A functional fiber spinning production monitoring system based on end-cloud cooperation

Abstract

This invention discloses a functional fiber spinning production monitoring system based on edge-cloud collaboration, belonging to the field of monitoring systems. The system includes: synchronously acquiring multi-dimensional microscopic physical signals for cleaning, spatiotemporal alignment, and feature extraction processing to obtain multi-dimensional feature data; fusing and calculating the multi-dimensional feature data based on preset dynamic weight coefficients to generate a particle distribution dispersion index, comparing it with a preset threshold, generating a control trigger signal, and uploading abnormal waveform data; reading registers to obtain the blower speed setpoint and heater status, and generating control commands by executing preset reverse compensation control logic in conjunction with the control trigger signal; iteratively updating the dynamic weight coefficients based on abnormal waveform data and offline quality detection data through data association and pattern mining. This invention provides a spinning monitoring method based on edge-cloud collaboration, capable of real-time sensing of fiber microstructure instability and synchronous reverse repair, and the system possesses self-evolution capabilities, which can improve the product quality and production stability of functional fibers.

Description

Technical Field

[0001] This invention relates to the field of monitoring systems, specifically a functional fiber spinning production monitoring system based on end-to-cloud collaboration. Background Technology

[0002] Functional fibers (such as conductive fibers and magnetic fibers) have important application value in high-end textiles, smart fabrics and special industrial materials. The production quality of such fibers depends not only on their macroscopic morphology, but more importantly on the uniform distribution of internal functional particles (such as nano-conductive particles) and the stability of their microstructure. During the spinning process, fluctuations in process parameters can easily lead to microscopic defects such as particle agglomeration and damage to the internal structure of the fiber, which in turn seriously affect the conductivity, mechanical strength and durability of the final product. Currently, industry monitoring of the spinning process mainly relies on online monitoring and closed-loop control of key process parameters (such as temperature, pressure, and drafting speed) and offline laboratory testing of finished products (such as electrical performance testing and strength testing). However, existing methods have the following limitations: First, online monitoring is mostly focused on macroscopic physical quantities, making it difficult to directly and in real-time capture changes in the microstructure of fibers and the distribution of functional particles; second, offline testing results lag significantly behind the production process and cannot be used for online intervention and quality correction; third, existing control strategies are mostly based on fixed rules or empirical models, lacking the ability to make autonomous decisions and dynamically optimize based on real-time microscopic conditions. Summary of the Invention

[0003] Based on the shortcomings of the prior art described above, the purpose of this invention is to provide a functional fiber spinning production monitoring system based on end-to-cloud collaboration to solve the aforementioned technical problems.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a functional fiber spinning production monitoring system based on end-to-cloud collaboration, comprising: a sensing module: synchronously acquiring multi-dimensional microscopic physical signals for cleaning, spatiotemporal alignment and feature extraction processing to obtain multi-dimensional feature data; Edge analysis module: Based on preset dynamic weight coefficients, it performs fusion calculations on multi-dimensional feature data, generates a particle distribution dispersion index, compares it with a preset threshold, generates a control trigger signal, and then uploads abnormal waveform data. Reverse execution module: Reads the register to obtain the set value of the blowing speed and the status of the heater, and combines the control trigger signal to generate control instructions by executing the preset reverse compensation control logic; Cloud-based evolution module: Based on abnormal waveform data and offline quality inspection data, it iteratively updates dynamic weight coefficients through data correlation and pattern mining.

[0005] The present invention is further configured such that the multi-dimensional microscopic physical signals specifically include: dielectric response signals, pressure signals, and surface micro-texture image signals.

[0006] The present invention is further configured such that the sensing module specifically includes: Digital filtering is performed on the dielectric response signal and the pressure signal respectively; Based on the preset fiber winding speed and the distance between the acquisition points of the pressure signal and the dielectric response signal, time shift compensation is performed on the filtered pressure signal. Pressure change characteristics are calculated based on the pressure signal after time shift compensation, and dielectric change characteristics are calculated based on the dielectric response signal.

[0007] The present invention is further configured such that the sensing module specifically includes: By combining pressure change characteristics, dielectric change characteristics, and surface texture characteristics, time-synchronized multidimensional feature data is formed.

[0008] The present invention is further configured such that the edge analysis module includes: Based on preset dynamic weighting coefficients, multidimensional feature data are fused and calculated to generate a particle distribution dispersion index that characterizes the internal structural state of fibers.

[0009] The present invention is further configured such that the edge analysis module further includes: The particle distribution dispersion index is compared with a preset safety threshold in real time. When the particle distribution dispersion index continuously exceeds the safety threshold for a predetermined duration, it is determined that microstructure instability has occurred, and the following two operations are performed simultaneously: Operation 1: The edge analysis module automatically generates a control trigger signal based on the determination result of microstructure instability, and sends the control trigger signal to the reverse execution module.

[0010] Operation 1: Upload the multi-dimensional microscopic physical signals corresponding to the microstructural instability as abnormal waveform data to the cloud evolution module.

[0011] The present invention is further configured such that the reverse execution module includes: Receives control trigger signals from the edge analysis module; Based on the control trigger signal, the current side blowing speed setpoint and heater status are read from the production system register to confirm the execution of the reverse compensation control logic.

