Textile production quality management method

Abstract

The present application relates to the technical field of image processing, and discloses a textile production quality management and control method, comprising: acquiring multi-modal data of the textile, the multi-modal data comprising fabric image data and multi-channel process parameter time series data; constructing a space-time graph attention network; extracting and fusing the spatial features of the fabric image data and the temporal features of the process parameter time series data to obtain a fusion representation vector; establishing a Gaussian process regression model, constructing a defect propagation field, and calculating a defect density gradient and a state transition probability; constructing a structural causal model of the production process to determine the intervention priority of the key process parameters; establishing an optimization model to solve the model to obtain an optimal process parameter adjustment combination and generate equipment control instructions. The present application overcomes the problem of superficial cause analysis caused by data silos in traditional methods by acquiring and deeply fusing multi-modal data of fabric images and process parameters to construct the internal mapping relationship between apparent quality and production process state.

Description

Technical Field

[0001] The present invention relates to the technical field of image processing, and in particular to a method for controlling the quality of textile production. Background Art

[0002] In the modern textile industry, product quality is a core factor determining a company's competitiveness. To ensure and improve textile quality, various monitoring systems are typically deployed on production lines. Existing quality control technologies rely primarily on two main branches: machine vision-based fabric surface defect detection, such as automated optical inspection (AOI) systems. These systems capture fabric images with high-speed cameras and use image processing algorithms to identify local defects such as broken warps, broken wefts, and stains. Sensor-based production process monitoring, such as statistical process control (SPC), monitors key process parameters such as yarn tension, machine speed, and temperature and humidity to determine whether the production process is stable. However, in practice, these two technologies often operate independently, creating data silos. Image inspection systems can only identify the "identity" of defects, while process parameter monitoring only reflects the "process status." The lack of effective and in-depth dynamic correlation between the two makes it difficult to directly link apparent quality issues with specific process parameter fluctuations. In addition, traditional defect detection and analysis is often limited to static, isolated defect points, and lacks a macroscopic characterization of the spatial distribution, aggregation, and evolution trends of defects on the fabric plane, making the tracing and prediction of quality problems extremely difficult and heavily dependent on the personal experience of front-line engineers.

[0003] Existing technologies also have significant limitations in fault diagnosis and process optimization. When quality anomalies are detected, traditional methods often use correlation analysis to identify possible causes, analyzing the statistical relationship between defect rates and changes in various process parameters. However, correlation does not equate to causation. Numerous process parameters are coupled and influence each other during production. Simple correlation analysis is likely to reveal spurious correlations, making it difficult to pinpoint the root drivers of quality fluctuations. When making process adjustments, operators often rely on experience or fixed rulebooks, making tentative adjustments to single variables or within a small range. This partial optimization approach, which treats the symptoms but not the root cause, ignores the complex physical constraints (such as tension balance) and synergistic effects between process parameters. This not only results in inefficient adjustments, but can also disrupt the overall balance of the production system through inappropriate interventions, potentially leading to new and more complex quality issues. Therefore, the industry urgently needs a new intelligent quality control approach that can deeply integrate multi-source data, reveal the causal mechanisms of quality issues, and provide globally optimal control strategies to overcome current technical bottlenecks. Summary of the Invention

[0004] The application provides a textile production quality management method to solve the problem that the image detection and parameter monitoring lack deep dynamic correlation in the prior art, and the apparent defects cannot be directly linked with specific process parameter fluctuations, so that the quality problem is difficult to trace.

[0005] The textile production quality management method of the application comprises the following steps:

[0006] The textile production quality management method of the application comprises the following steps:

[0007] Preferably, the construction of the spatio-temporal graph attention network comprises: dividing the fabric image into a plurality of image segmentation regions along the fabric running direction and the width direction as image nodes, and taking each process parameter sensor as a parameter node; establishing a connection edge between nodes based on the coverage relationship between the physical monitoring range of the sensor and the image segmentation region; assigning an image feature extracted by a pre-trained convolutional neural network to each image node, and assigning a time sequence feature extracted by a long short-term memory network to each parameter node as the initial representation of the node.

[0008] Preferably, the extraction and fusion of the spatial features of the fabric image data and the time features of the process parameter time sequence data by the spatio-temporal graph attention network to obtain the fusion representation vector of the fabric state comprises: updating the node representation by aggregating the feature information of the adjacent nodes of each node through the graph attention layer of the spatio-temporal graph attention network; and splicing all the updated node representations and inputting them into a fully connected network to obtain the fusion representation vector of the fabric state.

