A polyester spinning process path fast generation method and system based on order feature clustering

Abstract

The application relates to a polyester spinning process path fast generation method and system based on order feature clustering, which comprises the following steps: reading specification data of a target order and obtaining detection data of different batches of polyester raw materials to obtain order feature vectors and raw material feature vectors; a joint feature vector of the order and the raw material is constructed, and a joint feature matrix is constructed based on multiple joint feature vectors of all to-be-processed orders; a density clustering algorithm is used to cluster the multiple joint feature vectors in the joint feature matrix, and an order raw material clustering group is output; corresponding process path templates are matched for different order raw material clustering groups, and deviation compensation regulation and control of process parameters are performed on different to-be-processed orders in the same order raw material clustering group; quality data of actually produced products are collected as feedback samples, and function fitting adjustment is performed on the deviation compensation regulation and control of the process parameters. The application has the effect that the product quality of the same specification order produced under different raw material batches is improved.

Description

Technical Field

[0001] This application relates to the technical field of intelligent manufacturing of polyester spinning, and in particular to a method and system for rapid generation of polyester spinning process paths based on order feature clustering. Background Technology

[0002] In polyester spinning production, the process path, including the setting of parameters such as spinning temperature, cooling conditions, draw ratio, hot box temperature, and winding speed, directly determines the product quality. Currently, the mainstream method for generating process paths typically involves process engineers retrieving the closest process path from a historical process database based on the order's product type, fineness, and strength requirements. This allows for equipment selection and fine-tuning to meet the specific requirements of different product orders.

[0003] However, during the production scheduling process, the quality and specifications of the polyester raw materials used by enterprises can fluctuate from batch to batch. Often, orders with the same specifications and the same process will produce finished products with different specifications, leading to an increase in the number of unqualified products. Therefore, during production, most companies still rely on experience to allocate orders with high quality requirements to batches of raw materials with good viscosity, or to allocate orders with lenient quality requirements to raw materials with greater volatility. This human-find-materials model is inefficient, and it is difficult to optimize the overall process when there are many batches of raw materials and a variety of orders. This results in situations where high-grade raw materials are used by low-requirement orders, while high-requirement orders are forced to use volatile raw materials, which is unreasonable. This leads to resource misallocation and a decline in overall quality, so improvements are needed. Summary of the Invention

[0004] To improve the product quality of the same specification order produced under different raw material batches and reduce the risk of resource mismatch, this application provides a method and system for rapid generation of polyester spinning process paths based on order feature clustering.

[0005] The above-mentioned objective of this application is achieved through the following technical solution: A method for rapid generation of polyester spinning process paths based on order feature clustering, comprising the following steps: The system reads the specification data of the target order from the order management system and obtains the test data of different batches of polyester raw materials from the current raw material management system to obtain the order feature vector and the raw material feature vector. Based on the order feature vector and the raw material feature vector, a joint feature vector of orders and raw materials is constructed, and a joint feature matrix is ​​constructed based on multiple joint feature vectors of all orders to be processed; Density clustering algorithm is used to cluster multiple joint feature vectors in the joint feature matrix, and output order raw material cluster groups. Each order raw material cluster group contains joint feature vectors of several orders to be processed. Based on a pre-built process path template library, the corresponding process path template is matched for different order raw material clusters, and the deviation compensation and adjustment of process parameters are performed for different orders to be processed in the same order raw material cluster. The quality data of actual products are collected as feedback samples, and the deviation compensation and control of process parameters are adjusted by function fitting.

[0006] By adopting the above technical solution, in the data preparation stage, the solution performs anomaly detection on both order feature vectors and raw material feature vectors. Based on the process feasibility rule library of polyester spinning, it performs feasibility screening on each pair of order raw material combinations, ensuring that only physically reliable and process-feasible pairings can enter the subsequent clustering model. This eliminates invalid or unproducible combinations from the data source, avoiding distortion of the clustering structure due to a single erroneous data point, and laying a quality foundation for the stable operation of the entire system. Secondly, in the joint clustering stage, the solution weightedly integrates order specification parameters and raw material status parameters to construct a joint feature matrix. Addressing the uneven distribution density of order-raw material data in polyester spinning, it employs an adaptive neighborhood radius density clustering algorithm and optimizes uniformity based on indicators such as cluster size ratio and silhouette coefficient. This design ensures that each order raw material cluster in the clustering results has a reasonable sample size, high intra-group cohesion, and clear inter-group boundaries, avoiding the problem of large clusters swallowing small clusters or fragmented clustering. This results in orders within the same cluster being highly homogeneous in terms of quality requirements and raw material compatibility.

[0007] Furthermore, a process path template is matched to each order's raw material cluster. For individual orders within a group, the actual viscosity of the raw material deviates from the group's baseline viscosity. Key parameters such as draw ratio and oven temperature are dynamically adjusted using preset compensation rules. This compensation mechanism essentially maps minute differences in raw materials to precise corrections of process parameters, achieving fine-grained control from cluster-level baseline to order-level personalization. This effectively eliminates quality fluctuations caused by using different batches of raw materials for the same specification order. Finally, the compensation effect is quantitatively evaluated, and samples with poor compliance are used as training data to backfit and adjust the compensation function. This self-learning mechanism enables the compensation model to continuously approach the optimal mapping relationship and exhibits good generalization ability for newly added raw material types or order specifications. It reduces the workload of manually maintaining the process knowledge base, thus effectively solving the problems of poor process adaptability and low quality consistency caused by raw material viscosity fluctuations in polyester spinning production.

