A method for on-line control of the production quality of a cable protection tube

Abstract

The present application relates to a kind of production quality online control method of cable protection pipe, to solve the problem of real-time monitoring and response optimization of multiple source quality data.The method is by arranging sensor in the key link of production line to collect wall thickness standard deviation, concentricity offset, surface roughness mean and material melting index fluctuation and other multi-source time series characteristic data, generate synchronous multi-dimensional quality state vector using sliding time window and normalization processing, adopt shannon entropy to quantify system uncertainty and calculate entropy change rate, dynamically map to instruction response priority, realize the priority reordering of instruction to key control channel and combine anti-shock constraint, guarantee the smooth switching of control signal.Further introduce closed-loop adaptive learning mechanism, automatically correct baseline and driving threshold, improve the ability to deal with long-term process drift.The scheme realizes early warning and accurate intervention of quality risk, improves product consistency and process intelligent level.

Description

Technical Field

[0001] This invention relates to the field of continuous production process control technology for cable protection pipes, and in particular to an online quality control method for the production of cable protection pipes. Background Technology

[0002] Current quality control schemes in cable protection pipe production generally adopt instruction priority settings based on preset weights, manual calibration, or fixed anomaly thresholds. Once an anomaly warning is triggered, control instructions are executed sequentially according to a pre-set priority list. Mainstream solutions, such as melt pressure closed-loop control and traction speed feedback adjustment, map quality anomalies to fixed response actions through rule-based scheduling mechanisms. Some systems utilize statistical process control methods for deviation monitoring. However, these approaches heavily rely on historical experience and manual thresholds, are slow to identify multi-source process disturbances and complex quality instability states, struggle to capture real-time coupled fluctuations between quality characteristics, and lack dynamic adaptability in scenarios of sudden anomalies or transfers. As process complexity increases, quality anomalies exhibit highly multidimensional, time-varying, and coupled characteristics. Fixed instruction priority leads to delayed responses to critical anomalies and cannot adapt to quality pattern drift caused by raw material fluctuations and equipment wear.

[0003] Existing technologies have the following limitations: instruction priority decisions rely on pre-defined rules or set thresholds, making it difficult to dynamically adapt to real-time quality risk conditions; they lack in-depth analysis methods based on the inherent changing trends of multi-source real-time data, making it impossible to capture the critical process of the system transitioning from stability to instability, and instruction issuance is mostly a passive response rather than proactive intervention; they cannot self-optimize priority scheduling strategies as the production process evolves, nor can they form closed-loop learning feedback based on instruction execution results; they can only identify isolated single-point anomalies, making it difficult to achieve timely responses to the coupled instability of multi-dimensional quality characteristics and dynamic tracking of dominant anomaly factors. Summary of the Invention

[0004] This application provides an online quality control method for the production of cable protection pipes, which aims to solve one of the problems or issues of the prior art mentioned in the background section.

[0005] This application provides an online quality control method for the production of cable protection pipes, specifically including:

[0006] S1: Acquire multi-source quality characteristic time-series data during the continuous production process of cable protection pipes, and perform vector encapsulation on the multi-source quality characteristic time-series data based on a sliding time window to generate a multi-dimensional quality state vector sequence;

[0007] S2: Perform normalization mapping on the multidimensional mass state vector sequence, calculate the probability density histogram based on the normalized data distribution, and then perform uncertainty quantization operation on the probability density histogram to generate the current mass state Shannon entropy value.

[0008] S3: Perform first-order difference absolute value operation on the Shannon entropy value of the current mass state corresponding to multiple time windows collected continuously, and define the operation result as the mass state entropy change rate;

[0009] S4: Input the quality state entropy change rate into the dynamic mapping relationship rule for threshold comparison and judgment. If the quality state entropy change rate suddenly increases and continues to exceed the baseline bandwidth, generate an instruction reordering trigger signal; otherwise, generate a signal to maintain the original execution order.

[0010] S5: In response to the instruction reordering trigger signal, identify the dominant quality characteristic factor with the highest correlation to the quality state entropy change rate, and increase the priority of the control channel instruction corresponding to the dominant quality characteristic factor in the pending instruction queue, and generate a reordered control instruction execution sequence.

[0011] S6: Apply an instruction execution constraint buffer mechanism to the reordered control instruction execution sequence, and only schedule pending instructions that have not yet been issued to the actuator according to the reordered control instruction execution sequence, while keeping the instructions that have been issued to the actuator executed in the original order, thereby generating an anti-oscillation control instruction stream.

[0012] S7: Based on the execution results of the anti-vibration control command stream generated this time, the historical entropy baseline data corresponding to the dominant quality characteristic factor is updated synchronously to form closed-loop adaptive learning parameters.

[0013] The online quality control method for cable protection pipe production provided in this application has the following beneficial effects:

[0014] This solution constructs a multidimensional quality state vector within a sliding time window, combines normalization processing and Shannon entropy calculation, and introduces the absolute value of the first-order difference of the entropy value of the continuous time window as the quality state entropy change rate. This improves the foresight and sensitivity of process control and reduces the risk of anomaly warnings lagging behind actual defect formation. By establishing an adaptive mapping relationship between entropy change rate and control urgency, and integrating an execution constraint buffer mechanism, the instruction reordering process is automatically activated when the quality entropy change rate suddenly increases and continuously exceeds the baseline fluctuation range. Priority is given to control channel instructions most strongly associated with the entropy change dominant factor. At the same time, only instructions waiting to be issued that have not yet entered the execution queue are adjusted, while issued instructions are completed in their original order. This improves the accuracy of instruction response and system stability, and reduces the risk of actuator oscillation caused by frequent reverse adjustments. After each reordering, the historical entropy baseline of relevant features is updated synchronously, improving the system's long-term memory and dynamic adaptation capability to changes in the process environment, and reducing reliance on expert experience and manual parameter tuning. Attached Figure Description

[0015] Figure 1 This is a main flowchart of an online quality control method for the production of cable protection pipes.

[0016] Figure 2 This is a sub-flowchart of a method for online quality control in the production of cable protection pipes;

[0017] Figure 3 This is another sub-flowchart of a method for online quality control in the production of cable protection pipes. Detailed Implementation

[0018] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0019] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0020] like Figure 1 As shown, this application provides an online quality control method for the production of cable protection pipes, specifically including:

[0021] S1: Acquire multi-source quality characteristic time-series data during the continuous production process of cable protection pipes, and perform vector encapsulation on the multi-source quality characteristic time-series data based on a sliding time window to generate a multi-dimensional quality state vector sequence;

[0022] S2: Perform normalization mapping on the multidimensional mass state vector sequence, calculate the probability density histogram based on the normalized data distribution, and then perform uncertainty quantization operation on the probability density histogram to generate the current mass state Shannon entropy value.

[0023] S3: Perform first-order difference absolute value operation on the Shannon entropy value of the current mass state corresponding to multiple time windows collected continuously, and define the operation result as the mass state entropy change rate;

[0024] S4: Input the quality state entropy change rate into the dynamic mapping relationship rule for threshold comparison and judgment. If the quality state entropy change rate suddenly increases and continues to exceed the baseline bandwidth, generate an instruction reordering trigger signal; otherwise, generate a signal to maintain the original execution order.

[0025] S5: In response to the instruction reordering trigger signal, identify the dominant quality characteristic factor with the highest correlation to the quality state entropy change rate, and increase the priority of the control channel instruction corresponding to the dominant quality characteristic factor in the pending instruction queue, and generate a reordered control instruction execution sequence.

[0026] S6: Apply an instruction execution constraint buffer mechanism to the reordered control instruction execution sequence, and only schedule pending instructions that have not yet been issued to the actuator according to the reordered control instruction execution sequence, while keeping the instructions that have been issued to the actuator executed in the original order, thereby generating an anti-oscillation control instruction stream.

[0027] In one embodiment, this application provides an online quality control method for the production of cable protection pipes, which further includes:

[0028] S7: Based on the execution results of the anti-vibration control command stream generated this time, the historical entropy baseline data corresponding to the dominant quality characteristic factor is updated synchronously to form closed-loop adaptive learning parameters.

