misjudgment in traditional manual identification and poor adaptability of fixed calibration / parameters. Specifically, the glass type recognition accuracy is improved to 99.5%, and the misjudgment rate for special glass is reduced from 8%-12% to below 1%. Dynamic positioning calibration narrows the positioning deviation of special glass types such as UTG glass and sapphire to within ±0.03mm, reducing the scrap rate of special glass from 15% to below 2%. Dual-linkage parameter adjustment extends tool life by 25%, reduces the changeover debugging time for multi-variety orders from 15-20 minutes to 90 seconds, and increases the cutting rate to over 90%, balancing processing quality and production efficiency.

[0018] 3. By integrating multi-dimensional data, cutting schemes, dynamic parameters, and processing data, a full-cycle digital twin traceability system is constructed to address the pain points of traditional processing, such as isolated data, difficulty in quality traceability, slow response to anomalies, and poor resource coordination. On the one hand, the system achieves data linkage throughout the entire processing process, reducing the time for tracing quality issues from 2-4 hours to within 10 minutes. Combined with real-time anomaly warnings (such as immediate alerts when tool path deviation exceeds a threshold), the risk of batch scrapping is reduced by 90%. On the other hand, waste classification based on traceability identifiers and the push of dedicated recycling solutions increase the glass resource recycling rate from less than 60% to over 85%. At the same time, it automatically generates production analysis reports, providing managers with accurate decision-making basis and achieving closed-loop control and efficient resource coordination in the cutting process. Attached Figure Description

[0019] Figure 1 This is a flowchart of a machine vision-based intelligent glass cutting method; Figure 2 This is a flowchart illustrating the optimization process of a machine vision-based intelligent glass cutting method. Detailed Implementation

[0020] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.

[0021] It is understood that the term "a" should be understood as "at least one" or "one or more," that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.

[0022] like Figure 1 and Figure 2 As shown, multi-dimensional data of the glass is collected through a machine vision multi-dimensional fusion perception module; based on the multi-dimensional data and the glass type matched with a preset characteristic database, the calibration logic is dynamically activated to optimize the positioning information and mark the processing risk area, establishing a mapping relationship between the multi-dimensional data, the calibrated positioning information and the preset layout; based on the mapping relationship and the processing risk area, a customized layout scheme adapted to the glass is generated through the multi-objective optimization function of the dual-linkage optimization model of working conditions and glass characteristics; real-time acquisition of cutting machine working condition data, and output of dynamic parameter set through the dual-linkage optimization model based on the linkage relationship between glass type and working condition data; integrating multi-dimensional data, customized layout scheme, dynamic parameter set and processing process data to construct a full-cycle digital twin traceability system for closed-loop control of the cutting process, real-time early warning of processing anomalies and precise collaborative utilization of resources.

[0023] Furthermore, multi-dimensional data of the glass is collected through a machine vision multi-dimensional fusion perception module, as detailed below: Multi-dimensional feature collection is performed on the glass to be processed in the glass deep processing production workshop to obtain collected data; production events are associated and encapsulated based on production business logic to obtain encapsulated data; the encapsulated data is published to a preset central data platform based on the standardized interface of the workshop production system; real-time collection messages from the central data platform are received through message subscription, and message parsing is performed on the real-time collection messages to obtain parsed data; production context association is performed on the parsed data to obtain multi-dimensional data.

[0024] The machine vision multi-dimensional fusion perception module includes an anti-reflective image acquisition unit, a dynamic laser morphology detection unit, a high-precision photoelastic stress detection unit, and an integrated pressure control unit for collecting glass features in the glass deep processing system. The machine vision multi-dimensional fusion perception module is an integrated sensing system for the glass deep processing perception layer, located between the production equipment and the central data platform. It achieves full-dimensional acquisition of glass multi-dimensional data through real-time multi-dimensional feature acquisition and data preprocessing. The machine vision multi-dimensional fusion perception module records complete feature information of the surface, morphology, stress, and constraint pressure of the glass from the entry station to the completion of acquisition. The glass to be processed includes ordinary white glass (thickness 3-12mm), UTG ultra-thin glass (thickness 0.03-0.1mm), laminated glass, and sapphire substrate, etc.

[0025] In detail, multi-dimensional feature acquisition refers to the automatic recording of surface, morphology, stress, and constraint pressure feature data of the glass to be processed by various sensing units within the module. The acquired data includes surface characteristic data such as glass surface reflectivity, thickness, and texture defects (bubbles, scratches); three-dimensional morphological data such as glass three-dimensional morphology (micro-deformation, edge smoothness, interlayer offset); stress distribution data such as glass stress distribution (stress value, stress concentration area); and constraint pressure data such as lateral block constraint pressure. The angle signal switching compensation is calculated based on a first-order inertial filtering model. Specifically, it is calculated by subtracting the average angular velocity of the past three sampling periods (each sampling period is 1 millisecond) from the current angular velocity, with the error controlled within ±0.1 radians per second. The main source of error is sensor sampling noise, which is compensated for using a sliding window averaging method. Simultaneously, this module also collects relevant data during the acquisition process, such as acquisition start time, acquisition end time, acquisition equipment number, acquisition station information, and acquisition personnel team information. Production business logic refers to the logical relationship between each acquisition unit, acquisition station, and corresponding glass type. For example, when the acquisition unit is an anti-reflective image acquisition unit and the acquisition station is a surface inspection station, the corresponding acquisition task is surface characteristic acquisition. Production event association refers to associating the acquisition events corresponding to each acquisition task with the corresponding acquisition data. Acquisition events can be the start event of surface characteristic acquisition, the end event of surface characteristic acquisition, the entry event of 3D morphology acquisition, and the departure event of stress distribution acquisition, etc. The standardized interface can be the Open Platform Communications Unified Architecture (OPCW). The dArchitecture (OPCUA) interface, the central data platform is a data platform used to manage all glass acquisition data. The message subscription method can be the subscription method of message queues such as Java Message Service (JMS) or Message Queuing Telemetry Transport (MQTT). Message parsing refers to extracting the message type (such as surface characteristic message, three-dimensional shape message) from the header of the real-time acquired message, and extracting the parsed data from the payload according to the message type. Production context association refers to associating the parsed data with static data such as order number, glass specifications (thickness, size), and processing requirements of the central data platform to give it complete production context information.