[0012] The present invention is further configured such that the reverse execution module further includes: generating and issuing the following control instructions in sequence according to the reverse compensation control logic: Based on the reverse compensation control logic, control commands are generated and issued sequentially, including: By sending a command to the airflow regulation unit of the side blowing device, a first control command for reducing the side blowing speed is generated. By sending instructions to the drive unit of the infrared heater, a second control command is generated to start the infrared heater to produce transient thermal radiation. After a preset duration, a third control command is generated to restore the side-blowing air speed and the infrared heater to the state before receiving the control trigger signal.

[0013] The present invention is further configured such that the cloud evolution module includes: It receives abnormal waveform data from the edge analysis module and correlates and matches it with the corresponding batch of offline quality inspection data; Based on data association and pattern mining, we analyze the correlation between the signal patterns hidden in abnormal waveform data and the quality indicators in offline quality inspection data.

[0014] The present invention is further configured such that the cloud evolution module also includes: Based on the correlation pattern, the weight allocation of each feature in the dynamic weight coefficient is adjusted through a preset weight update strategy, and the updated dynamic weight coefficient is uploaded to the register to wait for the next cycle to be executed.

[0015] This invention provides a functional fiber spinning production monitoring system based on edge-cloud collaboration, the beneficial effects of which include: 1. Real-time perception of microstructural instability: Unlike traditional macroscopic parameter monitoring, this invention integrates multi-dimensional microscopic physical characteristics to diagnose online in real time whether the particle distribution inside the fiber is uniform, thus achieving early warning of microstructural collapse.

[0016] 2. Provides reverse online repair control: Breaking through the conventional control approach of "cooling and acceleration", when instability is detected, a counterintuitive control strategy of "cooling and heating" is adopted to actively repair the internal structure of the fiber online and improve the first-pass yield of the product.

[0017] 3. Establish a self-evolutionary mechanism for edge-cloud collaboration: Unlike monitoring systems with fixed thresholds, this invention enables the system to autonomously optimize judgment rules by associating production data with quality results, achieving continuous learning capabilities that become more accurate with use.

[0018] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 This is a schematic diagram illustrating the structure of a functional fiber spinning production monitoring system based on edge-cloud collaboration, as an exemplary embodiment of the present invention. Detailed Implementation

[0020] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0021] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the type, quantity and proportion of each component can be arbitrarily changed by a functional fiber spinning production monitoring system based on end-to-cloud collaboration, and its component layout may also be more complex.

[0022] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0023] Example: A functional fiber spinning production monitoring system based on edge-cloud collaboration, such as Figure 1 As shown, it includes: The sensing module synchronously collects multi-dimensional microscopic physical signals for cleaning, spatiotemporal alignment, and feature extraction to obtain multi-dimensional feature data. The edge analysis module fuses and calculates the multi-dimensional feature data based on preset dynamic weight coefficients, generates a particle distribution dispersion index, compares it with a preset threshold, generates a control trigger signal, and then uploads abnormal waveform data. The reverse execution module reads the register to obtain the blowing speed setpoint and heater status, and generates control commands by executing preset reverse compensation control logic in combination with the control trigger signal. The cloud evolution module iteratively updates the dynamic weight coefficients based on abnormal waveform data and offline quality detection data through data association and pattern mining. The present invention is further configured such that the multi-dimensional microscopic physical signals specifically include: dielectric response signals, pressure signals, and surface micro-texture image signals. Specifically, the dielectric response signal reflects the content and distribution continuity of functional particles inside the fiber. It is acquired by a microwave resonant cavity sensor installed at the exit of the spinning channel, and the changes in scattering parameters (standard network parameters used in microwave engineering to characterize the forward transmission gain and phase of two-port networks, belonging to the prior art) injected and detected by a vector network analyzer are used to obtain the original voltage sequence (this is the conventional result of data acquisition in the prior art, specifically including a set of discrete digital voltage values ​​arranged in time sequence, output by the analog-to-digital converter in the vector network analyzer, and proportional to the amplitude or phase value of the scattering parameters). Pressure signals characterize the melt rheological state and pressure fluctuations. They are collected by a high-frequency melt pressure sensor installed at the inlet of the spinning assembly, with the sampling frequency set to be at least five times the fundamental vibration frequency of the equipment. For example, when the equipment's basic vibration frequency is 10 Hz, the sampling frequency is set to 500 Hz to obtain a high-fidelity waveform. The surface micro-texture image signal records the micro-morphology of the fiber surface. It is acquired by a high-speed linear array industrial camera aligned with the fiber bundle and an infrared backlight. The camera's line frequency is synchronized with the winding speed to generate continuous images.