[0009] Preferably, the Gaussian process regression model is established to take the fusion feature vector as input, and a defect propagation field representing the spatial distribution probability of the fabric defects is constructed, including: taking the fusion feature vector as input and the defect type and location in the historical data as output for training; and using the trained model to predict the posterior probability distribution of the defect type for each image segmentation region, thereby constructing the defect propagation field.

[0010] Preferably, the defect density gradient and the state transition probability of the defect type between adjacent regions are calculated, including: calculating the change amount of the posterior probability of a specific defect type between adjacent image segmentation regions along the running direction of the fabric as the defect density gradient; and calculating the state transition probability of the defect type between adjacent image segmentation regions based on the posterior probability distribution.

[0011] Preferably, the intervention priority of the key process parameters is determined, including: learning a structural causal model between each process parameter and the defect density gradient from the historical data by using a causal discovery algorithm; quantifying the causal effect of each process parameter on the defect density gradient based on the structural causal model, and sorting the process parameters according to the absolute value of the causal effect to obtain the intervention priority.

[0012] Preferably, the objective function of the optimization model is to minimize the total entropy of the defect probability distribution of all image segmentation regions in the defect propagation field; and the constraint conditions of the optimization model include the value range constraint of each process parameter to be adjusted and the physical model constraint based on the yarn tension balance equation.

[0013] Preferably, the objective function of the optimization model is to minimize the total entropy of the defect probability distribution of all image segmentation regions in the defect propagation field; and the constraint conditions of the optimization model include the value range constraint of each process parameter to be adjusted and the physical model constraint based on the yarn tension balance equation.

[0014] Preferably, the optimal process parameter adjustment combination is obtained by solving the optimization model, including: using a sequential quadratic programming algorithm to solve the optimization model to obtain a set of optimal process parameters that minimize the objective function and satisfy all constraint conditions.

[0015] Preferably, the pre-trained convolutional neural network is a residual network.

[0016] The beneficial effects of the present application are: the present application acquires and deeply fuses the fabric image and the multi-modal process parameter data, constructs the internal and structured mapping relationship between the appearance quality and the production process state, and overcomes the problem of shallow cause analysis caused by data silos in traditional methods. By converting discrete defect points into continuous defect propagation fields for modeling, the spatial distribution trend of defects is quantified and predicted from a macro perspective, providing a forward-looking perspective for quality control. The present application uses causal inference to replace simple correlation analysis, which can accurately locate the root driving factors causing quality problems, significantly improving the accuracy of fault tracing. On this basis, the global optimization adjustment strategy constructed takes into account the integrity of the defect propagation field and the physical constraints between process parameters, and the parameter adjustment combination given not only can efficiently eliminate defects, but also can guarantee the overall stability of the production system, avoiding new problems caused by local and exploratory adjustments, thereby realizing fine and systematic control of textile production quality. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A flowchart of a textile production quality control method provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0018] Embodiments of the present application will be described in detail below, examples of which are shown in the accompanying drawings. The embodiments described below by reference to the drawings are exemplary and are intended to explain the present application, and cannot be understood as limiting the present application.

[0019] As shown in Figure 1 , the textile production quality control method provided by the embodiment of the present application specifically includes the following steps:

[0020] S1, acquiring multi-modal data at predetermined nodes of a textile production line, the multi-modal data including fabric image data continuously collected along the running direction of the fabric and multi-channel process parameter time series data collected synchronously with the fabric image data.

[0021] Specifically, a line array camera is installed at the entrance of the loom to continuously shoot fabric images, and tension, speed, and temperature sensors are deployed at positions such as the warp beam, the warp beam, and the take-up roller. A unified clock signal based on the main drive shaft encoder is used to stamp the synchronous time stamp for each frame of image data and the sensor readings at the corresponding time, forming a one-to-one image-parameter data pair.

[0022] S2, construct a spatio-temporal graph attention network, the nodes of the spatio-temporal graph attention network are process parameter sensors and image segmentation regions of the fabric image, the edges of the spatio-temporal graph attention network are determined based on a mapping relationship between physical positions of the sensors and the image regions; and spatial features of the fabric image data and time features of process parameter time series data are extracted and fused by using the spatio-temporal graph attention network to obtain a fusion representation vector of a fabric state.