[0029] Step S1: Acquire multi-source quality characteristic time-series data during the continuous production of cable protection pipes, and vectorize the multi-source quality characteristic time-series data based on a sliding time window to generate a multi-dimensional quality state vector sequence. The multi-source quality characteristic time-series data includes wall thickness standard deviation, concentricity offset, mean surface roughness, and material melt flow index fluctuation. Specifically, it includes:

[0030] S1.1: Synchronously trigger control is performed on the sensor array deployed at the extruder head, traction machine and vacuum shaping box to obtain the original multi-source quality characteristic time series data including wall thickness standard deviation, concentricity offset, surface roughness mean and material melt index fluctuation, forming a set of discrete sampling points with a unified timestamp.

[0031] Initialize the distributed sensor network communication links at the extruder head, traction machine, and vacuum shaping box, and establish a global clock synchronization mechanism based on the IEEE 1588 precision time protocol to ensure that the clock deviation of each node is controlled within the microsecond range, providing a unified time reference system for multi-source data fusion.

[0032] Hardware-level synchronous trigger pulse signals are sent to the ultrasonic wall thickness detection probe, laser concentricity scanner, online roughness meter, and melt flow index online monitoring module to force each sensor to start the data acquisition cycle at the same physical moment, eliminating sampling phase misalignment caused by internal sensor processing delays.

[0033] It receives raw analog or digital signals returned by each sensor and performs signal conditioning and analog-to-digital conversion operations on four dimensions: wall thickness standard deviation, concentricity offset, surface roughness mean, and material melt index fluctuation, respectively, to convert continuous physical quantities into discrete digital sampling sequences.

[0034] A high-precision global timestamp tag is attached to each collected discrete sampling point, with a timestamp resolution of no less than 1 millisecond, forming a binary data structure containing numerical and time information, ensuring that each set of quality feature data can be traced back to a specific production time.

[0035] Multi-source discrete sampling points with unified timestamps are aggregated and encapsulated to generate an initial set of original multi-source quality feature time series data. This set fully records the instantaneous state of each key quality indicator in the current production cycle, providing standardized input for subsequent timeline alignment processing.

[0036] S1.2: Perform time axis alignment processing on the discrete sampling point set based on a preset sliding step size to eliminate the transmission delay differences between multi-source signals and generate a multi-source quality feature synchronous data stream with strict time sequence correspondence.

[0037] The system receives a set of discrete sampling points with a unified timestamp but different physical transmission delays. This set includes heterogeneous data from multiple sources, such as wall thickness, concentricity, roughness, and melt flow index. The system analyzes the inherent hardware delay parameters of each sensor node and the signal transmission link delay constant to construct a multi-channel delay compensation matrix, providing a reference offset for subsequent alignment.

[0038] The time axis resampling grid is determined based on the preset sliding step size, and a standard time series index with equal intervals is generated as a unified time reference surface for multi-source data fusion, eliminating the time stamp asynchrony caused by slight fluctuations in sampling frequency.

[0039] Linear or spline interpolation is performed on each channel of quality feature time series data to map the original discrete sampling points to a standard time series index. By calculating the rate of change of slope between adjacent sampling points, the feature values ​​at non-sampling times are estimated, filling in the data gaps caused by time delay compensation, and generating a continuous and smooth single-channel reconstructed data stream.

[0040] The delay compensation matrix is ​​applied to perform phase correction on each single-channel reconstructed data stream. Data points that lag behind the standard time reference are shifted forward on the time axis, or data points that are ahead are shifted backward, to ensure that the multi-source features at the same physical moment strictly coincide on the logical time axis.

[0041] Integrity checks are performed on the phase-corrected multi-source data to remove outliers caused by excessive interpolation errors. The calibrated wall thickness standard deviation, concentricity offset, mean surface roughness, and material melt index fluctuation data are merged by timestamp to form a synchronous data stream of multi-source quality characteristics with strict temporal correspondence.

[0042] S1.3: Use a sliding time window of fixed length to perform a truncation operation on the multi-source quality feature synchronization data stream to extract local data fragments in the current time and its historical neighborhood, and generate multiple independent time window data subsets.

[0043] Receive a multi-source quality characteristic synchronous data stream after time axis alignment. This data stream contains a sequence of continuous sampling points in four dimensions: wall thickness standard deviation, concentricity offset, mean surface roughness, and material melt index fluctuation.

[0044] A fixed-length sliding time window with a window size W and a sliding step size S is set. The window size W is determined based on the response hysteresis characteristics of the cable protection pipe extrusion process to ensure coverage of a complete process disturbance cycle. The sliding step size S is smaller than the window size W to achieve overlapping sampling of data.

[0045] Based on the current system clock time t, a data segment of length W is extracted from the end of the synchronous data stream to construct a local data subset containing the time interval [t-W+1, t]. This subset fully preserves the quality state evolution information of the current time and its historical neighborhood.

[0046] The extracted local data subset is encapsulated into an independent time window data object, and the start and end timestamps of the time window are recorded to ensure the uniqueness and traceability of each time window data subset in terms of time sequence.

[0047] The time base is moved forward according to the preset sliding step size S, and the truncation and encapsulation operations are repeated to generate a series of independent time window data subset sequences that are continuous and partially overlapping on the time axis.

[0048] S1.4: Perform vector space mapping transformation on the wall thickness standard deviation, concentricity offset, surface roughness mean and material melt index fluctuation in each time window data subset to convert scalar time series data into numerical expression in high-dimensional space and generate single-window multidimensional quality feature vector.

[0049] Receive the independent time window data subset generated by S1.3. This subset contains four types of scalar time series data segments: wall thickness standard deviation, concentricity offset, surface roughness mean, and material melt index fluctuation.

[0050] Perform a dimension alignment check on each quality feature component within the time window data subset to confirm that the number of sampling points for each feature is consistent within the time window length N, and construct an initial 4×N dimensional original data matrix.

[0051] The Z-Score standardization method is used to perform zero-mean processing on each feature component, and the arithmetic mean and standard deviation of each feature within the current time window are calculated to eliminate the difference in numerical magnitude caused by different physical units.

[0052] The standardized feature data are arranged in rows, and the time axis is used as a column vector through transpose operation to construct a local state space mapping basis with time as the index and quality features as the dimension.

[0053] Based on the phase space reconstruction theory, the embedding dimension is set to 4 and the delay time is 1. The four standardized feature values ​​at the same time are combined into a four-dimensional state vector to realize the mapping transformation from one-dimensional time sequence signal to four-dimensional phase space point.

[0054] By iterating through all time points within the time window, corresponding four-dimensional state vectors are generated sequentially, forming a point cloud set that reflects the evolution trajectory of the quality state within the current window.

[0055] The covariance matrix is ​​calculated on the generated point cloud set to extract the correlation structure information between features, thereby enhancing the ability of vector representation to characterize multi-source coupling effects.

[0056] The eigenvalue decomposition results of the covariance matrix are incorporated into the state vector weight allocation to generate a weighted single-window multidimensional quality feature vector, highlighting the contribution of the dominant quality factor.

[0057] S1.5: Perform serialization and encapsulation processing on the continuously generated single-window multidimensional quality feature vectors in the order of time progression to construct a data structure that reflects the dynamic evolution of quality state and output a multidimensional quality state vector sequence.

[0058] Step S2: Perform normalization mapping on the multidimensional mass state vector sequence to eliminate dimensional differences, and calculate the probability density histogram based on the normalized data distribution. Then, perform uncertainty quantization on the probability density histogram to generate the Shannon entropy value of the current mass state. Specifically, the uncertainty quantization operation on the probability density histogram is implemented using the Shannon entropy calculation formula.

[0059] S2.1: Obtain the multidimensional quality state vector sequence, which includes original feature data such as wall thickness standard deviation, concentricity offset, mean surface roughness, and material melt index fluctuation. Based on range normalization processing, perform dimensionless mapping processing on each component in the multidimensional quality state vector sequence to eliminate the interference of different physical dimensions on probability distribution calculation and generate a normalized quality state vector sequence.

[0060] Receive the multidimensional quality state vector sequence, which contains raw feature data with different physical dimensions, such as wall thickness standard deviation, concentricity offset, mean surface roughness, and material melt index fluctuation.

[0061] The maximum and minimum values ​​of each quality characteristic component within a preset sliding time window are determined, and a dynamic boundary reference set for range normalization is constructed to adapt to the process drift characteristics in the production process.

