A computer-aided design system for apparel pattern making

Abstract

The application discloses a kind of clothing pattern computer-aided design systems, it is related to computer-aided design technical field, to the technical problem that only through static geometry file one-way transmission between existing clothing CAD system and numerical control cutting bed, cutting deformation data cannot be returned and reused, and pattern search only supports text keyword, leading to historical experience difficult to match the call of this system includes pattern drawing structure construction module, drawing structure semantic embedding module, structured vector database, similar search module, adaptive pre-compensation module and numerical control cutting bed execution layer, cutting bed execution layer is bound storage by sensor collection actual cutting deformation data and return, system is according to the deformation deviation data of historical similar pattern to the pre-compensation of new pattern's seam allowance, filament angle and alignment point parameter is adaptive, the application is used to improve clothing cutting first pass rate, reduce fabric loss and revision workload.

Description

Technical Field

[0001] This invention relates to the field of computer-aided design technology, specifically to a computer-aided design system for clothing patterns. Background Technology

[0002] In modern garment manufacturing, computer-aided design systems (CADS) and CNC cutting machines are core equipment for realizing digital production. CADS is used to complete the geometric contour design, grading, and layout of patterns. After layout, a static two-dimensional geometric processing file is generated. The CNC cutting machine then drives the cutting blade along a fixed trajectory to cut multiple layers of fabric based on this processing file. Garment fabrics are flexible anisotropic materials, with differences in their warp and weft elastic moduli, and they also exhibit heat shrinkage. During multi-layer fabric cutting, blade pressure, humidity changes, and batch variations in fabric can all cause the actual cut piece size to deviate from the design size. Existing technologies generally consider cutting deformation data to have only single-time correction value and not require systematic storage and reuse. Therefore, existing systems do not establish a data feedback path from the cutting machine to the design end. This application breaks through this understanding by establishing for the first time a two-way closed-loop mechanism of pattern diagram structural semantic embedding—historical deformation deviation binding storage—similarity-weighted pre-compensation. This allows historical deformation experience to be structurally reused for new patterns in terms of seam allowance, grain angle, and alignment point coordinates. Pre-compensation aims to achieve a first-cut pass rate improvement that existing unidirectional static solutions cannot achieve. Some cutting beds are equipped with visual sensors only for local correction during the cutting process; the correction data is discarded after cutting and not transmitted back to the design end. This results in the inability to systematically accumulate and reuse deformation errors that repeatedly occur in different batches and seasons of the same fabric. Furthermore, existing pattern databases only support searches based on filenames and text keywords, and cannot perform similarity searches based on the geometric topology of the pattern and the mechanical properties of the fabric. This makes it difficult to match and call historical deformation experience data, leading to repeated sampling and pattern revision work. The existing solutions include a computer-aided design terminal, a nesting calculation module, a static geometric file generation and transmission interface, a CNC cutting bed controller, and local visual correction sensors. These components are connected through a one-time, unidirectional file transmission method, and the cutting bed lacks a data feedback path to the design end. Therefore, we propose a computer-aided design system for apparel patterns. Summary of the Invention

[0003] The purpose of this invention is to provide a computer-aided design system for clothing patterns.

[0004] The technical problem this invention aims to solve is that existing computer-aided design systems for apparel and CNC cutting systems only exchange data unidirectionally through static geometric files, without establishing a feedback path between actual fabric cutting deformation data and design parameters. This results in the inability to systematically record and reuse deformation errors such as shrinkage and slippage caused by cutting the same batch and type of fabric in different seasons and with different layers. As a result, manual experience is required for repeated pattern revisions. Furthermore, existing pattern databases only support retrieval based on filenames and text keywords, and cannot perform similarity searches based on the geometric topology of the pattern and the mechanical properties of the fabric. Consequently, historical deformation experience data cannot be intelligently matched and retrieved, leading to a large amount of repetitive sampling and pattern revision work.

