CN-121997754-A - Yarn tensile property prediction and process adjustment system based on multi-source data fusion

CN121997754ACN 121997754 ACN121997754 ACN 121997754ACN-121997754-A

Abstract

The invention discloses a yarn tensile property prediction and process optimization system based on multi-source data fusion, relates to the technical field of high-end textile intelligent manufacturing and industrial artificial intelligent intersection, solves the technical problems that accurate and automatic quantitative characterization of a microstructure in a yarn is difficult, multi-source heterogeneous data such as fiber properties, yarn structures, production processes, equipment states and the like are difficult to deeply fuse, and a prediction mechanism capable of simultaneously predicting two mutually-restricted performance indexes of the tensile strength and the elongation at break of the yarn in a high precision manner is lacked.

Inventors

LIU ZHIHAI
LI RUITAO
DAI WENQUAN
Jiang e
YU XIAOWEN
XUE WENJUAN

Assignees

海阳市清鸿制衣有限公司

Dates

Publication Date: 20260508
Application Date: 20260128

Claims (10)

1. Yarn tensile property prediction and process optimization system based on multi-source data fusion is characterized by comprising the following modules: The fiber and structure quantization module is used for acquiring fiber property quantization data and yarn structure quantization data; The production process monitoring module is used for acquiring time sequence sensing data and real-time process parameter data in the production process; The fusion prediction model module is connected with the fiber and structure quantization module and the production process monitoring module and comprises a first feature extraction network, a second feature extraction network and a feature fusion unit, wherein the input of the first feature extraction network is the fiber property quantization data and yarn structure quantization data, the input of the second feature extraction network is the time sequence sensing data and real-time process parameter data, and the feature fusion unit is used for coupling a first feature vector output by the first feature extraction network and a second feature vector output by the second feature extraction network and outputting predicted values of yarn tensile strength and elongation at break; And the process tuning module is connected with the fusion prediction model module, and is used for setting double-layer optimization constraint by taking the predicted value as a performance index, and adjusting controllable parameters in the real-time process parameter data through an optimization algorithm so as to generate a process parameter recommended value.
2. The multi-source data fusion-based yarn tensile property prediction and process optimization system according to claim 1, wherein the fiber property quantification data comprise fiber types, lengths, fineness and strength, and the yarn structure quantification data comprise yarn twist, orientation angles of fiber arrangement and fiber radial distribution.
3. The multi-source data fusion-based yarn tensile property prediction and process tuning system of claim 2, wherein the orientation angle of the fiber arrangement and the radial distribution of the fibers comprises: synchronously collecting two vertical side views and one cross-sectional view of the same section of yarn sample, and carrying out spatial registration on the three views to construct a unified two-dimensional coordinate system; Based on the mass center position of the fiber in the cross-section view and the edge continuity of the fiber in two longitudinal side views, unique identity identification is allocated to each independent fiber, and cross-view tracking is performed under the two-dimensional coordinate system; The method comprises the steps of calculating the cross section area of each fiber by taking the geometric center of the yarn as an origin in the cross section view, and calculating the effective bearing contribution of the fiber to the yarn at the corresponding radial position by combining the three-dimensional true orientation angle, wherein the space included angle between the fitting axis of each fiber and the integral axial direction of the yarn is used as the three-dimensional true orientation angle of the fiber: , wherein, In order to efficiently carry the contribution value, Is the cross-sectional area of the fiber, Is a three-dimensional true orientation angle; And polymerizing three-dimensional real orientation angles of all fibers according to the identity marks and the effective bearing contribution values of the fibers and the radial positions of the fibers to respectively generate a three-dimensional orientation angle distribution field and a radial effective bearing distribution field in the yarns, and outputting the three-dimensional orientation angle distribution field and the radial effective bearing distribution field serving as quantized outputs of the orientation angles of the fiber arrangement and the radial distribution of the fibers.
4. The multi-source data fusion-based yarn tensile property prediction and process optimization system of claim 1, wherein the first feature extraction network outputs a first feature vector comprising: The first feature extraction network organizes the fiber property quantized data and yarn structure quantized data into three physical levels, including fiber level features, aggregate level features and interaction level features; The cross-level features are derived by performing cross calculation on the fiber level features and the aggregate level features, and comprise weighting statistics of the fiber strength and the orientation angle distribution and change rate of the fiber fineness distribution at radial positions; The first feature extraction network respectively processes the feature groups of the three physical levels through a fully-connected encoder and output-splices the feature groups to form the first feature vector.
5. The multi-source data fusion-based yarn tensile property prediction and process optimization system according to claim 1, wherein the second feature extraction network outputs a second feature vector, comprising: the second feature extraction network organizes the time sequence sensing data and the real-time process parameter data into two information flows, including a steady-state process flow and a dynamic disturbance flow; The dynamic disturbance flow is used for carrying out time-frequency analysis on the time-sequence sensing data and extracting the energy and the main frequency of a specific frequency band of the dynamic disturbance flow as disturbance characteristics; and the second feature extraction network processes the dynamic disturbance flow through a gating circulation unit network, and splices the final hidden state of the dynamic disturbance flow with the embedded vector of the steady-state process flow, and compresses the second feature extraction network through a full-connection layer to form the second feature vector.