[0024] The signal processing and feature extraction process is as follows: First, signal cleaning is performed. For dielectric response signals and pressure signals, digital band-stop filters are used to filter out specific high-frequency noise. For example, band-stop filters with cutoff frequencies of 45 Hz and 55 Hz are designed to eliminate 50 Hz power frequency interference. For surface micro-texture image signals, median filtering algorithm is used to remove noise. For example, a 3-pixel multiplied by 3-pixel neighborhood window is used to calculate the median. Then, spatiotemporal alignment is performed. Based on the fixed physical distance between the pressure sensor and the microwave resonant cavity sensor, for example, 1 meter, and the real-time winding speed read by the production line programmable logic controller, for example, 100 meters per second, the time lag difference is calculated as distance divided by speed, i.e., 10 milliseconds. The dielectric response signal acquired at the current moment is paired and aligned with the pressure signal acquired and cleaned 10 milliseconds ago in the data buffer to ensure that they correspond to the same fiber segment. Finally, feature extraction is performed. Calculate pressure change characteristics from the aligned pressure signal: Calculate the standard deviation of the first difference of the pressure signal sequence within a 100-millisecond time window, as the pressure change rate value. Calculate dielectric change characteristics from the dielectric response signal: Calculate the standard deviation of the first difference of the dielectric response signal sequence within the same 100-millisecond time window, as the dielectric change rate value. Calculate surface texture feature parameters from the filtered image signal: Select a rectangular analysis area in the main region of the fiber image, and apply the gray-level co-occurrence matrix algorithm to calculate texture contrast. For example, calculate the gray-level co-occurrence matrix with a pixel distance of 1 and an orientation of 0 degrees, and extract the contrast value. This value is used as the surface texture value; the closer the value is to 1, the smoother the surface.

[0025] It should be noted that the complete calculation process is as follows: The system synchronously triggers the microwave resonant cavity sensor, high-frequency pressure sensor, and high-speed linear array camera to acquire the original dielectric response voltage sequence, the original pressure voltage sequence, and the linear array scan image of the fiber surface, respectively. The two voltage sequences are cleaned using a preset digital band-stop filter, and the images are cleaned using a preset median filter. The time lag difference is calculated based on the preset sensor spacing and real-time winding speed, and the dielectric response signal is paired and aligned with the time-shifted pressure signal. The standard deviation of the difference within a 100-millisecond window is calculated for the time-shifted pressure signal, and the pressure change rate is output. The standard deviation of the difference within a 100-millisecond window is calculated for the dielectric response signal, and the dielectric change rate is output. The filtered fiber image is processed using a gray-level co-occurrence matrix algorithm in a preset analysis area to calculate texture contrast, and the surface texture value is output. The system creates a timestamp-synchronized data packet, encapsulates the pressure change rate value, dielectric change rate value, and surface texture value as a set of feature vectors into multi-dimensional feature data, and sends it to the edge analysis module.

[0026] The present invention is further configured such that the sensing module specifically includes performing digital filtering processing on the dielectric response signal and the pressure signal respectively; performing time-shift compensation on the filtered pressure signal according to a preset fiber winding speed and the distance between the acquisition points of the pressure signal and the dielectric response signal; calculating pressure change characteristics based on the time-shift compensated pressure signal; and calculating dielectric change characteristics based on the dielectric response signal. Specifically, the sensing module receives the original voltage sequence of the dielectric response signal from the microwave resonant cavity sensor and the original voltage sequence of the pressure signal from the high-frequency melt pressure sensor. First, digital filtering is performed: For the original voltage sequence of the pressure signal, a digital infinite impulse response bandstop filter with a preset center frequency of 10 Hz and a stopband width of 2 Hz is used to filter out periodic interference introduced by the vibration of the screw pump foundation; for the original voltage sequence of the dielectric response signal, a digital infinite impulse response bandstop filter with a preset center frequency of 50 Hz and a stopband width of 10 Hz is used to filter out power frequency electromagnetic interference, thus obtaining the clean pressure signal and the clean dielectric response signal. Next, time-shift compensation is performed: The preset distance between the sampling points of the pressure sensor and the dielectric sensor along the fiber running direction is fixed at 0.8 meters; the fiber winding speed is read in real time from the production line programmable logic controller, in the example 80 meters per second; the time lag is calculated as the sampling point distance divided by the winding speed, i.e., 0.01 seconds; the clean pressure signal is stored in a first-in-first-out data buffer. When processing the clean dielectric response signal at the current moment, the clean pressure signal data segment from 0.01 seconds ago is extracted from the buffer, aligning the two sets of signals in time to obtain the time-shifted pressure signal. Finally, feature calculations are performed: the analysis time window length is set to 100 milliseconds; The calculation process for pressure change characteristics is as follows: extract the data of the pressure signal within the most recent 100 milliseconds after time shift, calculate the first-order difference sequence of the difference between every two adjacent sampling points in the data segment, and then calculate the standard deviation of the first-order difference sequence. The obtained value is the pressure change characteristic. The calculation process for dielectric change characteristics is as follows: extract the data of the clean dielectric response signal within the same 100 milliseconds, calculate the first-order difference sequence of the data segment and calculate its standard deviation. The obtained value is the dielectric change characteristic. The complete workflow is as follows: the sensing module synchronously acquires the original voltage sequence of the dielectric response signal and the original voltage sequence of the pressure signal, respectively, and performs digital filtering through a preset band-stop filter. Based on the preset sampling point spacing and real-time winding speed, the time lag is calculated and time shift compensation is performed on the pressure signal. Based on a preset 100-millisecond time window, the standard deviation of the first-order difference sequence of the pressure signal after time shift is calculated as the pressure change feature, and the standard deviation of the first-order difference sequence of the clean dielectric response signal is calculated as the dielectric change feature. Finally, the pressure change feature and the dielectric change feature are output.

