Intelligent yarn breakage detection method and system

Abstract

This invention relates to the field of automated detection technology in the textile industry, specifically an intelligent yarn breakage detection method and system. The method includes: acquiring multiple channels of continuous analog voltage raw signals, and performing multi-channel consistency calibration on the raw signals to obtain continuous analog voltage calibration signals; performing analog-to-digital conversion and multi-level adaptive denoising on the calibration signals to obtain clean discrete signals; performing multi-scale hierarchical feature extraction and sensitive feature screening on the clean discrete signals to obtain sensitive feature vectors; and determining the location of the broken yarn path based on the sensitive feature vectors. By fully preserving the dynamic information of the yarn through continuous analog signals, eliminating inter-channel bias through consistency calibration, removing various interferences through multi-level adaptive denoising, enhancing state distinguishability through multi-scale feature extraction and supervised screening, and then performing time-series cumulative decision based on a test supermartingale, combined with precise yarn path positioning, this method solves the problem of insufficient accuracy in existing yarn breakage detection technologies.

Description

Technical Field

[0001] This invention relates to the field of automated inspection technology in the textile industry, specifically an intelligent yarn breakage detection method and system. Background Technology

[0002] Yarn breakage is one of the main types of malfunctions affecting production efficiency and product quality during high-speed operation of warp knitting machines. If a yarn breakage occurs and is not addressed promptly, it will lead to a continuous loss of yarn in that path, resulting in large-area, continuous defects in the fabric structure and irreparable waste. Therefore, real-time and accurate online monitoring of the status of each yarn path during warp knitting machine operation, and rapid location of the faulty yarn path when a breakage occurs, are essential technical means to ensure the continuity of warp knitting production, reduce the defect rate, and improve equipment utilization.

[0003] In the field of warp-knitted yarn condition detection, existing technologies are mainly divided into two categories: manual inspection and automatic detection based on simple sensing elements.

[0004] Manual inspection relies on operators walking along the machine, visually observing the yarn's movement from the coil head through the guide needles to the looping area. Its accuracy is limited by human visual acuity and fatigue tolerance; the missed detection rate rises sharply when multiple machines malfunction simultaneously. Inspection timeliness is constrained by the patrol cycle, making full-time coverage impossible. Furthermore, operator attention decreases after prolonged operation, leading to significant fluctuations in inspection reliability.

[0005] The mechanical contact probe detection solution installs a lightweight metal probe or yarn guide hook on each yarn path, coupled with a micro switch or capacitive proximity switch. During normal yarn feeding, the yarn tension maintains the probe in a balanced position, and the switch outputs a stable signal. After a yarn breakage, the tension disappears, and the probe triggers a switch level change due to gravity or spring reset. This signal is transmitted to the digital input module of the programmable logic controller (PLC) via a data acquisition cable. After polling or interrupting, the PLC drives an intermediate relay to cut off the main motor AC contactor according to the ladder logic, achieving an emergency stop, illuminating the alarm light, and triggering the buzzer.

[0006] The infrared photoelectric through-beam detection scheme uses an infrared light-emitting diode or laser diode as the transmitter and a photodiode as the receiver, mounted opposite each other on both sides of the yarn path. During normal yarn movement, the yarn blocks or attenuates the beam, and the receiver outputs an analog voltage or current signal; after a yarn break, the beam recovers, and the received signal changes. In current engineering practice, this analog signal is generally processed by a voltage comparator with a single threshold hard cutoff, only converting it into a high or low logic level and sending it to the I / O port of a PLC or microcontroller. The microcontroller captures the level transition and executes stop and alarm functions. Compared to mechanical solutions, the photoelectric method achieves non-contact detection and is suitable for high-speed scenarios.

[0007] However, the above-mentioned automatic detection schemes still have two common shortcomings: First, the signal processing is simple. Whether it is a mechanical on / off signal or a photoelectric single threshold comparison, both discard the continuous dynamic information contained in the analog signal, such as the occlusion depth, vibration amplitude, and rate of change. They only judge the two states of "with yarn" and "without yarn", and lack adaptability to complex working conditions such as instantaneous light intensity fluctuations of yarn, partial scattering, and slow time-varying drift of optical window transmittance caused by dust accumulation.

[0008] Secondly, the control logic is rigid. The common approach is to stop the machine immediately upon signal change, which lacks time-series analysis of signal change trends and comprehensive judgment of multiple conditions. This makes it prone to false alarms due to occasional interference such as flying fluff and mechanical vibration, resulting in unnecessary downtime and reduced equipment production efficiency.

[0009] Therefore, a more accurate intelligent yarn breakage detection method is urgently needed. Summary of the Invention

[0010] To address the shortcomings of existing technologies, this invention provides an intelligent yarn breakage detection method and system, which solves the problem of inaccurate yarn breakage detection in existing technologies.

[0011] To achieve the above objectives, the present invention provides an intelligent yarn breakage detection method, the method comprising: acquiring multiple continuous analog voltage raw signals, and performing multi-channel consistency calibration on the continuous analog voltage raw signals to obtain continuous analog voltage calibration signals; performing analog-to-digital conversion and multi-level adaptive denoising processing on the continuous analog voltage calibration signals to obtain clean discrete signals; performing multi-scale hierarchical feature extraction and sensitive feature screening on the clean discrete signals to obtain sensitive feature vectors; and determining the location of the yarn breakage path based on the sensitive feature vectors.

[0012] This invention acquires multiple continuous analog voltage raw signals and performs multi-channel consistency calibration to eliminate systematic baseline drift between multiple acquisition channels, fully preserving yarn dynamic information. After calibration, the signal undergoes analog-to-digital conversion and multi-level adaptive noise reduction to suppress various interferences caused by dust, power frequency electromagnetic fields, and mechanical vibrations, outputting a high-fidelity pure discrete signal. Then, multi-scale hierarchical feature extraction and sensitive feature screening are performed on the pure discrete signal to mine effective features that distinguish between normal and broken yarn states and eliminate redundant information. Finally, the location of the broken yarn path is determined based on the sensitive feature vector, improving the accuracy of yarn breakage detection in warp knitting machines.

[0013] Optionally, the step of performing multi-channel consistency calibration on the original continuous analog voltage signal to obtain a continuous analog voltage calibration signal includes: constructing a channel consistency calibration model based on the smoothness of the channel signal and the amplitude consistency between adjacent channels; and using the original continuous analog voltage signal as input, iteratively solving the channel consistency calibration model using the alternating direction multiplier method to obtain the continuous analog voltage calibration signal.

[0014] This invention constructs a channel consistency calibration model based on channel signal smoothness and amplitude consistency between adjacent channels. Using multiple continuous analog voltage raw signals as input, the channel consistency calibration model is iteratively solved using the alternating direction multiplier method, and the calibrated continuous analog voltage calibration signal is output. It takes into account the signal smoothness constraint of a single signal and the baseline matching constraint between multiple channels, weakens the systematic bias drift caused by aging differences in each photodetector, reduces the detection deviation introduced by the inconsistency of the baseline between channels, and improves the consistency of the baseline of each signal in the scenario of parallel detection of multiple yarns.

[0015] Optionally, the step of performing analog-to-digital conversion and multi-level adaptive denoising processing on the continuous analog voltage calibration signal to obtain a clean discrete signal includes: sequentially performing pre-amplification and anti-aliasing filtering on the continuous analog voltage calibration signal to obtain an analog domain conditioning signal; performing analog-to-digital conversion on the analog domain conditioning signal to obtain a discrete signal sequence; performing power frequency harmonic digital notch filtering on the discrete signal sequence to obtain a notched signal; performing variational mode decomposition on the notched signal to obtain multiple band-limited intrinsic mode functions; determining the denoised mode based on the band-limited intrinsic mode functions; and performing signal reconstruction and median filtering smoothing on the denoised mode to obtain a clean discrete signal.