[0008] Optionally: The step of constructing a joint feature vector of orders and raw materials based on order feature vectors and raw material feature vectors, and constructing a joint feature matrix based on multiple joint feature vectors of all orders to be processed, includes: Independent anomaly detection is performed on the order characteristic vector and raw material characteristic vector; anomalies in either the order characteristic vector or the raw material characteristic vector are marked and individually filtered out. After screening, the order characteristic vector and raw material characteristic vector are cross-paired and the pairing feasibility is evaluated. Through the preset process feasibility rule library, the fusion cases where the order characteristic vector and raw material characteristic vector cannot be matched are screened out. The order characteristic vector and raw material characteristic vector after independent anomaly detection and pairing feasibility assessment are weighted and fused to obtain a joint feature vector; Let D be the vector dimension in the joint feature vector, and M be the number of feasible pairings of order characteristic vector and raw material characteristic vector. Construct an M×D joint feature matrix.

[0009] By adopting the above technical solutions, the independent anomaly detection stage performs missing value, threshold range, and logical consistency checks on orders and raw materials respectively, identifying and eliminating abnormal samples with incomplete data or physical values ​​exceeding reasonable ranges. This prevents single erroneous data from contaminating subsequent models at the source. Simultaneously, a review waiting window is set for suspected anomalies to avoid false rejection due to temporary detection errors. Secondly, the cross-pairing feasibility assessment stage, based on a polyester spinning process feasibility rule library, performs pair-level verification on orders and raw materials that have passed independent detection. The rule library covers key process boundaries such as the matching constraints between raw material viscosity and order strength requirements, and the special requirements of fine denier yarn for raw material spinnability. Pairings that are normal in one dimension but lack production feasibility after combination are filtered out, effectively avoiding invalid combinations in the clustering model that are theoretically feasible but practically unproducible. Finally, the weighted fusion stage dynamically adjusts the weight coefficients of order features and raw material features according to different production scenarios and eliminates amplitude differences through secondary standardization, ensuring that the joint feature vector accurately reflects the relative importance of order demand and raw material supply under different conditions.

[0010] Optionally: The step of using density clustering algorithm to cluster multiple joint feature vectors in the joint feature matrix and outputting order raw material cluster groups includes: Euclidean distance is calculated for each sample point in the joint feature matrix. The distance value of the k-th neighbor is taken as the initial value. All sample points are sorted to obtain the distance curve. The density partition boundary is divided by the inflection point of the distance curve. Based on the density partition boundary, the neighborhood radius of each sample point in the joint feature matrix is ​​set, and the core points and boundary points of the sample points are determined by the DBSCAN clustering algorithm and the neighborhood radius. Starting from any unvisited core point, cluster expansion is performed based on density reachability, assigning all density reachable core points and their associated boundary points to the same order raw material cluster group, and outputting n initial cluster groups.

[0011] By adopting the above technical solution, in the parameter adaptive setting stage, the k-th neighbor distance is calculated for all sample points and a distance curve is plotted. The inflection point of the curve is used to identify different density regions in the data space. Unlike the traditional DBSCAN which uses a globally fixed neighborhood radius, it automatically divides the boundaries of high-density, medium-density, and low-density regions according to the distribution characteristics of the data itself. This makes the clustering parameters match the local data density and reduces the occurrence of over-segmentation of high-density regions or misjudgment of low-density regions as noise due to improper global parameters. In the core point and boundary point determination stage, differentiated neighborhood radii and minimum sample number thresholds are set for sample points in different density partitions. This allows samples in sparse regions to be included in the cluster, while the clustering of dense regions will not be over-expanded due to excessively large neighborhood radii. This achieves a better balance in the granularity control of the clustering results.

[0012] Optionally: After the step of dividing all density-reachable core points and their associated boundary points into the same order raw material cluster group and outputting n initial cluster groups, the following steps are also performed: The uniformity of the n initial clusters is quantitatively evaluated, and the initial clusters that do not meet the uniformity requirements are marked. Perform splitting operations on excessively large clusters and merging operations on excessively small clusters; Output multiple order raw material clusters and their corresponding feature profiles; and label the mapping table from each sample to the cluster group.

[0013] By adopting the above technical solutions, in the uniformity quantification evaluation stage, different types of non-uniform clusters, such as oversized, undersized, and oddly shaped clusters, are identified, providing a quantitative basis for subsequent targeted adjustments. In the splitting operation stage, for oversized clusters, spectral clustering or subspace clustering methods are used to identify natural substructures within them, and samples are redistributed into multiple sub-clusters, making the internal features of the split sub-clusters purer and more uniform. In the merging operation stage, for undersized clusters, their centroids are calculated and the nearest clusters in the feature space are found for merging. This operation integrates fragmented clusters with insufficient sample size and lack of statistical significance into nearby reasonable clusters, while retaining special orders that cannot be merged as independent markers, reducing the processing complexity of the model. By outputting the feature profile of each cluster and the mapping table from samples to clusters, the feature profile includes key information such as centroid coordinates, in-group radius, and representative samples, providing directly callable structured data for downstream process template matching.

[0014] Optionally: The step of matching corresponding process path templates for different order raw material clusters based on a pre-built process path template library includes: Based on historically successful production batches, a process path template library was established for different types of orders. Using a template matching engine, the feature profile of each order's raw material cluster is compared with each process path template in the process path template library for multi-dimensional similarity calculation. The template with the highest comprehensive score is selected as the benchmark process path for that cluster. The process path templates in the process path template library are dynamically maintained. When a new type of order appears that cannot be covered by the existing process path templates and is successfully produced, the process path corresponding to the new order is added as a new template.