[0062] A linear mapping transformation is performed on each component in the multidimensional mass state vector sequence to compress the original values ​​to the standard interval [0,1], eliminating the probability distribution calculation bias caused by the difference in magnitude.

[0063] The following range normalization formula is used to perform dimensionless processing on the j-th quality feature component within the i-th time window:

[0064]

[0065] in, The value is the normalized value. These are the original sampled values. and These are the minimum and maximum values ​​of the feature within the current statistical period, respectively.

[0066] By traversing all time windows and all feature dimensions, the standardization transformation of the entire sequence data is completed, generating a normalized quality state vector sequence with uniform numerical distribution and consistent dimensions.

[0067] S2.2: Receive the normalized quality state vector sequence, extract historical data segments at the current moment based on the sliding time window mechanism, discretize the numerical distribution within the historical data segments using an equal-width binning strategy, count the sample frequencies within each bin interval and convert them into relative frequencies, and generate probability density histogram data.

[0068] The system receives a multidimensional quality state vector sequence after range normalization, which contains dimensionless feature data such as wall thickness standard deviation and concentricity offset, and serves as the basic input source for probability distribution statistics.

[0069] Based on the sliding time window mechanism, a historical data segment consisting of the current moment and its N preceding sampling points is extracted from the normalized vector sequence to form an independent data subset for local statistical characteristic analysis, ensuring that the statistical window matches the process response delay.

[0070] An equal-width binning strategy is adopted to discretize the numerical distribution within the historical data segment. The normalized value range of [0,1] is uniformly divided into K non-overlapping sub-intervals. The width of each sub-interval is defined as 1 / K, and a histogram bins structure covering the entire value range is constructed.

[0071] Iterate through each sampling point in the historical data segment, determine the specific bin interval into which the value falls, and perform an accumulation operation on the counter of the corresponding interval to count the sample frequency in each bin interval, forming the original frequency distribution vector.

[0072] Divide the sample frequency of each bin interval by the total number of samples in the historical data segment to calculate the relative frequency of each bin interval. Ensure that the sum of the relative frequencies of all bins is strictly equal to 1, thus completing the normalization transformation from frequency to probability density.

[0073] The calculated relative frequency values ​​of each bin are arranged in bin order to generate a probability density histogram that represents the distribution pattern of quality characteristics within the current time window. This data directly reflects the central tendency and dispersion of the quality status.

[0074] S2.3: Read the relative frequency values ​​of each bin in the probability density histogram data, perform self-information calculation on each non-zero relative frequency value based on the natural logarithm operation rule, and sum the products of the self-information of all bins and their corresponding relative frequency values ​​to generate the original uncertainty metric value within the current time window.

[0075] Read the relative frequency values ​​of each bin in the probability density histogram data to construct a basic dataset for uncertainty quantification.

[0076] The system iterates through each bin interval in the histogram, filters out valid bins with relative frequency values ​​greater than zero, and excludes zero-probability terms to avoid divergence in logarithmic operations.

[0077] Based on the natural logarithm operation rules, self-information is calculated for the relative frequency value of each valid bin to quantify the degree of surprise in the occurrence of a single state.

[0078] The self-information of the i-th bin is calculated using the following formula:

[0079]

[0080] in, Let be the self-information of the i-th bin. Let ln be the relative frequency value of the i-th sub-bin, and ln represents the natural logarithm operation.

[0081] The self-information of each bin is multiplied by its corresponding relative frequency value to obtain the weighted contribution of that bin to the overall uncertainty.

[0082] The weighted contribution values ​​of all bins are summed to integrate local information entropy into a global metric.

[0083] The original uncertainty measure for the current time window is generated using the following formula:

[0084]

[0085] in, This is the original uncertainty measure. The total number of effective boxes, This is the weighted contribution value of the i-th bin.

[0086] S2.4: Obtain the original uncertainty metric value, perform a sign inversion operation on the original uncertainty metric value based on the Shannon entropy definition formula to correct the directionality, define the inverted result as the current quality state Shannon entropy value characterizing the degree of disorder of the quality system in the continuous production process of cable protection pipe, and output it to the next stage for entropy change rate calculation.

[0087] Obtain the original uncertainty metric value output from step S2.3. This value is the sum of the products of the self-information of each bin and the relative frequency. Its sign direction is opposite to the Shannon entropy definition, and it serves as the input data source for the sign inversion operation.

[0088] The standard definition formula of Shannon entropy is invoked to perform an algebraic sign inversion operation on the original uncertainty measure value to correct the negative value deviation caused by the logarithmic operation, ensuring that the entropy value is non-negative and conforms to the axiomatic system of information theory, and generating the inverted Shannon entropy value.

[0089] The sign correction and entropy calculation are performed using the following mathematical expression:

[0090]

[0091] in, The Shannon entropy value represents the current mass state. Let be the relative frequency of the i-th sub-bin. This represents the total number of boxes.

[0092] The calculated Shannon entropy value of the current quality state is defined as a quantitative index characterizing the dispersion of multidimensional quality characteristics such as wall thickness and concentricity and the degree of system disorder during the continuous production of cable protection pipes.

[0093] The current quality state Shannon entropy value is encapsulated using a data type, and a timestamp index of the current time window is added to generate a standardized entropy value data unit.

[0094] The standardized entropy data unit is output to step S3 as the basic input data for calculating the entropy change rate of the mass state.

[0095] Step S3: Based on the Shannon entropy values ​​of the current mass state corresponding to multiple continuously acquired time windows, perform a first-order difference absolute value operation, and define the operation result as the mass state entropy change rate. The mass state entropy change rate characterizes the system's transition trend from steady state to instability; and the mass state entropy change rate is used as the core driving parameter for instruction priority recalibration. Specifically, this includes:

[0096] S3.1: Obtain the current quality state Shannon entropy values ​​corresponding to multiple consecutive time windows arranged in chronological order, and perform time-series alignment processing on the current quality state Shannon entropy values ​​based on the sliding window mechanism to generate a standardized Shannon entropy value time-series data stream.

[0097] Receive the discrete set of Shannon entropy values ​​of the current quality state output from the preceding steps. This set contains multiple uncertainty quantization values ​​arranged by time window index.

[0098] A fixed-length time-series buffer queue is constructed. Newly added Shannon entropy values ​​are pushed to the tail of the queue, while expired data at the head of the queue is removed, maintaining the queue length consistent with the number of sliding time windows. Timestamp verification is performed on the Shannon entropy values ​​in the buffer queue to identify and mark time segments with transmission delays or data loss. For the marked time segments, linear interpolation is used to fill data gaps, ensuring the continuity of the time-series data. Monotonicity checks are performed on the completed Shannon entropy value sequence to remove non-physical extreme value jumps caused by transient sensor failures. The corrected Shannon entropy value sequence is resampled and aligned using a sliding window mechanism to generate a standardized Shannon entropy time-series data stream.

[0099] S3.2: Perform a difference operation on the values ​​of two adjacent time windows in the time-series data stream based on the standardized Shannon entropy values ​​to calculate the original Shannon entropy difference value set.

[0100] The system receives a normalized Shannon entropy sequence after time alignment processing. This sequence is calculated by a continuous sliding time window, and each data point represents a quantitative index of the disorder of the cable protection pipe quality status within the corresponding time window.

[0101] Extract the standardized Shannon entropy value H(t) corresponding to the current time t and the standardized Shannon entropy value H(t-1) corresponding to the previous time t-1 to establish the baseline data pair for the difference operation, ensuring the strict continuity of the time index to reflect the instantaneous change trend.

[0102] Perform a first-order forward difference operation, subtract the entropy value from the previous time step from the current entropy value, and calculate the algebraic difference between the two to obtain the instantaneous change in the Shannon entropy value. This change includes a positive or negative sign to indicate the direction of increase or decrease in uncertainty.

[0103] The difference calculation is performed using the following mathematical expression:

[0104]

[0105] in, This represents the original Shannon entropy difference value at time t. Represents the standardized Shannon entropy value at time t. The value represents the standardized Shannon entropy at time t-1.

[0106] By traversing the entire time series data stream of standardized Shannon entropy values, the above difference operation is repeated for each effective time step to generate a set of original Shannon entropy difference values ​​that matches the length of the original time series. This set completely records the instantaneous rate trajectory of the uncertainty evolution of the quality system.