[0016] This invention obtains an analog-domain conditioned signal by sequentially pre-amplifying and anti-aliasing filtering a continuous analog voltage calibration signal. The weak effective signal is amplified and spectral aliasing is avoided. Then, an analog-to-digital conversion is performed to obtain a discrete signal sequence. The discrete signal sequence is subjected to digital notch filtering for power frequency harmonics to filter out power frequency harmonic interference from high-voltage equipment in industrial sites. Subsequently, variational mode decomposition is performed on the notched signal to obtain multiple band-limited intrinsic mode functions. Based on the mode determination signal, noise and effective signals are separated by mode. Then, signal reconstruction and median filtering smoothing are performed to obtain a pure discrete signal. The step abrupt edges caused by yarn breakage are preserved, various mixed noise interferences are suppressed, and the reliability of the pure discrete signal is improved.

[0017] Optionally, performing variational mode decomposition on the notch-filtered signal to obtain multiple band-limited eigenmode functions includes: constructing a variational mode decomposition model; and using the variational mode decomposition model to decompose the notch-filtered signal to obtain multiple band-limited eigenmode functions.

[0018] This invention constructs a variational mode decomposition model and uses this model to decompose the notch-filtered signal to obtain several band-limited intrinsic mode functions. The mixed signal is adaptively decomposed into components of different frequency bands, realizing the initial separation of effective signal components and different types of noise components in the frequency domain. This avoids interference caused by the mutual coupling of signals of different frequencies, facilitates subsequent targeted identification and elimination of noise modes, and improves the resolvability of different frequency components of the signal.

[0019] Optionally, determining the denoised mode based on the band-limited intrinsic mode function (IEM) includes: extracting a noise reference set from the IEM and calculating the integrated empirical cumulative distribution function (ECV) of the noise reference set; calculating a first EVM based on the IEM; calculating a first CVM distance between the first EVM and the integrated EVM; extracting the dominant signal mode from the IEM based on the first CVM; sampling the dominant signal mode using a sliding window to obtain multiple local window sequences; calculating a second EVM based on the local window sequences; calculating a second CVM distance between the second EVM and the integrated EVM; and denoising the dominant signal mode based on the second CVM distance to obtain the denoised mode.

[0020] This invention extracts a noise reference set from the band-limited intrinsic mode function and calculates its integrated empirical cumulative distribution function. Then, it calculates the first empirical cumulative distribution function of each mode and the corresponding first CVM distance, thereby extracting the dominant signal mode. Next, it performs sliding window sampling on the dominant signal mode to obtain a local window sequence, calculates the second empirical cumulative distribution function and the second CVM distance of the local window, and uses the second CVM distance to complete denoising to obtain the denoised signal mode. Noise discrimination is completed in two layers from the global mode to the local window, retaining the effective details of yarn breakage and abrupt changes, avoiding information loss caused by directly filtering high frequencies, and improving the statistical accuracy of noise identification.

[0021] Optionally, the step of extracting sensitive features and selecting sensitive features from the pure discrete signal at multiple scales to obtain a sensitive feature vector includes: setting an averaging operator and a difference operator, and using the averaging operator and the difference operator to perform hierarchical decomposition on the pure discrete signal to obtain multiple hierarchical components; performing moving average coarsening processing on the hierarchical components to obtain multiple coarse-grained sequences; calculating the dispersion entropy value of each coarse-grained sequence, and using the dispersion entropy value to construct a hierarchical multi-scale dispersion entropy feature vector; obtaining a historical sample feature matrix, and performing spectral analysis on the historical sample feature matrix to obtain contribution weights; and selecting sensitive feature vectors from the hierarchical multi-scale dispersion entropy feature vector based on the contribution weights.

[0022] This invention decomposes a pure discrete signal into multiple hierarchical components by setting an averaging operator and a difference operator. It simultaneously captures the low-frequency trend and high-frequency abrupt change details of the signal. The hierarchical components are then subjected to moving average coarse-grained processing to obtain coarse-grained sequences, avoiding the statistical instability problem caused by traditional coarse-graining. The dispersion entropy value of each coarse-grained sequence is calculated to construct a hierarchical multi-scale dispersion entropy feature vector, which fully explores the signal complexity information at multiple scales. The feature matrix of historical samples is obtained and the contribution weight is obtained through spectral analysis. Sensitive feature vectors are selected based on the contribution weight, and redundant and invalid features are eliminated to improve the ability of features to distinguish abnormal yarn breakage states.

[0023] Optionally, determining the location of the broken yarn path based on the sensitive feature vector includes: obtaining the mean of the sensitive feature vector and the covariance matrix of the sensitive feature vector; calculating the Mahalanobis distance of the sensitive feature vector relative to the mean of the sensitive feature vector and the covariance matrix of the sensitive feature vector; obtaining an initial anomaly judgment threshold, and calculating the ratio at the current time based on the Mahalanobis distance and the initial anomaly judgment threshold; obtaining the test supermartingale at the previous time, and calculating the test supermartingale at the current time based on the ratio of the test supermartingale at the previous time and the current time; determining the broken yarn channel number based on the test supermartingale at the current time; and determining the location of the broken yarn path based on the broken yarn channel number.

[0024] This invention establishes a statistical benchmark for normal operating conditions by acquiring the mean and covariance matrix of sensitive feature vectors, calculates the Mahalanobis distance corresponding to the sensitive feature vectors to quantify the degree of signal deviation from the normal state, obtains the ratio at the current time by combining the initial anomaly judgment threshold, and then recursively calculates the test supermartingale at the current time using the test supermartingale at the previous time. Based on the test supermartingale, the yarn breakage channel number is determined, and the location of the yarn breakage is determined by the channel number. Sequential statistical testing is used to accumulate temporal evidence, suppressing false triggering caused by instantaneous interference, realizing automatic identification and physical location mapping of the yarn breakage channel, and improving the accuracy of yarn breakage judgment.

[0025] Optionally, the step of calculating the current test supermartingale based on the ratio of the previous test supermartingale to the current test supermartingale includes: determining that when the product of the previous test supermartingale and the ratio of the current test supermartingale is greater than a preset upper boundary threshold, calculating an enhancement factor; and updating the previous test supermartingale based on the enhancement factor and the ratio of the current test supermartingale to obtain the current test supermartingale.

[0026] This invention determines whether the product of the test supermartingale at the previous time step and the ratio at the current time step is greater than a preset upper boundary threshold. If the threshold is exceeded, an enhancement factor is calculated. The test supermartingale at the previous time step is then updated using the enhancement factor and the ratio at the current time step to obtain the test supermartingale at the current time step. This fully preserves evidence of temporal anomalies while preventing the supermartingale value from going out of bounds, takes into account the boundary constraints of statistical tests and the information utilization rate of anomalous signals, avoids the loss of detection sensitivity caused by simple truncation, and improves the response speed of sequential yarn breakage judgment.

[0027] Optionally, determining the location of the broken yarn path based on the broken yarn channel number includes: obtaining a mapping table from the number to the coordinates; and determining the location of the broken yarn path based on the broken yarn channel number using the mapping table from the number to the coordinates.

[0028] This invention obtains a mapping table from number to coordinate, and then determines the location of the broken yarn path by querying the mapping table based on the broken yarn channel number. It establishes a one-to-one correspondence between the acquisition channel and the actual spatial location of the yarn path on the machine. This eliminates the need for manual point-by-point inspection of each yarn path, and directly transforms the logical channel fault information into locatable physical location information on the equipment. This improves the intuitiveness and accuracy of locating the broken yarn fault and shortens the time spent on-site inspection and locating the faulty yarn path.