[0027] The present invention is further configured such that the sensing module specifically includes combining pressure change characteristics, dielectric change characteristics, and surface texture characteristic parameters to form time-synchronized multidimensional feature data. Specifically, the sensing module combines pressure change characteristics, dielectric change characteristics, and surface texture characteristic parameters to form time-synchronized multidimensional feature data. The pressure change characteristic is a scalar value characterizing the instantaneous fluctuation of melt pressure; the dielectric change characteristic is a scalar value characterizing the instantaneous fluctuation of fiber dielectric properties; the surface texture characteristic parameter is a scalar value characterizing the smoothness of the fiber surface; and the time-synchronized multidimensional feature data is a structured data packet containing the above three feature values ​​and a precise time identifier. The implementation process is based on a mechanism of unified timestamp and data buffer matching. The specific process is as follows: At the start of data acquisition, the system synchronously triggers the acquisition of a frame of raw pressure and dielectric voltage sequence and a frame of surface micro-texture image, and uniformly assigns a timestamp accurate to the microsecond level from a high-precision system clock. For example, the timestamp format is the number of seconds since the epoch plus microseconds. The pressure and dielectric signal processing thread and the image processing thread perform parallel calculations on the raw data carrying this timestamp. The pressure and dielectric signal processing thread sequentially performs preset digital band-stop filtering and time shift compensation based on fixed acquisition point spacing and real-time winding speed on the raw voltage sequence. Then, within the preset 100-millisecond analysis time window, the standard deviation of the first-order difference sequence of the signal is calculated respectively, and finally the pressure change characteristic value and dielectric change characteristic value corresponding to this timestamp are obtained. The image processing thread sequentially performs preset median filtering and texture analysis based on the gray-level co-occurrence matrix algorithm on the original image to calculate the surface texture feature parameter values ​​corresponding to this timestamp. The two processing threads store the calculation results in the form of "key-value pairs" into the signal feature buffer queue and the image feature buffer queue respectively. The storage format is, for example, `{timestamp: feature value list}`. The data packaging thread actively checks the header data of the signal feature buffer queue and the image feature buffer queue at fixed intervals (e.g., every 10 milliseconds). The packaging thread compares the timestamps carried in the header data of the two queues. When a matching timestamp is found, the packaging thread performs a data matching operation: retrieving the pressure change feature value and dielectric change feature value corresponding to that timestamp from the signal feature buffer queue, and retrieving the surface texture feature parameter value corresponding to that timestamp from the image feature buffer queue. Subsequently, the packaging thread combines the timestamp and the three feature values ​​into a complete data packet according to a predefined structured format. This data packet format can be defined as an array or object containing four fields: "timestamp," "pressure change feature," "dielectric change feature," and "surface texture feature," for example, `[1625097600.123456,0.15,0.08,0.92]`. This data packet is the time-synchronized multidimensional feature data. Finally, the data packaging thread sends this multidimensional feature data packet to the edge analysis module through the communication interface. If the timestamps at the heads of the two buffer queues are inconsistent, the data packaging thread will wait for the next cycle to perform a matching check to ensure that only complete data packets with strictly aligned timestamps are output.

[0028] The invention is further configured such that the edge analysis module includes, based on preset dynamic weighting coefficients, performing fusion calculations on multidimensional feature data to generate a particle distribution dispersion index characterizing the internal structural state of the fiber. Specifically, the edge analysis module receives a time-synchronized multidimensional feature data packet from the sensing module through a communication interface. This data packet contains a precise timestamp and three feature values: pressure change feature (a scalar characterizing instantaneous fluctuations in melt pressure), dielectric change feature (a scalar characterizing instantaneous fluctuations in fiber dielectric properties), and surface texture feature parameter (a scalar characterizing the smoothness of the fiber surface). An example data packet format is `[1625097600.123456, 0.15, 0.08, 0.92]`. Meanwhile, the edge analysis module loads dynamic weight coefficients corresponding to the current production formula from the cloud evolution module from local storage. These coefficients are a set of three preset weight values, corresponding to pressure change characteristics, dielectric change characteristics, and surface texture characteristics, respectively. They are used to quantify the relative importance of each feature to the final quality. The weight coefficients in the example are `[0.3, 0.5, 0.2]`, which means that the dielectric change characteristic has the highest weight under the current formula. The fusion calculation process employs a linear weighted summation algorithm: First, the pressure change feature value is multiplied by its corresponding weight to obtain the weighted pressure contribution; the dielectric change feature value is multiplied by its corresponding weight to obtain the weighted dielectric contribution; and the surface texture feature parameter value is multiplied by its corresponding weight to obtain the weighted texture contribution. Second, these three weighted contribution values ​​are summed to obtain a weighted sum. Subsequently, the weighted sum is multiplied by a preset scaling factor to generate the particle distribution dispersion index. This scaling factor is used to map the weighted sum to a convenient... Within the standard numerical range of the set threshold, the example scaling factor is `100`. The complete calculation example is as follows: For the example data, the weighted pressure contribution is `0.15*0.3=0.045`, the weighted dielectric contribution is `0.08*0.5=0.04`, and the weighted texture contribution is `0.92*0.2=0.184`; the weighted sum is `0.045+0.04+0.184=0.269`; and the particle distribution dispersion index is `0.269*100=26.9`. The generated particle distribution dispersion index is output in association with the data packet timestamp to characterize the uniformity and structural stability of the particle distribution within the fiber at the corresponding moment. A higher index value indicates greater structural instability. This index is compared in real-time with a preset safety threshold to determine the status. The safety threshold is set based on the statistical distribution of this index in historical normal production data; for example, a statistical upper limit can be used as the default threshold, with an example safety threshold of `30`. When the particle distribution dispersion index exceeds the safety threshold, the system determines that the fiber microstructure is unstable.