[0107] S3.3: Perform absolute value mapping processing on each original Shannon entropy difference value in the original Shannon entropy difference value set to eliminate interference from the direction of change and extract the unsigned Shannon entropy fluctuation intensity index sequence that characterizes the degree of change.

[0108] Receive the original set of Shannon entropy difference values, which contains the algebraic difference of Shannon entropy values ​​between adjacent time windows, reflecting the instantaneous direction and magnitude of the uncertainty change in the quality system.

[0109] Iterate through each numerical element in the original Shannon entropy difference set and identify the negative difference data that represents a decrease in entropy and the positive difference data that represents an increase in entropy.

[0110] The negative difference data in the set is inverted to convert it to a positive value; the positive difference data is kept unchanged to unify the measurement standard.

[0111] Based on the absolute value function mapping, all differenced data are processed to construct an unsigned volatility intensity index, the mathematical expression of which is as follows:

[0112]

[0113] in, Let H(t) be the unsigned Shannon entropy fluctuation intensity at time t, H(t) be the Shannon entropy value at the current time, and H(t-1) be the Shannon entropy value at the previous time.

[0114] The calculated absolute values ​​at each time point are rearranged according to the time series to generate an unsigned Shannon entropy fluctuation intensity index series.

[0115] S3.4: Perform trend smoothing filtering based on the unsigned Shannon entropy fluctuation intensity index sequence to remove high-frequency noise interference and generate a mass state entropy change rate that can accurately reflect the transition trend of the system from steady state to instability.

[0116] like Figure 2 As shown, in step S4: the quality state entropy change rate is input into the dynamic mapping relationship rule for threshold comparison and judgment. If the quality state entropy change rate suddenly increases and continues to exceed the baseline bandwidth, an instruction reordering trigger signal is generated; otherwise, a signal to maintain the original execution order is generated.

[0117] The construction of the dynamic mapping relationship rule includes: based on the historical entropy change rate of the quality state, extracting the baseline bandwidth parameter and dynamic threshold baseline using the sliding window statistical method, and constructing a dynamic mapping relationship rule that includes a steady-state interval, a transition interval and an unstable interval. The dynamic mapping relationship rule also defines the mapping relationship between the steady-state interval, the transition interval and the unstable interval and the low urgency level, the medium urgency level and the high urgency level, respectively.

[0118] The system receives the historical entropy change rate of continuously outputting data and stores it in the buffer in chronological order to form time-series data of the entropy change rate of the quality state, which serves as the input data source for sliding window statistics.

[0119] Construct a sliding statistical window of fixed length to cover the entropy change rate values ​​corresponding to the most recent consecutive time windows in the time series data of the quality state entropy change rate, and generate sliding statistical window configuration parameters.

[0120] Based on the sliding statistical window configuration parameters, the mean and standard deviation are calculated for the entropy change rate data within the sliding window, and the mean and standard deviation are determined using the following formulas:

[0121]

[0122]

[0123] in, Let be the mean of the rate of entropy change within the sliding window. The number of samples within the sliding window. Let be the rate of change of the entropy of the mass state in the i-th time window. The standard deviation of the rate of entropy change within the sliding window.

[0124] The mean and standard deviation of the entropy change rate within the sliding window constitute the dynamic threshold baseline parameters. Based on the mean and standard deviation of the entropy change rate within the sliding window in the dynamic threshold baseline parameters, a benchmark bandwidth parameter is set to divide the range of entropy change rate variation into three levels: steady-state interval, transition interval, and instability interval, generating interval definition parameters for the steady-state interval, transition interval, and instability interval. The steady-state interval is defined as the range of the mean entropy change rate within the sliding window plus or minus one time the standard deviation of the entropy change rate within the sliding window, i.e. This indicates that the system is operating smoothly; the transition interval is defined as the range of the mean of the entropy change rate within the sliding window plus or minus one to two times the standard deviation of the entropy change rate within the sliding window, i.e. This indicates a slight disturbance in the system; the instability interval is defined as the range exceeding twice the standard deviation of the entropy change rate within the sliding window, i.e. This indicates a significant anomaly in the system.

[0125] Based on the defined interval parameters, the steady-state interval, the transition interval, and the instability interval are mapped to low urgency, medium urgency, and high urgency levels, respectively. The reference bandwidth parameter is used as the trigger threshold to generate the dynamic mapping relationship rule. This rule possesses adaptive boundary characteristics, capable of updating the mean and standard deviation of the entropy change rate within the sliding window in real time as the sliding window progresses, thereby dynamically adjusting the boundary thresholds of each region to adapt to long-term process drift.

[0126] Step S4 specifically includes:

[0127] S4.1: Input the current real-time calculated mass state entropy change rate into the constructed dynamic mapping relationship rule between mass state entropy change rate and instruction response urgency, perform multi-interval membership degree matching operation, determine the urgency level interval to which the current mass state entropy change rate belongs, and output the current mass state entropy change rate matching result with urgency level label.

[0128] The system receives the current moment's quality state entropy change rate value output from the preceding steps. This value characterizes the instantaneous rate of uncertainty evolution in the cable protection pipe production system.

[0129] Receive the mass state entropy change rate calculated in real time at the current moment from the output of step S3, as well as the interval definition parameters of the steady-state interval, transition interval, and unstable interval in the dynamic mapping relationship rule.

[0130] The input real-time quality state entropy change rate is compared with the boundary threshold of each interval in the dynamic mapping relationship rule to determine the relative position distribution of the value on the number axis, and the comparison results of the quality state entropy change rate and each interval boundary are generated.

[0131] Based on the fuzzy logic membership function, the membership values ​​of the current entropy change rate with respect to the steady-state interval, transition interval, and unstable interval are calculated according to the comparison results, thereby generating the membership values ​​of the steady-state interval, transition interval, and unstable interval.

[0132] Using the maximum membership degree principle, the largest membership degree value is selected from the steady-state interval membership degree value, the transition interval membership degree value, and the unstable interval membership degree value. The interval identifier corresponding to this value is determined as the dominant urgency level of the current quality state, and a dominant urgency level determination result is generated.

[0133] Based on the determination of the dominant urgency level, if the maximum membership degree corresponds to the steady-state interval, it is marked as low urgency; if it corresponds to the transition interval, it is marked as medium urgency; and if it corresponds to the unstable interval, it is marked as high urgency, thus generating an urgency level label.

[0134] The current timestamp, the quality state entropy change rate, and the urgency level label are encapsulated in a structured manner to generate a current quality state entropy change rate matching result with an urgency level label.

[0135] S4.2: Based on the current quality state entropy change rate matching result with urgency level label, call the duration counter to accumulate the duration of the sudden increase in quality state entropy change rate, and logically compare the accumulated duration value with the preset minimum duration bandwidth threshold to output the valid sudden increase judgment flag.

[0136] Receive the current quality state entropy change rate matching result with an urgency level label, initialize the duration counter, and set a zeroing threshold. When the matching result indicates that the entropy change rate is in the transition or unstable range, start the timing accumulation logic; if the matching result returns to the steady-state range, immediately reset the counter to zero.

[0137] The system clock pulse signal is acquired, and the duration of the non-steady-state region is discretized and accumulated with a preset sampling period as the step size. After each sampling period, the counter value is increased by one unit time slice, forming cumulative time series data reflecting the persistence of the abnormal state.

[0138] The system reads the pre-stored minimum sustained bandwidth threshold parameter, which is calibrated based on the combined thermal inertia time and mechanical response delay of the cable protection pipe extrusion process. It then compares the real-time accumulated duration value with the minimum sustained bandwidth threshold to determine whether the sudden increase possesses physically sustainable characteristics.

[0139] If the cumulative duration is greater than or equal to the minimum sustained bandwidth threshold, the current entropy rate surge is determined to be a valid process fluctuation without noise interference, and a valid surge determination flag with a logic high level is generated. If the cumulative duration is less than the threshold, it is determined to be instantaneous noise or a minor disturbance, and the flag is kept at a logic low level to suppress false triggering.

[0140] S4.3: Execute branch decision processing based on the logical state of the valid burst judgment flag. If the valid burst judgment flag indicates a sudden increase in the quality state entropy change rate and continues to exceed the baseline bandwidth, activate the instruction reordering trigger signal generation logic. Otherwise, activate the original order execution signal generation logic and output the initial control scheduling decision signal to be issued.