6. The multi-source data fusion-based yarn tensile property prediction and process optimization system according to claim 1, wherein the feature fusion unit comprises: inputting the first feature vector into a bottleneck hypothesis network to output a multidimensional bottleneck hypothesis vector Each dimension of the multi-dimensional bottleneck hypothesis vector corresponds to a potential performance constraint pattern hypothesis; inputting the second feature vector into an evidence extraction network to output a process evidence vector For quantifying the intensity of evidence supporting or negating various bottleneck hypotheses; calculating a matching degree matrix: , wherein, Obtaining a matching degree matrix for the outer product operation Each element of (2) Represent the first Bottleneck hypothesis and the first Matching degree of class process evidence, and matching degree matrix Singular value decomposition is carried out, and left singular vectors and right singular vectors corresponding to the maximum singular value are respectively used as a common-knowledge bottleneck mode Evidence of consensus process ; Splicing all vectors to obtain a final fusion feature vector , wherein, As a first feature vector of the set of features, As a second feature vector of the set of features, For the purpose of the transposition, Is a Sigmoid function.
7. The multi-source data fusion-based yarn tensile property prediction and process tuning system of claim 1, wherein the fusion feature vector comprises: The fusion feature vector Inputting the yarn tensile strength predicted value and the elongation at break predicted value into a double-channel decoupling predicted head, wherein the predicted head is used for generating the yarn tensile strength predicted value and the elongation at break predicted value; A structural reference path for inputting the first feature vector to a reference feedforward network and outputting a structural reference vector , , wherein, To base tensile strength based on fiber and structural data, A process correction path for inputting the fusion characteristic vector to a correction feedforward network and outputting a process correction matrix The method specifically comprises the following steps: Wherein each element Representing performance in process information pair In the correction of (a) and performance The associated coupling component is used to determine the coupling ratio, Is of the performance For the coupling component in the self-correction, Is of the performance Performance of pair The coupling component in the correction is such that, Is of the performance Performance of pair The coupling component in the correction is such that, A coupling component in the self-correction for performance; according to the structural reference vector And process correction matrix And calculating a final tensile strength correction amount and an elongation at break correction amount, and adding the structural reference and the process correction to obtain a final predicted value.
8. The multi-source data fusion-based yarn tensile property prediction and process tuning system of claim 7, wherein the calculating the final tensile strength modifier and elongation at break modifier, adding the structural reference to the process modifier to obtain the final predicted value, comprises: The final tensile strength correction amount and the elongation at break correction amount are obtained through linear transformation, and specifically: I.e. ; Wherein, the method comprises the steps of, In order to correct the amount of tensile strength, In order to correct the amount of elongation at break, Is a transposition; adding the structural reference and the process correction to obtain a final predicted value: ; wherein, the method comprises the steps of, As a predicted value of the tensile strength of the yarn, Is a predicted value of elongation at break.
9. The multi-source data fusion-based yarn tensile property prediction and process tuning system of claim 7, wherein the reference feed-forward network, the correction feed-forward network, and all network parameters involved in the fusion feature vector are combined end-to-end supervised training by minimizing a composite loss function using a sample dataset with a true strength tag and a true elongation tag; the composite loss function is specifically: ; ; ; Wherein, the As a function of the composite loss, The loss is predicted for the reference and, In order to ultimately predict the loss, In order to be a regular term of the coupling strength, In the case of a true intensity label, In order for the elongation to be a true label, Is a preset regularized intensity coefficient.
10. The multi-source data fusion-based yarn tensile property prediction and process tuning system of claim 1, wherein the process tuning module comprises: The method comprises the steps of taking controllable parameters in real-time process parameter data as decision variables to construct an optimization problem comprising three targets, wherein the three targets comprise a first target for maximizing the predicted tensile strength value of the yarn, a second target for maximizing the predicted elongation at break value and a third target for minimizing a process disturbance index, and the process disturbance index is obtained by calculating components related to dynamic disturbance in a second feature vector output by a second feature extraction network; The double-layer optimization constraint comprises a first-layer constraint which is a numerical range allowed by technological parameters, a second-layer constraint which is a structural consistency constraint, and a random candidate technological parameter set which is required to be input into the fiber and structure quantization module and the first characteristic extraction network together with current fiber property quantization data, wherein the Euclidean distance between the predicted yarn structure quantization data and a preset ideal structure reference vector is smaller than a set threshold value; Adopting Bayesian optimization framework to search decision variable space iteratively under the condition of meeting the double-layer optimization constraint, and using acquisition function in each iteration step Selecting a next evaluation point from the candidate set, the acquisition function , wherein, And (3) with The mean value and standard deviation of the comprehensive utility predictions of the agent model to the first target and the second target are respectively, A predicted value of the process disturbance index for the evaluation point, And (3) with Is a balance coefficient; and after the optimization process is finished, outputting a group of non-dominant pareto optimal solution sets on the first target, the second target and the third target, wherein each solution in the optimal solution sets comprises a group of technological parameter recommended values and three item label predicted values corresponding to the technological parameter recommended values.