[0029] In another aspect, the present invention provides an intelligent yarn breakage detection system, comprising: a processor, an input device, an output device, and a memory, wherein the processor, the input device, the output device, and the memory are interconnected, wherein the memory is used to store a computer program, the computer program including program instructions, and the processor is configured to invoke the program instructions to execute an intelligent yarn breakage detection method according to any of the preceding aspects of the present invention.

[0030] The present invention provides an intelligent yarn breakage detection system with a compact structure, stable performance, high integration and simple configuration. It can stably execute the intelligent yarn breakage detection method provided in the preceding aspect of the present invention, further improving the overall applicability and practical application capability of the present invention. Attached Figure Description

[0031] Figure 1 This is a flowchart of an intelligent yarn breakage detection method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of an intelligent yarn breakage detection system according to an embodiment of the present invention. Detailed Implementation

[0032] Specific embodiments of the present invention will now be described in detail. It should be noted that the embodiments described herein are for illustrative purposes only and are not intended to limit the invention. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that these specific details are not necessary to practice the invention. In other instances, well-known circuits, software, or methods have not been specifically described to avoid obscuring the invention.

[0033] Throughout this specification, references to "an embodiment," "an embodiment," "an example," or "an example" mean that a particular feature, structure, or characteristic described in connection with that embodiment or example is included in at least one embodiment of the invention. Therefore, the phrases "in an embodiment," "in an embodiment," "an example," or "an example" appearing in various places throughout the specification do not necessarily refer to the same embodiment or example. Furthermore, specific features, structures, or characteristics can be combined in one or more embodiments or examples in any suitable combination and / or sub-combination. Moreover, those skilled in the art will understand that the illustrations provided herein are for illustrative purposes and are not necessarily drawn to scale.

[0034] Please see Figure 1 In order to solve the problems in the prior art, in an alternative embodiment, such as Figure 1 The intelligent yarn breakage detection method shown includes the following steps: Step S1: Acquire multiple continuous analog voltage raw signals and perform multi-channel consistency calibration on the continuous analog voltage raw signals to obtain continuous analog voltage calibration signals; In this embodiment, a system bracket is installed on the frame of one side of the warp knitting machine's yarn running area using special bolts, ensuring the bracket is level and stable. A laser transmitter is mounted on the bracket, and a photoelectric receiver is installed at a corresponding position on the other side of the yarn. The angle and height of both are adjusted to ensure the laser beam illuminates the yarn's running trajectory. The system is powered by an industrial standard DC 24V power supply provided by a standard 24V switching power supply within the warp knitting machine's control cabinet. In terms of hardware configuration, each yarn path is independently equipped with a set of transmitter and receiver channels, forming a parallel acquisition architecture.

[0035] During operation, the laser emitter projects a high-precision laser beam onto the yarn's trajectory, while the photodetector monitors the beam's blocking, reflection, or transmission in real time, converting the light intensity signal into a continuous analog voltage signal. The original continuous analog voltage signal of the i-th channel is modeled as follows:

[0036] in, For the first The continuous analog voltage raw signal of the channel is generated by the first... The data is acquired in real time by the channel photoelectric receiver. For the first The photoelectric conversion gain of the channel can be calibrated by the ratio of the output voltage of the receiving end of the channel to the intensity of the laser emitted light when there is no yarn and the optical window is clean. The laser emission intensity can be set according to the laser's nominal output intensity. This is the attenuation coefficient of the yarn to the laser (related to the yarn material and linear density). The equivalent blocking thickness of the yarn in the optical path (which varies with time) is determined by the yarn diameter, number of yarns, coverage area of ​​the light spot, and yarn vibration displacement during normal yarn movement. When the yarn breaks, the equivalent blocking thickness is zero. The transmittance of the optical window (affected by dust and exhibiting slow time-varying characteristics) can be estimated online by the ratio of the received light intensity when there is no yarn to the clean reference light intensity; The equivalent noise at the receiver (including shot noise and thermal noise) is obtained by statistically analyzing the mean and variance of the sampled data at the receiver when there is no yarn. Compared with the traditional infrared beam-based method that only outputs a "yarn / no yarn" switching signal, this method of acquiring the original continuous analog voltage signal retains richer dynamic information about the yarn.

[0037] The process of obtaining a continuous analog voltage calibration signal by performing multi-channel consistency calibration on the original continuous analog voltage signal specifically includes the following sub-steps: Step S101: Based on the smoothness of the channel signal and the amplitude consistency between adjacent channels, construct a channel consistency calibration model.

[0038] In this embodiment, to improve signal quality from the source and overcome the limitations of independent acquisition of a single channel, this step does not simply involve direct sampling channel by channel, but rather... Treating the received signal as a whole, a perceptual variational problem is constructed, i.e., the channel consistency calibration model satisfies the following formula:

[0039] in, For the first The channel's continuous analog voltage raw signal, i.e., the raw voltage signal directly output by the photodetector without any calibration or filtering, This represents the total number of yarn paths. The Dirac impulse function; The imaginary unit; For the first The angular frequency of the channel, from the first angular frequency under normal yarn feeding conditions... The main peak frequency of the original signal power spectrum of the channel was determined offline. The weights used to adjust the balance between smoothness constraints and baseline consistency constraints can be determined through grid search or cross-validation. For the first The channel's continuous analog voltage raw signal; The average of the logarithmic amplitude difference between adjacent channels under normal historical operating conditions is obtained by collecting multiple raw signals and calculating the time average of the logarithmic amplitude difference between adjacent channels when the equipment is running normally. For time The partial derivative operator; That is, to perform convolution operations; for The square of the norm; in the formula To constrain the smoothness of signals in each channel, Baseline consistency constraints between adjacent channels.

[0040] Step S102: Using the original continuous analog voltage signal as input, the channel consistency calibration model is iteratively solved using the alternating direction multiplier method to obtain the continuous analog voltage calibration signal.

[0041] In this embodiment, the objective function in the constructed channel consistency calibration model is used as the optimization objective, and the original raw signals of multiple continuous analog voltages are used as the initial values ​​for iteration. The alternating direction multiplier method is employed for iterative solution. The alternating direction multiplier method transforms the objective function into an augmented Lagrange function by introducing Lagrange multipliers, and decomposes the original problem into multiple separable subproblems by introducing auxiliary variables. These subproblems then alternately update the continuous analog voltage signal variables of each channel. The iteration continues until the objective function value meets the preset convergence condition or reaches the preset maximum number of iterations, using auxiliary variables and Lagrange multipliers. The resulting M-channel signals after convergence are the continuous analog voltage calibration signals after inter-channel consistency calibration. These calibration signals simultaneously satisfy the smoothness constraints of each channel signal and the amplitude consistency constraints between adjacent channels. This multi-channel joint solution effectively suppresses the systematic bias drift between channels caused by inconsistent aging rates of multiple receivers, ensuring baseline consistency for each signal in subsequent processing.

[0042] Step S2: Perform analog-to-digital conversion and multi-level adaptive denoising processing on the continuous analog voltage calibration signal to obtain a clean discrete signal.

[0043] The process of performing analog-to-digital conversion and multi-level adaptive denoising on the continuous analog voltage calibration signal to obtain a clean discrete signal specifically includes the following sub-steps: Step S201: The continuous analog voltage calibration signal is sequentially pre-amplified and anti-aliasing filtered to obtain an analog domain conditioning signal.

