A method for analyzing and processing textile yarn production quality detection data

Abstract

The application relates to the technical field of textile equipment data processing, and discloses a kind of analysis and processing method of textile yarn production quality detection data, the method comprises: obtaining quality time sequence signal and binary state signal, determine discrete time delay offset based on cross-correlation operation to generate phase alignment sequence;Set dynamic sliding window to calculate the first-order backward difference of the trace ratio of the eigenvalue matrix, determine the total spindle position dimension of spatial calculation cluster;Combine the sequence and state signal to generate effective space-time observation matrix and perform singular value decomposition, extract the left singular vector and separate out the residual matrix;Calculate Mahalanobis distance based on the residual matrix, send yarn cutting instruction when exceeding limit;Use left singular vector to construct autoregressive model, send early warning when endpoint deviates from reference surface.The application realizes feature decoupling of yarn quality anomaly and mechanical wear trend at the data processing level, reduces the false positive rate of defects.

Description

Technical Field

[0001] This invention relates to the field of textile equipment data processing technology, specifically a method for analyzing and processing textile yarn production quality inspection data. Background Technology

[0002] In the textile yarn production process, sensors are installed at each spindle node of the yarn frame to collect yarn quality inspection data. Existing data analysis and processing methods mostly rely on the independent detection of the time sequence signal of a single spindle and the determination of whether there are defects in the yarn by setting a fixed amplitude threshold. However, multi-spindle textile equipment adopts an integrated rigid frame structure, and there is physical and mechanical coupling between the spindles. The common-mode mechanical shock waves generated during equipment operation will be transmitted along the frame, causing the independent quality signal of the target spindle to be affected by the spatial coupling interference of the operating state of adjacent spindles. This detection method based on independent single channels and fixed thresholds makes it difficult to distinguish between the actual sudden changes in yarn defects and the signal fluctuations caused by environmental mechanical shock waves, thus producing false defect reports.

[0003] Meanwhile, due to the communication polling mechanism of the underlying hardware, there is an asynchronous delay in the data acquisition of each spindle position, which causes the data of spatial nodes to be phase misaligned on the time axis. Existing spatial correlation analysis methods often ignore this underlying time domain deviation, and when processing machine data that includes changes in start-stop status, they fail to shield the baseline drift noise caused by the dead zone of the stopped spindle, resulting in distorted spatial feature extraction results. In addition, existing detection logic mixes yarn quality abnormalities with wear and degradation of underlying mechanical components, and fails to effectively decouple transient yarn defect features from long-term mechanical low-frequency resonant modes at the signal processing level. This affects the accuracy of yarn cutting execution commands and makes it difficult to achieve objective early warning of equipment mechanical degradation trends. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for analyzing and processing quality inspection data in textile yarn production. This method solves the problems of high false alarm rates for defects caused by multi-spindle mechanical shock wave coupling and asynchronous sampling in existing textile yarn inspection methods, as well as the difficulty in effectively separating mechanical wear and degradation trends.

[0005] To achieve the above objectives, the present invention provides the following technical solution: This invention provides a method for analyzing and processing textile yarn production quality inspection data, which relies on hardware equipment including a data acquisition unit, an edge gateway, a yarn cutting actuator, and an alarm terminal, and includes: The edge gateway acquires the normalized quality time-series signal and the binarized state signal, and determines the discrete time delay offset based on cross-correlation operation to generate a phase-aligned sequence. Suppose that the first-order backward difference is used to calculate the trace ratio of the eigenvalues ​​of the covariance matrix using a dynamic sliding window, and then determine the total spindle dimension of the spatial computation cluster. An effective spatiotemporal observation matrix is ​​generated by combining the phase alignment sequence and state signal corresponding to the total spindle dimension. Singular value decomposition is performed on the effective spatiotemporal observation matrix to extract the left singular vector and separate the residual matrix; The Mahalanobis distance is calculated based on the residual matrix, and a command is sent to the yarn cutting actuator when the limit is exceeded; an autoregressive model is constructed using the left singular vector, and a warning is sent to the alarm terminal when the endpoint deviates from the reference plane.

[0006] Furthermore, the edge gateway acquires the normalized quality time-series signal and the binarized state signal, and determines the discrete time delay offset based on cross-correlation operation to generate a phase-aligned sequence, specifically including: The normalized quality timing signal of each spindle position after conversion is input into the configured digital low-pass filter, and the environmental noise is filtered out according to the preset attenuation stopband, and the low-frequency envelope signal is separated and retained. Select the target spindle position as the time axis alignment reference benchmark to extract the corresponding reference low-frequency envelope signal, and synchronously extract the corresponding adjacent low-frequency envelope signal for adjacent physical spindle positions. The discrete time delay offset variable is introduced to calculate the envelope cross-correlation function between the reference channel and the signal sequences of each adjacent channel. A global extremum search is performed on the obtained envelope cross-correlation function to find the optimal discrete time delay offset corresponding to the maximum peak value of the envelope cross-correlation function. The normalized quality time-series signal is shifted and resampled along the time axis using the optimal discrete time delay offset to output a phase-aligned sequence.

[0007] Furthermore, the process of calculating the envelope cross-correlation function between the reference channel and the signal sequences of each adjacent channel specifically involves obtaining the total number of discrete sampling points within a set single data processing time window, the low-frequency envelope signal data of the target spindle under the discrete time index, and the low-frequency envelope signal data of the adjacent spindles after applying the time delay offset. The envelope cross-correlation function under the discrete time delay offset is obtained by accumulating the discrete time index.

[0008] Furthermore, the step of using a dynamic sliding window to calculate the first-order backward difference of the eigenvalue trace ratio of the covariance matrix to determine the total spindle dimension of the spatial computational cluster specifically includes: With the target spindle as the physical center node, discrete spatial dynamic sliding windows are simultaneously set to adjacent spindles on both sides of the rack. Multi-path phase alignment sequences within the corresponding spatial range are extracted and assembled to generate a multi-dimensional spatial exploration observation matrix. The space trial observation matrix is ​​multiplied by its own transpose to generate a trial covariance matrix. A symmetric matrix eigenvalue decomposition operation is performed on the trial covariance matrix to extract the largest eigenvalue in the discrete eigenvalue sequence and divide it by the sum of all eigenvalues ​​to obtain the eigenvalue trace ratio that characterizes the energy concentration of the common mode seismic wave. The discrete expansion radius is increased sequentially by a fixed increment, and the continuously changing eigenvalue trace ratio data is recorded. The eigenvalue trace ratio obtained from two adjacent iterations is calculated using the difference equation using a first-order backward difference. When the first-order backward difference value is detected to be negative for three consecutive times or the current sliding window radius reaches the upper limit of the set maximum expansion radius, the increase of the sliding window radius is terminated, and the total spindle dimension of the spatial calculation cluster is determined by combining the boundary locking parameters and the target spindle center position.