[0026] Specifically, based on process identifiers, multi-dimensional data is cleaned and standardized to obtain standardized multi-dimensional glass data. This includes: classifying the acquisition types of the multi-dimensional data (acquisition types correspond to multi-dimensional data types, namely surface characteristic acquisition, three-dimensional morphology acquisition, stress distribution acquisition, and constraint pressure acquisition), and adding acquisition tags to the multi-dimensional data based on the acquisition type classification results; verifying the integrity of the multi-dimensional data based on the acquisition tags, and deduplicating the multi-dimensional data after integrity verification based on process identifiers and acquisition time difference thresholds to obtain deduplicated data; using time-series spline interpolation to complete the deduplicated data, and adopting a node selection strategy of densifying stress concentration areas and uniformly spacing non-concentration areas to obtain completed data; and performing local density anomaly detection on the completed data based on a time-series sliding window, using the Local Outlier Factor (LOF) algorithm, with the neighborhood size k value set to 15 (based on the distribution characteristics of glass processing data samples, verified by 1000+ batches of data, k=15 yields the best anomaly detection accuracy). The noise window data is obtained, and the threshold setting rule for local density anomaly detection is "based on the local density mean of historical normal data plus 2 times the standard deviation", that is, threshold = local density mean of historical normal data + 2 × local density standard deviation of historical normal data. Window data exceeding this threshold is judged as noise window data, and the smooth data corresponding to the noise window data is fitted in the time-series sliding window using the least squares method. The smooth data is used to replace the completed data to obtain denoised data. The denoised data is unified based on the preset data unit, and the data structure of the denoised data after unit unification is standardized based on the preset field template to obtain standardized glass multi-dimensional data.

[0027] Among them, data acquisition type classification refers to determining the acquisition type corresponding to each data based on information such as acquisition events and acquisition stations in multi-dimensional data. Acquisition types include surface characteristic acquisition, three-dimensional morphology acquisition, stress distribution acquisition, and constraint pressure acquisition, etc. Acquisition labels are the labels corresponding to each type of acquisition task. Integrity verification refers to determining whether there is data containing all data fields in the multi-dimensional data corresponding to the acquisition label for each acquisition type's attributes and the complete data fields corresponding to that acquisition type (e.g., surface characteristic acquisition needs to include reflectivity, thickness, and texture defect fields). If so, the data is complete. If not, the missing data field is considered incomplete data; the process identifier refers to the unique identifier corresponding to the glass to be processed in each acquisition process, which can be the acquisition process card number, such as G251015001.01.02 / B. Data deduplication refers to deduplicating data based on the duplication of process identifiers and the time difference between two consecutive acquisition events of the same type within the same process identifier. That is, when the process identifiers of two acquisition tasks are duplicated, they are determined to be duplicate data and deduplicated. When the time difference between two acquisition events of the same type within the same process identifier is less than the preset acquisition time difference threshold, they are determined to be duplicate data and deduplicated.

[0028] In detail, data completion using time-series spline interpolation involves filtering out missing data from the deduplicated data, generating a corresponding time data sequence based on the data type of the missing data (such as surface reflectivity, stress value) and the context acquisition sequence, and then completing the missing data in the time data sequence using spline interpolation. Local outlier factor (LOF) algorithms can be used for local density anomaly detection. For example, for micro-deformation values ​​in three-dimensional morphological data, LOF algorithms can identify outliers exceeding the normal density range and mark them as noise window data.

[0029] Specifically, by utilizing a machine vision multi-dimensional fusion perception module to acquire multi-dimensional data, real-time and full-dimensional acquisition of glass features can be achieved. Through a series of data processing methods such as event encapsulation, acquisition annotation, and data cleaning, high-quality governance of the acquired data is realized. By using process identifiers and acquisition time difference thresholds for deduplication, event redundancy caused by repeated reporting by acquisition units is avoided. Missing data is filled in by time-series spline interpolation, and data noise reduction is achieved by combining local density anomaly detection and least squares smoothing. This can significantly improve the continuity and stability of the data, enhance the feature expression ability of the data, and provide a high-quality data foundation for subsequent glass type identification, positioning calibration, and cutting scheme optimization. The system-level safety boundary model is constructed as follows: After ASIL_D decomposition, a hardware dual redundancy combined with software fault tolerance design is adopted, with a single-point fault tolerance coverage of no less than 90%, covering key nodes such as sensors, actuators, and communication links; the multi-core lockstep integration method adopts a 2-core lockstep architecture, with core 1 and core 2 executing instructions synchronously, comparing the instruction results every 10 milliseconds, and triggering a reset mechanism if a deviation occurs to ensure the consistency of instruction execution; the ECU hardware resource requirements are clearly defined as: Flash memory no less than 128 megabytes, RAM no less than 32 megabytes, CPU computing power no less than 1 gigahertz, and must support floating-point operations to meet the timeliness requirements of real-time data processing and control instruction generation.