[0044] In this embodiment, the continuous analog voltage calibration signal output after calibration is still a weak electrical signal, requiring pre-amplification to increase the signal amplitude and meet the input range requirements of the subsequent analog-to-digital converter. The pre-amplification uses a transimpedance amplifier to convert the weak current signal output from the photodetector into a voltage signal, and sets... The gain is on the order of magnitude. The amplified signal is then subjected to anti-aliasing filtering using an eighth-order elliptic low-pass filter with a cutoff frequency set to 400Hz. This filters out high-frequency noise components above the cutoff frequency, preventing spectral aliasing during subsequent analog-to-digital conversion and contaminating the effective signal band. Both pre-amplification and anti-aliasing filtering are performed in the analog domain, and the output signal remains a continuous analog signal, i.e., an analog-domain conditioned signal, for use in subsequent analog-to-digital conversion steps.

[0045] Step S202: Perform analog-to-digital conversion on the analog domain conditioning signal to obtain a discrete signal sequence.

[0046] In this embodiment, the analog domain conditioning signal is input to the analog-to-digital converter, and the continuous analog voltage signal is sampled at equal intervals at a sampling frequency of 1kHz. The voltage amplitude of each sampling point is quantized into a digital quantity, thereby converting the time-continuous and amplitude-continuous analog signal into a time-discrete and amplitude-quantized digital signal sequence, resulting in a discrete signal sequence for subsequent digital domain signal processing.

[0047] Step S203: Perform power frequency harmonic digital notch filtering on the discrete signal sequence to obtain the notched signal.

[0048] In this embodiment, high-power equipment such as the warp knitting machine's frequency converter and servo driver generate 50Hz power frequency interference and its odd harmonics (150Hz, 250Hz, 350Hz). This interference is superimposed on the photoelectric signal through spatial coupling or power line conduction, severely reducing the accuracy of subsequent feature extraction. To suppress this interference, this step uses a digital notch filter to filter the discrete signal sequence. The digital notch filter consists of four cascaded digital notch filters, used to filter out the 50Hz frequency and its odd harmonics (150Hz, 250Hz, 350Hz), respectively. Its transfer function is:

[0049] in, For transfer functions, For unit delay operators, For two-unit delay operators, It is the power frequency base frequency. Sampling rate, Notch bandwidth factor These correspond to the fundamental frequency and the third, fifth, and seventh harmonics, respectively. Through the cascaded processing of the above four notch filters, the amplitudes of the 50Hz, 150Hz, 250Hz, and 350Hz frequency components in the discrete signal sequence are suppressed to less than 10% of the original amplitudes, while the amplitudes within the effective signal band do not show significant attenuation, resulting in the notch-filtered signal.

[0050] Step S204: Perform variational mode decomposition on the notch-filtered signal to obtain multiple band-limited eigenmode functions.

[0051] The variational mode decomposition of the notch-filtered signal to obtain multiple band-limited eigenmode functions specifically includes the following sub-steps: Step S20401: Construct the variational mode decomposition model.

[0052] The variational mode decomposition model satisfies the following formula:

[0053] in, For the first The first channel obtained by decomposition A band-limited eigenmode function For the first Channel 1 The center frequency of a band-limited intrinsic mode function The number of decomposition layers, For time The partial derivative operator, Let be the Dirac impulse function. The imaginary unit, For convolution operations, for The square of the norm, For the first The signal after notch filtering in the channel, The constraints are as follows: the model aims to optimize the center frequency band of each mode to be as compact as possible, and the constraint is that the sum of all modes equals the notch-filtered signal. In this embodiment, the number of decomposition layers is... .

[0054] Step S20402: The notch-filtered signal is decomposed using the variational mode decomposition model to obtain multiple band-limited eigenmode functions.

[0055] In this embodiment, the notch-filtered signal Input a variational mode decomposition model, with the optimization objective being the most compact center frequency band limit of each mode, and the sum of all modes equal to... As constraints, the variational mode decomposition model is iteratively solved using the alternating direction multiplier method, and the mode functions are updated iteratively. and its corresponding center frequency The iteration continues until a preset convergence condition is met or a preset maximum number of iterations is reached. After convergence, the notch-filtered signal... Adaptively decomposed into There are band-limited intrinsic mode functions, and the modes are arranged in order of increasing center frequency.

[0056] Step S205: Determine the denoised mode of the signal based on the band-limited intrinsic mode function.

[0057] The determination of the denoised mode based on the band-limited intrinsic mode function specifically includes the following sub-steps: Step S20501: Extract the noise reference set from the band-limited intrinsic mode function and calculate the integrated empirical cumulative distribution function of the noise reference set.

[0058] In this embodiment, the decomposition obtained The band-limited intrinsic mode functions (IMFs) are arranged from low to high center frequency. The low-frequency modes mainly contain the effective components of the yarn vibration signal, while the high-frequency modes mainly contain noise components. The effective energy of the warp knitting machine yarn vibration signal is concentrated below 200Hz, while the center frequencies of the last three modes after VMD decomposition are usually distributed in the frequency band above 250Hz, which does not overlap with the effective signal frequency band. Therefore, the last three modes can be regarded as a noise reference set (i.e.,...). Extract all data points from the noise reference set, that is, merge the data points of the three modes into one dataset, and then uniformly calculate its empirical cumulative distribution function to obtain the integrated empirical cumulative distribution function of the noise reference set. .

[0059] Step S20502: Calculate the first empirical cumulative distribution function based on the band-limited intrinsic mode function.

[0060] In this embodiment, the steps are decomposed to obtain... For each band-limited intrinsic mode function, all data points are extracted from each mode, and the empirical cumulative distribution function of each mode is calculated, denoted as the first empirical cumulative distribution function. ,in For channel index, This is the modal number.

[0061] Step S20503: Calculate the first CVM distance between the first empirical cumulative distribution function and the integrated empirical cumulative distribution function.

[0062] In this embodiment, the first empirical cumulative distribution function for each mode Calculate the integrated empirical cumulative distribution function of each set of noise references. The Cramér-von Mises distance (CVM distance) between them is denoted as the first CVM distance. The CVM distance measures the overall difference between two empirical cumulative distribution functions. A smaller value indicates that the empirical cumulative distribution of that mode is closer to the distribution of the noise reference set, meaning that the mode contains more noise components. The formula for calculating the first CVM distance is:

[0063] Its numerical calculation form is as follows:

[0064] in, For the first Channel 1 The first CVM distance of each modality For the first Channel 1 The first empirical cumulative distribution function of each mode, Let be the integrated empirical cumulative distribution function of the noisy reference set. For integration variables, Indicates about Integral measure, The sequence length for each modality, For summation index variables, For the first Channel 1 The modalities sorted by numerical value are the first... Each sample value, For the first The empirical cumulative probability value corresponding to each sorting position is used to calculate the first CVM distance for each band-limited intrinsic mode function through the above calculation. This is used to distinguish between signal-dominant and noise-dominant modes in subsequent steps.

[0065] The first CVM distance corresponding to each band-limited intrinsic mode function is obtained through the above calculations. This is used to distinguish between signal-dominant and noise-dominant modes in subsequent steps.

[0066] Step S20504: Extract the dominant signal mode from the band-limited intrinsic mode function based on the first CVM distance.

[0067] In this embodiment, the calculated first CVM distance for each modality is used. (in ), calculate the slope of the first CVM distance between adjacent modes ,in The smaller the first CVM distance, the closer the statistical distribution of that mode is to the noise distribution. The location where the CVM distance changes most between adjacent modes is the boundary between the signal-dominant mode and the noise-dominant mode. (Location slope) Modal number when the maximum value is obtained ,determination The mode is the dominant mode of the signal and is retained. The noise-dominant modes are eliminated, thus obtaining the signal-dominant mode set. Meanwhile, to avoid noise reference sets ( If any mode in the above dynamic determination is misclassified as the dominant mode due to numerical deviation in the algorithm, and if the above dynamic determination boundary point is not met... Then take As the upper bound of the signal mode, i.e. the first To the The mode is not retained as the dominant mode of the signal, but its local window coefficients still have the opportunity to participate independently in the denoising decision in subsequent steps.