[0009] Furthermore, when calculating the eigenvalue trace ratio that characterizes the energy concentration of the common-mode seismic wave, the value of the eigenvalue trace ratio is equal to the quotient of the largest eigenvalue parameter at the top of the trial covariance matrix and the sum of all eigenvalue parameters arranged in descending order corresponding to the total number of physical spindle nodes included in the current sliding window range.

[0010] Furthermore, the generation of an effective spatiotemporal observation matrix by combining the phase alignment sequence and state signal corresponding to the total spindle dimension specifically includes: The discrete time series corresponding to each machine slot are arranged row by row downwards according to the topological order of the physical node arrangement, generating an original spatiotemporal observation matrix with a row dimension equal to the total number of slot dimensions and a column dimension equal to the number of discrete sampling points in a single window. The discrete state data within the corresponding time window is converted into a discrete state row vector with the same length as the number of discrete sampling points in a single window. The dynamic topology mask matrix is ​​constructed by splicing and combining the data according to the row dimension mapping order. The element values ​​are mapped to 0 when the physical working condition is stopped and to 1 when the physical working condition is normal. Perform the Hadamard product operation between the original spatiotemporal observation matrix and the dynamic topology mask matrix to cut off the data association between the dead zone baseline drift noise of the shutdown spindle position and the overall spatial feature extraction process, and output an effective spatiotemporal observation matrix constrained by the real-time electromechanical topology state.

[0011] Furthermore, the step of performing singular value decomposition on the effective spatiotemporal observation matrix, extracting the left singular vector, and separating the residual matrix specifically includes: Scan the dynamic topological mask matrix corresponding to the effective spatiotemporal observation matrix, set the total number of physical row vectors with all state elements equal to 1 as the effective rank parameter, generate a decentralized observation matrix based on the arithmetic mean of non-zero effective elements, and generate a spatial autocorrelation covariance matrix by performing matrix product operation between the decentralized observation matrix and its own transpose matrix. An adaptive rank-reduced singular value decomposition operation of the spatial autocorrelation covariance matrix is ​​performed using the effective rank parameter as the truncation order. The left singular vector corresponding to the maximum singular value is extracted to generate the transpose vector, and a rank-one feature projection operator is constructed. The common-mode background noise matrix is ​​generated by performing matrix multiplication mapping operation between the rank-one feature projection operator and the effective spatiotemporal observation matrix. The element-level subtraction separation calculation between the effective spatiotemporal observation matrix and the common-mode background noise matrix is ​​performed. The result is then subjected to Hadamard product operation with the dynamic topology mask matrix to eliminate artifacts. The residual matrix stripped of common interferences in the space environment is output.

[0012] Furthermore, the step of calculating the Mahalanobis distance based on the residual matrix and issuing a command to the yarn cutting actuator when the limit is exceeded specifically includes: When the target spindle is in normal operating condition within the current time window, the discrete time series corresponding to the target spindle in the residual matrix is ​​extracted, and phase space reconstruction is performed according to the set embedding dimension parameters to generate a multidimensional state vector. Set a local sliding window on the time axis, extract the historical multidimensional state vector within the local sliding window point by point, calculate the mean vector and the sample covariance matrix, and add a constant regularization perturbation term to the main diagonal elements of the sample covariance matrix. The Mahalanobis distance test statistic is calculated using the multidimensional state vector at the current moment and the corrected sample covariance matrix. The Mahalanobis distance test statistic is then compared with the anomaly judgment threshold parameter in a polling manner. When the value is greater than the anomaly judgment threshold parameter, a low-level control message is generated to issue a physical cutting command for the abnormal yarn to the yarn cutting actuator.

[0013] Furthermore, the anomaly detection threshold parameter is based on the chi-square distribution mathematical model and combined with the embedding dimension value set by the phase space reconstruction as a fixed constant determined by looking up the distribution degrees of freedom table; the calculated value of the Mahalanobis distance test statistic is equal to the product of the transpose of the deviation vector between the current multidimensional state vector and the arithmetic mean vector, the inverse matrix of the multidimensional state vector sample covariance matrix after regularization, and the deviation vector between the current multidimensional state vector and the arithmetic mean vector.

[0014] Furthermore, the step of using left singular vectors to build an autoregressive model and issuing a warning to the alarm terminal when the endpoint deviates from the reference plane specifically includes: The transpose of the first-order left singular vector corresponding to the maximum singular value is extracted and multiplied with the effective spatiotemporal observation matrix to generate a time evolution vector characterizing the overall physical wear trend of the spatial computing cluster. The baseline order parameter determined by performing a minimum search based on the Akaike information content criterion function is read, and a time series autoregressive model of a fixed order is constructed using the time evolution vector. Extract the multidimensional autoregressive coefficient vector generated by the convergence of the autoregressive model, and map it into the established high-dimensional parameter space with corresponding fixed dimensions to generate the coordinates of the current endpoint; The system calls the preset health baseline envelope equation established by multivariate Gaussian mixture clustering analysis, calculates the shortest Euclidean distance from the current endpoint coordinates to the baseline envelope, and generates a mechanical degradation early warning signal when the shortest Euclidean distance is greater than the set safety redundancy threshold and sends it to the alarm terminal.

[0015] The specific innovative principles and effects of this invention are as follows: By calculating the cross-correlation function of the low-frequency envelope and performing an extreme value search, the time-domain deviation introduced by the communication polling mechanism is obtained. A translation resampling operation is performed to align the time axis of the spatial nodes and eliminate the phase misalignment caused by asynchronous sampling. The local eigenvalue trace ratio is calculated using a dynamic sliding window. The spatial damping attenuation boundary of the mechanical stress wave of the rigid frame is determined by the negative state of the first-order backward difference, thereby adaptively determining the spatial calculation cluster dimension that matches the actual rigid transmission structure. A dynamic topological mask matrix is ​​constructed by combining the binarized state signal. The Hadamard product operation is used to cut off the spindle node data in the shutdown and broken state in the numerical calculation domain, eliminating the interference of the shutdown dead zone baseline drift on the spatial correlation analysis.

[0016] In the signal analysis stage, a reduced-rank singular value decomposition is performed on the effective spatiotemporal observation matrix, and the principal components of the low-frequency mechanical resonance mode of the whole machine are extracted using the first and first maximum singular values. A rank-one projection operator is constructed to reconstruct the common-mode background noise matrix, and the residual matrix representing the independent mechanical waveform of the target spindle is separated by subtraction operation. The time evolution vector reflecting the overall physical wear trend of the spatial computing cluster is extracted using the background noise generated by reconstruction, and the mechanical degradation early warning is realized by tracking the endpoint coordinate evolution trajectory in the parameter space using an autoregressive model. At the same time, the residual sequence is extracted and the Mahalanobis distance test statistic is calculated in a local sliding window, and defect judgment is performed in the data domain where common-mode vibration interference is eliminated. The above processing realizes the physical state decoupling between yarn quality anomaly detection and underlying mechanical wear degradation early warning.