[0030] Furthermore, based on multi-dimensional data and a preset characteristic database matching glass types, the calibration logic is dynamically activated to optimize positioning information and mark processing risk areas, establishing a mapping relationship between multi-dimensional data, post-calibration positioning information, and the preset layout. The specific details are as follows: Feature matching analysis is performed on standardized glass multi-dimensional data based on multi-dimensional data and preset characteristic database matching rules to obtain glass type identification results. Specifically, this process includes: extracting glass feature event groups corresponding to each process identifier from the standardized glass multi-dimensional data using process identifiers as the primary key, and extracting core feature parameters from each glass feature event group to obtain a core feature set; sorting the core feature set according to the collection type to obtain a core feature sequence set, and calculating the similarity sequence set between the core feature sequence set and the glass features of each type in the preset characteristic database; grouping the similarity sequence set according to the collection label to obtain a similarity feature set; and performing multi-dimensional threshold judgment on each similarity feature group in the similarity feature set to obtain a similarity judgment dataset. This similarity judgment dataset includes a similarity mean set, a similarity median set, a similarity standard deviation set, and a similarity attainment index set, indicating whether the similarity meets the threshold. The index set is calculated as "the ratio of the number of qualifying features in a single similarity feature group to the total number of features", that is, the qualifying index = (the number of features with similarity greater than the corresponding feature qualifying threshold) / the total number of features in the group. When the qualifying index is ≥0.8, the feature group is considered to be a valid match. Glass type matching is performed on the similarity feature group set based on the similarity judgment dataset to obtain the glass type identification result. The glass type identification result and the similarity judgment dataset are used as glass type data. If there is no matching glass type, an initial adaptation scheme for adapting multi-dimensional data is generated based on glass types with similar parameters. The specific method is "nearest neighbor interpolation method based on similarity ranking": the top 3 glass types in similarity ranking are selected as reference samples, and the initial parameters are calculated by weighted interpolation according to the similarity ratio (initial parameters = summation (reference sample parameters × corresponding similarity ratio)) to form an initial adaptation scheme. After actual processing and verification that it meets the quality requirements, it is updated to the preset feature database.

[0031] Among them, the glass feature event group refers to all glass feature events in a complete acquisition process. The core feature parameters refer to the key feature indicators corresponding to each acquisition type. For example, the core parameters corresponding to surface characteristics are surface reflectivity and thickness, and the core parameters corresponding to three-dimensional morphology are micro-deformation and edge smoothness. Calculating the similarity sequence set between the core feature sequence set and the glass feature of each type in the preset feature database means calculating the cosine similarity between each core feature parameter and the feature threshold of the corresponding type of glass in the database to obtain the similarity value of each core parameter. Glass type matching refers to the matching analysis of the corresponding glass type based on the degree of similarity of each core feature parameter in the corresponding similarity feature group. For example, the similarity feature groups in the similarity feature group set are selected one by one as the target similarity feature group, the average similarity of the target similarity feature group is used as the target similarity average, the similarity threshold is determined based on the target similarity average, all core feature parameters in the similarity feature group that are greater than the similarity threshold are used as matching criteria features, and the glass type corresponding to the matching criteria features is used as the target glass type, thus forming the glass type recognition result.

[0032] In this embodiment of the invention, by aggregating glass feature events based on process identifiers, core feature parameter groups containing complete acquisition paths can be extracted, thereby improving the accuracy of type matching. By grouping acquisition types and performing similarity statistics under the same acquisition dimension, multi-dimensional feature similarity features can be obtained, reflecting the stability and accuracy of glass type matching.

[0033] Positioning calibration data is obtained by performing positioning optimization analysis on standardized glass multi-dimensional data and glass type data based on dynamic positioning calibration rules. Specifically, this includes: extracting positioning feature parameters from each glass feature event group to obtain a positioning feature set; sorting the positioning feature set according to the acquisition time order to obtain a positioning feature sequence set, and calculating the initial positioning deviation sequence set corresponding to the positioning feature sequence set; grouping the initial positioning deviation sequence set according to the acquisition label to obtain a positioning deviation set; and obtaining the historical positioning deviation set corresponding to the positioning deviation set. The dynamic update mechanism for the positioning calibration threshold is as follows: using a sliding time window (window length set to 50 batches of processing data), combined with a threshold attenuation coefficient of 0.98 (to avoid the cumulative impact of historical abnormal data), the threshold is updated in real time. The update formula is: current threshold = (average threshold within the historical window × attenuation coefficient + current batch calibration requirement value × (1 - attenuation coefficient)). Based on historical positioning deviation sets and glass type data, corresponding positioning calibration thresholds are generated for each positioning deviation set. The positioning deviation sets are then calibrated and filtered based on these thresholds to obtain a set of deviations to be calibrated, and a positioning calibration strategy table is generated based on this set. The calibration effect prediction analysis is performed on each positioning deviation set within the set of deviations to be calibrated, resulting in a calibration prediction dataset. The "deviation correction rate" in the calibration prediction dataset is calculated using a weighted moving average method: Correction rate = (deviation correction amount of the last 3 calibrations × sum of corresponding weights) / total weights, where the weight for the most recent calibration is 0.5, the weight for the calibration two years ago is 0.3, and the weight for the calibration two years ago is 0.2, with a total weight of 1.0. The positioning calibration strategy table and the calibration prediction dataset are used as positioning calibration data. Simultaneously, based on stress distribution data and glass type, processing risk areas such as stress concentration areas and high-hardness areas are marked, and the coordinate difference between the calibrated positioning information and the preset layout is calculated, establishing a mapping relationship between multi-dimensional data, calibrated positioning information, and the preset layout.