[0068] Step S20505: A sliding window is used to sample the dominant mode of the signal to obtain multiple local window sequences.

[0069] In this embodiment, for each extracted dominant signal mode Since some local segments with similar statistical characteristics to noise may still be mixed in, further refined denoising processing is required. To achieve segment-by-segment discrimination of local segments, this step uses a sliding window sampling method to divide each dominant signal mode into multiple local window sequences. The window length is... Each sampling point (corresponding to a 32ms time length) starts from the beginning of the sequence and moves point by point in a sliding manner with a step size of 1 sampling point. Each move extracts a sample of length 1. The local window sequence is obtained by using a sliding method with a step size of 1 sampling point. This method can maximize the preservation of continuity and redundancy information between adjacent windows, avoid the omission of signal segments due to window jumping, and ensure that the details of yarn breakage and sudden changes are completely preserved during the sliding process. Through the above sampling, each dominant signal mode is divided into multiple local window sequences, so that subsequent steps can independently perform statistical tests on the noise distribution of each local window to achieve refined noise reduction processing.

[0070] Step S20506: Calculate the second empirical cumulative distribution function based on the local window sequence.

[0071] In this embodiment, the empirical cumulative distribution function (ECF) is calculated for each sampled local window sequence. The ECF is calculated based on the amplitude distribution of all sampling points within the window, reflecting the statistical distribution characteristics of the local window sequence without requiring any prior assumptions about the data distribution, thus exhibiting good adaptability and robustness. For local window sequences with statistical characteristics consistent with the noise reference set, their ECF will be highly similar to the integrated ECF of the noise reference set, while local window sequences containing details of signal abrupt changes will show significant deviations in distribution. The calculated ECF is denoted as the second ECF, used for subsequent comparison with the integrated ECF of the noise reference set to quantitatively evaluate the noise properties of each local window sequence.

[0072] Step S20507: Calculate the second CVM distance between the second empirical cumulative distribution function and the integrated empirical cumulative distribution function.

[0073] In this embodiment, for each local window sequence, the CVM distance between its second empirical cumulative distribution function and the integrated empirical cumulative distribution function of the noise reference set is calculated and denoted as the second CVM distance. This distance is calculated in the same way as the first CVM distance, i.e., by integrating the overall difference between the two empirical cumulative distribution functions. The smaller the value, the closer the statistical distribution of the local window sequence is to the distribution of the noise reference set, indicating that the window is dominated by noise components; the larger the value, the more significant the difference between the statistical distribution of the local window sequence and the noise reference set, indicating that the window may contain important information such as signal abrupt changes. By calculating the second CVM distance, a quantitative basis for the decision to retain or zero out each local window sequence is provided.

[0074] Step S20508: Denoise the dominant mode of the signal based on the second CVM distance to obtain the denoised mode of the signal.

[0075] In this embodiment, a significance level is set. The corresponding decision threshold can be obtained by looking up the CVM test critical value table. The second CVM distance for each local window sequence. Perform the judgment: If This indicates that the local window sequence is statistically indistinguishable from noise, and it is determined that it follows a noise distribution; therefore, all coefficients of the window are set to zero. This indicates that the statistical characteristics of the local window sequence significantly deviate from the noise distribution, thus determining that it contains effective signal components, and all coefficients of the window are retained unchanged. This decision rule can be formally expressed as:

[0076] in, For the first Channel 1 The first mode A sequence of local windows, For the first Channel 1 The first mode The second CVM distance of a local window sequence The decision threshold for the CVM test. This indicates an assignment operation, that is, assigning the value on the right side of the arrow to the variable on the left side. This indicates that all coefficients of the local window sequence are set to zero. Through the above window-by-window discrimination and zeroing process, it is possible to accurately remove local segments that are statistically indistinguishable from noise while preserving the details of yarn breakage abrupt changes (i.e., windows where the CVM distance is significantly greater than the threshold), thus obtaining a finely denoised mode set. That is, the denoised mode of the signal, which is used in the subsequent signal reconstruction step.

[0077] Step S206: After denoising the signal, perform signal reconstruction and median filtering smoothing to obtain a clean discrete signal.

[0078] In this embodiment, the modal coefficients obtained after fine denoising are summed to complete signal reconstruction, resulting in a reconstructed signal. :

[0079] in, For the first Channel 1 The finely denoised mode index in discrete time The amplitude at that point, This marks the boundary between the signal-dominant mode and the noise-dominant mode. For the first After channel reconstruction, index in discrete time The signal amplitude at that point. The reconstructed signal. The structural noise in the signal has been effectively removed, leaving only a small number of random pulses with extremely low amplitude. To eliminate these residual random pulses without damaging the abrupt edge of the yarn breakage, this step uses a three-point median filter to reconstruct the signal. A gentle smoothing process is performed by taking each sampling point as the center and forming a window of three points: the sampling point itself and the sampling points before and after it. The median value within the window is used as the output value for that point. The window is then traversed point by point throughout the entire signal sequence. This three-point median filtering effectively removes isolated impulse noise while preserving the signal's step edges to the greatest extent possible, thus protecting the integrity of the signal abrupt change when a yarn breakage occurs. After median filtering and smoothing, the final clean discrete signal is obtained. .

[0080] Step S3: Perform multi-scale hierarchical feature extraction and sensitive feature screening on the pure discrete signal to obtain a sensitive feature vector.

[0081] The process of extracting sensitive features from the pure discrete signal using multi-scale hierarchical features and filtering sensitive features to obtain a sensitive feature vector specifically includes the following sub-steps: Step S301: Set the averaging operator and the difference operator, and use the averaging operator and the difference operator to perform hierarchical decomposition on the pure discrete signal to obtain multiple hierarchical components.

[0082] In this embodiment, in order to comprehensively extract feature information of different frequency bands from a pure discrete signal, an averaging operator is set. Sum difference operator Among them, the average operator Used to extract low-frequency information from signals, reflecting the overall trend of the signal; differential operator Used to extract high-frequency information from signals and capture details of local abrupt changes in the signal. For adjacent data points in the input sequence, For the first Data points, For the first Data points. For pure discrete signals. A hierarchical decomposition is performed, with each layer using the two operators described above to process each node of the previous layer, generating two child nodes: a low-frequency node and a high-frequency node. This progressively decomposes the signal into components of different frequency bands. In this embodiment, the number of decomposition layers is... After three levels of decomposition, we obtain Each level component ,in, For channel index, The number of levels in the hierarchical decomposition. Number the nodes. For the first Passage After hierarchical decomposition, the node numbers are: The hierarchical components are decomposed as described above. This hierarchical decomposition preserves both the low-frequency trend and high-frequency details of the signal, providing a rich information foundation for subsequent multi-scale feature extraction.

[0083] Step S302: Perform moving average coarse-graining processing on the hierarchical components to obtain multiple coarse-grained sequences.