[0017] This invention provides a method for analyzing and processing quality inspection data in textile yarn production. It has the following beneficial effects: 1. This invention extracts low-frequency envelope signals and calculates the envelope cross-correlation function between channels. It then uses global extremum search to determine the optimal discrete time delay offset to perform translational resampling on the data, thereby achieving time axis alignment between adjacent physical spindles. This process directly eliminates the asynchronous sampling phase misalignment introduced by the underlying hardware communication polling mechanism, providing an accurate time-domain basis for subsequent spatial correlation analysis of multi-spindle data.

[0018] 2. This invention employs a dynamic sliding window mechanism to calculate the first-order backward difference of the trace ratio of the eigenvalues ​​of the covariance matrix. It can adaptively determine the total spindle dimension of the spatial calculation cluster based on the spatial attenuation boundary of the mechanical stress wave, avoiding the limitations caused by artificially setting a fixed spatial dimension. At the same time, it constructs a dynamic topological mask matrix in conjunction with the real-time equipment status and performs Hadamard product operations, cutting off the data association of the stopped spindles at the numerical calculation level and blocking the interference of the dead zone baseline drift of the stoppage on the feature extraction of the normally operating spindles.

[0019] 3. This invention performs reduced-rank singular value decomposition on the effective spatiotemporal observation matrix, constructs common-mode background noise using the extracted maximum singular value component, and separates the residual matrix. This method uses the residual matrix to calculate Mahalanobis distance to perform yarn defect judgment, and at the same time uses the extracted common-mode background evolution vector combined with an autoregressive model to perform mechanical degradation early warning. At the data processing level, transient yarn quality anomalies and long-term mechanical wear trends are physically decoupled, reducing the false alarm rate of defects caused by common-mode mechanical shock waves. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall hardware topology of the system of the present invention; Figure 2 This is a schematic diagram of the overall control execution flow of the method of the present invention; Figure 3 This is a schematic diagram of the multi-channel sensor data acquisition and phase alignment processing flow of the present invention; Figure 4 This is a schematic diagram of the spatial computing cluster partitioning and physical node topology reconstruction process of the present invention; Figure 5 This is a schematic diagram of the spatiotemporal observation matrix generation and dynamic topology mask operation process of the present invention; Figure 6 This is a schematic diagram of the calculation process for reduced-rank singular value decomposition and common-mode background noise separation in this invention; Figure 7 This is a schematic diagram of the dual-path control process for abnormal defect removal judgment and mechanical degradation early warning in this invention; Figure 8 This is a schematic diagram comparing the signal noise reduction and separation based on orthogonal projection of the present invention. In this diagram, a is a schematic diagram of the original observation signal waveform before removing common interference from the space environment, and b is a schematic diagram of the single-spindle independent residual waveform output after removing common interference from the space environment. Figure 9 This is a schematic diagram of the mechanical degradation evolution trajectory and health baseline envelope in the high-dimensional parameter space of this invention; Figure 10 This is a schematic diagram illustrating the experimental results of comparing the frequency of defect detection over a long testing period.

[0021] The components include: 1. Data acquisition unit; 2. Fieldbus; 3. Edge gateway; 4. Yarn cutting actuator; and 5. Alarm terminal. Detailed Implementation

[0022] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] See attached document Figure 1 This invention provides a method for analyzing and processing textile yarn production quality inspection data. This method relies on an underlying hardware system. The hardware system includes a data acquisition unit 10, a fieldbus 20, an edge gateway 30, a yarn cutting actuator 40, and an alarm terminal 50, all arranged on the spinning machine frame.

[0024] The data acquisition unit 10 includes quality detection sensors and operating condition contacts installed at each spindle position; the fieldbus 20 is communicatively connected to the data acquisition unit 10 and is used to transmit timing signals and discrete status signals of each spindle position; the edge gateway 30 establishes hardware electrical connections with the fieldbus 20, the yarn cutting actuator 40, and the alarm terminal 50, respectively. The edge gateway 30 is internally configured with a microprocessor and a memory; the method provided by this invention is stored in the memory in the form of program instructions and is called and run by the microprocessor.

[0025] See attached document Figure 2 This invention provides a method for analyzing and processing textile yarn production quality inspection data, comprising the following steps: S100, the edge gateway 30 acquires the original normalized quality time-series signal sequence and binary chemical condition status signal of each spindle position within a set time window through the fieldbus 20, extracts the low-frequency envelope of the time-series signal of each channel, performs envelope cross-correlation function operation based on the target spindle position, determines the discrete time delay offset by searching the peak value of the cross-correlation function, and performs translation resampling on the original normalized quality time-series signal sequence using the discrete time delay offset to generate a phase-aligned sequence; S200, the edge gateway 30 sets up dynamic sliding windows on both sides with the target spindle as the center, increases the sliding window radius in a step manner to extract the spatial trial matrix and calculate the covariance matrix of the spatial trial matrix, extracts the energy concentration trace ratio of the eigenvalue sequence of the covariance matrix, performs first-order backward difference calculation on the energy concentration trace ratio, and stops increasing the sliding window radius when the first-order backward difference value is negative for 3 consecutive times or the current sliding window radius reaches the set maximum expansion radius upper limit value. The total spindle dimension of the spatial computing cluster is determined by the sliding window radius or the maximum expansion radius upper limit value corresponding to the moment before the first negative value mutation. S300, edge gateway 30 extracts the phase alignment sequence corresponding to the total spindle dimension to construct the original spatiotemporal observation matrix, extracts the binary chemical condition state signal corresponding to each spindle in the spatial computing cluster to construct a dynamic topology mask matrix with the same dimension as the original spatiotemporal observation matrix, performs the Hadamard product operation between the original spatiotemporal observation matrix and the dynamic topology mask matrix to generate an effective spatiotemporal observation matrix; S400, the edge gateway 30 counts the number of effective running spindles in the spatial computing cluster and sets it as the effective rank parameter. For the effective spatiotemporal observation matrix, it calculates the spatial autocorrelation covariance matrix. Under the constraint of the effective rank parameter, it performs singular value decomposition, extracts the left singular vector corresponding to the maximum singular value to construct the common mode background noise matrix, subtracts the effective spatiotemporal observation matrix from the common mode background noise matrix, and performs Hadamard product operation with the dynamic topology mask matrix to separate the residual matrix. S500, edge gateway 30 receives residual matrix, effective spatiotemporal observation matrix, left singular vector and dynamic topology mask matrix and performs control mapping in parallel. For the first path, it reads the discrete state data corresponding to the target spindle in the dynamic topology mask matrix. When the target spindle is in normal operating condition within the current time window, it extracts the time series sequence corresponding to the target spindle in the residual matrix and calculates the Mahalanobis distance test statistic of the time series sequence within the local sliding window. When the Mahalanobis distance test statistic exceeds the judgment threshold, a defect removal command is sent to the yarn cutting execution mechanism 40. The time evolution vector is extracted by using the left singular vector and the effective spatiotemporal observation matrix, and the preset benchmark order parameter is read to construct an autoregressive model. The multidimensional autoregressive coefficient vector is extracted to generate the current endpoint coordinates of the fixed high-dimensional parameter space. When the current endpoint coordinates deviate from the benchmark envelope, a mechanical degradation early warning signal is sent to the alarm terminal 50.