[0034] Among them, the positioning feature parameters refer to the positioning correlation indicators corresponding to each acquisition type, such as the edge coordinates in the three-dimensional shape and the center position marker in the surface characteristics. The initial positioning deviation sequence set corresponding to the positioning feature sequence set is calculated by calculating the difference between the current positioning feature parameters and the theoretical positioning parameters of the preset map according to the acquisition time sequence, so as to obtain the initial positioning deviation of each acquisition node. The historical positioning deviation set is the positioning deviation set corresponding to the normal acquisition of the same type of glass in the past time period. The positioning calibration threshold generated for each positioning deviation set based on the historical positioning deviation set and glass type data is based on the mean of each historical positioning deviation set and 1. The sum of five standard deviations, combined with the rigidity characteristics of the glass type (e.g., UTG glass has low rigidity and requires a smaller calibration threshold), generates the positioning calibration threshold for the corresponding acquisition node. Calibration screening refers to comparing the magnitude of each initial positioning deviation in each positioning deviation group according to the positioning calibration threshold corresponding to each positioning deviation group. Initial positioning deviations greater than the corresponding positioning calibration threshold are collected as deviations to be calibrated and grouped into a set of deviations to be calibrated. The acquisition nodes corresponding to the deviations to be calibrated are selected as key calibration nodes and compiled into a positioning calibration strategy table. Calibration effect prediction analysis refers to analyzing the deviation correction rate of different calibration strategies in each acquisition node and the degree of impact on subsequent cutting.

[0035] In detail, by setting positioning calibration thresholds based on historical positioning deviations and glass type, flexible calibration threshold settings can be achieved for each acquisition node, enabling global positioning optimization analysis and improving calibration stability for local abnormal positioning. By effectively distinguishing between core feature parameters and positioning feature parameters, refined feature-positioning correlation analysis can be achieved, thereby improving the accuracy of subsequent cutting scheme design.

[0036] Furthermore, based on the mapping relationship and processing risk areas, a customized layout scheme adapted to the glass is generated through a multi-objective optimization function of a dual-linkage optimization model of working conditions and glass properties. The specific details are as follows: Based on standardized multi-dimensional glass data, glass type data, and positioning calibration data, a multi-objective optimization function of a dual-linkage optimization model of working conditions and glass characteristics is used to model the cutting scheme for the glass to be processed, resulting in customized nesting scheme data. Specifically, customized nesting scheme data refers to a set of cutting parameters adapted to the type and positioning state of the glass to be processed during deep glass processing. This data includes parameters such as toolpath layout, cutting speed, and safe edge distance, and contains scheme annotations, such as toolpath spacing for high-hardness glass and toolpath sequence for ultra-thin glass.

[0037] Specifically, based on standardized multi-dimensional glass data, glass type data, and positioning calibration data, a cutting scheme model is created for the glass to be processed, resulting in customized layout scheme data. This includes: extracting multi-dimensional cutting features from the glass to be processed based on standardized multi-dimensional glass data, glass type data, and positioning calibration data to obtain cutting time sequence features; modeling the graph structure and cutting context of the glass to be processed based on the cutting time sequence features and preset cutting rules to obtain a cutting association graph structure; using glass type data as type labels and type labels as supervision signals, a neural network model containing a graph attention layer is trained based on a multi-objective optimization function (including cutting rate, consumable adaptability, and glass type-specific constraints) of a dual-linkage optimization model of working conditions and glass characteristics. The graph neural network model is set as a 3-layer network structure, with 64 nodes in the input layer, 128 nodes in the hidden layer, and 32 nodes in the output layer. The learning rate is set to 0.001 (optimized and determined using a grid search method within the range [0.0001, 0.01]). The cutting scheme generation model is obtained; based on the cutting scheme generation model, the cutting association graph structure is forward propagated to obtain the attention cutting parameter weights. The physical meaning of the attention weights is "the sensitivity of the corresponding node features to the cutting parameters": the higher the weight of a node, the higher the sensitivity of the glass features (such as hardness, stress value) corresponding to that node to the adjustment of cutting parameters (such as speed, pressure), and the parameter settings should be optimized based on this feature first; the attention cutting parameter weights are optimized based on the graph neural network interpreter to obtain customized layout scheme data.

[0038] In detail, because standardized glass multi-dimensional data can reflect the basic characteristics of glass, and glass type data and positioning calibration data can reflect the glass's adaptation requirements and positioning deviations, which are strongly correlated with the cutting scheme, combining the cutting sequence characteristics of the three types of data can improve the accuracy of the scheme design. At the same time, the weight allocation of cutting rate, consumable adaptability and glass type-specific constraints in the multi-objective optimization function can be dynamically adjusted according to actual production needs.