[0005] To address the aforementioned technical problems, this invention provides a computer-aided design system for garment patterns, comprising a CAD design module, a CNC instruction generation module, and a CNC cutting bed execution layer. The CNC cutting bed execution layer drives a cutting blade to cut fabric according to CNC machining instructions issued by the CNC instruction generation module. The system further includes: a pattern diagram structure construction module, used to receive the original pattern data output by the CAD design module, define key points and alignment points of the cut piece contours as nodes, define the stitching topological connections between nodes as edges, and assign geometric feature vectors and fabric attribute feature vectors to the nodes and edges respectively, thereby constructing a pattern diagram structure; a graph structure semantic embedding module, used to input the pattern diagram structure into a graph neural network encoder for neighborhood feature aggregation and graph-level pooling processing to obtain a global semantic embedding vector for the pattern; and a structured vector database, used to bind and store the global semantic embedding vector with the corresponding original pattern data and fabric constitutive parameters. A pattern semantic index record is formed. A similarity retrieval module is used to respond to new pattern design requests, extract the global semantic embedding vector of the new pattern, and retrieve historical pattern semantic index records with similarity exceeding a preset threshold in the structured vector database to obtain a recall historical pattern set. An adaptive pre-compensation module is used to calculate a prediction compensation vector based on the historical deformation deviation data and similarity corresponding to the recall historical pattern set, and apply it to the seam allowance parameters, thread angle parameters, and key alignment point coordinate parameters of the new pattern. The CNC cutting bed execution layer also includes a vision sensor and a tension sensor, which are used to collect the measured coordinates and interlayer slippage of each key point of the cut piece during the cutting process. A deviation acquisition and feedback module is used to compare the measured coordinates with the CAD theoretical design coordinates to obtain the deformation deviation vector, bind it with the semantic index record of the pattern used in this cutting, and then send it back to the structured vector database to form a data closed loop between the design end and the execution end.

[0006] Compared with the prior art, the beneficial effects of the present invention by adopting the above technical solution are as follows:

[0007] 1. This invention establishes a feedback path from the measured deformation data of the CNC cutting bed to the structured vector database, forming a continuous closed-loop accumulation of cutting deviation data, thereby reducing the number of repeated trial cuttings and pattern revisions;

[0008] 2. This invention improves the accuracy and efficiency of searching for historical similar patterns by abstracting the pattern into a joint graph structure that includes geometric features, topological relationships and fabric constitutive parameters and generating semantic embedding vectors for similarity retrieval, replacing the search method based on filenames and keywords.

[0009] 3. This invention generates a predictive compensation vector by weighted fusion of deformation deviation data based on similar historical patterns, and corrects the seam allowance, thread angle, and alignment point parameters before the CNC command is issued, thereby improving the first-cut pass rate and reducing fabric waste. Attached Figure Description

[0010] Figure 1 This is a schematic diagram of the overall closed-loop business process of the system in this embodiment of the invention;

[0011] Figure 2 This is a schematic diagram of the pre-compensation processing sub-process of the new version in an embodiment of the present invention;

[0012] Figure 3 This is a schematic diagram of the data return and storage sub-process of the cutting bed in an embodiment of the present invention. Detailed Implementation

[0013] The following is in conjunction with the appendix Figure 1-3 The specific embodiments of the present invention will be further described below. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0014] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0015] The computer-aided design system for clothing patterns provided in this embodiment includes, in terms of structure, a CAD design module, a pattern diagram structure construction module, a diagram structure semantic embedding module, a structured vector database, a similarity retrieval module, an adaptive pre-compensation module, a CNC instruction generation module, and a CNC cutting bed execution layer. The CNC cutting bed execution layer includes a vision sensor, a tension sensor, a cutting blade execution mechanism, and a deviation acquisition and feedback module. The connection relationship between the modules is as follows: the CAD design module outputs the original pattern data to the pattern diagram structure construction module; the pattern diagram structure construction module outputs the diagram structure data to the diagram structure semantic embedding module. The semantic embedding vector, along with the original pattern data and fabric constitutive parameters, is stored in a structured vector database. The similarity retrieval module retrieves similar historical patterns and their deviation data from the structured vector database and transmits them to the adaptive pre-compensation module. The adaptive pre-compensation module corrects the pattern parameters and outputs them to the CNC instruction generation module. The CNC instruction generation module generates CNC machining instructions and sends them to the CNC cutting bed execution layer. During the cutting process, the CNC cutting bed execution layer collects actual deformation data through sensors, processes it through the deviation acquisition and feedback module, and then transmits it back to the structured vector database for binding and storage, forming a complete closed loop.