[0026] See attached document Figure 3 In specific implementations, step S100 provided by the present invention may include the following steps: S101, the edge gateway 30 receives the multi-channel data stream continuously uploaded by the underlying sensor network, and continuously polls and collects the original normalized quality time-series signal sequence and the corresponding binary working condition status signal of each machine spindle position through the fieldbus 20. In this embodiment, the set time window for collecting the original normalized quality time-series signal sequence and the corresponding binary working condition status signal is 5 to 15 seconds. The status word in the control register is extracted and the working condition status signal of the corresponding spindle position is assigned a value of 1 when it is in the mechanical operation and stretching state and assigned a value of 0 when it is in the abnormal head breakage and shutdown state. In this embodiment, the mathematical operation library inside the microprocessor can be called to use the extreme value linear mapping algorithm to scale and map the physical dimensions to a preset fixed interval. In this embodiment, the preset fixed interval is 0 to 1. This front-end normalized numerical conversion processing process is a well-known technology in the field.

[0027] S102, the edge gateway 30 calls the digital low-pass filter configured in its internal memory, and sequentially inputs the converted original normalized quality timing signal sequence of each spindle position into the digital low-pass filter for digital filtering and noise reduction processing. According to the preset attenuation stopband, in this embodiment, the value range of the preset attenuation stopband attenuation value is -60 to -40dB. The high-frequency environmental noise and feather interference data components are filtered out, and the low-frequency envelope signal that can accurately characterize the low-frequency mechanical shock wave transmission characteristics of the rigid body of the spinning frame is separated and retained. In this embodiment, combined with the general resonant frequency distribution characteristics of the rigid transmission structure of the spinning frame and the field experience measurement results, the cutoff frequency parameter of the digital low-pass filter is limited to the range of 10 to 50Hz.

[0028] S103, the edge gateway 30 selects a pre-set target spindle as the time axis alignment reference and extracts the corresponding reference low-frequency envelope signal. For adjacent physical spindles participating in the data analysis, the corresponding adjacent low-frequency envelope signals are extracted synchronously. Within the time window data range, a discrete time delay offset variable is introduced to calculate the envelope cross-correlation function between the reference channel and the signal sequences of each adjacent channel. The mathematical expression corresponding to the function's solution process is: ; In the formula, For target position and adjacent neighbor Discrete time delay offset The envelope cross-correlation function; The physical index number for the set target location; The physical index number of adjacent physical locations participating in the data analysis; The discrete time delay offset parameter is introduced; For discrete-time index variables; The total number of frequency sampling points within a single data processing time window; Indexing the target scale in discrete time Low-frequency envelope signal data below; For adjacent bits, apply a time delay offset The low-frequency envelope signal data afterwards.

[0029] S104, the edge gateway 30 performs a global extremum search algorithm on the obtained envelope cross-correlation function within the set variable range, and combines the inherent physical delay constraint of the fieldbus 20 communication polling cycle to adjust the discrete time delay offset. The upper limit of the search interval is forcibly limited to within 5% of the total number of discrete sampling points to prevent the algorithm iteration from getting stuck in the optimization dead zone. The optimal discrete delay offset is found to maximize the peak value of the envelope cross-correlation function. This optimal discrete delay offset is extracted as the time phase compensation parameter for each data channel. A time-axis shift resampling operation is directly performed on the original normalized quality time-series signal sequence to output a phase-aligned sequence. The corresponding algebraic mapping formula is: ; ; In the formula, The optimal discrete time delay offset corresponding to the maximum peak value of the envelope cross-correlation function; In the parameters A mathematical operator that finds the value of the independent variable within the search interval that maximizes the objective function; For target position and adjacent neighbor Discrete time delay offset The envelope cross-correlation function; The final output in discrete time index after applying the translation resampling operation The phase-aligned sequence data below; This refers to the node index number of each physical location on the machine. For discrete-time indexing; For each suspected node at the time corresponding to the translation resampling operation The original normalized quality time-series signal data below; To achieve the optimal discrete time delay offset, the software-level sequence translation and alignment operation eliminates the time axis misalignment caused by the polling time difference of communication nodes.

[0030] See attached document Figure 4 In specific implementations, step S200 provided by the present invention may include the following steps: S201, the edge gateway 30 sets discrete spatial dynamic sliding windows synchronously to adjacent spindles on both sides of the rack with the set target spindle as the physical center node. During the algorithm initialization phase, the discrete expansion radius is set to 1 by default and the multi-channel phase alignment sequence within the corresponding spatial range is extracted. The multi-channel phase alignment sequence is vertically spliced ​​and assembled according to the physical node topology order to generate a multi-dimensional spatial trial observation matrix. The matrix multiplication operation instruction built into the microprocessor is called to perform the product operation operation of the spatial trial observation matrix and its own transpose matrix, and the trial covariance matrix reflecting the spatial correlation distribution between the data channels of each spindle in the local area is calculated and generated.

[0031] S202, the edge gateway 30 performs symmetric matrix eigenvalue decomposition on the obtained trial covariance matrix, extracts all non-negative eigenvalues ​​obtained from the decomposition, and sorts them in descending order of numerical value to generate a discrete eigenvalue sequence. Based on the physical prior condition that the common-mode seismic wave energy of the rigid body of the physical frame is mainly concentrated on the largest eigencomponent, the largest eigenvalue in the discrete eigenvalue sequence is extracted. The largest eigenvalue is divided by the sum of all eigenvalues ​​of the trial covariance matrix to obtain the eigenvalue trace ratio, which represents the concentration of common-mode seismic wave energy. In this embodiment, the Jacobi iteration algorithm in the linear algebra operation library can be called to solve the eigenvalues ​​of the real symmetric matrix. This basic matrix decomposition operation is a well-known technology in the art, and the expression for calculating the eigenvalue trace ratio is as follows: ; In the formula, The discrete expansion radius takes the value of The corresponding eigenvalue trace ratio parameter; For spatial dynamic weakening, the discrete expansion radius parameter; To explore the largest eigenvalue parameter that is first in the covariance matrix after sorting in descending order; For the discrete space index parameter of the summation operation; This represents the total number of physical suspicious nodes contained within the current sliding window range; The discrete feature sequence arranged in descending order is the first... Each eigenvalue parameter.