[0039] Specifically, based on standardized glass multi-dimensional data, glass type data, and positioning calibration data, multi-dimensional cutting features are extracted from the glass to be processed to obtain cutting time-series features. This includes: using process identifiers as the primary key, extracting glass feature event groups corresponding to each process identifier from the standardized glass multi-dimensional data, and sorting the glass feature events in each glass feature event group according to the acquisition time order to obtain a glass feature sequence set; performing multi-scale temporal convolution on the glass type data and positioning calibration data based on the glass feature sequence set to obtain type temporal features and positioning temporal features; from... Glass parameter sequence sets are extracted from the glass feature sequence set, and temporal features are extracted from the glass parameter sequence set based on a bidirectional long short-term memory network to obtain parameter temporal features. The type temporal features, positioning temporal features, and parameter temporal features are concatenated into cutting temporal features. The concatenation logic adopts weighted concatenation, with a weight of 0.4 for type temporal features, 0.3 for positioning temporal features, and 0.3 for parameter temporal features. The weight allocation is based on the Pearson correlation coefficient between each feature and the cutting scheme (the higher the correlation coefficient, the greater the weight). After multiple batches of verification, the scheme adaptation accuracy is the best under this weight allocation.

[0040] Among them, multi-scale temporal convolution refers to performing temporal convolution based on temporal convolution kernels of different time scales according to the order of glass type data in the glass feature sequence set to obtain type temporal features, and performing temporal convolution based on temporal convolution kernels of different scales according to the order of positioning calibration data in the glass feature sequence set to obtain positioning temporal features. Each glass parameter in the glass parameter sequence set corresponds to the core attribute of the glass to be processed in each glass feature event, which can be stress value, micro-deformation, thickness, hardness, etc. There is a strong correlation between glass parameters and the parameter settings of the cutting scheme. Therefore, the parameter temporal features can reflect this correlation, thereby improving the accuracy of the scheme design.

[0041] In detail, based on the cutting time sequence features and preset cutting rules, graph structure modeling and cutting context modeling are performed on the glass to be processed to obtain a cutting association graph structure. This includes: generating a cutting graph node feature set by using the cutting time sequence features as graph node features, and generating a node edge set corresponding to the cutting graph node feature set based on the sequence order of the glass feature sequence set; aggregating the first few acquisition events corresponding to the glass to be processed in the glass feature sequence set into a pre-cutting event sequence set, and performing cutting context modeling on the glass to be processed based on the positioning calibration data and glass type data corresponding to the pre-cutting event sequence set to obtain a cutting context event sequence set; embedding cutting features into the node edge set based on the cutting context event sequence set to obtain an edge feature set, and generating a cutting association graph structure based on the cutting graph node feature set, node edge set, and edge feature set; during the modeling process, processing risk areas must be avoided to ensure the safety of the toolpath layout.

[0042] Here, sequence order refers to the logical order between the acquisition of various features during the glass deep processing acquisition process. The first few acquisition events can be the first five acquisition events. By analyzing the positioning data and type data of the glass to be processed in the acquisition stage, the influence of glass characteristics on the cutting scheme can be explored, thereby improving the accuracy of the scheme design. Cutting context modeling refers to labeling the corresponding positioning calibration data and glass type data of each glass to be processed, and labeling the association weights of the corresponding positioning calibration data, glass type data and cutting parameters, thereby enhancing the context feature association of the three. Cutting feature embedding refers to transforming the glass characteristics and feature associations corresponding to the cutting context event sequence set into cutting features in vector form, and using the cutting features as the corresponding edge features.

[0043] In detail, when training a neural network model containing a graph attention layer, the attention weight allocation rule for the graph attention layer is "introducing glass type embedding vectors as prior information," performing a dot product operation between the glass type embedding vectors and the node feature vectors, and then weighting the result before participating in the attention weight calculation, i.e., attention weight = (Weight matrix multiplied by [node feature vector + prior weight coefficient multiplied by glass type embedding vector]), where the prior weight coefficient is 0.3. The loss function of mean squared error is used for training, and data augmentation is used to handle sample imbalance. The graph neural network interpreter can be a graph neural network generating interpretation (GNNExplainer) or a higher-order interpretation of graph neural networks via relevant paths (Graph-LRP). Scheme optimization refers to analyzing the potential correlation between cutting parameters and glass characteristics and positioning deviation. The specific implementation method of potential correlation analysis is "feature correlation mining based on mutual information entropy": calculate the mutual information entropy between cutting parameters and glass characteristics and positioning deviation. When the mutual information entropy is ≥0.6, it is determined to be a strong correlation. Based on the strong correlation, the cutting parameters are optimized in a targeted manner. For example, through analysis, it is found that when the stress value of a certain type of glass exceeds a certain range, the cutting speed needs to be reduced by 15% to avoid edge chipping. Then, the cutting speed setting can be optimized for the stress parameters of this type of glass to generate a customized scheme.

[0044] In this embodiment of the invention, multi-dimensional cutting features are constructed by combining standardized multi-dimensional glass data, glass type data, and positioning calibration data. Based on these multi-dimensional cutting features, a cutting association graph structure is generated. This allows for the acquisition of contextual feature representations of the glass to be processed in the feature dimension, type dimension, and positioning dimension. By obtaining dynamic patterns of glass characteristics, type adaptation, and positioning deviation, and combining contextual modeling, the sensitivity of the graph structure to cutting semantics is enhanced. By using glass type data as a supervisory signal to train the graph attention network, the attention weights can reflect the contribution distribution of cutting parameters in the graph structure. Ultimately, the interpretable optimization of the cutting scheme based on features, type, and positioning is achieved, thereby significantly improving the accuracy and adaptability of customized layout schemes.