Description

Yarn tensile property prediction and process adjustment system based on multi-source data fusion Technical Field The invention belongs to the technical field of high-end textile intelligent manufacturing and industrial artificial intelligence intersection, and particularly relates to a yarn tensile property prediction and process optimization system based on multi-source data fusion. Background The tensile strength and elongation at break of the yarn are key mechanical indexes for determining the quality of the final product and the performance of a downstream textile product, and in the traditional technology, the two performances are mainly obtained through off-line physical tests, and the result is lagged and cannot be used for real-time control and optimization of the production process. In the prior art, partial researches try to predict the yarn performance by utilizing partial process parameters or sensing data in the production process and combining a machine learning method, however, the technical problems that the prior art mostly depends on limited process parameters (twist degree and speed) or simple on-line sensing signals, the fundamental factors for determining the yarn performance, namely fiber self properties (strength and fineness) and the microstructure (fiber arrangement, orientation and radial distribution) in the yarn are ignored, the model is known to be natural due to splitting, prediction precision and generalization capability encounter bottlenecks, in few researches for attempting to fuse various data, a simple characteristic splicing or shallow layer fusion is usually adopted, a complex and nonlinear interaction influence mechanism between static properties of fibers/structures and dynamic disturbance in the production process cannot be deeply mined, so that information fusion is insufficient, the model is difficult to cope with complex and changeable actual working conditions, the existing prediction model is mostly a black box, the prediction result has poor interpretability, the existing prediction result is difficult to be directly and reliably used for guiding the reverse optimization of the production process parameters, the process optimization often depends on engineering experience or simply meets the single-objective search, and the intelligent optimizing capability under the condition of the constraint of multiple performance and the intelligent structure is lacked. Therefore, a system for predicting the tensile property of yarns and optimizing the process based on multi-source data fusion is needed. Disclosure of Invention The invention aims to at least solve one of the technical problems in the prior art, and therefore, the invention provides a yarn tensile property prediction and process optimization system based on multi-source data fusion, which is used for solving the following technical problems: The method is difficult to accurately and automatically quantitatively characterize microstructures (orientation angles and radial distribution of fiber arrangement) in yarns, so that key intrinsic information for determining macroscopic mechanical properties of the yarns is lost, an accurate structure-performance correlation model is difficult to build, multi-source heterogeneous data such as intrinsically static fiber properties, yarn structures, dynamically changeable production processes and equipment states are difficult to deeply fuse, information utilization of a prediction model is insufficient, generalization capability is weak, a prediction mechanism capable of simultaneously predicting two mutually restricted performance indexes such as tensile strength and elongation at break of the yarns with high precision is lacking, a multi-target intelligent optimization method taking the high-precision prediction model as a core and simultaneously considering production process stability and microstructure rationality of the yarns is also difficult to reversely recommend an optimal or near optimal solution capable of comprehensively improving product quality and production robustness from a mass of process parameter combinations with high efficiency and reliability. In order to solve the problems, the invention provides a yarn tensile property prediction and process optimization system based on multi-source data fusion, which comprises the following modules: The fiber and structure quantization module is used for acquiring fiber property quantization data and yarn structure quantization data; The production process monitoring module is used for acquiring time sequence sensing data and real-time process parameter data in the production process; The fusion prediction model module is connected with the fiber and structure quantization module and the production process monitoring module and comprises a first feature extraction network, a second feature extraction network and a feature fusion unit, wherein the input of the first feature extraction network is the fiber prope