[0032] S203, the edge gateway 30 increases the discrete expansion radius value successively in fixed increments of 1, and simultaneously sets the upper limit value of the maximum expansion radius that matches the physical length of the rack. In this embodiment, the upper limit value of the maximum expansion radius ranges from 10 to 24 spindle nodes. The matrix construction and eigenvalue decomposition process is executed cyclically, and the eigenvalue trace ratio data that changes continuously with the radius expansion is recorded. Based on the difference equation, the eigenvalue trace ratio obtained in two adjacent iterations is calculated using a first-order backward difference. The state of the first-order backward difference value is used as the convergence stopping condition for determining the spatial damping attenuation boundary of the rack mechanical stress wave. Its algebraic expression for the difference calculation is: ; In the formula, For the discrete expansion radius at the th The first-order backward difference value calculated in the next iteration; For the discrete expansion radius parameter of the spatial dynamic sliding window; The discrete expansion radius takes the value of The eigenvalue filter ratio parameter at time; The discrete expansion radius takes the value of The eigenvalue filter ratio parameter corresponding to the previous iteration cycle; The rigid body of the rack's physical damping consumption causes independent white noise to be introduced into the out-of-bounds spindle position, resulting in a negative value for the first-order backward difference value. When the edge gateway 30 detects that the first-order backward difference value is negative for three consecutive times or the current sliding window radius reaches the set maximum expansion radius upper limit, the algorithm iteration process of increasing the sliding window radius is terminated. When the algorithm iteration process is terminated due to three consecutive negative values, the sliding window radius corresponding to the moment before the first negative value change is extracted as the boundary locking parameter. When the algorithm iteration process is terminated due to reaching the maximum expansion radius upper limit, the maximum expansion radius upper limit is directly extracted as the boundary locking parameter. The total spindle dimension of the spatial computing cluster is determined by combining the boundary locking parameter and the target spindle center position.

[0033] See attached document Figure 5 In specific implementations, step S300 provided by the present invention may include the following steps: S301, the edge gateway 30 extracts the multi-path phase alignment sequence within the corresponding physical region based on the total spindle dimension of the spatial computing cluster determined in the aforementioned steps. It arranges the discrete time sequences corresponding to each machine spindle position row by row downwards according to the topological order of the physical node arrangement. It allocates continuous address segments in the microprocessor's memory space to generate an original spatiotemporal observation matrix with a row dimension equal to the total spindle dimension and a column dimension equal to the number of discrete sampling points in a single window. Each row of discrete vectors in the matrix carries the dynamic quality fluctuation data characteristics of a single mechanical node within the observation time window under specific physical spatial coordinates. In this embodiment, the two-dimensional array interface of the underlying data structure can be called to complete the matrix encapsulation and mapping of the discrete sequence. This basic data assembly operation is a well-known technology in the field.

[0034] S302, the edge gateway 30 synchronously extracts the binary working condition status signals corresponding to each physical spindle within the spatial computing cluster. According to the preset mapping rules, it converts the discrete state data within the time window into discrete state row vectors with the same length as the number of discrete sampling points in a single window to respond to the transient jumps in the start-stop state of the spindle within a single time window. According to the same row dimension mapping order, it vertically splices and combines the discrete state row vectors corresponding to all nodes in the spatial computing cluster to construct a dynamic topology mask matrix whose overall row dimension and column dimension match the original spatiotemporal observation matrix. When the spindle is in the physical working condition of being cut off and stopped at a certain discrete sampling moment within the time window, the element value of the corresponding row of the dynamic topology mask matrix and the column where the moment is located is mapped to 0, and the value is mapped to 1 when the working condition is normal operation.

[0035] S303, the edge gateway 30 calls the built-in digital processing unit of the microprocessor to perform the Hadamard product operation on the original spatiotemporal observation matrix and the dynamic topology mask matrix. Using the element-by-element multiplication logic of the corresponding matrix positions, it sets the sensor data corresponding to the physical shutdown state of the spindle at that moment to 0 in the numerical computation domain, severing the data association between the dead zone baseline drift noise of the shutdown spindle and the overall spatial feature extraction process. It outputs an effective spatiotemporal observation matrix constrained by the real-time electromechanical topology state. The corresponding matrix algebraic mapping expression is: ; In the formula, This is the effective spatiotemporal observation matrix generated after filtering and isolating operating conditions; This is the original spatiotemporal observation matrix generated by sequentially assembling the phase-aligned sequences of each adjacent node; For the Hadamard product operator of two matrices of the same dimension at the level of corresponding coordinate elements; This is a dynamic topology mask matrix that is expanded based on the real-time running status of nodes, and the values ​​of the elements in this matrix are limited to 0 and 1.

[0036] See attached document Figure 6In specific implementations, step S400 provided by the present invention may include the following steps: S401, the edge gateway 30 scans and reads the numerical status of each discrete coordinate position of the dynamic topology mask matrix by row, counts the total number of physical row vectors with all state elements being 1 within the time window, sets the total number of physical row vectors as the effective rank parameter of the number of currently running real-time physical load nodes, calculates the arithmetic mean of the non-zero effective elements in each data row of the effective spatiotemporal observation matrix, when it is detected that all state elements in the time window corresponding to a certain data row are zero, the arithmetic mean of that row is set to zero by default, extracts each data row of the effective spatiotemporal observation matrix and subtracts the corresponding arithmetic mean, and performs Hadamard product operation with the dynamic topology mask matrix again to generate a decentralized observation matrix; The microprocessor's hardware floating-point unit is invoked to perform matrix multiplication of the decentralized observation matrix and its own transpose matrix. This product is then divided by the number of discrete sampling points in a single window to generate a spatial autocorrelation covariance matrix containing the cross-linking characteristics of signals from normally operating nodes. In this embodiment, the built-in mathematical operation library can be invoked to perform mean biasing of multidimensional arrays and matrix multiplication and accumulation operations. This basic multi-channel statistical operation is a well-known technique in the field.

[0037] S402, the edge gateway 30 passes the obtained spatial autocorrelation covariance matrix as an input parameter to the singular value decomposition algorithm firmware module. During the effective rank parameter value determination stage, if the parameter is less than or equal to 1, an abnormal truncation mechanism is directly triggered, terminating the subsequent data analysis process within the current time window and throwing an abnormal running status flag. If the effective rank parameter is greater than 1, the adaptive rank-reduced singular value decomposition operation of the spatial autocorrelation covariance matrix is ​​performed using the effective rank parameter as the truncation order of the singular value calculation result. The largest singular value at the top of the decomposition calculation output and the corresponding left singular vector are extracted. The left singular vector is used as the core principal component feature vector of the low-frequency mechanical resonance mode of the entire rigid frame. The algebraic expression corresponding to its rank-reduced singular value decomposition process is: ; In the formula, This is the spatial autocorrelation covariance matrix generated from the effective spatiotemporal observation matrix; It is the left singular matrix containing the orthogonal singular vectors of each order generated by the decomposition; It is a non-negative real diagonal matrix with singular values ​​arranged in descending order on its diagonal. It is the transpose of the right singular matrix containing orthogonal right singular vectors of all orders.