[0045] Furthermore, real-time data collection of the cutting machine's operating conditions is conducted. Based on the correlation between glass type and operating condition data, a dynamic parameter set is output through a dual-linkage optimization model, as detailed below: Real-time acquisition of cutting machine operating data is used. Based on the linkage between glass type and operating data, a dynamic parameter set is output through a dual-linkage optimization model of operating conditions and glass characteristics. The dynamic parameter set refers to a set of cutting control parameters adapted to the current glass type and the real-time operating conditions of the cutting machine. It includes core parameters such as cutting speed, cutting pressure, tool path indentation, and tool cooling flow rate. These parameters can be adjusted in real time according to changes in operating conditions, solving the problems of poor adaptability and large fluctuations in processing quality caused by traditional fixed parameters. The dual-linkage optimization model of operating conditions and glass characteristics is a model that takes both glass type (static characteristics, such as hardness, brittleness, and thickness) and operating data (dynamic status, such as tool wear and equipment operating status) as inputs and outputs adaptive parameters through algorithms. This is different from the traditional fixed parameter mode that only relies on glass type, and achieves dual control of static type adaptation and dynamic operating condition adaptation.

[0046] When collecting real-time data on the cutting machine's operating conditions, the machine's built-in and external sensing units collect comprehensive operating condition data at preset frequencies to ensure data coverage of the entire processing scenario and real-time availability. First, the focus is on collecting core operating condition data. For tool status, a professional detection unit captures anomalies such as tool wear and temperature changes. For equipment operation, key data such as spindle speed and load are collected to identify issues like speed fluctuations and overload. For the processing environment, data on worktable vibration and regional temperature and humidity are collected to mitigate environmental interference. For processing feedback, a vision unit acquires real-time information on the cutting edge quality, providing a basis for subsequent parameter optimization. The collected raw operating condition data is then preprocessed. Noise is removed through filtering, data format and units are standardized, and outliers that significantly exceed physical logic are initially screened, forming a standardized operating condition dataset. This ensures the data can be directly input into subsequent correlation models, avoiding model input deviations caused by messy raw data.

[0047] Based on historical processing data (batch processing records covering multiple types of glass), a correlation model is constructed between glass type, working condition data, and parameter adjustment rules. The core logic is to mine the inherent relationship among these three elements through historical data to achieve precise mapping of parameter adjustments. The model input layer explicitly includes the glass type (such as ordinary clear glass, UTG ultra-thin glass, high-hardness special glass, etc.) and a standardized working condition dataset. The output layer contains specific parameter adjustment rules (including the direction and range of parameter adjustment). Different working condition thresholds and parameter adjustment ranges are set for each type to address the differences in physical properties of different glass types. For example, for highly brittle glass types, stricter tool wear thresholds and gentler parameter adjustment ranges are set to avoid breakage due to parameter fluctuations; for high-hardness glass types, higher pressure adjustment upper limits are adopted to ensure cutting results. During the model training phase, the model parameters are optimized by utilizing historical processing data on working conditions, the correspondence between parameter adjustments and processing quality. The gradient descent algorithm for the parameter adjustment rules uses the Adam optimizer, with the first momentum parameter coefficient = 0.9, the second coefficient = 0.999, and the weight decay coefficient = 0.0001. The parameter adjustment rules are optimized by minimizing the processing quality loss function (loss value = 1 - cutting pass rate + edge breakage rate × 0.5) to ensure that the output adjustment rules have high adaptability. At the same time, an incremental learning mechanism is set up to automatically update the model parameters every time a certain amount of new processing data is accumulated, ensuring that the model can adapt to new glass types or changes in cutting machine working conditions.

[0048] When standardized operating condition data exceeds the preset threshold for the corresponding glass type, the associated model generates parameter adjustment instructions in real time and sends them to the cutting machine control system via the industrial communication link to achieve real-time updates of the dynamic parameter set. The specific logic consists of three steps: The first step is anomaly detection, comparing the standardized operating condition data with the differential threshold for the current glass type, and classifying the anomaly into minor, severe, etc., according to the extent of the exceedance, thus clarifying the adjustment priority; The second step is graded adjustment, following the principle of "fine-tuning for minor anomalies and emphasizing for severe anomalies." For minor anomalies, core parameters are adjusted slightly (such as fine-tuning the cutting speed and pressure), while for severe anomalies, not only are the parameters increased, but the core parameters are also adjusted significantly. The adjustment of parameters may also trigger the equipment protection mechanism, which is defined as follows: for minor anomalies, processing continues and parameters are adjusted in real time; for severe anomalies, the trigger condition is that the working data exceeds the threshold by 30% or more or there are three consecutive sampling cycles of anomalies. At this time, processing is immediately suspended, the cooling system starts the maximum cooling flow within 50ms, and processing is resumed after the working data recovers to within the threshold and is confirmed by manual inspection to avoid deterioration of processing quality; the third step is effect verification. After parameter adjustment, the adjustment effect is verified by real-time processing feedback data (such as cutting edge quality). If the expected result is not achieved, a second adjustment is triggered until the processing status returns to stability.