[0084] In this embodiment, multi-scale coarsening is performed on each level component, using a moving average coarsening method for each scale factor. Based on window length The sliding window moves point-by-point across the hierarchical component sequence with a step size of one sampling point. The arithmetic mean of all sampling points within the window is calculated as the coarsened value at that position. Let the length of the signal sequence within a single analysis window be... Then the scaling factor is taken as At that time, coarse-grained sequence The calculation formula is:

[0085] in, For the first The number of channels and hierarchical decomposition levels is Node number is , scale factor is The coarse-grained sequence in the first The value at each position, For the position index of the coarse-grained sequence, For summation index variables, For the first The number of channels and hierarchical decomposition levels is Node number is The hierarchical components in the first The amplitude of each sampling point The length of the signal sequence. The number of levels in the hierarchical decomposition. Scale factor. The range of values ​​is This moving average coarsening method calculates the average value point by point using a sliding window, ensuring that the length of the coarsened sequence remains essentially unchanged from the length of the original hierarchical component sequence. This overcomes the shortcomings of traditional coarsening methods, such as the sharp decrease in sequence length with increasing scale, statistical instability of short sequences, and decreased reliability of calculated entropy values. It provides a stable and reliable input sequence for subsequent dispersive entropy calculations.

[0086] Step S303: Calculate the dispersion entropy value of each coarse-grained sequence, and construct a hierarchical multi-scale dispersion entropy feature vector using the dispersion entropy value.

[0087] In this embodiment, for each coarse-grained sequence Calculate the dispersion entropy value for each. Dispersion entropy is a measure of the complexity and disorder of a time series. Its core idea is to map the sequence to a finite number of categories and quantify the sequence's complexity by analyzing the probability of occurrence of each category. Compared to sample entropy and permutation entropy, dispersion entropy has the advantages of being insensitive to abrupt changes and having high computational efficiency. When calculating dispersion entropy, the embedding dimension is set. Number of categories Specifically, for each coarse-grained sequence, it is first mapped to... There are several categories, and then based on the embedding dimension... Construct embedding vectors, calculate the probability distribution of each category pattern, and finally calculate the dispersive entropy value according to the definition of information entropy. .in, For channel index, To determine the number of levels in the hierarchical decomposition. Number the nodes. As a scale factor, This represents the dispersive entropy function. It represents the components of each level across the entire spectrum. The dispersion entropy values ​​calculated under each scale factor are arranged in order, with each level component corresponding to... There are a total of dispersion entropy values. Each level of components, from which each channel can obtain [the following] components. 3D feature vector ,in For the first Each level component in the scale factor The dispersion entropy value is calculated below. This feature vector simultaneously integrates multi-band information (low-frequency trends and high-frequency details) from hierarchical decomposition and multi-time-scale information from multi-scale expansion, which can comprehensively characterize the dynamic characteristics of the signal and provide a rich feature foundation for subsequent sensitive feature screening.

[0088] Step S304: Obtain the historical sample feature matrix and perform spectral analysis on the historical sample feature matrix to obtain the contribution weight.

[0089] In this embodiment, to select the most discriminative sensitive features for yarn breakage from the 64-dimensional hierarchical multi-scale dispersion entropy feature vector, it is first necessary to obtain the historical sample feature matrix. The historical sample feature matrix... During the offline debugging phase of the equipment, historical data of the warp knitting machine under normal yarn feeding and simulated yarn breakage conditions were pre-built and collected. A 64-dimensional hierarchical multi-scale dispersion entropy feature vector was extracted from each sample, and the feature vectors of all samples were arranged row-wise to form... 3D historical sample feature matrix ,in The total number of historical samples is given, and each sample is labeled with its category (normal state or yarn breakage state). After obtaining the feature matrix of the historical samples, a supervised feature selection algorithm based on Fisher discriminant analysis is used to perform discriminant analysis on the feature matrix. The discriminant analysis solves the following generalized eigenvalue problem:

[0090] in, The inter-class scatter matrix, The scatter matrix is ​​within the class. The weight vector represents the feature dimension. These are generalized eigenvalues. By solving the above generalized eigenvalue problem, we obtain the weight vectors corresponding to each feature dimension. The values ​​of each element in the weight vector Each element in the array corresponds to a feature dimension, and the value of the feature dimension reflects its ability to distinguish between the normal state and the broken yarn state. That is, the larger the value, the greater the contribution of the feature dimension to distinguishing between the two states, and the smaller the value, the smaller the contribution of the feature dimension to distinguish between the two states. Thus, the contribution weight of each feature dimension is obtained, which is used to screen sensitive features in subsequent steps.

[0091] Step S305: Based on the contribution weight, filter sensitive feature vectors from the hierarchical multi-scale dispersion entropy feature vectors.

[0092] In this embodiment, the 64 feature dimensions are sorted from largest to smallest according to their contribution weight. The larger the contribution weight, the greater the contribution of the feature dimension to distinguishing between the normal state and the yarn breakage state, and the stronger the discriminative power. The top 16 feature dimensions with the largest contribution weights are selected, and the feature values ​​corresponding to these selected dimensions are extracted from the 64-dimensional hierarchical multi-scale dispersion entropy feature vector of each channel. These are combined into a 16-dimensional sensitive feature vector. Through the above screening, the most discriminative sensitive information in the original features is retained, while redundant features with low contribution to state distinction are eliminated. This effectively reduces the dimensionality of the feature space, alleviates the computational burden of subsequent decision-making, and enhances the distinguishability between the normal state and the yarn breakage state in the feature space, providing high-quality feature input for subsequent anomaly judgment.

[0093] Step S4: Determine the location of the broken yarn path based on the sensitive feature vector.

[0094] The determination of the location of the broken yarn path based on the sensitive feature vector specifically includes the following sub-steps: Step S401: Obtain the mean of the sensitive feature vector and the covariance matrix of the sensitive feature vector.

[0095] In this embodiment, during the offline debugging phase of the equipment, data is collected on the continuous operation of the warp knitting machine under normal yarn feeding conditions. The sensitivity feature vectors of each channel in each analysis window are denoted as follows: ,in For channel index, This is a normal window number. , This represents the total number of analysis windows under normal operating conditions. Then, for each channel, the mean vector of its sensitive feature vector under normal conditions is calculated. Covariance Matrix The calculation formula is:

[0096]

[0097] in, For the first The mean vector of the sensitive feature vectors of normal samples in the channel represents the central position of the feature distribution under normal conditions; For the first The covariance matrix of the sensitive feature vectors of normal channel samples represents the correlation and dispersion among the feature dimensions under normal conditions. The calculated mean vector... Covariance Matrix Pre-stored in the system storage unit, it is used to calculate the Mahalanobis distance of the current sensitive feature vector during the real-time detection phase, so as to quantify the degree to which the current state deviates from the normal baseline.

[0098] It should be noted that if the number of normal samples included in the statistics... Smaller than the sensitive feature dimension or covariance matrix If a singularity occurs, the covariance matrix is ​​regularized before inverting it. The inverse is performed instead of the original covariance matrix, where , for The identity matrix ensures that the Mahalanobis distance can be calculated stably.

[0099] Meanwhile, considering the transmittance of the optical window Due to its slow time-varying characteristics, the signal baseline may gradually drift after long-term operation. The system is set to update online. Hours, every time When the duration or the host computer triggers a recalibration command, it automatically collects continuous data under the current normal yarn feeding state. The data in each analysis window is recalculated and updated. , and initial anomaly detection threshold This is to adapt to baseline drift and avoid the accumulation of false alarms during long-term operation.

[0100] Step S402: Calculate the Mahalanobis distance between the sensitive feature vector and the mean and covariance matrix of the sensitive feature vector.