[0023] Specifically, a single fabric image is divided into M*N grid-shaped image segmentation regions as image nodes, and each process parameter sensor is a parameter node. If a sensor physically affects an image segmentation region, an edge is established between the corresponding nodes. For example, the left warp beam tension sensor node establishes edges with all image nodes, but the connection weight with the left image segmentation region node is higher. Image features of each image segmentation region are extracted by using a residual network ResNet, and time series features of each process parameter are extracted by using a long short-term memory network LSTM, and then the features are used as initial node embeddings and input into a graph attention network GAT, neighbor node information is aggregated by a multi-head attention mechanism, and finally a fusion representation vector of each image segmentation region node is output.

[0024] S3, a Gaussian process regression model is established, the fusion representation vector is used as input, a defect propagation field representing spatial distribution probability of fabric defects is constructed, and a defect density gradient and a state transition probability of a defect type between adjacent image segmentation regions are calculated.

[0025] Specifically, the fusion representation vector output by the spatio-temporal graph attention network is used as input of the Gaussian process regression model, and a two-dimensional coordinate of the fabric is used as an index. The Gaussian process regression model outputs a mean value and a variance of a probability of defect occurrence at each coordinate point, and these probability values jointly constitute a continuous defect propagation field. The partial derivative of the mean value function of the defect propagation field is calculated, and a defect density gradient along the warp and weft directions of the fabric is obtained. Based on the joint probability distribution characteristics of the Gaussian process, the posterior probability of the defect type of any two adjacent image segmentation regions is calculated, and a transition probability matrix of the defect type from one image segmentation region to another image segmentation region is obtained.

[0026] S4, a structural causal model of a production process is constructed based on historical data, causal effects of each process parameter on the defect density gradient are quantified by causal inference, and an intervention priority of a key process parameter is determined.

[0027] Specifically, first, an initial causal relationship directed acyclic graph is constructed based on textile process knowledge, in which nodes contain each process parameter and defect density gradient. Then, using a PC algorithm or LiNGAM model, the graph structure is verified and corrected using historical production data to obtain the final structural causal model. Using the backdoor adjustment formula, the average causal effect value of the defect density gradient when each process parameter changes by one unit is calculated by simulating intervention on the structural causal model. The effect value is sorted from large to small, and the intervention priority list of each process parameter is obtained.

[0028] S5, determine the key process parameters to be adjusted according to the intervention priority, establish an optimization model with the minimum overall entropy of the defect propagation field as the optimization objective and the physical model containing the yarn tension balance equation as the constraint, solve the optimization model to obtain the optimal process parameter adjustment combination, and generate device control instructions.

[0029] Specifically, the top N key process parameters in the intervention priority list are selected as decision variables. The overall entropy of the defect propagation field is obtained by calculating the Shannon entropy of the defect probability distribution at each point in the field and integrating it. The tension relationship of the yarn during transmission between each component of the loom, such as the balance relationship between the warp tension, the weaving tension and the winding tension, is expressed as a set of algebraic or differential equations as physical constraints. Then a nonlinear programming problem is constructed, and the sequential quadratic programming algorithm or interior point method is used to solve it to obtain a set of optimal process parameter settings that minimize the objective function and satisfy all constraints. Finally, this set of settings is converted into control signals for corresponding actuators such as frequency converters or servo motors and sent to the device controller.

[0030] In an optional embodiment, a space-time graph attention network is constructed, including:

[0031] The fabric image is divided into multiple image segmentation regions along the fabric running direction and the width direction as image nodes, and each process parameter sensor is taken as a parameter node; the connection edges between nodes are established based on the coverage relationship between the sensor physical monitoring range and the image segmentation region; the image features extracted by the pre-trained convolutional neural network are assigned to each image node, and the time series features extracted by the long short-term memory network are assigned to each parameter node as the initial representation of the node.

[0032] Specifically, this step aims to unify the multi-source heterogeneous production data into a graph structure. For example, for a fabric image with a width of 1920 pixels and a length of 1080 pixels along the running direction, it can be uniformly divided into 192 by 108 image segmentation regions with a size of 10 by 10 pixels along the running direction and the width direction, and each image segmentation region is taken as an image node. At the same time, assuming that 5 sensors are deployed on the production line to monitor the warp tension, weft tension, loom speed, let-off rate and temperature respectively, then the 5 sensors are taken as parameter nodes.