[0049] By utilizing full-cycle processing performance data, the parameter adjustment logic of the correlation model is continuously optimized, forming a closed loop of data collection, parameter adjustment, performance feedback, and model optimization. First, core evaluation indicators (such as cutting pass rate, edge chipping rate, tool life, and dimensional accuracy) are set, and the compliance status of these indicators is automatically calculated after each batch of processing. The loss function of the tool life prediction model uses exponentially weighted mean square error, i.e., loss value = summation (the batch number raised to the power of the exponential weight coefficient multiplied by the square of (predicted life - actual life)). The exponential weight coefficient is set to 0.9, and the batch number is the order of processing batches. The sequence numbering is then used to improve the stability of long-term life prediction through this loss function. If the processing indicators of a certain type of glass fail to meet the standards continuously, the parameter adjustment rules corresponding to the correlation model are traced back to analyze whether there is a rule adaptation bias (such as insufficient parameter adjustment under a certain working condition leading to quality problems). After optimizing the adjustment rules for the problem, a certain batch of the same type of glass is selected for verification. If the optimized indicators meet the standards, the new rules are solidified into the correlation model. If they do not meet the standards, the root cause of the problem is re-examined (such as whether key working condition dimensions are omitted, insufficient consideration of glass type characteristics, etc.) until the adjustment rules meet the processing requirements.

[0050] Furthermore, by integrating multi-dimensional data, customized layout schemes, dynamic parameter sets, and processing data, a full-cycle digital twin traceability system is constructed to achieve closed-loop control of the cutting process, real-time early warning of processing anomalies, and precise collaborative utilization of resources. Specific details are as follows: By integrating multi-dimensional data, customized layout schemes, dynamic parameter sets, and processing data, a full-cycle digital twin traceability system is constructed. This system enables closed-loop control of the cutting process, real-time early warning of processing anomalies, and precise collaborative utilization of resources. Based on production logic, a data acquisition-cutting analysis chart interface is generated corresponding to standardized multi-dimensional glass data, glass type data, positioning calibration data, customized layout scheme data, and dynamic parameter sets. This chart interface is then visualized. Specifically, after calculating the above data, a corresponding data acquisition-cutting analysis chart interface needs to be generated to more intuitively display the data and facilitate administrator analysis of data acquisition quality and the rationality of the cutting scheme.

[0051] Specifically, based on production logic, a data acquisition-cutting analysis interface is generated corresponding to standardized glass multi-dimensional data, glass type data, positioning calibration data, customized layout scheme data, and dynamic parameter sets. This includes: using the process identifier as the primary key, linking and integrating standardized glass multi-dimensional data, glass type data, positioning calibration data, customized layout scheme data, and dynamic parameter sets to obtain integrated analysis data; adding acquisition event information (including acquisition duration and equipment status) to the integrated analysis data based on the multi-dimensional data to obtain enhanced analysis data; and mapping the enhanced analysis data to corresponding chart types based on the production logic and the data type of the enhanced analysis data to obtain a chart configuration rule set; for example, a scatter plot is used for standardized glass multi-dimensional data. The customized nesting scheme for stress distribution is displayed using a 3D model to show the toolpath layout, and the dynamic parameter set is displayed in a table to show the cutting speed and pressure. Based on preset user roles (workshop technicians, quality control personnel, etc.), a Kanban layout template is configured for the chart configuration rule set to obtain the Kanban structure template. According to the chart configuration rule set, the enhanced analysis data is instantiated into charts (using ECharts or D3.js methods) to obtain the analysis chart set, and dynamic interactive logic (including time range filtering and toolpath trajectory comparison pop-ups) is added to the analysis chart set to obtain the dynamic analysis chart set. According to the Kanban structure template, the dynamic analysis chart set is assembled into an interface (using Vue.js and UI libraries) to obtain the acquisition-cutting analysis chart interface.

[0052] Among them, event information collection refers to adding collection duration, collection equipment status, collection anomaly records, etc.; production logic refers to the logic of classifying each collection-cutting link in production (such as analyzing the differences in layout schemes for different glass types and the dynamic parameter adaptation effect); mapping corresponding chart types refers to assigning corresponding enhanced analysis data chart types, data field bindings, color schemes, and titles to various production logics; for processing anomaly warnings, the charts must support "real-time comparison of actual toolpath trajectories with the theoretical toolpaths of customized layout schemes," and trigger pop-up warnings when the deviation exceeds the threshold; for precise collaborative utilization of resources, it is necessary to push exclusive recycling schemes based on waste data associated with traceability identifiers, combined with glass types, and display recycling benefits in the report; user roles refer to the production roles of users, which can be workshop technicians, quality control personnel, and production managers, etc.; and the Kanban layout template refers to setting the combination rules, position, and size of various charts in the chart configuration rule set.

[0053] Specifically, chart instantiation refers to mapping the corresponding data into a chart. This can be done using ECharts or D3.js. Adding dynamic interactive logic means adding features such as time range filtering, feature filtering, and parameter detail pop-ups to the instantiated chart (e.g., clicking "Dynamic Parameter Table" allows viewing parameter adjustment history). For example, the data collection time period can be filtered based on the filter method, and a click event listener can be used to display complete collection-slicing data details for a specific glass (including multi-dimensional data, layout scheme, and dynamic parameters). Interface assembly refers to assembling multiple independent chart components into a complete Kanban board according to the combination rules, positions, and sizes defined by the Kanban structure template. This can be done using the front-end framework VUE.js and UI libraries such as ElementUI and AntDesign.

[0054] Specifically, through chart mapping based on production logic, messy acquisition-cutting data can be transformed into an intuitive and interactive chart pattern, making it easy for users to quickly grasp the matching relationship between glass features and cutting schemes, locate acquisition anomalies, and optimize cutting schemes. This facilitates subsequent parameter adjustments for acquisition units or cutting equipment, improving the efficiency of data application. At the same time, the full-cycle digital twin traceability system enables traceability of the processing process and traceability of quality problems, supporting closed-loop management of the cutting process.