[0038] S403, the edge gateway 30 extracts the left singular vector corresponding to the maximum singular value calculated, generates the corresponding transpose vector, constructs a rank-one feature projection operator, and controls the arithmetic logic unit to perform a linear algebraic matrix multiplication mapping operation between this rank-one feature projection operator and the effective spatiotemporal observation matrix. It extracts the orthogonal projection operation result to generate a common-mode background noise matrix containing the common seismic wave characteristics of the machine. In the corresponding dimensional coordinate system, it performs element-wise subtraction separation calculation between the effective spatiotemporal observation matrix and the common-mode background noise matrix. The result of the subtraction separation calculation is then multiplied again with the dynamic topology mask matrix to eliminate projection mapping artifacts in the shutdown dead zone. The output is a residual matrix representing the independent mechanical waveform of a single spindle after stripping away common spatial environment interference. The corresponding algebraic expressions for background reconstruction and residual separation are: ; ; In the formula, This is the cospectral background noise matrix reconstructed through orthogonal projection operations; This is the first left singular vector corresponding to the largest singular value in the singular value decomposition result; It is the transpose of the first-order left singular vector; This is the residual matrix output after removing common interferences from the space environment; This is the effective spatiotemporal observation matrix input after being filtered and isolated based on operating conditions; This is the common-spectral background noise matrix reconstructed through orthogonal projection operations.

[0039] See attached document Figure 7 In specific implementations, step S500 provided by the present invention may include the following steps: S501, the edge gateway 30 receives the residual matrix, effective spatiotemporal observation matrix, left singular vector and dynamic topology mask matrix output from the previous steps, and opens two independent threads inside the microprocessor to execute two-way control mapping operations in parallel. For the defect removal control link of the first path, it reads the discrete state data corresponding to the target spindle in the dynamic topology mask matrix. When the target spindle is continuously in normal operating condition within the current time window, it extracts the discrete time series corresponding to the target spindle in the residual matrix. When the target spindle has an abnormal shutdown condition within the current time window, the defect removal judgment calculation process of the current time window is skipped directly. Based on the set embedding dimension parameter, in this embodiment, the set embedding dimension parameter ranges from 3 to 7. The phase space reconstruction is performed on the discrete time series to generate a multidimensional state vector, and a local sliding window is set on the time axis. In this embodiment, the data length of the local sliding window ranges from 100 to 500 discrete sampling points. The historical multidimensional state vector within the local sliding window is extracted point by point to calculate the mean vector and the sample covariance matrix. A constant regularization perturbation term is added to the main diagonal elements of the sample covariance matrix. In this embodiment, the value of the constant regularization perturbation term ranges from 0.000001 to 0.0001, ensuring that the sample covariance matrix satisfies the full-rank invertibility condition. The Mahalanobis distance test statistic is calculated using the multidimensional state vector at the current moment and the corrected sample covariance matrix. The corresponding calculation expression is: ; In the formula, At the current discrete moment on the time axis The calculated Mahalanobis distance test statistic; For discrete time point index variables on the time axis; For the current moment A multidimensional state vector generated by reconstructing phase space; It is the arithmetic mean vector of all historical multidimensional state vectors within the local sliding window; The transpose vector generated by the deviation motion between the current multidimensional state vector and the arithmetic mean vector; It is the inverse matrix of the covariance matrix of the multidimensional state vector samples within the local sliding window after regularization.

[0040] S502, the edge gateway 30 extracts the anomaly judgment threshold parameter stored in the system configuration register. The anomaly judgment threshold parameter is a fixed constant determined by looking up a table based on the chi-square distribution mathematical model and the embedding dimension value set by the phase space reconstruction as the distribution degree of freedom and the significance level in the preset range of 0.01 to 0.05. The Mahalanobis distance test statistic generated point by point is compared with the anomaly judgment threshold parameter in a round-robin manner. When the value of the Mahalanobis distance test statistic is greater than the anomaly judgment threshold parameter, a low-level control message containing the physical address number of the target spindle is generated. The low-level control message is sent to the yarn cutting actuator 40 through the fieldbus, which controls the electromagnetic relay coil inside the yarn cutting actuator 40 to be energized and drive the mechanical cutter to execute a physical cutting command on the abnormal yarn of the corresponding spindle.

[0041] S503, the edge gateway 30 extracts the transpose of the first-order left singular vector corresponding to the maximum singular value of the mechanical degradation early warning control link of the second path, multiplies it with the effective spatiotemporal observation matrix, and extracts the result as a time evolution vector representing the overall physical wear trend of the spatial computing cluster. It reads the preset baseline order parameter from the microprocessor memory. This baseline order parameter is a fixed order determined during the factory modeling stage by calling the Akaike information content criterion function to perform a discrete minimum value search calculation within a candidate order range limited to 1 to 10. Using this time evolution vector, a fixed-order time series autoregressive model is constructed in the microprocessor memory based on the baseline order parameter. The corresponding autoregressive model construction equation and the Akaike information content criterion solution expression used in the aforementioned factory modeling stage are as follows: ; ; In the formula, The time evolution vector at the current time step Data values; For discrete time point index parameters on the time axis; Sum of index parameters for the historical order of the autoregressive model; The total order parameter of the autoregressive model; In the autoregressive model, the first... The model coefficient parameters corresponding to the order; For the time evolution vector at historical moments Data values; For following a normal distribution and corresponding to time... The white noise residual parameters, For the autoregressive model with order parameter of The cost function value of Akaike information content was calculated in real time; The total order parameter of the autoregressive model; This represents the total number of physical parameters of the data points used to construct the autoregressive time evolution vector; It is a logarithmic operator with the natural constant as its base; For the order parameter is The variance estimate of the prediction error in the time-fit simulation.

[0042] S504, the edge gateway 30 extracts the multidimensional autoregressive coefficient vector generated by the autoregressive model in the current analysis cycle based on the benchmark order parameter, establishes a corresponding fixed-dimensional high-dimensional parameter space with each order autoregressive coefficient as the coordinate axis, and maps the current multidimensional autoregressive coefficient vector to generate the current endpoint coordinates in the high-dimensional parameter space, and calls the preset benchmark envelope surface equation in the microprocessor non-volatile memory. In this embodiment, the preset reference envelope equation is a healthy envelope model uniquely established for the fixed reference order parameter during the machine tool manufacturing modeling stage. The edge gateway 30 directly extracts the preset reference envelope equation corresponding to the fixed spatial dimension from the memory. The aforementioned reference envelope equation is established by extracting the historical multidimensional autoregressive coefficient vector of the machine tool under the new state of the machine tool after long-term no-load operation and performing multivariate Gaussian mixture clustering analysis. This basic state boundary modeling method is a well-known technology in the field. The shortest Euclidean distance from the current endpoint coordinates to the reference envelope is calculated as the mechanical degradation index value. When the shortest Euclidean distance is greater than the set safety redundancy threshold, in this embodiment, the set safety redundancy threshold ranges from 2.5 to 3.5. A mechanical degradation early warning signal is generated and sent to the alarm terminal 50 via the communication port, triggering the alarm terminal 50 to activate the internal sound and light prompt circuit.