[0033] Specifically, the connection relationship between nodes reflects the influence range in the physical world. For example, a warp tension sensor monitors the area within the range of 500 to 1000 pixels in the width direction of the fabric, and the parameter node corresponding to the sensor will establish a connection edge with all image nodes located in this width range. Subsequently, initial features are assigned to the nodes. For each image node, a 2048-dimensional feature vector of a 10 by 10 pixel image block is extracted using a ResNet50 network pre-trained on the ImageNet dataset. For each parameter node, assuming that it collects data every second within the past 60 seconds, forming a time series with a length of 60, after processing by a long short-term memory network, a 128-dimensional feature vector is generated as its initial representation.

[0034] In an optional embodiment, the spatial features of the fabric image data and the temporal features of the process parameter time series data are extracted and fused by the spatio-temporal graph attention network to obtain a fusion representation vector of the fabric state, comprising: aggregating the feature information of the adjacent nodes of each node by the graph attention layer of the spatio-temporal graph attention network to update the node representation; concatenating all updated node representations and inputting them into a fully connected network to obtain the fusion representation vector of the fabric state.

[0035] Specifically, the graph attention layer is the core of information aggregation. For any node in the graph, for example, an image node representing a small area in the center of the fabric, this layer calculates the contribution of all its adjacent nodes, including the image nodes above, below, left and right of it and the process parameter nodes that may cover this area, to the update of its features. For example, if an adjacent tension sensor node shows abnormal fluctuations, the graph attention mechanism will give the features of the parameter node a higher weight, so that more information about the tension anomaly is integrated when updating the representation of the current image node.

[0036] After the iterative update of one or more layers of graph attention layers, each node obtains a new representation that integrates local spatio-temporal neighborhood information. Assuming there are 20736 image nodes and 5 parameter nodes in the graph, the representation of each node is updated to 256 dimensions. At this point, the 20741 256-dimensional vectors are concatenated in a predetermined order to form a very high-dimensional long vector. This long vector is then input into a fully connected network consisting of three hidden layers. The network performs non-linear transformation and dimension reduction layer by layer, and finally outputs a fusion representation vector of, for example, 1024 dimensions. The fusion representation vector highly condenses the visual information of the entire fabric image at the current time and the dynamic information of all key process parameters.

[0037] In an optional embodiment, the Gaussian process regression model is established to take the fusion representation vector as input to construct a defect propagation field representing the spatial distribution probability of fabric defects, including: training with the fusion representation vector as input and the defect type and location in the historical data as output; using the trained model to predict the posterior probability distribution of the defect type for each current image segmentation region, thereby constructing the defect propagation field.

[0038] Specifically, the training stage uses historical production data. 10000 samples are selected from the historical database, each corresponding to a 1024-dimensional fusion representation vector at a time point and a real defect map corresponding to the time point. The map records the defect type of each image segmentation region, for example, using numerical coding 0 to represent normal, 1 to represent broken warp, and 2 to represent weft skew. The model is trained with the fusion representation vector as input and the defect map as output to learn the complex mapping relationship between the two.

[0039] When a new production data produces a current fusion representation vector, the trained Gaussian process regression model will make a prediction. The unique feature of the model is that it does not directly output a certain defect type, but outputs a posterior probability distribution of the defect type for each image segmentation region. For example, for the i-th row and j-th column image segmentation region, the model may predict that the probability of being a broken warp is 70%, the probability of being a weft skew is 20%, and the probability of being normal is 10%. Combining the probability distributions of all image segmentation regions forms a defect propagation field covering the entire fabric surface, which visually displays the spatial distribution trend and severity of each type of defect on the fabric in the form of a probability cloud map.

[0040] In an optional embodiment, the calculation of the defect density gradient and the state transition probability of the defect type between adjacent regions includes: calculating the change amount of the posterior probability of a specific defect type between adjacent image segmentation regions along the running direction of the fabric as the defect density gradient; and calculating the state transition probability of the defect type between adjacent regions based on the posterior probability distribution.

[0041] Specifically, the defect density gradient is used to quantify the speed of defect evolution along the production direction. Assuming the fabric runs along the Y-axis direction, we focus on the broken end defects. If the average posterior probability of broken end in a row of segmentation region with Y-coordinate 100 is forty percent, while in the adjacent row with Y-coordinate 101, the average probability rises to forty-five percent, then the density gradient of broken end defects at this point is five percent. High gradient values indicate that defects are rapidly developing or spreading in this area, which is an area of concern.