[0141] The effective surge determination flag bit output in step S4.2 is received. The effective surge determination flag bit is a Boolean logic variable used to characterize whether the current quality state entropy change rate meets the triggering condition of a surge and persistence, and serves as the input data source for branch decision processing.

[0142] Read the logic level state of the valid burst judgment flag, execute the conditional branch judgment logic, and generate the branch decision result.

[0143] If the branch decision result indicates that the valid burst judgment flag is true, then the instruction reordering trigger path is entered, an instruction reordering trigger signal object is instantiated, the signal type is set to high priority interrupt request, the trigger source is marked as quality state entropy change rate exceeding limit event, and an instruction reordering trigger signal is generated.

[0144] If the branch decision result indicates that the valid burst judgment flag is false, then the original execution path is entered, the original execution signal object is instantiated, the signal type is set to normal polling response, the execution strategy is marked as queue first-in-first-out principle, and the original execution signal is generated.

[0145] The instruction reordering trigger signal or the original order execution signal is encapsulated into a standard control and scheduling decision data packet, and the current timestamp, signal type identifier and source data check code are appended to generate the initial control and scheduling decision signal to be issued.

[0146] S4.4: Perform pulse shaping and level standardization processing on the initial control scheduling decision signal to be issued to eliminate signal jitter interference, generate the final determined instruction reordering trigger signal or maintain the original execution order signal, and transmit it to the subsequent instruction priority promotion operation module to drive the dynamic adjustment of the control instruction queue.

[0147] like Figure 3 As shown, step S5: In response to the instruction reordering trigger signal, identify the dominant quality characteristic factor with the highest correlation to the quality state entropy change rate, and increase the priority of the control channel instruction corresponding to the dominant quality characteristic factor in the pending instruction queue, generating a reordered control instruction execution sequence. Specifically, this includes:

[0148] S5.1: Based on the instruction reordering trigger signal, obtain the multidimensional quality state vector sequence and the corresponding quality state entropy change rate within the current time window, and use the Pearson correlation coefficient calculation formula to calculate the linear correlation degree between each component in the multidimensional quality state vector sequence and the quality state entropy change rate, and generate a correlation degree feature vector containing the correlation strength values ​​of the wall thickness standard deviation, concentricity offset, surface roughness mean, and material melt index fluctuation.

[0149] Upon receiving the command reordering trigger signal, the system retrieves the multidimensional mass state vector sequence within the current sliding time window and the synchronously calculated scalar value of the mass state entropy rate of change from the data buffer. The multidimensional mass state vector sequence is deconstructed, separating four independent one-dimensional time-series components: wall thickness standard deviation, concentricity offset, mean surface roughness, and material melt index fluctuation. For each mass feature component, the system extracts the corresponding set of data points within the same time window as the mass state entropy rate of change, ensuring strict alignment between the sample size and timestamps. The Pearson correlation coefficient formula is used to calculate the linear correlation coefficient between the time-series data of each mass feature component and the mass state entropy rate of change. The Pearson correlation coefficient formula is as follows:

[0150]

[0151] in, The Pearson correlation coefficient is used. Let i be a specific quality feature value of the i-th sampling point. Let i be the rate of change of the entropy of the mass state corresponding to the i-th sampling point. This represents the mean of the quality characteristic within the window. Let be the mean of the rate of change of entropy of the mass state within the window. This represents the total number of sampling points within the window.

[0152] The Pearson correlation coefficients, i.e., the linear correlation coefficients, are calculated for each of the four components: standard deviation of wall thickness, concentricity offset, mean surface roughness, and material melt index fluctuation. The four calculated Pearson correlation coefficients are then arranged in characteristic order and encapsulated to form a correlation feature vector.

[0153] S5.2: Based on the correlation feature vector, identify the correlation strength value with the largest value, and define the quality feature component corresponding to the largest correlation strength value as the dominant quality feature factor. At the same time, extract the control channel identifier bound to the dominant quality feature factor in the process control mapping table and generate the target control channel index.

[0154] The correlation feature vector is received, which includes the linear correlation strength values ​​between four components—wall thickness standard deviation, concentricity offset, surface roughness mean, and material melt index fluctuation—and the rate of change of entropy of mass state.

[0155] Perform a traversal scan operation on the correlation feature vector, initialize the maximum value variable to negative infinity, and set the dominant feature index to null, in preparation for extreme value search.

[0156] Read each association strength value in the vector sequentially and compare it with the maximum value variable of the current record. If the current value is greater than the maximum value variable, update the maximum value variable to the current value and update the dominant feature index to the position identifier of the current component.

[0157] After traversing all components, the final maximum value variable and its corresponding dominant feature index are locked. This index uniquely points to the quality feature component that contributes the most to the change in system uncertainty, and it is defined as the dominant quality feature factor.

[0158] The pre-stored process control mapping table is invoked. This table establishes a static binding relationship between quality characteristics and control channels, using the name or type of the dominant quality characteristic factor as the query key.

[0159] Retrieve records in the process control mapping table that perfectly match the dominant quality characteristic factor, and extract the control channel identifier bound to the record. This identifier corresponds to a specific actuator interface address or logical channel number.

[0160] The extracted control channel identifier is encapsulated into a target control channel index, which serves as the sole addressing basis for subsequent instruction filtering and priority adjustment.

[0161] S5.3: Based on the target control channel index, scan all the original control commands to be sent to the actuator in the command queue, filter out the target commands whose control channel identifiers completely match the target control channel index, and mark the remaining mismatched control commands as regular commands to be sent, and generate a command classification set.

[0162] The system receives the target control channel index and the pending command queue data. The pending command queue contains several raw control command objects to be issued to the extruder, traction machine, or shaping box actuator. It iterates through each command object in the pending command queue, extracting its control channel identifier field, which uniquely corresponds to a specific process control loop. The extracted control channel identifier is compared with the target control channel index using a string-based exact match to determine if they are completely identical. If the match is true, the current command object is marked as a target pending command and logically removed from its original queue position, stored in a temporarily constructed high-priority command buffer. If the match is false, the current command object is marked as a regular pending command, retaining its original timestamp sequence attribute, and stored in a regular command buffer. This iteration continues until all command objects in the pending command queue have been traversed, ensuring no omissions. The command objects in the high-priority command buffer and the regular command buffer are structurally encapsulated to generate a command category set containing subsets of target pending commands and regular pending commands.

[0163] S5.4: Based on the instruction classification set, adjust the execution priority weight parameter of the target instruction to be issued to the highest priority level, while keeping the original priority weight parameter of the regular instruction to be issued unchanged, and generate a hybrid instruction set with a new priority label.

[0164] The instruction classification set output from step S5.3 is received. The instruction classification set includes target pending instructions and regular pending instructions, which serve as the input data source for priority weight adjustment.

[0165] Extract the control channel identifier and raw execution parameters of the target instruction to be sent.

[0166] Instantiate the priority weight register in memory and initialize the default weight value to the standard level value.

[0167] The priority weight field corresponding to the target instruction to be dispatched is located at the specified address in the register. The highest priority level value is written to overwrite the original standard level value, establishing the instruction's first response right in the scheduling queue.

[0168] Iterate through the regular pending instructions and read the current priority weight parameter of each instruction. Compare the weight of the regular instruction with the preset standard level threshold to confirm that it is not in a locked or promoted state, and generate an unlocked status confirmation flag for the regular pending instructions.

[0169] Based on the confirmation result of the regular pending instructions, the original priority weight parameters of the regular pending instructions are kept unchanged, their relative temporal position in the queue is maintained, and a set of regular pending instructions after maintenance is generated.

[0170] The target pending instructions in the updated priority weight register are combined with the set of regular pending instructions after being retained and encapsulated to construct a hybrid instruction set data structure with new priority labels, thereby generating a hybrid instruction set with new priority labels.

[0171] S5.5: Based on the hybrid instruction set with the new priority label, perform a sequence reorganization operation arranged in descending priority order, move the adjusted target pending instructions to the top of the queue, then arrange the regular pending instructions in sequence, and output the reordered control instruction execution sequence after the timing logic is updated.

[0172] The hybrid instruction set with new priority labels is received from the output of step S5.4. The hybrid instruction set with new priority labels includes target pending instructions marked as the highest priority and regular pending instructions that retain their original weights.