[0016] In the pattern diagram construction stage, the garment pattern is abstracted into a graph structure. The node set in the graph structure corresponds to the key points and alignment points of the cut piece outline. The node feature vector contains the two-dimensional coordinates, local curvature and seam allowance of the node. The edge set corresponds to the stitching topological connection relationship between the nodes. The edge feature vector contains the stitching type, splicing direction, warp elastic modulus of the fabric, weft elastic modulus of the fabric and heat shrinkage rate.

[0017] In the graph structure semantic embedding stage, the constructed pattern graph structure is input into the pre-trained graph neural network encoder, and the layer-by-layer update formula for the hidden layer representation of the nodes is as follows:

[0018] ;

[0019] in, Represents a node In the The hidden layer feature vector of a layer is initialized by taking the feature vector of that node. Represents a node The set of adjacent nodes in a graph structure Represents a node For nodes The attention aggregation weight is used to characterize the contribution of the neighboring node to semantic fusion. Indicates the first The learnable weight matrix of the layer, This represents a vector concatenation operation. Represents a non-linear activation function;

[0020] After completion After layer aggregation, graph-level pooling is performed on the hidden layer representations of all nodes to obtain the global semantic embedding vector for this pattern:

[0021] ;

[0022] in, Indicates the layout The global semantic embedding vector, The graph-level pooling function can be implemented using summation, mean, or attention weighting. Represents a node go through The final hidden layer feature vector obtained after layer aggregation is assigned a unique semantic identifier to the pattern. The global semantic embedding vector, the original pattern file and the fabric constitutive parameters are bound and stored in the structured vector database to form a pattern semantic index record.

[0023] In the reverse feedback stage of actual deviation measurement on the cutting bed, the CNC cutting bed execution layer uses deployed vision sensors and tension sensors to collect the actual coordinates of each key point of the cut piece and the amount of interlayer slippage in real time during the cutting process. These coordinates are then registered and compared with the coordinates of the CAD theoretical design to calculate the deformation deviation vector.

[0024] ;

[0025] in, Indicates the first Deformation deviation vector at key points This indicates the actual coordinates of the key point as measured by the sensor. This represents the theoretical coordinates of the key point in the original CAD design. The set of deformation deviation vectors of all key points is filtered, denoised, and normalized to obtain the standardized deviation data of this batch of cutting. This data is then bound and packaged together with the semantic identifier of the pattern used in this cutting, the fabric batch number, and the number of cutting layers. The data is then sent back and stored in the structured vector database and appended to the corresponding pattern semantic index record to form a continuous accumulation of the historical deformation deviation dataset of this pattern.

[0026] In the similarity retrieval and adaptive pre-compensation stage, when a designer completes a new pattern design or submits a layout task, the system extracts the graph structure of the new pattern and generates its global semantic embedding vector. The similarity between the semantic embedding vectors of the new pattern and the historical pattern in the database is calculated using cosine similarity.

[0027] ;

[0028] in, This represents the global semantic embedding vector of the new version. Represents the first in the database Global semantic embedding vectors of each historical version. This represents the similarity score between the two. The system retrieves historical versions with similarity exceeding a preset threshold, obtaining a quantity of... The recall history pattern set is used to calculate the normalized weights and prediction compensation vectors using the following formula:

[0029] ;

[0030] in, Indicates the first Normalized weights for each recall history version. Indicates the first Standardized historical deviation data corresponding to each recalled historical version. This means that for the predicted compensation vector calculated for the new pattern, the system decomposes the predicted compensation vector and applies it to the seam allowance parameter, thread angle parameter and key alignment point coordinate parameter of the new pattern, generates the pattern parameters after compensation correction, and then generates CNC machining instructions and sends them to the cutting knife execution mechanism of the CNC cutting bed execution layer, driving the cutting knife to cut the multi-layer stacked fabric along the set trajectory. The measured data generated during the cutting process continues to be fed back to the structured vector database according to the aforementioned reverse feedback process, and is bound and stored with the semantic identifier of the current pattern, thereby forming a continuous closed loop iteration of data between the design end and the execution end.