[0015] By adopting the above technical solution, a process path template library is established based on historically successful production batches. Each template contains a complete combination of process parameters such as spinning temperature, draw ratio, and winding speed. These templates are derived from proven production practices, ensuring the reliability and executability of the benchmark process path. Secondly, a template matching engine performs multi-dimensional similarity calculations between the feature profiles of each order's raw material cluster and the templates in the template library. Taking into account the matching degree between order specifications, raw material status characteristics, and the applicable scope of the template, the template with the highest comprehensive score is selected as the benchmark process path for that cluster. This mechanism transforms the complex problem of setting process parameters into an efficient similarity retrieval problem, significantly improving the generation speed of process paths. The process path template library is continuously updated. When a new type of order appears that cannot be covered by existing templates, the system automatically adds the corresponding process path as a new template after the new order is successfully produced. This allows for continuous enrichment and expansion with production practice, constantly improving the applicability of the process path templates.

[0016] Optionally: the step of adjusting the process parameters to compensate for deviations in different orders to be processed within the same raw material cluster group includes: Obtain any order in the raw material cluster and its actual allocated raw material viscosity value; calculate the viscosity deviation between the raw material viscosity value and the reference viscosity of the cluster to which the order belongs; Based on a pre-established database of compensation mapping relationships between viscosity deviation values ​​and process parameter compensation amounts, the initial compensation amount of the process parameters in the process path template corresponding to the order is obtained. Based on the specifications of the current order products, the initial compensation amount is scaled to obtain the final deviation compensation amount, and the process parameters in the process path template are adjusted based on the deviation compensation amount.

[0017] By adopting the above technical solution, the viscosity value of the raw material actually allocated in the order is obtained and compared with the benchmark viscosity of the cluster group to which the order belongs, and the viscosity deviation value is calculated. This deviation value quantifies the degree of difference between the current raw material and the standard state of the cluster group, providing accurate input parameters for subsequent compensation. Secondly, in the compensation amount acquisition stage, based on the pre-established mapping relationship library between viscosity deviation value and process parameter compensation amount, the initial compensation amount corresponding to the current viscosity deviation is quickly retrieved. After outputting the initial compensation amount, in the compensation amount scaling stage, the solution adaptively scales the initial compensation amount according to the product specification information of the current order. This mapping relationship library is derived from regression analysis of historical production data, which improves the empirical basis and reliability of the compensation amount. Moreover, different product specifications have different sensitivities to process parameter adjustments. Fine denier yarns require more precise compensation ranges, while coarse denier yarns have a relatively wider tolerance range. Therefore, scaling the compensation amount allows the compensation amount to be personalized according to the order characteristics, making the compensation of process parameters more accurate.

[0018] Optionally: The step of collecting quality data of actual produced products as feedback samples and adjusting the deviation compensation control of process parameters by function fitting includes: Collect actual product quality data and compare it with order requirement data to calculate the comparison data. The system performs validity screening on the comparison data of multiple orders, removes invalid samples, and outputs valid feedback samples. Based on the feedback samples, the compensation function in the compensation mapping relation library is adjusted, and the scaling function that scales the initial compensation amount is fitted and optimized.

[0019] By adopting the above technical solution, the actual product quality data is collected and compared with the order requirements item by item to calculate the degree of deviation of each indicator, providing an objective basis for subsequent model adjustment. The validity screening of the indicator comparison data of multiple orders is carried out to remove invalid samples caused by non-process reasons such as detection errors and equipment failures, so that the feedback samples entering the model optimization stage truly reflect the correlation between the compensation effect and the quality result. Furthermore, in the function fitting optimization stage, the solution is based on the screened valid feedback samples and simultaneously fits and adjusts the compensation function and scaling function in the compensation mapping relationship library. The compensation function determines the mapping relationship between viscosity deviation and process parameter compensation amount, while the scaling function controls the compensation magnitude correction for different specifications of orders. The synergistic optimization of the two functions enables the entire compensation model to continuously approach the optimal mapping relationship and has good generalization ability for newly emerging raw material batch types and order specifications.

[0020] The second objective of this invention is achieved through the following technical solution: A rapid generation system for polyester spinning process paths based on order feature clustering, comprising: The vector generation module is used to read the specification data of the target order from the order management system and obtain the test data of different batches of polyester raw materials from the current raw material management system to obtain the order feature vector and the raw material feature vector. The matrix construction module is used to construct a joint feature vector of orders and raw materials based on the order feature vector and the raw material feature vector, and to construct a joint feature matrix based on multiple joint feature vectors of all orders to be processed; The clustering module is used to cluster multiple joint feature vectors in the joint feature matrix using the density clustering algorithm, and outputs order raw material cluster groups. Each order raw material cluster group contains joint feature vectors of several orders to be processed. The compensation and control module is used to match the corresponding process path templates for different order raw material clusters based on a pre-built process path template library, and to perform deviation compensation and control of process parameters for different orders to be processed in the same order raw material cluster. The feedback module is used to collect quality data of actual products as feedback samples and adjust the deviation compensation control of process parameters by function fitting.

[0021] The above-mentioned objective three of this application is achieved through the following technical solution: A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for rapid generation of polyester spinning process paths based on order feature clustering.

[0022] The fourth objective of this application is achieved through the following technical solution: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method for rapidly generating polyester spinning process paths based on order feature clustering.