[0042] The state transition probability describes the propagation characteristics of defects from another perspective. Based on historical data and the current posterior probability distribution, the probability of state transition from one region to its downstream adjacent region can be calculated. For example, by analyzing a large amount of data, it is found that when the probability of a region appearing broken end is higher than eighty percent, the probability of its downstream adjacent region appearing broken end reaches ninety percent, while the probability of appearing weft skew is only one percent. This state transition probability matrix reveals the mutual influence and propagation law between different defect types, providing a quantitative basis for predicting the future trend of defects.

[0043] In an optional embodiment, determining the intervention priority of the key process parameters comprises: learning a structural causal model between each process parameter and the defect density gradient from historical data using a causal discovery algorithm; quantifying the causal effect of each process parameter on the defect density gradient based on the structural causal model by causal inference, and sorting the process parameters according to the absolute value of the causal effect to obtain the intervention priority.

[0044] Specifically, this step aims to fundamentally find the cause of defect deterioration. Collect the sequences of each process parameter, such as warp tension, loom speed, etc., and the corresponding sequences of broken end defect density gradient changes over time in historical data. Using causal discovery techniques such as PC algorithm or FCI algorithm, analyze these time series data, and a directed graph can be constructed to represent the causal relationship between them. For example, the model may find that the change of warp tension is the direct cause of the change of broken end density gradient, while the loom speed indirectly affects the defect by affecting the warp tension.

[0045] After the causal structure model is established, causal inference techniques are used to quantify the strength of this influence. For example, using do-calculus, it can be calculated that increasing the warp tension by 1 Newton will cause the broken warp density gradient to decrease by an average of 0.05 units, under the ideal intervention that all other parameters remain unchanged. Similarly, it is calculated that decreasing the loom speed by 10 revolutions per minute has a causal effect of decreasing the broken warp density gradient by 0.02 units. Comparing the absolute values of these two causal effects, 0.05 is greater than 0.02, so the intervention priority of adjusting the warp tension is higher than adjusting the loom speed. By ranking the causal effects of all process parameters, a clear list of intervention priorities is obtained.

[0046] In an optional embodiment, the objective function of the optimization model is to minimize the total entropy of the flaw probability distribution of all image segmentation regions in the flaw propagation field; the constraint conditions of the optimization model include the value range constraints of each process parameter to be adjusted, and the physical model constraints based on the yarn tension balance equation.

[0047] Specifically, the objective of the optimization model is to make the fabric quality state most certainly tend to be flawless. Information entropy is an indicator of uncertainty, and the more dispersed the flaw probability distribution of a region, the higher its entropy value, indicating that its state is more uncertain. For example, a region has a probability of 50% of being broken warp and 50% of being normal, and its entropy value is high. The optimization objective is to adjust the process parameters so that this distribution tends to have a normal probability of 99% and a broken warp probability of 1%, at which time the entropy value is very low. Minimizing the total entropy of all regions is to drive the entire fabric surface to be in a high-probability, certain state of being flawless.

[0048] The optimization process must comply with the limitations of actual production, the first type being parameter range constraints, for example, the adjustment range of warp tension must be between 20 and 50 Newtons, which is determined by equipment capacity and material bearing limits; the second type being physical law constraints, for example, the yarn tension balance equation is a physical formula describing the interdependent relationship between the warp feed rate, loom speed, and warp tension. When finding the optimal parameter combination, the values of these three parameters must always satisfy this equation, ensuring that the optimization result is physically feasible and stable, and avoiding the adjustment of a parameter that violates physical laws and causes more serious problems.

[0049] The implementation principle of the textile production quality control method of the embodiment of the present application is: by obtaining and deeply fusing the fabric image and the process parameter multi-modal data, the internal and structured mapping relationship between the apparent quality and the production process state is constructed, and the problem of superficial cause analysis caused by data silos in the traditional quality control method is overcome. By converting discrete defect points into continuous defect propagation field for modeling, the spatial distribution trend of defects can be quantified and predicted from a macroscopic point of view, providing a forward-looking perspective for quality control. In addition, the simple correlation analysis is replaced by the causal inference, which can accurately locate the root driving factors leading to quality problems, significantly improving the accuracy of fault tracing. Moreover, the constructed global optimization adjustment strategy comprehensively considers the integrity of the defect propagation field and the physical constraints between the process parameters, and the parameter adjustment combination given can not only efficiently eliminate defects, but also guarantee the overall stability of the production system, avoid new problems caused by trial and error adjustment, and thus realize the fine and systematic control of the production quality of textiles.

[0050] Although the embodiments of the present application have been shown and described above, it can be understood that the above-mentioned embodiments are exemplary and cannot be understood as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-mentioned embodiments within the scope of the present application.