[0031] In another implementation, the pattern semantic embedding generation method does not require the construction of an explicit graph structure. Instead, the pattern outline vector image is rendered as a two-dimensional raster image, and combined with the stitching relationship annotation layer and the fabric attribute numerical layer, the image feature vector is extracted by a convolutional neural network as the semantic embedding representation of the pattern. The subsequent retrieval and compensation process is consistent with the aforementioned implementation method. This method can be used as a supplement to the graph structure modeling method in scenarios where it is difficult to obtain fine topological connection relationships.

[0032] In another implementation, instead of storing all historical pattern semantic embedding vectors and deviation data in a central structured vector database, an edge computing unit can be deployed locally on the CNC cutting bed to store pattern semantic indexes and deviation data subsets related to the historical processing tasks of that cutting bed. This enables localized similarity retrieval and adaptive pre-compensation, and periodically performs incremental data synchronization with the central database. This approach can be used in production line environments with limited network conditions.

[0033] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, any modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope defined by the claims of the present invention.

Claims

1. A computer-aided design system for clothing patterns, characterized in that, The system includes a CAD design module, a CNC instruction generation module, and a CNC cutting bed execution layer. The CNC cutting bed execution layer is used to drive a cutting blade to cut the fabric according to the CNC machining instructions issued by the CNC instruction generation module. The system is characterized by further including: The pattern diagram structure construction module is used to receive the original pattern data output by the CAD design module, define the key points and alignment points of the cut piece outline as nodes, define the sewing topological connection relationship between nodes as edges, and assign geometric feature vectors and fabric attribute feature vectors to the nodes and edges respectively to construct the pattern diagram structure. The graph structure semantic embedding module is used to input the layout graph structure into the graph neural network encoder and perform neighborhood feature aggregation and graph-level pooling to obtain the global semantic embedding vector of the layout. A structured vector database is used to bind and store the global semantic embedding vector with the corresponding pattern original data and fabric constitutive parameters to form a pattern semantic index record; The similarity retrieval module is used to respond to a new pattern design request, extract the global semantic embedding vector of the new pattern, retrieve historical pattern semantic index records in the structured vector database whose similarity to the global semantic embedding vector exceeds a preset threshold, and obtain a recall set of historical patterns. The adaptive pre-compensation module is used to calculate a prediction compensation vector based on the historical deformation deviation data corresponding to each historical pattern in the recalled historical pattern set and the similarity between each historical pattern and the new pattern. The prediction compensation vector is then applied to the seam allowance parameter, thread angle parameter and key alignment point coordinate parameter of the new pattern to obtain the compensated and corrected pattern parameters, and output to the CNC instruction generation module. The CNC cutting bed execution layer also includes a vision sensor and a tension sensor, which are used to collect the measured coordinates of each key point of the cut piece and the amount of interlayer slippage during the cutting process; The deviation acquisition and feedback module is used to compare the measured coordinates of each key point of the cut piece with the corresponding CAD theoretical design coordinates to obtain the deformation deviation vector. After binding the deformation deviation vector with the semantic index record of the pattern used in this cutting, it is sent back and stored in the structured vector database to form a data closed loop between the design end and the execution end.

2. The computer-aided design system for clothing patterns according to claim 1, characterized in that: The geometric feature vector of the node includes the node's two-dimensional coordinates, local curvature, and seam allowance. The fabric attribute feature vector of the edge includes the stitching type, splicing direction, warp elastic modulus, weft elastic modulus, and heat shrinkage rate.

3. The computer-aided design system for clothing patterns according to claim 1, characterized in that: In the graph structure semantic embedding module, the hidden feature vector of a node at layer k+1 is obtained as follows: the hidden feature vector of the node at layer k is concatenated with the weighted sum of the hidden feature vectors of its neighboring nodes at layer k, and then multiplied by the learnable weight matrix and processed by a nonlinear activation function. The weights used in the weighted summation are the attention aggregation weights of each neighboring node to that node, which are used to characterize the degree of contribution of the neighboring node to semantic fusion.