[0023] In summary, this application includes at least one of the following beneficial technical effects: Anomaly detection is performed on both order feature vectors and raw material feature vectors to ensure that only physically reliable and technologically feasible pairings can enter the subsequent clustering model. Order specification parameters and raw material state parameters are weighted and fused to construct a joint feature matrix, making orders within the same cluster highly homogeneous in terms of quality requirements and raw material compatibility. A process path template is matched to each order's raw material cluster group, and key parameters such as draw ratio and hot box temperature are dynamically adjusted according to preset compensation rules for deviations between the actual viscosity of raw materials in individual orders within the group and the group's benchmark viscosity. The essence of this compensation mechanism is to map the small differences in raw materials into precise corrections of process parameters, realizing fine control from cluster group-level benchmark to order-level personalization. This effectively eliminates quality fluctuations caused by the use of different batches of raw materials for the same specification order. The compensation effect is quantitatively evaluated, and the compensation model can continuously approach the optimal mapping relationship. At the same time, it has good generalization ability for new raw material types or order specifications, reducing the workload of manually maintaining the process knowledge base. Therefore, it effectively solves the problem of poor process adaptability and low quality consistency caused by raw material viscosity fluctuations in polyester spinning production. Pairs that are normal in one dimension but lack production feasibility after combination are filtered out, effectively avoiding invalid combinations in the clustering model that are theoretically feasible but practically unproducible. Finally, the weighted fusion stage dynamically adjusts the weight coefficients of order features and raw material features according to different production scenarios, and eliminates amplitude differences through secondary standardization, so that the joint feature vector accurately reflects the relative importance of order demand and raw material supply under different conditions; In the core point and boundary point determination stage, differentiated neighborhood radii and minimum sample number thresholds are set for sample points in different density partitions, so that samples in sparse areas have the opportunity to be included in clusters, while clusters in dense areas will not be over-expanded due to excessively large neighborhood radii, thus achieving a better balance in the granularity control of clustering results. In the merging process, for excessively small clusters, their centroids are calculated and the nearest clusters in the feature space are found for merging. This operation integrates fragmented clusters with insufficient sample size and lack of statistical significance into nearby reasonable clusters, while retaining special orders that cannot be merged as independent labels, thus reducing the processing complexity of the model. By outputting the feature profile of each cluster and the mapping table from samples to clusters, the feature profile contains key information such as centroid coordinates, in-group radius, and representative samples, providing directly callable structured data for downstream process template matching. Attached Figure Description

[0024] Figure 1 This is a flowchart of an embodiment of a method for rapid generation of polyester spinning process paths based on order feature clustering in this application; Figure 2This is a schematic diagram illustrating the clustering process in an embodiment of a method for rapidly generating polyester spinning process paths based on order feature clustering, as described in this application. Figure 3 This is a schematic diagram of the deviation compensation and control principle in an embodiment of a rapid generation method for polyester spinning process path based on order feature clustering in this application; Figure 4 This is a schematic block diagram of a computer device according to this application. Detailed Implementation

[0025] The following is in conjunction with the appendix Figure 1-4 This application will be described in further detail.

[0026] In one embodiment, such as Figure 1 As shown, this application discloses a method for rapid generation of polyester spinning process paths based on order feature clustering, which specifically includes the following steps: S10: Read the specification data of the target order from the order management system and obtain the test data of different batches of polyester raw materials from the current raw material management system to obtain the order feature vector and the raw material feature vector; In this embodiment, the order specification data includes product type, such as polyester pre-oriented yarn, stretch textured yarn and fully stretched yarn; nominal fineness, number of filaments, target breaking strength requirement, target breaking elongation range, target boiling water shrinkage range and dyeing uniformity requirement. After normalizing the above feature values, an order feature vector is formed.

[0027] The test data of polyester raw materials include, but are not limited to, intrinsic viscosity and its deviation from the reference viscosity, such as 0.68 dL / g, end carboxyl group content, number of aggregated particles, melting point, etc., and the raw material feature vector after normalization.

[0028] S20: Based on the order feature vector and the raw material feature vector, construct a joint feature vector of orders and raw materials, and based on multiple joint feature vectors of all orders to be processed, construct a joint feature matrix; In this embodiment, step S20 includes the following steps: S21: Perform independent anomaly detection on the order characteristic vector and raw material characteristic vector; mark and individually filter out orders with anomalies or raw material characteristic vectors with anomalies. S22: After screening, the order characteristic vector and the raw material characteristic vector are cross-paired and the pairing feasibility is evaluated. Through the preset process feasibility rule library, the fusion cases where the order characteristic vector and the raw material characteristic vector cannot be matched are screened out. S23: Weighted fusion of the order characteristic vector and raw material characteristic vector after independent anomaly detection and pairing feasibility assessment to obtain a joint feature vector; S24: Let the vector dimension in the joint feature vector be D, and the number of feasible order feature vector and raw material feature vector pairings be M. Construct an M×D joint feature matrix.

[0029] The independent anomaly detection of order feature vectors includes missing value detection, threshold range detection, logical consistency detection, and duplicate order detection. Missing value detection involves marking any key feature, such as target breaking strength or nominal fineness, as an abnormal order and refusing it from entering the joint matrix if it is empty. Threshold range detection includes detecting the range of target breaking strength; if it exceeds the range, it is marked and manual review is requested. Logical consistency detection involves detecting and marking an abnormal ratio of nominal fineness to the number of filaments. Duplicate order detection involves determining whether the order feature vector is exactly the same as an existing order feature vector; if it is a duplicate order, one is retained.

[0030] Independent anomaly detection of raw material feature vectors includes viscosity value reasonableness detection, viscosity mutation detection, and data integrity detection. If the viscosity exceeds the acceptable range for polyester spinning, it is marked as an abnormal raw material batch. Viscosity mutation occurs when the difference between two consecutive test values ​​of the same batch of raw material is too large.

[0031] Feasibility matching involves cross-matching and assessing the feasibility of orders and raw materials based on the first step of independent anomaly detection, eliminating those pairings that are normal in one dimension but are not feasible when combined.