[0029] The present invention is further configured such that the edge analysis module further includes comparing the particle distribution dispersion index with a preset safety threshold in real time. When the particle distribution dispersion index continuously exceeds the safety threshold for a predetermined duration, it is determined that microstructure instability has occurred, and the following two operations are executed simultaneously: Operation 1: The edge analysis module automatically generates a control trigger signal based on the determination result of microstructure instability, and sends the control trigger signal to the reverse execution module; Operation 2: The multi-dimensional microphysical signal corresponding to the determination of microstructure instability is uploaded as abnormal waveform data to the cloud evolution module. Specifically, the edge analysis module performs an instant comparison of each particle distribution dispersion index calculated in real time. The particle distribution dispersion index is a comprehensive indicator characterizing the stability of the internal structure of the fiber. The preset safety threshold is a critical standard value for judging whether the microstructure of the fiber is unstable. This threshold is set based on the statistical distribution of the particle distribution dispersion index in historical normal production data. For example, it can be set to the upper limit of twice the standard deviation of the average value of the index in historical normal data. A default example value is `30`. The preset duration is the length of time used to confirm the persistence of the abnormal state. This duration setting should be sufficient to filter out instantaneous random interference in the production process. A default example value is `50 milliseconds`.

[0030] The edge analysis module internally sets up a continuous over-limit timer, and the specific execution process is as follows: Whenever a new particle distribution dispersion index is generated, it is immediately compared with a preset safety threshold. If the current particle distribution dispersion index is greater than the preset safety threshold, the value of the continuous over-limit timer is increased by one calculation cycle (1 millisecond in the example). If the current particle distribution dispersion index is less than or equal to the preset safety threshold, the continuous over-limit timer is immediately reset to zero. The edge analysis module continuously monitors the value of the continuous over-limit timer. When the value reaches or exceeds a predetermined duration (50 milliseconds in the example), the edge analysis module determines that microstructure instability has occurred. When the determination occurs, the edge analysis module records the start and end times of the instability period. In the example, the start time is the current time minus 50 milliseconds, and the end time is the current time. Upon determining that microstructural instability has occurred, the edge analysis module simultaneously performs the following two operations. The first operation is to automatically generate and send a control trigger signal. Based on the determination result of microstructural instability, the edge analysis module immediately generates a digital message of control trigger signal containing a specific instruction identifier and a timestamp. The edge analysis module sends this control trigger signal message to the reverse execution module through a low-latency industrial real-time communication network. The second operation is to prepare and upload abnormal waveform data. The edge analysis module extracts all the original multi-dimensional microscopic physical signal data saved in the recorded unstable period (in the example, from "current time - 50 milliseconds" to "current time") from the internal circular data buffer. This data includes the original dielectric response voltage sequence, the original pressure voltage sequence, and the original fiber surface linear array image data within this time period. The edge analysis module packages these original signal data, the corresponding production batch information, and the unstable period marker together to form a complete data file. The edge analysis module uploads this data packet as abnormal waveform data to the cloud evolution module through the network communication protocol.

[0031] The present invention is further configured such that the reverse execution module includes receiving a control trigger signal from the edge analysis module, and based on the control trigger signal, reading the current side blowing speed setting value and heater status in the production system register to confirm the execution of reverse compensation control logic. Specifically, when the edge analysis module determines that the microstructure is unstable, it sends a control trigger signal digital message in a specific format. An example message content is {"COMMAND":"TRIGGER_REVERSE","TIMESTAMP":1625097600123}. This message contains a predefined command identifier and a timestamp. After receiving the network data stream, the reverse execution module parses it according to the preset message format (JSON format), verifies whether the command identifier is consistent with the predefined start command (TRIGGER_REVERSE), and verifies the validity of the timestamp, thereby confirming that a valid control trigger signal has been received. After confirming the signal is valid, the reverse execution module immediately accesses the internal registers of the programmable logic controller (PLC) of the production system via the industrial communication protocol to obtain the current process status. The reverse execution module has preset specific register addresses to be accessed. For example, the register address corresponding to the side blowing speed setting value is "DB100.DBD10", and the register address corresponding to the heater status is "DB101.DBX0.0". The reverse execution module, as the communication master station (Modbus TCP client or OPCUA client), initiates a synchronous read request to the PLC, which is the slave station, to obtain the current value in the above addresses. The value returned by the read operation is received and stored by the reverse execution module. For example, the current side blowing speed setting value is "5.0 m / s", and the current heater status is "0" (representing the off state). The reverse execution module saves these values ​​as the reference state parameters for subsequent control logic execution. After completing the status reading, the reverse execution module performs the final confirmation and start-up preparation of the control logic. First, the reverse execution module checks its internal working mode flag to ensure that it is in an "idle" state where new instructions can be executed. Next, the reverse execution module writes the current side blowing speed setpoint and the current heater status, two reference parameters, into protected internal variables and locks these variables to prevent them from being modified during the control process. Finally, the reverse execution module atomically switches the internal working mode flag from the "idle" state to the "reverse compensation execution" state. This flag switch completes the confirmation of the execution of the reverse compensation control logic, marking the reverse execution module's formal entry into the active control stage, and then activates the subsequent control instruction sequence generation and issuance process.