4. The computer-aided design system for clothing patterns according to claim 1, characterized in that: The similarity retrieval module uses cosine similarity to calculate the similarity between the global semantic embedding vector of the new version and the global semantic embedding vector of each historical version in the structured vector database. The cosine similarity is the ratio of the dot product of the two vectors to the product of the magnitudes of the two vectors.

5. The computer-aided design system for clothing patterns according to claim 1, characterized in that: The adaptive pre-compensation module calculates normalized weights based on the similarity between each recalled historical version and the new version. The normalized weight of each recalled historical version is the ratio of the similarity of that historical version to the sum of the similarities of all recalled historical versions. The adaptive pre-compensation module performs weighted summation on the historical deformation deviation data corresponding to each recalled historical version according to the normalized weights to obtain the prediction compensation vector.

6. The computer-aided design system for clothing patterns according to claim 1, characterized in that: The deformation deviation vector is the difference between the measured coordinates of the key point of the cut piece and the coordinates of the key point in the CAD theoretical design. The deviation acquisition and feedback module is also used to bind the fabric batch number and the number of cutting layers corresponding to the deformation deviation vector with the pattern semantic index record and then send them back to the structured vector database.

7. The computer-aided design system for clothing patterns according to claim 1, characterized in that: The CNC cutting bed execution layer also includes a cutting blade execution mechanism, which is used to drive the cutting blade to cut multi-layer stacked fabric along a set trajectory according to the CNC machining instructions issued by the CNC instruction generation module based on the compensated and corrected pattern parameters.

8. A computer-aided design method for clothing patterns, characterized in that, include: Obtain the original pattern data, define the key points and alignment points of the cut piece outline as nodes, define the sewing topological connection relationship between nodes as edges, and assign geometric feature vectors and fabric attribute feature vectors to nodes and edges respectively to construct the pattern diagram structure. The pattern diagram structure is input into the graph neural network encoder and subjected to neighborhood feature aggregation and graph-level pooling to obtain the global semantic embedding vector of the pattern. The global semantic embedding vector is bound to the corresponding pattern original data and fabric constitutive parameters and stored in the structured vector database to form a pattern semantic index record; In response to a new pattern design request, the global semantic embedding vector of the new pattern is extracted, and historical pattern semantic index records with a similarity exceeding a preset threshold to the global semantic embedding vector are retrieved from the structured vector database to obtain a recall set of historical patterns. Based on the historical deformation deviation data corresponding to each historical pattern in the recalled historical pattern set and the similarity between each historical pattern and the new pattern, a prediction compensation vector is calculated, and the prediction compensation vector is applied to the seam allowance parameter, thread angle parameter and key alignment point coordinate parameter of the new pattern to obtain the compensated and corrected pattern parameters. Based on the compensated and corrected pattern parameters, CNC machining instructions are generated, and the CNC machining instructions are sent to the CNC cutting bed execution layer to drive the cutting blade to cut the fabric; During the cutting process, the measured coordinates and interlayer slip of each key point of the cut piece are collected by visual sensors and tension sensors. The measured coordinates are compared with the corresponding CAD theoretical design coordinates to obtain the deformation deviation vector. After binding the deformation deviation vector with the semantic index record of the pattern used in this cutting, it is sent back and stored in the structured vector database, forming a data closed loop between the design end and the execution end.

9. The computer-aided design method for clothing patterns according to claim 8, characterized in that: The similarity between the retrieved vector and the global semantic embedding vector is calculated using cosine similarity, which is the ratio of the dot product of the global semantic embedding vector of the new version and the global semantic embedding vector of the historical version to the product of their magnitudes.

10. A computer-aided design method for clothing patterns according to claim 8, characterized in that, The prediction compensation vector is calculated as follows: normalized weights are calculated based on the similarity between each recalled historical version and the new version. The normalized weight of each recalled historical version is the ratio of the similarity of that historical version to the sum of the similarities of all recalled historical versions. Then, the historical deformation deviation data corresponding to each recalled historical version are weighted and summed according to the normalized weights.

Images