[0173] Extract the priority weight parameters and timestamp information of each instruction in the mixed instruction set, and construct a list of instruction tuples to be sorted.

[0174] The instruction tuple list is sorted in descending order using a quicksort algorithm, with priority weight as the primary sorting key and instruction generation timestamp as the secondary sorting key. When two instructions are found to have the same priority weight, their timestamp values ​​are compared, and the instruction with the earlier timestamp is placed first, ensuring that instructions of the same priority follow the first-in-first-out principle, thus generating a descending order list of instruction tuples.

[0175] Traverse the sorted instruction list, read the instruction content, control channel identifier and execution parameters in sequence, encapsulate them into ordered control instruction objects, and generate an ordered sequence of control instruction objects.

[0176] The ordered control instruction objects are stored sequentially into a first-in-first-out queue structure, forming a reordered control instruction execution sequence that strictly follows the descending priority order.

[0177] Step S6: Apply an instruction execution constraint buffer mechanism to the reordered control instruction execution sequence. Only instructions yet to be issued to the actuators are scheduled according to the reordered control instruction execution sequence, while instructions already issued to the actuators are executed in their original order, generating an anti-oscillation control instruction flow. Specifically, this includes:

[0178] S6.1: Obtain the reordered control instruction execution sequence and the real-time instruction status monitoring list. The real-time instruction status monitoring list contains the current transmission stage identifier of each control channel instruction. Based on the current transmission stage identifier, perform execution status discrimination processing on each instruction to be processed in the reordered control instruction execution sequence to generate an instruction classification set with execution attribute tags. The execution attribute tags clearly distinguish between in-transit instructions that have been sent to the executor and pending instructions that have not yet been sent to the executor.

[0179] The system receives the reordered control instruction execution sequence generated in step S5. This sequence contains instructions to be sent with adjusted priorities and their corresponding control channel identifiers. Simultaneously, it reads the real-time instruction status monitoring list, which is periodically updated by the underlying communication driver layer. This list records the current transmission stage identifier of each control channel instruction, including states such as idle, packaging, sending, executing, and completed. The idle, packaging, and sending states are categorized as 'not yet sent,' while the executing and completed states are categorized as 'sent.'

[0180] Iterate through each instruction in the reordered control instruction execution sequence and extract its bound unique control channel identifier. Use this identifier to search the real-time instruction status monitoring list and retrieve the current transmission stage identifier value for the corresponding channel.

[0181] Based on the preset state determination logic, the acquired transmission stage identifiers are binarized and classified. If the transmission stage identifier indicates that the instruction has entered the physical executor buffer or is being parsed and executed by the hardware, the instruction is marked as an "in-transit instruction" and given the uninterruptible attribute; if the transmission stage identifier indicates that the instruction is still residing in the software scheduling queue and has not entered the transmission buffer, the instruction is marked as a "pending instruction" and given the reschedulable attribute.

[0182] Instructions with execution attribute tags are encapsulated along with their original control parameters, timestamps, and priority weights to construct a set of instruction categories with execution attribute tags. This set clearly distinguishes between the subset of instructions in transit that need to maintain atomic execution and the subset of instructions to be executed that can be dynamically adjusted.

[0183] S6.2: Extract the subset of instructions marked as in-transit instructions from the instruction classification set, perform lock-and-hold processing on the subset of in-transit instructions based on the principle of instruction atomicity execution, generate the original order execution instruction stream, ensure that the control channel of the initiated physical action is not disturbed by subsequent reordering logic, and maintain the continuous adjustment process of the current process parameters until the current round of action is completed.

[0184] Receive the instruction classification set output in step S6.1, and extract the instruction subset marked as in-transit instructions. The in-transit instruction subset includes control channel instructions that have been issued to the actuator but have not yet completed physical actions and their current transmission stage identifiers, which serve as the input data source for lock-and-hold processing.

[0185] For each instruction in the in-transit instruction subset, perform an atomic state locking operation, read the current value of the control channel register corresponding to the instruction, mark it as an uninterruptible execution state, block overwrite requests from reordering logic, and generate an atomically locked in-transit instruction set.

[0186] Based on the principle of atomic execution of instructions, a buffer for original-order execution instruction stream is constructed. Instructions in transit that are locked are stored in the buffer according to their original timestamps, maintaining the temporal dependencies between instructions and generating an original-order execution instruction stream.

[0187] The system monitors the feedback signals of the executors corresponding to each in-transit instruction. When the physical action completion flag is detected or the preset maximum execution time limit is reached, the atomicity lock state of the instruction is released, and the instruction is removed from the original execution instruction stream to generate an updated original execution instruction stream.

[0188] S6.3: Extract the subset of instructions marked as pending instructions from the instruction classification set, and perform queue reorganization mapping processing on the subset of pending instructions according to the priority order defined in the reordered control instruction execution sequence to generate a high-priority front-end queue of pending instructions, thereby adjusting the control channel instructions associated with the dominant quality characteristic factor to the top of the queue.

[0189] The subset of instructions marked as pending instructions in the received instruction classification set includes pending instructions corresponding to the target control channel and pending instructions for the regular control channel.

[0190] For each instruction in the subset of instructions to be issued, its associated quality feature priority weight parameters are parsed to construct a list of metadata for instructions to be sorted.

[0191] The priority-based quicksort algorithm is used to sort the list of instruction metadata in descending order to ensure that high-priority instructions are at the front of the queue.

[0192] During the sorting process, if multiple instructions are detected to have the same highest priority weight, their original relative positions are maintained according to the order of their timestamps, following the first-in-first-out principle to avoid logical competition between instructions of the same priority.

[0193] The sorted instruction sequence is mapped into the address space of the physical execution queue, generating a queue of instructions to be scheduled with high priority.

[0194] S6.4: Receive the original execution instruction stream and the high-priority pre-scheduled instruction queue, perform time-slice synchronization fusion processing to splice the execution timing of the two and resolve conflicts, and generate a mixed control instruction sequence containing complete execution timing information to ensure seamless connection of new priority-optimized instructions after the completion of in-transit instructions, and avoid time breakpoints in control signals.

[0195] The original execution instruction stream output from step S6.2 and the high-priority pre-scheduled instruction queue output from step S6.3 are received as input data sources for timing splicing and conflict resolution.

[0196] Extract the estimated completion timestamps of the instructions in transit from the original execution instruction stream, and the earliest executable timestamps of the instructions to be dispatched from the high-priority pre-scheduled instruction queue, to generate a parameter set of completion times for instructions in transit and executable times for instructions to be dispatched.

[0197] Based on the sampling period of the cable protection pipe extrusion process, a discrete time slice sequence is constructed, and the control period is divided into several equal-length time slots to generate the discrete time slice sequence.

[0198] Calculate the time difference between the remaining execution time of the in-transit instruction and the next control cycle in the parameter set of the in-transit instruction completion time and the executable time of the pending instruction, determine the safety buffer window for instruction switching, and generate the safety buffer window parameters.

[0199] If the safety buffer window is greater than zero, then the window is marked as an instruction handover area on the timeline to accommodate the parameter warm-up phase of new instructions, and an instruction handover area marker is generated.

[0200] The first high-priority instruction in the queue of instructions to be scheduled is processed for parameter smoothing to generate a transition trajectory from the current executor state to the target set value.

[0201] The intermediate parameter values ​​of the transition trajectory in each time slice are calculated using linear interpolation to ensure that the rate of change of physical quantities is within the allowable range of the actuator, and the interpolation parameter values ​​for each time slice are generated.

[0202] The interpolation parameter values ​​of each time slice are mapped to the time slice corresponding to the instruction handover area to generate a seamless parameter update sequence.

[0203] The regular pending instructions in the high-priority front-end scheduling instruction queue, excluding the first high-priority instruction, are delayed and rearranged, and their start execution time is shifted to the first available time slice after the transition trajectory ends, generating a delayed and rearranged set of regular instructions.

[0204] Check the overlap between the parameter update sequence and the delayed rearranged regular instruction set on the time axis, identify whether there are multiple assignment conflicts in the same control channel within the same time slice, and generate conflict detection results.

[0205] If the conflict detection result indicates that a conflict exists, the parameter values ​​of the high-priority instructions are retained according to the priority label, redundant fragments of the low-priority instructions are removed, and a complete mixed control instruction sequence after conflict resolution is generated.