[0043] Application Examples: Please refer to the appendix. Figure 8 - Appendix Figure 10 To better understand the technical solution of this invention, the following explanation uses the application scenario of a ring spinning machine in a textile mill as an example: In this embodiment, a ring spinning machine in a textile mill is selected as the physical operating carrier. The 50th spindle in the middle of the frame is set as the target spindle for continuous data acquisition within a 10-second time window. The edge gateway 30 receives the timing signal and the corresponding binary chemical condition status signal uploaded by the underlying quality detection sensor. It calls the internally configured digital low-pass filter to perform filtering operation on the discrete sequence according to the preset stopband attenuation value, extracts the corresponding reference low-frequency envelope signal, and introduces the discrete time delay offset variable to calculate the envelope cross-correlation function. The optimal discrete time delay offset corresponding to the maximum peak value is extracted as the time phase compensation parameter. The translation and resampling operation in the time axis direction is performed to output the phase-aligned sequence.

[0044] Edge gateway 30 sets discrete spatial dynamic sliding windows on both sides with target spindle position 50 as the physical center node. The discrete expansion radius value is increased successively by a fixed increment of 1 and first-order backward difference calculation is performed. When the first-order backward difference value is detected to be negative for 3 consecutive times, the corresponding sliding window radius is recorded as 12. Combined with the boundary locking parameter, the total spindle position dimension of the spatial calculation cluster is determined to be 25 physical nodes. The phase alignment sequence and binary chemical condition state signal corresponding to the total spindle position dimension are extracted to construct the original spatiotemporal observation matrix and dynamic topology mask matrix. The hardware processor is used to perform the Hadamard product operation of the two to generate an effective spatiotemporal observation matrix that isolates shutdown interference.

[0045] Edge gateway 30 counts the number of valid operating spindles within the current time window and performs a reduced-rank singular value decomposition of the spatial autocorrelation covariance matrix using this value as the truncation order. It extracts the left singular vector corresponding to the maximum singular value, constructs a rank-one feature projection operator, and generates a common-mode background noise matrix. It separates the residual matrix representing the independent mechanical waveform of a single spindle through element-level subtraction and mask mapping. Within a local sliding window, it calculates the Mahalanobis distance test statistic for the time series data of the 50th target spindle. It polls and compares the statistic value with the fixed constant for anomaly judgment configured by the system. When the statistic value reaches the set cutting limit, it controls the internal action of the yarn cutting actuator 40 and cuts the current yarn.

[0046] For the mechanical degradation early warning control link of the second path, the edge gateway 30 generates a time evolution vector by multiplying the transpose of the left singular vector with the effective spatiotemporal observation matrix. Based on the preset benchmark order parameters, it constructs a multi-order time series autoregressive model and extracts a multi-dimensional autoregressive coefficient vector. It maps and generates the current endpoint coordinates in the constructed high-dimensional parameter space. It calls the preset machine health envelope equation in the non-volatile memory to calculate the shortest Euclidean distance from the current endpoint coordinates to the benchmark envelope. After comparison, the value of the shortest Euclidean distance is found to be 3.8, which is greater than the set safety redundancy threshold upper limit. The edge gateway 30 generates a mechanical degradation early warning signal and sends it to the alarm terminal 50 and activates the internal audio-visual prompt circuit.

[0047] Based on the physical operating platform of the above application embodiments, a comparative test experimental group was established. The traditional time series fixed threshold detection method was set as the control group, and the analysis and processing method provided by the present invention was set as the experimental group. Yarn quality detection data and electromechanical operating status parameters were continuously recorded during a 30-day cycle of full-load operation of the machine. During this test cycle, a total of 500 yarn defects and 15 mechanical bearing wear and degradation events were marked by manual offline sampling and verification. The accurate detection rate of defects, the false alarm rate of downtime artifacts, and the average early warning time of mechanical degradation were set as quantitative evaluation indicators. All multi-dimensional log records generated during the comparative test were summarized to generate an experimental statistical feature dataset.

[0048] The collected experimental statistical feature dataset was standardized and organized. The physical frequency parameters corresponding to various test judgment results were extracted and percentage indicators were calculated. Multiple sets of discrete data were structured and input into a two-dimensional data table to generate a quantitative comparison matrix. The measurement values ​​of the two algorithms under three evaluation dimensions were extracted to generate a quantitative comparison table of the experimental effects of the two detection methods, as shown in Table 1.

[0049] Table 1: Quantitative Comparison of Experimental Effects of Two Detection Methods Based on the quantitative values ​​recorded in Table 1, the defect detection accuracy rate of the experimental group after introducing time delay resampling and common-mode noise orthogonal stripping was 98.7%, while that of the control group was 82.4%. The false alarm rate of shutdown artifacts after the experimental group implemented the physical isolation mechanism of dead zone baseline drift by dynamic topology mask matrix was 0.8%, while that of the control group was 15.6%. The average warning lead time of the mechanical degradation early warning model established by the experimental group based on autoregressive parameter spatial trajectory tracking was 7.2 days, while that of the control group was 1.5 days.

[0050] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for analyzing and processing quality inspection data in textile yarn production, characterized in that, It operates based on hardware equipment including a data acquisition unit, an edge gateway, a yarn cutting actuator, and an alarm terminal, including: The edge gateway acquires the normalized quality time-series signal and the binarized state signal, and determines the discrete time delay offset based on cross-correlation operation to generate a phase-aligned sequence. With the target spindle as the physical center node, discrete spatial dynamic sliding windows are simultaneously set to adjacent spindles on both sides of the rack. Multi-path phase alignment sequences within the corresponding spatial range are extracted and assembled to generate a multi-dimensional spatial exploration observation matrix. The space trial observation matrix is ​​multiplied by its own transpose to generate a trial covariance matrix. A symmetric matrix eigenvalue decomposition operation is performed on the trial covariance matrix to extract the largest eigenvalue in the discrete eigenvalue sequence and divide it by the sum of all eigenvalues ​​to obtain the eigenvalue trace ratio that characterizes the energy concentration of the common mode seismic wave. The discrete expansion radius is increased sequentially by a fixed increment, and the continuously changing eigenvalue trace ratio data is recorded. The eigenvalue trace ratio obtained from two adjacent iterations is calculated using the difference equation using a first-order backward difference. When the first-order backward difference value is detected to be negative for three consecutive times or the current sliding window radius reaches the set maximum expansion radius limit, the increase of the sliding window radius is stopped, and the total spindle dimension of the spatial calculation cluster is determined by combining the boundary locking parameters and the target spindle center position. The discrete time series corresponding to each machine slot are arranged row by row downwards according to the topological order of the physical node arrangement, generating an original spatiotemporal observation matrix with a row dimension equal to the total number of slot dimensions and a column dimension equal to the number of discrete sampling points in a single window. The discrete state data within the corresponding time window is converted into a discrete state row vector with the same length as the number of discrete sampling points in a single window. The dynamic topology mask matrix is ​​constructed by splicing and combining the data according to the row dimension mapping order. The element values ​​are mapped to 0 when the physical working condition is stopped and to 1 when the physical working condition is normal. Perform the Hadamard product operation between the original spatiotemporal observation matrix and the dynamic topology mask matrix to cut off the data association between the dead zone baseline drift noise of the shutdown spindle position and the overall spatial feature extraction process, and output an effective spatiotemporal observation matrix constrained by the real-time electromechanical topology state. Singular value decomposition is performed on the effective spatiotemporal observation matrix to extract the left singular vector and separate the residual matrix; The Mahalanobis distance is calculated based on the residual matrix, and a command is sent to the yarn cutting actuator when the limit is exceeded; an autoregressive model is established using the left singular vector, and an early warning is sent to the alarm terminal when the endpoint deviates from the reference plane.