[0032] The present invention is further configured such that the reverse execution module further includes: generating and issuing the following control commands in sequence according to the reverse compensation control logic, including generating a first control command for reducing the side blowing speed by sending a command to the airflow adjustment unit of the side blowing device; generating a second control command for starting the infrared heater to generate transient thermal radiation by sending a command to the drive unit of the infrared heater; and generating a third control command for restoring the side blowing speed and the infrared heater to the state before receiving the control trigger signal after a preset duration. Specifically, the reverse compensation control logic is a series of ordered control steps preset in the reverse execution module for repairing the instability of the fiber microstructure. This logic is started after the reverse execution module confirms execution. The preset duration is a key time parameter in the reverse compensation control logic, defined as the time interval between starting the infrared heater and starting the state restoration operation. This interval must ensure that the fiber experiences transient thermal effects without being overheated. The default example value is `150 milliseconds`. After the reverse execution module switches its internal working mode flag to "Reverse Compensation Execution in Progress," it immediately begins to perform the following operations in sequence: First, it generates and issues a first control command to reduce the side-blowing wind speed. The reverse execution module reads the reference side-blowing wind speed setting value (default example value is 5.0 m / s) from the locked internal reference status parameters. Based on a preset wind reduction ratio (default example value is 30%, which can be determined based on process testing and debugging), it calculates the target wind speed. The calculation formula is: target wind speed equals the reference side-blowing wind speed setting value multiplied by (1 minus the wind reduction ratio). After calculating the target wind speed, the reverse execution module generates a digital write command containing the inverter station address, the target frequency register address, and the target frequency value converted from the target wind speed, according to the communication protocol of the side-blowing device inverter. The reverse execution module sends this digital write command to the side-blowing device inverter through the industrial communication network, driving the inverter to adjust the wind speed to the target value. The second step involves generating and issuing a second control command to activate the infrared heater and generate transient thermal radiation. After the first control command is issued, the reverse execution module immediately generates a start command for the infrared heater. If the infrared heater is controlled by a digital switch, the reverse execution module, through its integrated digital output card, sets a specific output channel connected to the heater's power control circuit to a high level, energizing the heater circuit and enabling it to operate at full power. Simultaneously, the reverse execution module starts a high-precision timer and sets its timing length to a preset duration (150 milliseconds). This timer is used to control the heating duration. The third step involves generating and issuing a third control command to restore the side-blowing air speed and infrared heater to their states before receiving the control trigger signal. When the timer started in the second step reaches its preset duration (150 milliseconds), the timer generates an interrupt signal. The reverse execution module responds to this interrupt and simultaneously executes two recovery operations: First, it generates an air speed recovery command, which rewrites the locked reference side-blowing air speed setting (5.0 m / s) into the inverter's target frequency register, commanding the inverter to restore the air speed to its original setting. Second, it generates a heater shutdown command, which cuts off the infrared heater power supply by setting the output channel controlling the heater on the digital output card to a low level. The reverse execution module issues the air speed recovery command through the communication network and simultaneously changes the state of the digital output channel, thereby restoring the side-blowing air speed and infrared heater state to the reference state recorded before receiving the control trigger signal. After the recovery command is executed, the reverse execution module changes its internal working mode flag from "reverse compensation in progress" back to "idle," and the entire reverse compensation control sequence ends.