[0206] S6.5: Perform smoothness verification and format encapsulation processing on the hybrid control instruction sequence, convert the verified sequence into the final executable data packet based on the preset communication protocol standard, generate an anti-oscillation control instruction stream and send it to the underlying execution unit, and complete the safe transition from logical reordering to physical execution.

[0207] Step S7: Based on the execution results of the generated anti-vibration control command stream, synchronously update the historical entropy baseline data of the corresponding dominant quality characteristic factors to form closed-loop adaptive learning parameters, which will be used for benchmark comparison and dynamic evaluation of the quality state entropy change rate in the next round. Specifically, this includes:

[0208] S7.1: Obtain multi-source quality feature time-series data and corresponding control channel response records after the execution of the anti-vibration control command stream. Use timestamp alignment processing to perform spatiotemporal correlation matching between the multi-source quality feature time-series data and the control channel response records to generate a causal correlation dataset with control effect labels, so as to clarify the specific impact trajectory of this control action on key indicators such as wall thickness standard deviation and concentricity offset.

[0209] The system receives multi-source quality characteristic time-series data and corresponding control channel response records after the anti-vibration control command stream is executed. The multi-source quality characteristic time-series data includes real-time sampled values ​​of key indicators such as wall thickness standard deviation and concentricity offset. The control channel response records include actuator action parameters such as mold temperature adjustment and traction speed compensation values.

[0210] A unified timestamp index is extracted from the multi-source quality characteristic time series data and the control channel response record. Based on the linear interpolation method, the discrete sampling points with small time deviations are resampled to eliminate the time axis misalignment caused by the inconsistency between the sensor acquisition frequency and the controller execution frequency, and generate a synchronized data sequence with strict time alignment characteristics.

[0211] A sliding time window is constructed, starting from the moment the control command is issued, to capture the synchronous data sequence within a preset duration, and the time dimension is mapped to the spatial dimension to form a multidimensional spatiotemporal data matrix containing input variables (control actions) and output variables (quality response).

[0212] The nonlinear dependency between the control channel response record and the time series data of each quality characteristic is calculated using the mutual information method, and the strength of the causal association is quantified by the following formula:

[0213]

[0214] in, To control the channel response variables, For quality characteristic response variables, For joint probability distribution, and These are the marginal probability distributions.

[0215] Based on the mutual information calculation results, feature pairs that are significantly higher than the noise threshold are selected and marked as valid causal association pairs. False correlation data caused by random fluctuations are removed, and a causal association dataset with control effect labels is generated.

[0216] S7.2: Based on the causal association dataset, extract the evolution sequence of the dominant quality characteristic factor within the control window, use the sliding window statistical method to calculate the local fluctuation variance and trend slope of the evolution sequence, and generate a quality recovery efficiency index that characterizes the effectiveness of control, so as to quantify the degree of suppression of system uncertainty by the current control strategy.

[0217] We receive a causal correlation dataset labeled with control effects and extract the temporal evolution sequence of the dominant quality characteristic factors within the control window. This sequence records the trajectory of key quality indicators from the moment the anti-oscillation control command flow is issued until the end of the system response stability interval.

[0218] A sliding window statistical mechanism is applied to the extracted evolutionary sequences, with a window length covering at least three sampling periods to capture local dynamic characteristics. Within each sliding window, the dispersion of data points relative to the window mean is calculated to obtain the local fluctuation variance.

[0219] The local variance is calculated using the following formula:

[0220]

[0221] in, The number of samples within the window. Let i be the value of the i-th sampling point. This is the arithmetic mean of the data within the window.

[0222] Based on data within the same sliding window, a linear trend line is fitted using the least squares method to calculate the trend slope of the evolutionary sequence. This slope characterizes the convergence or divergence rate of the quality characteristics under controlled intervention.

[0223] The trend slope is calculated using the following formula:

[0224]

[0225] in, The timestamp corresponding to the i-th sampling point. The slope of the fitted line.

[0226] The calculated local volatility variance and the absolute value of the trend slope are weighted and fused to construct a quality recovery effectiveness index. The weighting coefficients are dynamically adjusted based on the backpropagation error of historical control effects to strengthen the positive incentive for fast convergence and low volatility states.

[0227] S7.3: The quality recovery efficiency index is used to perform confidence-weighted correction on the pre-stored historical entropy baseline data. The current quality state Shannon entropy value and the quality recovery efficiency index are fused through exponential weighted moving average processing to generate an updated dynamic entropy baseline reference value, thereby eliminating the defect that the static baseline cannot adapt to long-term process drift.

[0228] The quality recovery performance index output from step S7.2 and the Shannon entropy value of the current quality state output from step S2.4 are received as the input data source for dynamic baseline correction.

[0229] A baseline update model is constructed based on exponentially weighted moving average processing. A forgetting factor is set to balance the stability of historical data with the sensitivity of the current state, thus generating the baseline update model.

[0230] The quality recovery efficiency index is mapped to a confidence weight coefficient, which reflects the reliability of the current control action in suppressing system uncertainty, thus generating the confidence weight coefficient.

[0231] The updated dynamic entropy baseline reference value is calculated using the following formula:

[0232]

[0233] in, The dynamic entropy baseline reference value updated at time t. An adaptive forgetting factor determined based on the quality recovery efficacy index. The current mass state is represented by the Shannon entropy value. This is the historical entropy baseline data from the previous moment.

[0234] Based on the numerical range of the quality recovery performance index, the size of the adaptive forgetting factor is dynamically adjusted. When the performance index shows that the control is effective, the adaptive forgetting factor is increased to quickly respond to process changes; conversely, the adaptive forgetting factor is decreased to maintain baseline stability, thus generating the adjusted adaptive forgetting factor.

[0235] The current quality state Shannon entropy value and the historical entropy baseline data are weighted and fused according to the adjusted adaptive forgetting factor to eliminate baseline drift noise caused by data fluctuations at a single time point, and generate a noise-reduced dynamic entropy baseline reference value.

[0236] The noise-reduced dynamic entropy baseline reference value is used as the updated dynamic entropy baseline reference value, which includes the latest process state characteristics, to replace the original static preset threshold.

[0237] S7.4: Based on the updated dynamic entropy baseline reference value, reconstruct the benchmark bandwidth threshold of the quality state entropy change rate, perform an adaptive boundary expansion operation to generate a new version of the benchmark bandwidth range containing a safety margin, and write the new version of the benchmark bandwidth range into the instruction reordering trigger criterion storage unit of the process control module to complete the instantiation update of the closed-loop adaptive learning parameters.

[0238] The updated dynamic entropy baseline reference value output from step S7.3 is received. This value represents the latest benchmark level of quality system stability under current process conditions. Based on the dynamic entropy baseline reference value, a historically preset safety margin coefficient is extracted. This safety margin coefficient is used to accommodate sensor noise and minor process fluctuations, preventing frequent instruction reordering due to oversensitivity.

[0239] A linear expansion method is used to weight the dynamic entropy baseline reference value with the safety margin coefficient to construct a new reference bandwidth range including upper and lower limits. Specifically, the upper and lower thresholds of the new reference bandwidth are calculated using the following formulas:

[0240]

[0241] in, The upper limit threshold, This is the updated dynamic entropy baseline reference value. This is the safety margin coefficient.

[0242]

[0243] in, The lower threshold is... This is the lower limit safety margin coefficient, which is usually taken to be slightly less than α to allow for a normal drop in entropy.

[0244] The calculated upper and lower thresholds are combined to generate a closed new reference bandwidth range data object. An integrity check is performed on this data object to ensure that the upper threshold is greater than the lower threshold and that both are non-negative real numbers, thus conforming to the physical constraint of non-negativity of entropy values.

[0245] The non-volatile storage interface of the process control module is invoked to write the verified new version of the baseline bandwidth range data into the instruction reordering trigger criterion storage unit. This overwrites the original static threshold parameters, completing the instantiation update of the closed-loop adaptive learning parameters, enabling the entropy rate comparison in the next round to have dynamic adaptability.

[0246] S7.5: Call the updated closed-loop adaptive learning parameters to initialize the next round of multidimensional quality state vector sequence monitoring process, and use the new version of the benchmark bandwidth range as the dynamic comparison benchmark for calculating the quality state entropy change rate, so as to realize the logical closed loop from single control execution to long-term strategy optimization, and ensure the accuracy and timeliness of subsequent instruction priority recalibration of core driving parameters.