[0032] Furthermore, raw materials and orders that have passed independent anomaly detection and cross-pairing evaluation are weighted and fused. Different weight coefficients are assigned to the order feature vector and raw material feature vector according to different polyester spinning scenarios, including routine production, periods of raw material shortage, high-end orders, and scenarios with significant batch variations in raw materials. The fused joint feature vector is then standardized a second time to eliminate the magnitude differences introduced by the scaling of the weight coefficients. S30: Density clustering algorithm is used to cluster multiple joint feature vectors in the joint feature matrix, and the order raw material cluster group is output. Each order raw material cluster group contains the joint feature vectors of several orders to be processed. Reference Figure 2 In this embodiment, step S30 includes the following steps: S31: Calculate the Euclidean distance for each sample point in the joint feature matrix, take the distance value of the k-th neighbor as the initial value, sort all sample points to obtain the distance curve, and divide the density partition boundary through the inflection point of the distance curve. S32: Based on the density partition boundary, set the neighborhood radius of each sample point in the joint feature matrix, and use the DBSCAN clustering algorithm and neighborhood radius to determine the core points and boundary points of the sample points; S33: Starting from any unvisited core point, perform cluster expansion through density reachability relationships, assign all density reachable core points and their associated boundary points to the same order raw material cluster group, and output n initial cluster groups.

[0033] Specifically, in the traditional DBSCAN algorithm, the neighborhood radius and minimum number of samples are globally fixed. However, this approach cannot adapt to the uneven data distribution density in the joint feature matrix of polyester spinning, i.e., some regions have dense order-raw material combinations, while others are sparse. This can easily lead to problems such as large clusters swallowing small clusters or sparse points being isolated in the clustering results. In this embodiment, a local density adaptive strategy is adopted to dynamically calculate different clustering parameters for different feature space regions. Here, the distance value of its k-th nearest neighbor is taken and denoted as k_dist, where k is a preset initial value, and in this embodiment k=4~6; the k_dist values ​​of all sample points are arranged in ascending order, and a k-distance curve is plotted; multiple inflection points in the curve are identified, and each inflection point corresponds to a density partition boundary.

[0034] The adaptive parameter positioning for density zoning is shown in Table 1 below:

[0035] Near the partition boundaries, linear interpolation is used to calculate the smoothed parameter values ​​to avoid abrupt changes in clustering results at the partition boundaries. For boundary points, the actual ε used is obtained by weighted averaging of the ε values ​​of the left and right partitions. The weight is inversely proportional to the distance from the point to the partition center. Adaptive parameter setting solves the failure problem of traditional DBSCAN on data with uneven density and lays the foundation for the uniformity of subsequent cluster groups.

[0036] Furthermore, for each sample point, the number of sample points falling into the hypersphere centered on the sample point and with the radius of the neighborhood is calculated using the local neighborhood radius of the point's location. The local minimum number of samples, MinPts, at the point's location is used as the decision threshold. Boundary points are defined as points that are not core points but fall within the neighborhood radius of a certain core point.

[0037] The output order raw material clusters contain a list of sample points and their core and boundary markers within each cluster.

[0038] Furthermore, after step S33, the following steps are also performed: S34: Perform a uniformity quantification evaluation on the n initial cluster groups, and mark the initial cluster groups that do not meet the uniformity requirements; S35: Perform a split operation on excessively large clusters and a merge operation on excessively small clusters; S36: Output multiple order raw material clusters and their corresponding feature profiles; and mark the mapping table from each sample to the cluster group.

[0039] The uniformity requirements stipulate that excessively large clusters, excessively small clusters, and oddly shaped order material clusters all need to be adjusted. An excessively large cluster refers to an order material cluster containing too many samples, for example, exceeding 40% of the total samples, and may also contain multiple naturally complete sub-order material clusters. An excessively small cluster refers to an order material cluster with too few samples; in this embodiment, it is set to less than the minimum sample size, which lacks statistical significance. An oddly shaped cluster refers to an order material cluster whose shape in the feature space is too elongated or irregular, which is not conducive to subsequent template allocation.

[0040] For order raw material clusters marked as oversized, the subspace clustering method is used to identify their internal natural substructures. The sample points in the oversized order raw material clusters are reassigned to 2 to 3 sub-clusters. The size of each sub-order raw material cluster is then verified to be below the maximum size threshold. If it is not satisfied, the subspace clustering method is used to split the clusters.

[0041] For the order raw material cluster group marked as too small, calculate the centroid of the order raw material cluster group; Find the other order raw material clusters that are closest to the centroid of the current order raw material cluster in the joint feature matrix, specifically measured by the Euclidean distance between centroids; further merge the order raw material clusters that are too small into the other order raw material clusters that are closest to the centroid; verify whether the average silhouette coefficient of the new group after merging decreases by more than 0.1; if it decreases too much, abandon the merging, mark the small cluster as a special order group, and keep it separately; the sub-clusters generated after splitting are used as new order raw material clusters and retained in the clustering results.

[0042] The feature profile of the order raw material cluster includes the number of samples within the group, the centroid within the group, the most representative sample within the group (i.e., the sample point closest to the centroid), the order feature centroid, the raw material feature centroid, the radius within the group, and the quality label within the group.

[0043] S40: Based on a pre-built process path template library, match the corresponding process path template for different order raw material clusters, and perform deviation compensation and adjustment of process parameters for different orders to be processed in the same order raw material cluster. Reference Figure 3 In this embodiment, the process path template matching step includes: S41: Based on historically successful production batches, establish a process path template library for different types of orders; S42: Using a template matching engine, the feature profile of each order's raw material cluster is compared with each process path template in the process path template library to perform multi-dimensional similarity calculations. The template with the highest comprehensive score is selected as the benchmark process path for that cluster. S43: Dynamically maintain the process path templates in the process path template library. When a new type of order appears that cannot be covered by the existing process path template and is successfully produced, add the process path corresponding to the new order as a new template.