[0033] The present invention is further configured such that the cloud evolution module includes receiving abnormal waveform data from the edge analysis module and performing correlation matching with the corresponding batch of offline quality inspection data. Based on data correlation and pattern mining, it analyzes the correlation between the signal patterns hidden in the abnormal waveform data and the quality indicators in the offline quality inspection data. Specifically, the cloud evolution module continuously receives abnormal waveform data from the edge analysis module. The abnormal waveform data is a structured data packet containing the production batch number, the timestamp range of the abnormal occurrence period, and the original multi-dimensional microscopic physical signals collected during that period. Simultaneously, the cloud evolution module obtains the corresponding batch of offline quality inspection data from the laboratory information management system. The offline quality inspection data consists of quantitative performance indicators of the fiber finished product after standard laboratory testing, such as conductivity, breaking strength, and wash resistance, and is strictly bound to the production batch number. The cloud-based evolution module first performs a data association and matching operation. It parses the abnormal waveform data packets and extracts the production batch number and abnormal time period information. Then, in the cloud-based relational database, the cloud-based evolution module performs a database association query operation with the production batch number as the unique association key. It accurately associates abnormal waveform data records with the same production batch number with offline quality inspection data records to form a complete data sample. For example, the abnormal waveform data with the production batch number "LOT20231028001" and the abnormal time period "10:15:30.000 to 10:15:30.050" is associated with the offline quality inspection result of the same batch, "conductivity: 78 Siemens / meter, tensile strength: 4.2 centinewtons / dtex", as a single sample. After the association is completed, the cloud evolution module performs signal pattern feature extraction. The cloud evolution module performs secondary analysis on the original signal in the abnormal waveform data packet in the associated sample, and calculates a set of predefined time domain and frequency domain feature values ​​to form a signal pattern feature vector describing the abnormal event. The specific extracted features include: for the original dielectric response signal, calculating its zero-crossing rate during the abnormal period. This feature is used to quantify the frequency of signal polarity changes and can identify sawtooth wave patterns; for the original pressure signal, performing a fast Fourier transform to extract the amplitude of the dominant frequency in its spectrum. This feature is used to quantify the intensity of periodic pressure disturbances; for the original fiber surface image, calculating its local binary mode variance. This feature is used to quantify the non-uniformity of surface texture. Finally, the cloud-based evolution module mines association patterns based on accumulated associated samples. The cloud-based evolution module periodically (by default, every 1000 new samples) inputs a batch of associated samples into the data mining algorithm. The input of each sample is a signal pattern feature vector, and the output is a quality label (e.g., “conductivity qualified” or “conductivity unqualified”, the judgment threshold can be preset to conductivity ≥ 80 Siemens / meter). The cloud-based evolution module uses the decision tree algorithm, a supervised learning algorithm, to train the samples. The decision tree algorithm automatically analyzes the relationship between feature vectors and quality labels, generating an interpretable tree model from which explicit association rules can be extracted. For example, a generated rule might be stated as: if "the zero-crossing rate of the dielectric signal is >10 Hz" and "the amplitude of the main frequency of the pressure signal is <2.0 kPa", then predict "conductivity is unqualified". This rule has a confidence level of 85% (i.e., 85% of the samples meeting this condition are indeed unqualified in terms of conductivity) and a support level of 15% (i.e., 15% of the total samples meet this condition). This "if-then" form of logical judgment is a mined association rule, and the cloud-based evolution module subsequently stores this association rule in a structured format in a knowledge base for later optimization of dynamic weight coefficients.

[0034] The invention is further configured such that the cloud-based evolution module also includes adjusting the weight allocation of each feature in the dynamic weight coefficient based on the correlation pattern and a preset weight update strategy, and uploading the updated dynamic weight coefficient to a register for execution in the next cycle. Specifically, the cloud-based evolution module adjusts the dynamic weight coefficient in real time according to the correlation pattern obtained by data mining. This coefficient is an array containing three weight values, corresponding to the pressure change feature, dielectric change feature, and surface texture feature parameters used by the edge analysis module, respectively, for weighted fusion calculation. The initial value of the dynamic weight coefficient is set according to the production formula; the default initial value is `[0.30, 0.50, 0.20]`. The correlation pattern is a judgment obtained by the cloud-based evolution module through data mining and expressed in the form of logical rules, which describes the stable correspondence between specific signal pattern feature combinations and specific quality defects. The preset weight update strategy is a set of predefined algorithm rules used to quantify the information in the correlation pattern into the adjustment amount of the dynamic weight coefficient. Its core parameters include a preset gain factor and a preset smoothing factor. The adjustment process begins with the analysis and feature mapping of association patterns. The cloud-based evolution module analyzes the newly discovered association rules and extracts all conditional clauses from the rule section. Each conditional clause contains a signal pattern feature, a comparison operator, and a threshold. Subsequently, through a preset feature mapping table, the signal pattern features used in the conditions are mapped to one of the three fusion features used by the edge analysis module. The feature mapping is as follows: the signal pattern feature "dielectric signal zero-crossing rate" is mapped to the fusion feature "dielectric change feature"; the signal pattern feature "pressure signal dominant frequency amplitude" is mapped to the fusion feature "pressure change feature"; and the signal pattern feature "image texture variance" is mapped to the fusion feature "surface texture feature parameters". At the same time, the confidence and support of the rule are recorded. Next, the weight adjustment is calculated according to the weight update strategy. The cloud evolution module maintains an initial responsibility score of zero for each of the three fusion features. For each association rule to be processed, the fusion feature mapped to its condition clause is checked. For each mapped fusion feature, its responsibility score is increased by an increment equal to the rule's confidence multiplied by its support, and then multiplied by a preset gain factor. For example, for a rule with a confidence of 85% and a support of 15%, if its condition is mapped to "dielectric change feature", the responsibility score of this feature increases by `0.85*0.15*1.0=0.1275`. The responsibility scores of fusion features not mapped by the rule's condition remain unchanged. After processing a batch of new association rules, the updated responsibility scores of the three fusion features are obtained, for example, `[0.1275,0.1275,0.14]`. Subsequently, weight calculation and smoothing updates are performed. First, the vector composed of the three responsibility scores is converted into an ideal weight distribution vector with a sum of 1 using the Softmax function, for example, `[0.32, 0.32, 0.36]`. Then, the exponential moving average algorithm is used to fuse the ideal weights with the old dynamic weight coefficients to generate new, smoothly transitioning dynamic weight coefficients. The specific calculation formula is: New weight = Smoothing factor × Ideal weight + (1 - Smoothing factor) × Old weight. Based on example data, the new weights for pressure features are: `0.1 * 0.32 + 0.9 * 0.30 = 0.302`; for dielectric features, `0.1 * 0.32 + 0.9 * 0.50 = 0.482`; for texture features, `0.1 * 0.36 + 0.9 * 0.20 = 0.216`. Finally, the updated dynamic weight coefficients are approximately `[0.30, 0.48, 0.22]`. Finally, the cloud evolution module generates and issues the update command. It encapsulates the calculated new dynamic weight coefficients (`[0.30,0.48,0.22]`), the device identifier of the target edge analysis module, the corresponding production recipe code, and the activation command into a structured configuration update message. Through a secure network communication protocol, it publishes this message to the specified topic. The target edge analysis module subscribes to the topic, verifies its validity after receiving the message, and writes the new dynamic weight coefficients into its local non-volatile storage configuration area, replacing the original coefficients. According to the command requirements, the edge analysis module will load and use the updated coefficients in the next production monitoring cycle to perform fusion calculation of the particle distribution dispersion index.