2. The method for analyzing and processing textile yarn production quality inspection data according to claim 1, characterized in that, The edge gateway acquires normalized quality time-series signals and binarized state signals, determines discrete time delay offsets based on cross-correlation operations, and generates a phase-aligned sequence, specifically including: The normalized quality timing signal of each spindle position after conversion is input into the configured digital low-pass filter, and the environmental noise is filtered out according to the preset attenuation stopband, and the low-frequency envelope signal is separated and retained. Select the target spindle position as the time axis alignment reference benchmark to extract the corresponding reference low-frequency envelope signal, and synchronously extract the corresponding adjacent low-frequency envelope signal for adjacent physical spindle positions. The discrete time delay offset variable is introduced to calculate the envelope cross-correlation function between the reference channel and the signal sequences of each adjacent channel. A global extremum search is performed on the obtained envelope cross-correlation function to find the optimal discrete time delay offset corresponding to the maximum peak value of the envelope cross-correlation function. The normalized quality time-series signal is shifted and resampled along the time axis using the optimal discrete time delay offset to output a phase-aligned sequence.

3. The method for analyzing and processing textile yarn production quality inspection data according to claim 2, characterized in that, The process of calculating the envelope cross-correlation function between the reference channel and the signal sequences of each adjacent channel specifically involves obtaining the total number of discrete sampling points within a set single data processing time window, the low-frequency envelope signal data of the target spindle under the discrete time index, and the low-frequency envelope signal data of the adjacent spindles after applying the time delay offset. The envelope cross-correlation function under the discrete time delay offset is obtained by accumulating the discrete time index.

4. The method for analyzing and processing textile yarn production quality inspection data according to claim 1, characterized in that, When calculating the eigenvalue trace ratio, which characterizes the energy concentration of the common-mode seismic wave, the value of the eigenvalue trace ratio is equal to the quotient of the largest eigenvalue parameter at the top of the trial covariance matrix and the sum of all eigenvalue parameters arranged in descending order corresponding to the total number of physical spindle nodes included in the current sliding window range.

5. The method for analyzing and processing textile yarn production quality inspection data according to claim 1, characterized in that, The process of performing singular value decomposition on the effective spatiotemporal observation matrix, extracting the left singular vector, and separating the residual matrix specifically includes: Scan the dynamic topological mask matrix corresponding to the effective spatiotemporal observation matrix, set the total number of physical row vectors with all state elements equal to 1 as the effective rank parameter, generate a decentralized observation matrix based on the arithmetic mean of non-zero effective elements, and generate a spatial autocorrelation covariance matrix by performing matrix product operation between the decentralized observation matrix and its own transpose matrix. An adaptive rank-reduced singular value decomposition operation of the spatial autocorrelation covariance matrix is ​​performed using the effective rank parameter as the truncation order. The left singular vector corresponding to the maximum singular value is extracted to generate the transpose vector, and a rank-one feature projection operator is constructed. The common-mode background noise matrix is ​​generated by performing matrix multiplication mapping operation between the rank-one feature projection operator and the effective spatiotemporal observation matrix. The element-level subtraction separation calculation between the effective spatiotemporal observation matrix and the common-mode background noise matrix is ​​performed. The result is then subjected to Hadamard product operation with the dynamic topology mask matrix to eliminate artifacts. The residual matrix stripped of common interferences in the space environment is output.

6. The method for analyzing and processing textile yarn production quality inspection data according to claim 1, characterized in that, The process of calculating the Mahalanobis distance based on the residual matrix and issuing a command to the yarn cutting actuator when the limit is exceeded specifically includes: When the target spindle is in normal operating condition within the current time window, the discrete time series corresponding to the target spindle in the residual matrix is ​​extracted, and phase space reconstruction is performed according to the set embedding dimension parameters to generate a multidimensional state vector. Set a local sliding window on the time axis, extract the historical multidimensional state vector within the local sliding window point by point, calculate the mean vector and the sample covariance matrix, and add a constant regularization perturbation term to the main diagonal elements of the sample covariance matrix. The Mahalanobis distance test statistic is calculated using the multidimensional state vector at the current moment and the corrected sample covariance matrix. The Mahalanobis distance test statistic is then compared with the anomaly judgment threshold parameter in a polling manner. When the value is greater than the anomaly judgment threshold parameter, a low-level control message is generated to issue a physical cutting command for the abnormal yarn to the yarn cutting actuator.

7. The method for analyzing and processing textile yarn production quality inspection data according to claim 6, characterized in that, The anomaly detection threshold parameter is based on the chi-square distribution mathematical model and is a fixed constant determined by looking up the embedded dimension value set by the phase space reconstruction as the distribution degree of freedom. The calculated value of the Mahalanobis distance test statistic is equal to the product of the transpose of the deviation vector between the current multidimensional state vector and the arithmetic mean vector, the inverse matrix of the multidimensional state vector sample covariance matrix after regularization, and the deviation vector between the current multidimensional state vector and the arithmetic mean vector.

8. The method for analyzing and processing textile yarn production quality inspection data according to claim 1, characterized in that, The method of building an autoregressive model using left singular vectors and issuing a warning to the alarm terminal when the endpoint deviates from the reference plane specifically includes: The transpose of the first-order left singular vector corresponding to the maximum singular value is extracted and multiplied with the effective spatiotemporal observation matrix to generate a time evolution vector characterizing the overall physical wear trend of the spatial computing cluster. The baseline order parameter determined by performing a minimum search based on the Akaike information content criterion function is read, and a time series autoregressive model of a fixed order is constructed using the time evolution vector. Extract the multidimensional autoregressive coefficient vector generated by the convergence of the autoregressive model, and map it into the established high-dimensional parameter space with corresponding fixed dimensions to generate the coordinates of the current endpoint; The system calls the preset health baseline envelope equation established by multivariate Gaussian mixture clustering analysis, calculates the shortest Euclidean distance from the current endpoint coordinates to the baseline envelope, and generates a mechanical degradation early warning signal when the shortest Euclidean distance is greater than the set safety redundancy threshold and sends it to the alarm terminal.

Images