[0101] In this embodiment, the Mahalanobis distance satisfies the following formula:

[0102] in, For the first Channel 1 Mahalanobis distance for each real-time analysis window For the first Channel 1 A 16-dimensional sensitive feature vector for each real-time analysis window. For the first The mean vector of the sensitive feature vectors of normal samples in the channel. For the first The covariance matrix of the sensitive feature vectors of normal channel samples. for The inverse matrix, This represents the transpose operation of a matrix. Mahalanobis distance is a multivariate statistical distance measure. Unlike Euclidean distance, Mahalanobis distance normalizes the variance of each feature dimension and the correlation between dimensions through the covariance matrix. This eliminates the influence of dimensional differences and correlations between different feature dimensions, truly reflecting the statistical distance between the current sample and the distribution center of the normal sample. A larger Mahalanobis distance indicates that the current sensitive feature vector deviates further from the distribution center of the normal state, and the greater the probability of an abnormal yarn state at the current moment; a smaller Mahalanobis distance indicates that the current state is closer to the normal yarn flow state. Through the above calculation, the 16-dimensional sensitive feature vector is compressed into a scalar value that can intuitively reflect the degree of state deviation. .

[0103] Step S403: Obtain the initial anomaly determination threshold, and calculate the ratio at the current time based on the Mahalanobis distance and the initial anomaly determination threshold.

[0104] In this embodiment, based on the sensitive feature vectors of the collected Q normal operating condition analysis windows, the Mahalanobis distance is calculated for each normal window according to the Mahalanobis distance calculation method, resulting in a total of Q Mahalanobis distance values ​​under normal conditions. Then, these Q Mahalanobis distance values ​​are statistically analyzed to calculate their mean and standard deviation. Based on the statistical distribution of the Mahalanobis distance under normal conditions, an initial anomaly judgment threshold is set equal to the mean plus three times the standard deviation. This threshold serves as the engineering statistical upper bound of the Mahalanobis distance under normal conditions. In this embodiment, it is specifically set using the mean of the Mahalanobis distance of normal samples plus three times the standard deviation. This setting does not rely on the normality assumption of the signal distribution but only serves as an empirical statistical boundary for distinguishing between normal and abnormal conditions.

[0105] The ratio at the current time satisfies the following formula:

[0106] in, For the first Channel 1 The ratio of the current time in each real-time analysis window. For the first Channel 1 Mahalanobis distance for each real-time analysis window For the first The initial anomaly detection threshold of the channel, the ratio at the current time. It reflects the relative degree to which the current state deviates from normal operating conditions. When the current Mahalanobis distance exceeds the initial anomaly detection threshold, it means that the current state is statistically deviating from the normal range; when When the current Mahalanobis distance does not exceed the initial anomaly detection threshold, it means that the current state is within the normal range.

[0107] Step S404: Obtain the test supermartingale of the previous time step, and calculate the test supermartingale of the current time step based on the ratio of the test supermartingale of the previous time step to the current time step.

[0108] In this embodiment, during the real-time operation of the warp knitting machine, the previous moment (i.e., the first moment) is read from the storage unit. Test supermartingale value (in multiple analysis windows) Used to compare with the current time. Together, we calculate the test supermartingale at the current time, for the initial time (i.e. The supermartingale was not found in the previous test, so an initial value needs to be preset. Once the system completes the calculation of the supermartingale for the current moment, it stores the result in a storage unit as the previous supermartingale value for the next moment, thus achieving a recursive update of the supermartingale in time sequence. Through this "read-calculate-store-read for next moment" loop mechanism, it ensures that the supermartingale calculation at each moment is based on all historical accumulated information, and there is no need to repeatedly calculate historical data, effectively improving the computational efficiency of online detection.

[0109] The calculation of the current-time test supermartingale based on the ratio of the previous-time test supermartingale to the current-time test supermartingale specifically includes the following sub-steps: Step S40401: If the product of the supermartingale tested at the previous time moment and the ratio at the current time moment is greater than a preset upper boundary threshold, calculate the boost factor.

[0110] In this embodiment, in each real-time analysis window Perform the following judgment: Test the supermartingale in the previous time step. Ratio to the current time Multiply, if the product Greater than the preset upper boundary threshold This triggers the calculation of the boost factor, with a preset upper boundary threshold. Significance level of the target Relatedly, in this embodiment, we take The upper boundary threshold is 100. When the product exceeds this upper boundary threshold, directly multiplying the test supermartingale from the previous time step to the ratio of the current time step to update the test supermartingale will cause the supermartingale value to exceed the limit. Therefore, it is necessary to calculate the boosting factor. Enhancement Factor The equation that yields the conditional expectation equal to 1 after truncation is obtained by numerically solving it, i.e., solving the equation. ,in This represents the expected conditions under normal operating conditions; Indicates as of the date A collection of historical information from each analysis window; This is a truncation function whose first input is the lifting factor. The second input is the product of the ratio at the current time step and the supermartingale tested at the previous time step. Its function is to truncate the ratio at the current time after the boost if it exceeds the upper boundary threshold. Specifically, during the solution process, under normal operating conditions... Approximately follows the degree of freedom of The chi-square distribution can be used to determine... The cumulative distribution function is obtained by using the bisection method in the interval. Search until the left side of the equation is The absolute deviation is less than Stop iteration when the time is right. Conversely, if Then there is no need to calculate the improvement factor, just let At this point, the truncation function The first branch ( It takes effect immediately, preserving the original evidence value.

[0111] Through the above processing, it is possible to maintain the supermartingale property of the test (i.e., the false alarm rate is strictly controlled at the significance level). Under the premise of (the following), we can make the most of the effective information in the ratio at the current moment, avoid information loss due to simple truncation, and achieve a logarithmically optimal fast response.

[0112] Step S40402: Update the test supermartingale of the previous time step according to the ratio of the enhancement factor and the current time step to obtain the test supermartingale of the current time step.

[0113] In this embodiment, the calculated improvement factor is used. Ratio at the current time Perform boosting processing, and then calculate the boosted current time ratio. Input truncation function The evidence value is obtained by truncating the original evidence. Truncation function The definition of is:

[0114] in, The evidence value to be truncated (i.e., the enhanced ratio of the current time) ), This is the current test supermartingale value. When... When, retain the original evidence value; when At that time, the evidence value was truncated to This ensures that the updated test supermartingale does not exceed the preset upper boundary threshold. Finally, the test supermartingale from the previous moment is... Multiplying the truncated evidence value yields the test supermartingale at the current moment. :

[0115] in, For the first The updated test supermartingale value is displayed in the analysis window. This update mechanism ensures that the test supermartingale maintains its supermartingale properties (i.e., under normal operating conditions). Under the premise of ), when the boost factor is triggered, it can approach the upper boundary at a log-optimal speed, thereby achieving the dual goals of strictly controllable false alarm rate and rapid response.

[0116] Step S405: Determine the yarn breakage channel number based on the supermartingar test at the current moment.

[0117] In this embodiment, the supermartingale will be tested at the current moment. With preset upper boundary threshold Compare. If Then determine the first A yarn breakage event has occurred in the channel. Confirm the yarn breakage event and immediately lock the current channel number. ,like If no yarn breakage has occurred in the current channel, monitoring will continue, and the supermartingale will be tested at the current time. The supermartingale from the previous time step is stored as the data for the next analysis window. Through this window-by-window recursive decision mechanism, yarn breakage events can be confirmed promptly and accurately, and the yarn breakage channel can be locked, while ensuring a strictly controlled false alarm rate. Once a yarn breakage is confirmed, the latched yarn breakage channel number will serve as input for subsequent yarn path position mapping. Simultaneously, the system records the yarn breakage time, providing data support for the statistical analysis of subsequent yarn breakage events.

[0118] Step S406: Determine the location of the broken yarn path according to the broken yarn channel number.

[0119] Determining the location of the broken yarn path based on the broken yarn channel number specifically includes the following sub-steps: Step S40601: Obtain the mapping table from number to coordinate.