[0044] The criteria for judging excellent batches are that the key quality indicators such as the product’s breaking strength, breaking elongation, boiling water shrinkage, and dyeing uniformity all meet the order requirements and there are no records of downgrading. For each group, the process path parameters of all excellent batches within the group are statistically analyzed. The average value of each process parameter is calculated as the baseline parameter of the template for that group. At the same time, the standard deviation of each parameter is calculated as the allowable fluctuation range of the template. The order feature centroid of the group, the template baseline parameter, and the allowable fluctuation range are stored together to form a complete process path template.

[0045] Determine the common feature dimension used for similarity calculation. This dimension should exist in both the feature profile of the cluster group and the feature description of the process path template, including but not limited to product variety, nominal fineness, target breaking strength, target breaking elongation, dyeing uniformity requirements, boiling water shrinkage requirements, and special processing requirements.

[0046] Furthermore, the process parameter deviation compensation and control steps include: S44: Obtain any order in the raw material cluster and its actual allocated raw material viscosity value; calculate the viscosity deviation between the raw material viscosity value and the reference viscosity of the order's cluster; S45: Based on the pre-established compensation mapping relationship library between viscosity deviation values ​​and process parameter compensation amounts, obtain the initial compensation amount of the process parameters in the process path template corresponding to the order; S46: Based on the specifications of the current order products, scale the initial compensation amount to obtain the final deviation compensation amount, and adjust the process parameters in the process path template through the deviation compensation amount.

[0047] This involves establishing a compensation mapping relationship library based on historical production data through regression analysis or experimental design. Specifically, this includes extracting historical production batch data from the MES or production database. Each record should simultaneously include the actual raw material viscosity used in that batch and the corresponding reference viscosity, and calculating the deviation δ. It also includes the actual process parameters used in that batch, such as spinning temperature, draw ratio, hot box temperature, and winding speed. Finally, it includes the quality test results of that batch of products at the time of leaving the factory, such as breaking strength, elongation, and dyeing uniformity.

[0048] Further statistical methods were used to screen the input variables that were most strongly correlated with the compensation amounts of the process parameters. Specifically, this included calculating the Pearson correlation coefficients between each candidate input variable (deviation δ, fineness, strength requirements, etc.) and each output compensation amount, and eliminating variables with insignificant correlations; and automatically selecting the combination of independent variables that had the strongest explanatory power for the dependent variable by using a stepwise introduction and elimination method. A multi-source linear regression model is selected, suitable for scenarios where there is an approximately linear relationship between δ and the compensation amount. The model form is: ΔP = β0 + β1·δ + β2·X1 + β3·X2 + ε; where β0 is the intercept, β1, β2, and β3 are the regression coefficients, X1 and X2 are other influencing factors (such as fineness grade and strength grade), and ε is the random error term. The least squares method is used to minimize the sum of squared residuals between the predicted compensation amount and the actual compensation amount used.

[0049] After fitting, the goodness-of-fit R-squared of the model needs to be used to assess its reliability. 2 Test, R 2 The closer the correlation coefficient R is to 1, the higher the model's explanatory power for historical data. 2 In this embodiment, a value of 0.95 or higher is required for the model to be considered reliable.

[0050] Finally, the multi-source linear regression model was converted into a compensation mapping relationship and stored as structured data, as shown in Table 2 below:

[0051] By substituting the δ and specification parameters of the current order into the corresponding regression equation, the initial compensation amount can be quickly calculated.

[0052] The mapping relationship library records the correspondence between viscosity deviation values ​​and compensation amounts for various process parameters under different order types. Each mapping record includes at least an order type label, viscosity deviation range, and compensation amounts for various process parameters, such as draw ratio compensation value, hot box temperature compensation value, and winding speed compensation value. The mapping relationship library is stored in the database for easy querying.

[0053] Using the order type label and viscosity deviation value as the query keys, retrieve matching records from the compensation mapping relationship database; If the viscosity deviation value δ precisely matches a certain interval, the compensation amount corresponding to that interval is read directly; if the viscosity deviation value δ falls between two intervals, the compensation amount is calculated using a linear interpolation method; specifically: ΔP=ΔPlower+(δ-δlower) / (δupper-δlower)×(ΔPupper-ΔPlower); Where δlower and δupper are the boundary values ​​of adjacent intervals, and ΔPlower and ΔPupper are the corresponding compensation amounts; The compensation values ​​of each process parameter obtained through querying or interpolation are organized into an initial compensation value set: ΔPinitial={ΔTspin,ΔDR,ΔThot,Δvwind,...}, where ΔTspin is the spinning temperature compensation amount, ΔDR is the draw ratio compensation amount, ΔThot is the hot box temperature compensation amount, and Δvwind is the winding speed compensation amount.

[0054] Extract key parameters affecting compensation sensitivity from the current order's specification information; including the following sensitivity factors: Fineness factor: Fine denier yarns with a fineness of ≤50 dtex are sensitive to parameter changes, and the compensation range should be appropriately reduced; Coarse denier yarns with a fineness of ≥200 dtex are more tolerant, and the compensation range can be normal or slightly increased.

[0055] Monofilament number factor: Porous filaments with a monofilament number ≥96f are sensitive to changes in spinning temperature, and the compensation range needs to be reduced; Strength requirement factor: Orders with high strength requirements respond significantly to draw ratio compensation, and the compensation range can be executed normally; Elongation requirement factor: Narrow range elongation requirement is sensitive to hot box temperature compensation, and the compensation range needs to be appropriately reduced.