[0055] The processes described above with reference to the flowcharts in the embodiments disclosed in this invention can be implemented as computer software programs. The embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wire segments, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless segments, wire segments, optical fibers, RF, etc., or any suitable combination thereof.

[0056] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0057] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been shown and explained in the embodiments. Without departing from the stated principles, the implementation of the present invention may have any variations or modifications.

Claims

1. A machine vision-based intelligent glass cutting method, characterized in that, The method includes: Multi-dimensional data of the glass is collected through a machine vision multi-dimensional fusion perception module; Glass type matching is performed based on multi-dimensional data and a preset characteristic database. The calibration logic is dynamically activated to optimize the positioning information and mark the processing risk area. A mapping relationship between multi-dimensional data, post-calibration positioning information and preset layout is established. Based on the mapping relationship and processing risk area, a customized layout scheme adapted to the glass is generated through the multi-objective optimization function of the dual-linkage optimization model of working conditions and glass properties. Real-time acquisition of cutting machine operating data; based on the linkage relationship between glass type and operating data, output dynamic parameter set through dual linkage optimization model; By integrating multi-dimensional data, customized sampling schemes, dynamic parameter sets, and processing data, a full-cycle digital twin traceability system is constructed.

2. The intelligent glass cutting method based on machine vision according to claim 1, characterized in that, The specific process for obtaining the multi-dimensional data is as follows: The surface reflectivity, thickness, and texture features of the glass surface are acquired by the image acquisition unit to obtain surface characteristic data; The entire glass area is scanned by a laser morphology detection unit to reconstruct three-dimensional morphology data and capture information on glass micro-deformation and edge smoothness. The photoelastic stress detection unit identifies areas of stress concentration in the glass and outputs stress distribution data. The lateral constraint pressure of the glass is detected by an integrated pressure control unit, and the constraint pressure data is output. Event correlation and encapsulation are performed on surface characteristic data, three-dimensional morphology data, stress distribution data, and constraint pressure data to form multi-dimensional data.

3. The intelligent glass cutting method based on machine vision according to claim 2, characterized in that, The specific process for obtaining the mapping relationship is as follows: Extract the core feature parameters from the multi-dimensional data, perform cosine similarity calculation with the preset feature database, and take the type with similarity exceeding the preset threshold as the glass recognition result; Based on the identified glass type, the calibration logic of the corresponding detection unit is dynamically activated, and the initial positioning information is iteratively corrected according to the preset accuracy threshold to obtain the calibrated positioning information. Calculate the coordinate difference between the calibrated positioning information and the preset map, and establish a mapping relationship between multi-dimensional data, calibrated positioning information and preset map; Based on stress distribution data and glass type, processing risk areas are marked and associated with mapping relationships.

4. The intelligent glass cutting method based on machine vision according to claim 3, characterized in that, The specific process for obtaining the customized layout scheme is as follows: Based on the glass thickness, hardness data and processing risk areas in the mapping relationship, plan the toolpath spacing and layout sequence; By optimizing the distribution of finished products through a nesting algorithm, processing risk areas can be avoided, thereby improving the cutting rate and the lifespan of consumables.

5. The intelligent glass cutting method based on machine vision according to claim 4, characterized in that, The specific process for obtaining the dynamic parameter set is as follows: Establish a correlation model between glass type, operating condition data, and parameter adjustment rules, and set differentiated parameter adjustment thresholds for different glass types; When the operating data is detected to be outside the preset range, the cutting speed, pressure and toolpath indentation are adjusted in real time based on the correlation model. The parameter adjustment logic of the correlation model is optimized based on feedback from actual processing results to improve parameter adaptation accuracy.

6. The intelligent glass cutting method based on machine vision according to claim 5, characterized in that, The specific process of obtaining the full-cycle digital twin traceability system is as follows: A standardized interface library enables data integration between multi-dimensional data, customized layout schemes, dynamic parameter sets, and production management, warehousing, and recycling systems. Assign a unique traceability identifier to each cutting task, link the entire process data from original piece positioning and cutting execution to finished product warehousing and waste recycling, and build a digital twin model; Based on the digital twin model, closed-loop control of the cutting process, real-time early warning of processing anomalies, and precise collaborative utilization of resources are achieved.

7. The intelligent glass cutting method based on machine vision according to claim 6, characterized in that, The full-cycle digital twin traceability system enables real-time early warning of processing anomalies and precise collaborative utilization of resources. The specific process is as follows: The actual toolpath trajectory during the cutting process is compared with the theoretical toolpath of the customized nesting scheme in real time. When the deviation exceeds the preset threshold, an abnormal warning is triggered and adjustment suggestions are pushed. Based on the waste data associated with traceability tags, and combined with the glass type, a customized recycling plan is pushed.

8. The intelligent glass cutting method based on machine vision according to claim 7, characterized in that, The preset characteristic database contains core parameters of various types of glass. When there is no matching glass type, an initial adaptation scheme is generated based on glass types with similar parameters to adapt to multi-dimensional data. After actual processing verification that it meets the quality requirements, it is updated to the characteristic database.

9. A machine vision-based intelligent glass cutting system, characterized in that, The system is used to execute a machine vision-based intelligent glass cutting method as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that is executed by a processor to implement a machine vision-based intelligent glass cutting method as described in any one of claims 1-8.

Images