[0247] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0248] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The “multiple” mentioned in the embodiments of this application refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0249] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should all be covered 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 method for online quality control in the production of cable protection pipes, characterized in that, include: S1: Acquire multi-source quality characteristic time-series data during the continuous production process of cable protection pipes, and perform vector encapsulation on the multi-source quality characteristic time-series data based on a sliding time window to generate a multi-dimensional quality state vector sequence; S2: Perform normalization mapping on the multidimensional mass state vector sequence, calculate the probability density histogram based on the normalized data distribution, and then perform uncertainty quantization operation on the probability density histogram to generate the current mass state Shannon entropy value. S3: Perform first-order difference absolute value operation on the Shannon entropy value of the current mass state corresponding to multiple time windows collected continuously, and define the operation result as the mass state entropy change rate; S4: Input the quality state entropy change rate into the dynamic mapping relationship rule for threshold comparison and judgment. If the quality state entropy change rate suddenly increases and continues to exceed the baseline bandwidth, generate an instruction reordering trigger signal; otherwise, generate a signal to maintain the original execution order. S5: In response to the instruction reordering trigger signal, identify the dominant quality characteristic factor with the highest correlation to the quality state entropy change rate, and increase the priority of the control channel instruction corresponding to the dominant quality characteristic factor in the pending instruction queue, and generate a reordered control instruction execution sequence. S6: Apply an instruction execution constraint buffer mechanism to the reordered control instruction execution sequence, and only schedule pending instructions that have not yet been issued to the actuator according to the reordered control instruction execution sequence, while keeping the instructions that have been issued to the actuator executed in the original order, thereby generating an anti-oscillation control instruction stream.

2. The online quality control method for cable protection pipe production according to claim 1, characterized in that, Also includes: S7: Based on the execution results of the anti-vibration control command stream generated this time, the historical entropy baseline data corresponding to the dominant quality characteristic factor is updated synchronously to form closed-loop adaptive learning parameters.

3. The online quality control method for cable protection pipe production according to claim 1, characterized in that, S3 specifically includes: Obtain the current quality state Shannon entropy values ​​corresponding to multiple consecutive time windows arranged in chronological order, and perform time-series alignment processing on the current quality state Shannon entropy values ​​based on the sliding window mechanism to generate a standardized Shannon entropy value time-series data stream; Based on the values ​​of two adjacent time windows in the time-series data stream of the standardized Shannon entropy, a difference operation is performed to calculate the original Shannon entropy difference value set; For each original Shannon entropy difference value in the original Shannon entropy difference value set, perform absolute value mapping processing to extract the unsigned Shannon entropy fluctuation intensity index sequence; Trend smoothing filtering is performed on the unsigned Shannon entropy fluctuation intensity index sequence to generate the mass state entropy change rate.

4. The online quality control method for cable protection pipe production according to claim 1, characterized in that, S4 specifically includes: The current real-time calculated rate of change of the quality state entropy is input into the constructed dynamic mapping relationship rule, and a multi-interval membership degree matching operation is performed to determine the urgency level interval to which the current rate of change of the quality state entropy belongs, and the matching result of the current rate of change of the quality state entropy with urgency level label is output. Based on the matching results of the current quality state entropy change rate with urgency level labels, the duration of the sudden increase in quality state entropy change rate is accumulated and statistically analyzed. The accumulated duration value is logically compared with the preset minimum duration bandwidth threshold, and a valid sudden increase judgment flag is output. According to the logical state of the effective surge determination flag, branch decision processing is performed. If the effective surge determination flag indicates that the quality state entropy change rate has surged and continues to exceed the reference bandwidth, the instruction reordering trigger signal generation logic is activated. Otherwise, the original order execution signal generation logic is activated, and the initial control scheduling decision signal to be issued is output. The initial control scheduling decision signal to be issued is subjected to pulse shaping and level standardization processing to generate an instruction reordering trigger signal or a signal to maintain the original execution order.

5. The online quality control method for cable protection pipe production according to claim 4, characterized in that, The construction of the dynamic mapping relationship rules includes: based on the historical entropy change rate of the quality state, extracting the baseline bandwidth parameter and dynamic threshold baseline using the sliding window statistical method, and constructing dynamic mapping relationship rules that include steady-state interval, transition interval and unstable interval.

6. The online quality control method for cable protection pipe production according to claim 5, characterized in that, Based on the historical entropy change rate of the quality state, a reference bandwidth parameter and a dynamic threshold baseline are extracted using a sliding window statistical method. A dynamic mapping relationship rule is then constructed, encompassing the steady-state interval, transition interval, and instability interval, including: Receive the historical quality state entropy change rate of continuous output, store it in the buffer in time order to form time series data of quality state entropy change rate; Construct a sliding statistical window of fixed length to cover the entropy change rate values ​​corresponding to the most recent consecutive time windows in the time series data of the quality state entropy change rate, and generate sliding statistical window configuration parameters; Based on the sliding statistical window configuration parameters, the mean and standard deviation of the entropy change rate data within the sliding window are calculated to obtain the mean and standard deviation of the entropy change rate within the sliding window, which constitute the state threshold baseline parameters. Based on the mean and standard deviation of the entropy change rate within the sliding window in the dynamic threshold baseline parameters, a benchmark bandwidth parameter is set, and the range of entropy change rate variation is divided into the steady-state interval, the transition interval, and the unstable interval, generating interval definition parameters for the steady-state interval, the transition interval, and the unstable interval. Based on the interval definition parameters, the steady-state interval, the transition interval, and the unstable interval are mapped to low urgency, medium urgency, and high urgency levels, respectively, and the reference bandwidth parameter is used as the trigger threshold to generate the dynamic mapping relationship rules.

7. The online quality control method for cable protection pipe production according to claim 5, characterized in that, The steady-state interval is defined as the range of the mean entropy change rate within the sliding window plus or minus one standard deviation of the entropy change rate within the sliding window; the transition interval is defined as the range of the mean entropy change rate within the sliding window plus or minus one to two standard deviations of the entropy change rate within the sliding window; the instability interval is defined as the range exceeding two standard deviations of the entropy change rate within the sliding window.

8. The online quality control method for cable protection pipe production according to claim 1, characterized in that, The multi-source quality characteristic time series data includes wall thickness standard deviation, concentricity offset, mean surface roughness, and material melt index fluctuation.

9. The online quality control method for cable protection pipe production according to claim 8, characterized in that, S5 specifically includes: Based on the instruction reordering trigger signal, the multidimensional quality state vector sequence and the corresponding quality state entropy change rate within the current time window are obtained. The linear correlation degree between each component in the multidimensional quality state vector sequence and the quality state entropy change rate is calculated to generate a correlation degree feature vector containing the correlation strength values ​​of the wall thickness standard deviation, concentricity offset, surface roughness mean and material melt index fluctuation. Based on the correlation feature vector, the correlation strength value with the largest value is identified, and the quality feature component corresponding to the largest correlation strength value is defined as the dominant quality feature factor. At the same time, the control channel identifier bound to the dominant quality feature factor in the process control mapping table is extracted to generate the target control channel index. Based on the target control channel index, scan all the original control commands to be sent to the actuator in the command queue, filter out the target commands to be sent whose control channel identifiers completely match the target control channel index, and mark the remaining mismatched control commands as regular commands to be sent, thus generating a command classification set. Based on the instruction classification set, the execution priority weight parameter of the target instruction to be issued is adjusted to the highest priority level, while keeping the original priority weight parameter of the regular instruction to be issued unchanged, thus generating a hybrid instruction set with a new priority label; Based on the hybrid instruction set with the new priority label, a sequence reorganization operation is performed in descending priority order. The adjusted target pending instructions are moved to the top of the queue, and then the regular pending instructions are arranged in sequence. The reordered control instruction execution sequence after the timing logic is updated is output.

10. The online quality control method for cable protection pipe production according to claim 9, characterized in that, The linear correlation between each component in the multidimensional quality state vector sequence and the rate of change of quality state entropy is calculated, including: using the Pearson correlation coefficient calculation formula to calculate the linear correlation coefficient between the time series data of each quality feature component and the rate of change of quality state entropy.

Images