[0056] The above sensitivity factors are weighted and fused to obtain the scaling comprehensive coefficient λ. Each item in the initial compensation set is multiplied by the comprehensive scaling coefficient λ to obtain the final deviation compensation amount ΔPfinal, which includes: ΔPfinal = λ·ΔPinitial; Finally, adjust the process parameters in the process path template, including: Pfinal = Pbase + ΔPfinal; where Pbase is the original parameter value in the process path template.

[0057] S50: Collect quality data of actual produced products as feedback samples, and adjust the deviation compensation control of process parameters by function fitting.

[0058] Specifically, step S50 includes the following steps: S51: Collect actual product quality data and compare it with order requirement data to calculate the comparison data of indicators; S52: Perform validity screening on the comparison data of multiple orders' metrics, remove invalid samples, and output valid feedback samples; S53: Based on the feedback samples, adjust the compensation function in the compensation mapping relationship library, and fit and optimize the scaling function that scales the initial compensation amount.

[0059] Specifically, this includes defining the functions that need to be optimized, including compensation functions and scaling functions.

[0060] The compensation function fcomp(δ, type) describes the mapping relationship between the viscosity deviation δ and the initial compensation amount ΔPinitia of the process parameters. This function can be expressed as a piecewise linear function, a polynomial function, or a lightweight neural network, with the specific form chosen based on the characteristics of the historical data.

[0061] The scaling function gscale(spec) describes the mapping relationship between the order specification information spec, including fineness, number of monofilaments, strength requirements, etc., and the comprehensive scaling factor λ. This function can be expressed as a weighted linear function or a nonlinear regression model.

[0062] For each valid feedback sample, extract the viscosity deviation value δ and the order specification feature vector spec corresponding to that sample; the actual compensation amount ΔPused and scaling factor λused used in the production process of that sample; and further optimize the compensation function fcomp by using the viscosity deviation value δ and the order type label as input features and the actual initial compensation amount ΔPused as the output target; and by using the least squares method or gradient descent method to adjust the parameters of the compensation function to minimize the error between the function output and the actual compensation amount used. After optimization, calculate the prediction error of the updated compensation function on the validation set. If the error is significantly reduced, replace the old function with the new function.

[0063] Based on the selected positive training samples, the scaling function gscale is fitted and optimized, including taking the order specification feature vector spec as the input feature and the comprehensive scaling coefficient λused actually used in the sample as the output target; for the weighted linear scaling function, the weight coefficients w1, w2, w3, w4 of each sensitivity factor are updated through multiple linear regression; for the non-linear scaling function, neural network fitting or decision tree regression can be used to update the parameters.

[0064] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0065] In one embodiment, a rapid generation system for polyester spinning process paths based on order feature clustering is provided. This rapid generation system for polyester spinning process paths based on order feature clustering corresponds to the rapid generation method for polyester spinning process paths based on order feature clustering described in the previous embodiment. The rapid generation system for polyester spinning process paths based on order feature clustering includes: The vector generation module is used to read the specification data of the target order from the order management system and obtain the test data of different batches of polyester raw materials from the current raw material management system to obtain the order feature vector and the raw material feature vector. The matrix construction module is used to construct a joint feature vector of orders and raw materials based on the order feature vector and the raw material feature vector, and to construct a joint feature matrix based on multiple joint feature vectors of all orders to be processed; The clustering module is used to cluster multiple joint feature vectors in the joint feature matrix using the density clustering algorithm, and outputs order raw material cluster groups. Each order raw material cluster group contains joint feature vectors of several orders to be processed. The compensation and control module is used to match the corresponding process path templates for different order raw material clusters based on a pre-built process path template library, and to perform deviation compensation and control of process parameters for different orders to be processed in the same order raw material cluster. The feedback module is used to collect quality data of actual products as feedback samples and adjust the deviation compensation control of process parameters by function fitting.

[0066] Specific limitations regarding the rapid generation system for polyester spinning process paths based on order feature clustering can be found in the limitations of the rapid generation method for polyester spinning process paths based on order feature clustering described above, and will not be repeated here. Each module in the aforementioned rapid generation system for polyester spinning process paths based on order feature clustering can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0067] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a method for rapid generation of polyester spinning process paths based on order feature clustering.

[0068] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements a method for rapidly generating polyester spinning process paths based on order feature clustering.

[0069] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements a method for rapidly generating polyester spinning process paths based on order feature clustering.

[0070] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0071] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0072] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for rapid generation of polyester spinning process paths based on order feature clustering, characterized in that, Including the following steps: The system reads the specification data of the target order from the order management system and obtains the test data of different batches of polyester raw materials from the current raw material management system to obtain the order feature vector and the raw material feature vector. Based on the order feature vector and the raw material feature vector, a joint feature vector of orders and raw materials is constructed, and a joint feature matrix is ​​constructed based on multiple joint feature vectors of all orders to be processed; Density clustering algorithm is used to cluster multiple joint feature vectors in the joint feature matrix, and output order raw material cluster groups. Each order raw material cluster group contains joint feature vectors of several orders to be processed. Based on a pre-built process path template library, the corresponding process path template is matched for different order raw material clusters, and the deviation compensation and adjustment of process parameters are performed for different orders to be processed in the same order raw material cluster. The quality data of actual products are collected as feedback samples, and the deviation compensation and control of process parameters are adjusted by function fitting. The step of using density clustering algorithm to cluster multiple joint feature vectors in the joint feature matrix and outputting order raw material cluster groups includes: Euclidean distance is calculated for each sample point in the joint feature matrix. The distance value of the k-th neighbor is taken as the initial value. All sample points are sorted to obtain the distance curve. The density partition boundary is divided by the inflection point of the distance curve. Based on the density partition boundary, the neighborhood radius of each sample point in the joint feature matrix is ​​set, and the core points and boundary points of the sample points are determined by the DBSCAN clustering algorithm and the neighborhood radius. Starting from any unvisited core point, cluster expansion is performed based on density reachability, assigning all density reachable core points and their associated boundary points to the same order raw material cluster group, and outputting n initial cluster groups.