Claims

1. A textile production quality control method, characterized in that: The method comprises the following steps: acquiring multimodal data at predetermined nodes of a textile production line, the multimodal data comprising fabric image data continuously collected along a fabric running direction and multi-channel process parameter time series data collected synchronously with the fabric image data; constructing a spatiotemporal graph attention network, wherein the nodes of the spatiotemporal graph attention network are process parameter sensors and image segmentation regions of the fabric image, and the edges of the spatiotemporal graph attention network are determined based on a mapping relationship between the physical position of the sensor and the image region; and utilizing the spatiotemporal graph attention network to extract and fuse the spatial features of the fabric image data with the temporal features of the process parameter time series data to obtain a fused representation vector of the fabric state. Establishing a Gaussian process regression model, taking the fused representation vector as input, constructing a defect propagation field that represents the spatial distribution probability of fabric defects, and calculating the defect density gradient and the state transition probability of defect types between adjacent regions; A structural causal model of the production process is constructed based on historical data. The causal effect of each process parameter on the defect density gradient is quantified through causal inference, and the intervention priority of key process parameters is determined. The key process parameters to be adjusted are determined according to the intervention priority, and an optimization model is established with minimizing the overall entropy of the defect propagation field as the optimization goal and with a physical model containing the yarn tension balance equation as the constraint. The optimization model is solved to obtain the optimal process parameter adjustment combination, and equipment control instructions are generated.

2. The textile production quality control method according to claim 1, characterized in that: The construction of the spatiotemporal graph attention network includes: dividing the fabric image along the fabric running direction and width direction into multiple image segmentation areas as image nodes, and each process parameter sensor as a parameter node; establishing connecting edges between nodes based on the coverage relationship between the sensor's physical monitoring range and the image segmentation area; assigning image features extracted by a pre-trained convolutional neural network to each image node, and assigning time series features extracted by a long short-term memory network to each parameter node as the initial representation of the node.

3. The textile production quality control method according to claim 2, characterized in that: The method of extracting and fusing the spatial features of the fabric image data and the temporal features of the process parameter time series data using the spatiotemporal graph attention network to obtain a fused representation vector of the fabric state includes: Through the graph attention layer of the spatiotemporal graph attention network, the feature information of the adjacent nodes of each node is aggregated to update the node representation; all updated node representations are spliced ​​and input into the fully connected network to obtain the fused representation vector of the fabric state.

4. The textile production quality control method according to claim 1, characterized in that: The Gaussian process regression model is established, and the fused representation vector is used as input to construct a defect propagation field representing the spatial distribution probability of fabric defects, including: The fused representation vector is used as input and the defect type and location in the historical data is used as output for training; the trained model is used to predict the posterior probability distribution of the defect type for each current image segmentation area, thereby constructing the defect propagation field.

5. The textile production quality control method according to claim 4, characterized in that: The calculation of the defect density gradient and the state transition probability of the defect type between adjacent areas includes: Along the fabric running direction, the change in the posterior probability of a specific defect type between adjacent image segmentation areas is calculated as the defect density gradient; based on the posterior probability distribution, the state transition probability of the defect type between adjacent image segmentation areas is calculated.

6. The textile production quality control method according to claim 5, characterized in that: The determination of intervention priorities for key process parameters includes: A causal discovery algorithm is used to learn a structural causal model between each process parameter and the defect density gradient from historical data. Based on the structural causal model, the causal effect of each process parameter on the defect density gradient is quantified through causal inference, and the process parameters are sorted according to the absolute value of the causal effect to obtain the intervention priority.

7. The textile production quality control method according to claim 6, characterized in that: The causal discovery algorithm is a PC algorithm or a FCI algorithm.

8. The textile production quality control method according to claim 1, characterized in that: The objective function of the optimization model is to minimize the total entropy of the defect probability distribution of all image segmentation areas in the defect propagation field; the constraints of the optimization model include the value range constraints of each process parameter to be adjusted and the physical model constraints based on the yarn tension balance equation.

9. The textile production quality control method according to claim 8, characterized in that: Solving the optimization model to obtain the optimal process parameter adjustment combination includes: using a sequential quadratic programming algorithm to solve the optimization model to obtain a set of optimal process parameters that minimize the objective function and meet all constraints.

10. The textile production quality control method according to claim 2, characterized in that: The pre-trained convolutional neural network is a residual network.

Images