[0035] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A functional fiber spinning production monitoring system based on edge-cloud collaboration, characterized in that, include: The sensing module synchronously acquires multi-dimensional microscopic physical signals, performs cleaning, spatiotemporal alignment, and feature extraction processing to obtain multi-dimensional feature data; Edge analysis module: Based on preset dynamic weight coefficients, it performs fusion calculations on multi-dimensional feature data, generates a particle distribution dispersion index, compares it with a preset threshold, generates a control trigger signal, and then uploads abnormal waveform data. Reverse execution module: Reads the register to obtain the set value of the blowing speed and the status of the heater, and combines the control trigger signal to generate control instructions by executing the preset reverse compensation control logic; Cloud-based evolution module: Based on abnormal waveform data and offline quality inspection data, it iteratively updates dynamic weight coefficients through data correlation and pattern mining.

2. The functional fiber spinning production monitoring system based on edge-cloud collaboration according to claim 1, characterized in that, The multi-dimensional microscopic physical signals specifically include: dielectric response signals, pressure signals, and surface micro-texture image signals.

3. The functional fiber spinning production monitoring system based on edge-cloud collaboration according to claim 2, characterized in that, The sensing module specifically includes: Digital filtering is performed on the dielectric response signal and the pressure signal respectively; Based on the preset fiber winding speed and the distance between the acquisition points of the pressure signal and the dielectric response signal, time shift compensation is performed on the filtered pressure signal. Pressure change characteristics are calculated based on the pressure signal after time shift compensation, and dielectric change characteristics are calculated based on the dielectric response signal.

4. The functional fiber spinning production monitoring system based on edge-cloud collaboration according to claim 3, characterized in that, The sensing module specifically also includes: By combining pressure change characteristics, dielectric change characteristics, and surface texture characteristics, time-synchronized multidimensional feature data is formed.

5. A functional fiber spinning production monitoring system based on edge-cloud collaboration according to claim 1, characterized in that, The edge analysis module includes: Based on preset dynamic weighting coefficients, multidimensional feature data are fused and calculated to generate a particle distribution dispersion index that characterizes the internal structural state of fibers.

6. The functional fiber spinning production monitoring system based on end-to-cloud collaboration according to claim 5, characterized in that, The edge analysis module also includes: The particle distribution dispersion index is compared with a preset safety threshold in real time. When the particle distribution dispersion index continuously exceeds the safety threshold for a predetermined duration, it is determined that microstructure instability has occurred, and the following two operations are performed simultaneously: Operation 1: The edge analysis module automatically generates a control trigger signal based on the determination result of microstructural instability, and sends the control trigger signal to the reverse execution module: Operation 2: Upload the multi-dimensional microscopic physical signals corresponding to the determination of microstructural instability as abnormal waveform data to the cloud evolution module.

7. A functional fiber spinning production monitoring system based on edge-cloud collaboration according to claim 6, characterized in that, The reverse execution module includes: Receives control trigger signals from the edge analysis module; Based on the control trigger signal, the current side blowing speed setpoint and heater status are read from the production system register to confirm the execution of the reverse compensation control logic.

8. A functional fiber spinning production monitoring system based on end-to-cloud collaboration according to claim 7, characterized in that, The reverse execution module further includes: generating and issuing the following control commands sequentially according to the reverse compensation control logic: Based on the reverse compensation control logic, control commands are generated and issued sequentially, including: By sending a command to the airflow regulation unit of the side blowing device, a first control command for reducing the side blowing speed is generated. By sending instructions to the drive unit of the infrared heater, a second control command is generated to start the infrared heater to produce transient thermal radiation. After a preset duration, a third control command is generated to restore the side-blowing air speed and the infrared heater to the state before receiving the control trigger signal.

9. A functional fiber spinning production monitoring system based on edge-cloud collaboration according to claim 1, characterized in that, The cloud-based evolution module includes: It receives abnormal waveform data from the edge analysis module and correlates and matches it with the corresponding batch of offline quality inspection data; Based on data association and pattern mining, we analyze the correlation between the signal patterns hidden in abnormal waveform data and the quality indicators in offline quality inspection data.

10. A functional fiber spinning production monitoring system based on end-to-cloud collaboration according to claim 9, characterized in that, The cloud-based evolution module also includes: Based on the correlation pattern, the weight allocation of each feature in the dynamic weight coefficient is adjusted through a preset weight update strategy, and the updated dynamic weight coefficient is uploaded to the register to wait for the next cycle to be executed.

Images