2. The method for rapid generation of polyester spinning process paths based on order feature clustering according to claim 1, characterized in that, The steps of constructing a joint feature vector of orders and raw materials based on order feature vectors and raw material feature vectors, and constructing a joint feature matrix based on multiple joint feature vectors of all orders to be processed, include: Independent anomaly detection is performed on the order characteristic vector and raw material characteristic vector; anomalies in either the order characteristic vector or the raw material characteristic vector are marked and individually filtered out. After screening, the order characteristic vector and raw material characteristic vector are cross-paired and the pairing feasibility is evaluated. Through the preset process feasibility rule library, the fusion cases where the order characteristic vector and raw material characteristic vector cannot be matched are screened out. The order characteristic vector and raw material characteristic vector after independent anomaly detection and pairing feasibility assessment are weighted and fused to obtain a joint feature vector; Let D be the vector dimension in the joint feature vector, and M be the number of feasible pairings of order characteristic vector and raw material characteristic vector. Construct an M×D joint feature matrix.

3. The method for rapid generation of polyester spinning process paths based on order feature clustering according to claim 1, characterized in that, After the step of dividing all density-reachable core points and their associated boundary points into the same order raw material cluster group and outputting n initial cluster groups, the following steps are also performed: The uniformity of the n initial clusters is quantitatively evaluated, and the initial clusters that do not meet the uniformity requirements are marked. Perform splitting operations on excessively large clusters and merging operations on excessively small clusters; Output multiple order raw material clusters and their corresponding feature profiles; And label each sample with a mapping table to a cluster group.

4. The method for rapid generation of polyester spinning process paths based on order feature clustering according to claim 1, characterized in that, The step of matching corresponding process path templates for different order raw material clusters based on a pre-built process path template library includes: Based on historically successful production batches, a process path template library was established for different types of orders. Using a template matching engine, the feature profile of each order's raw material cluster is compared with each process path template in the process path template library for multi-dimensional similarity calculation. The template with the highest comprehensive score is selected as the benchmark process path for that cluster. The process path templates in the process path template library are dynamically maintained. When a new type of order appears that cannot be covered by the existing process path templates and is successfully produced, the process path corresponding to the new order is added as a new template.

5. The method for rapid generation of polyester spinning process paths based on order feature clustering according to claim 4, characterized in that, The step of adjusting and compensating for deviations in process parameters for different orders to be processed within the same raw material cluster group includes: Obtain any order in the raw material cluster and its actual allocated raw material viscosity value; calculate the viscosity deviation between the raw material viscosity value and the reference viscosity of the order's cluster; Based on a pre-established database of compensation mapping relationships between viscosity deviation values ​​and process parameter compensation amounts, the initial compensation amount of the process parameters in the process path template corresponding to the order is obtained. Based on the specifications of the current order products, the initial compensation amount is scaled to obtain the final deviation compensation amount, and the process parameters in the process path template are adjusted based on the deviation compensation amount.

6. The method for rapid generation of polyester spinning process paths based on order feature clustering according to claim 1, characterized in that, The step of collecting quality data of actual produced products as feedback samples and adjusting the deviation compensation control of process parameters by function fitting includes: Collect actual product quality data and compare it with order requirement data to calculate the comparison data. The system performs validity screening on the comparison data of multiple orders, removes invalid samples, and outputs valid feedback samples. Based on the feedback samples, the compensation function in the compensation mapping relation library is adjusted, and the scaling function that scales the initial compensation amount is fitted and optimized.

7. A rapid generation system for polyester spinning process paths based on order feature clustering, characterized in that, include: The vector generation module is used to read the specification data of the target order from the order management system and obtain the test data of different batches of polyester raw materials from the current raw material management system to obtain the order feature vector and the raw material feature vector. The matrix construction module is used to construct a joint feature vector of orders and raw materials based on the order feature vector and the raw material feature vector, and to construct a joint feature matrix based on multiple joint feature vectors of all orders to be processed; The clustering module is used to cluster multiple joint feature vectors in the joint feature matrix using a density clustering algorithm, and output order raw material cluster groups. Each order raw material cluster group contains joint feature vectors of several orders to be processed. The module includes calculating the Euclidean distance for each sample point in the joint feature matrix, taking the distance value of the k-th neighbor as the initial value, sorting all sample points to obtain the distance curve, and dividing the density partition boundary by the inflection point of the distance curve. Based on the density partition boundary, the neighborhood radius of each sample point in the joint feature matrix is ​​set, and the core points and boundary points of the sample points are determined by the DBSCAN clustering algorithm and the neighborhood radius. Starting from any unvisited core point, cluster expansion is performed based on density reachability, assigning all density reachable core points and their associated boundary points to the same order raw material cluster group, and outputting n initial cluster groups; The compensation and control module is used to match the corresponding process path templates for different order raw material clusters based on a pre-built process path template library, and to perform deviation compensation and control of process parameters for different orders to be processed in the same order raw material cluster. The feedback module is used to collect quality data of actual products as feedback samples and adjust the deviation compensation control of process parameters by function fitting.

8. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the rapid generation method for polyester spinning process paths based on order feature clustering as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the rapid generation method for polyester spinning process paths based on order feature clustering as described in any one of claims 1 to 6.

Images