[0120] In this embodiment, for each photoelectric receiving channel corresponding to a yarn path, the spatial coordinates of its actual mechanical installation position on the warp knitting machine (such as the lateral distance and height from the left side of the machine frame) are measured, and the channel number is associated with the corresponding spatial coordinates to form a mapping table from channel number to yarn path position. This mapping table is stored in the system storage unit in the form of a lookup table, with a one-to-one correspondence between channel number and spatial coordinate, ensuring that each channel number uniquely corresponds to a yarn path position coordinate. Through this mapping table, the system can quickly query and locate the corresponding yarn path position after determining the broken yarn channel number, providing accurate spatial guidance for subsequent on-site troubleshooting.

[0121] Step S40602: Determine the location of the broken yarn path using the mapping table from the broken yarn path number.

[0122] In this embodiment, after confirming a yarn break and locking the broken yarn channel number, the system uses the broken yarn channel number as an index to query the mapping table from number to coordinate. The system then extracts the yarn path location coordinates uniquely corresponding to the channel number from the mapping table to obtain the location of the broken yarn path. These yarn path location coordinates are presented as spatial coordinates on the host computer monitoring interface. On-site operators can quickly locate the corresponding faulty yarn path on the warp knitting machine using these coordinates, eliminating the need for path-by-path troubleshooting and significantly shortening the downtime for recovery after a yarn breakage. When multiple channels experience simultaneous yarn breaks, the system queries and maps each channel number, outputting the yarn path location coordinates corresponding to each broken yarn channel, providing on-site maintenance personnel with complete broken yarn location information.

[0123] like Figure 2 As shown, in another aspect, the present invention also provides an intelligent yarn breakage detection system, including: a processor, an input device, an output device, and a memory, wherein the processor, the input device, the output device, and the memory are interconnected, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to call the program instructions to execute the relevant steps of the relevant embodiments of the intelligent yarn breakage detection method of the present invention.

[0124] This invention provides an intelligent yarn breakage detection system. The functional components can be integrated into a single processing unit, or each component can exist independently, or two or more components can be integrated into one unit. The integrated components can be implemented in hardware or software.

[0125] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A smart yarn breakage detection method, characterized in that, The method includes: Acquire multiple channels of continuous analog voltage raw signals, and perform multi-channel consistency calibration on the continuous analog voltage raw signals to obtain continuous analog voltage calibration signals; The continuous analog voltage calibration signal is subjected to analog-to-digital conversion and multi-level adaptive denoising processing to obtain a clean discrete signal; Sensitive feature vectors are obtained by performing multi-scale hierarchical feature extraction and sensitive feature screening on the pure discrete signal; The location of the broken yarn path is determined based on the aforementioned sensitive feature vector.

2. The intelligent yarn breakage detection method according to claim 1, characterized in that, The process of performing multi-channel consistency calibration on the original continuous analog voltage signal to obtain the continuous analog voltage calibration signal includes: Based on the smoothness of channel signals and the amplitude consistency between adjacent channels, a channel consistency calibration model is constructed. The continuous analog voltage calibration signal is obtained by iteratively solving the channel consistency calibration model using the alternating direction multiplier method with the original continuous analog voltage signal as input.

3. The intelligent yarn breakage detection method according to claim 1, characterized in that, The step of performing analog-to-digital conversion and multi-level adaptive denoising processing on the continuous analog voltage calibration signal to obtain a clean discrete signal includes: The continuous analog voltage calibration signal is sequentially pre-amplified and anti-aliasing filtered to obtain an analog domain conditioning signal; The analog-domain conditioned signal is converted from analog to digital to obtain a discrete signal sequence; The discrete signal sequence is subjected to power frequency harmonic digital notch filtering to obtain the notched signal; Variational mode decomposition is performed on the notched signal to obtain multiple band-limited eigenmode functions; Determine the denoised mode of the signal based on the band-limited intrinsic mode function; After denoising, the signal is reconstructed and smoothed by median filtering to obtain a clean discrete signal.

4. The intelligent yarn breakage detection method according to claim 3, characterized in that, The variational mode decomposition of the notched signal yields multiple band-limited eigenmode functions, including: Construct a variational mode decomposition model; The notch-filtered signal is decomposed using the variational mode decomposition model to obtain multiple band-limited eigenmode functions.

5. The intelligent yarn breakage detection method according to claim 3, characterized in that, The determination of the denoised mode based on the band-limited intrinsic mode function includes: Extract a noise reference set from the band-limited intrinsic mode function and calculate the integrated empirical cumulative distribution function of the noise reference set; Calculate the first empirical cumulative distribution function based on the band-limited intrinsic mode function; Calculate the first CVM distance between the first empirical cumulative distribution function and the ensemble empirical cumulative distribution function; Based on the first CVM distance, the dominant signal mode is extracted from the band-limited intrinsic mode function; A sliding window is used to sample the dominant mode of the signal to obtain multiple local window sequences; Calculate the second empirical cumulative distribution function based on the local window sequence; Calculate the second CVM distance between the second empirical cumulative distribution function and the ensemble empirical cumulative distribution function; The signal dominant mode is denoised based on the second CVM distance to obtain the denoised mode.

6. The intelligent yarn breakage detection method according to claim 1, characterized in that, The process of performing multi-scale hierarchical feature extraction and sensitive feature screening on the pure discrete signal to obtain a sensitive feature vector includes: An averaging operator and a difference operator are set up, and the averaging operator and the difference operator are used to perform hierarchical decomposition on the pure discrete signal to obtain multiple hierarchical components; The hierarchical components are subjected to moving average coarsening to obtain multiple coarsened sequences; Calculate the dispersion entropy value of each coarse-grained sequence, and construct a hierarchical multi-scale dispersion entropy feature vector using the dispersion entropy value; Obtain the historical sample feature matrix, and perform spectral analysis on the historical sample feature matrix to obtain the contribution weights; Sensitive feature vectors are selected from the hierarchical multi-scale dispersion entropy feature vectors based on the contribution weight.

7. The intelligent yarn breakage detection method according to claim 1, characterized in that, Determining the location of the broken yarn path based on the sensitive feature vector includes: Obtain the mean and covariance matrix of the sensitive feature vectors; Calculate the Mahalanobis distance between the sensitive feature vector and the mean and covariance matrix of the sensitive feature vector; Obtain the initial anomaly detection threshold, and calculate the ratio at the current time based on the Mahalanobis distance and the initial anomaly detection threshold; Obtain the test supermartingale of the previous time step, and calculate the test supermartingale of the current time step based on the ratio of the test supermartingale of the previous time step to the test supermartingale of the current time step; The yarn breakage channel number is determined based on the supermartingale test at the current moment; The location of the broken yarn path is determined based on the broken yarn channel number.

8. The intelligent yarn breakage detection method according to claim 7, characterized in that, The step of calculating the current time test supermartingale based on the ratio of the previous time test supermartingale to the current time includes: When the product of the supermartingale tested at the previous time step and the ratio at the current time step is greater than a preset upper boundary threshold, the improvement factor is calculated. The test supermartingale at the previous time step is updated based on the ratio of the enhancement factor to the current time step to obtain the test supermartingale at the current time step.

9. The intelligent yarn breakage detection method according to claim 7, characterized in that, Determining the location of the broken yarn path based on the broken yarn path number includes: Obtain the mapping table from numbers to coordinates; The location of the broken yarn path is determined using the mapping table from the broken yarn path number to the coordinates.

10. An intelligent yarn breakage detection system, characterized in that, include: The system includes a processor, an input device, an output device, and a memory, all interconnected. The memory stores a computer program, which includes program instructions. The processor is configured to invoke the program instructions to execute an intelligent yarn breakage detection method as described in any one of claims 1 to 9.

Images