Method, system and storage medium for intelligent control of spinning frame spindle speed

Abstract

The application provides a method and system for intelligent control of spindle speed of a spinning frame and a storage medium, comprising collecting yarn tension, temperature and humidity, front roller speed and spindle speed to construct a multi-dimensional state vector, calculating tension fluctuation rate, breakage frequency and tension gradient; determining the grid granularity according to the historical path divergence factor of the last period to discretize the multi-dimensional vector to generate an initial state transition matrix; correcting the matrix according to the double threshold values of the tension fluctuation rate and the breakage frequency, solving the breakage margin factor, setting the cross weight combined with the fusion results of the temperature and humidity and the front roller speed variance, and obtaining the multi-parameter state transition matrix by weighting; comparing the state deviation before and after the update to update the divergence factor, inversely projecting to obtain the target transition matrix, correcting the step vector combined with the cross weight, and calculating the candidate spindle speed adjustment amount by the inner product of the two; if the tension gradient is over the limit, the adjustment amount is rolled back and corrected combined with the margin factor and the divergence factor, and the instruction is issued to control the operation of the spinning frame.

Description

Technical Field

[0001] This application belongs to the field of intelligent control, and in particular relates to a method, system and storage medium for intelligent control of spindle speed of a spinning frame. Background Technology

[0002] The spindle speed of a ring spinning machine directly affects the stability of spinning tension, and abnormal tension is an important cause of yarn breakage and reduced yarn quality. Therefore, the rational control of the spindle speed of a ring spinning machine has always been a key research direction in the spinning field.

[0003] Existing spinning frame spindle speed control technologies mostly rely on single-factor feedback or preset static curves. Common solutions typically use yarn tension as the core reference quantity, adjusting the spindle speed according to a preset step size when the sensor detects tension exceeding the limit; other solutions use the front roller speed as an auxiliary parameter, optimizing the spindle speed by constructing a two-dimensional matrix. However, the actual spinning process is a complex system influenced by the environment, machinery, and material properties. Workshop temperature and humidity, as well as the historical frequency of yarn breakage, all affect the current tension state. Existing methods, when constructing state transition models, usually use a uniform mesh partitioning method to discretize continuous variables, making it difficult to dynamically scale the mesh granularity according to the actual path divergence of the system, thus limiting the accuracy of state mapping.

[0004] Existing matrix-based control strategies often rely on pre-set empirical weights when fusing multiple parameters, failing to adequately consider the fluctuations in each parameter. The influence of temperature, humidity, and roller speed on tension changes with the spinning stage; static weight allocation cannot accurately represent the correlation between parameters, nor can it effectively correct the control tendencies of the state transition matrix under different operating conditions. Furthermore, during spindle speed adjustment, existing systems often exhibit unidirectional optimization output. When encountering abnormal states such as extreme tension gradients, they lack fault-tolerance and rollback mechanisms, easily leading to over-adjustment of spindle speed, exacerbating yarn tension oscillations, and failing to effectively suppress yarn breakage.

[0005] Therefore, in view of the problems of low integration of multi-dimensional environment and mechanical parameters and lack of rollback adjustment mechanism under abnormal conditions in existing spindle speed control methods, there is an urgent need to provide a spindle speed control method for ring spinning machines based on an improved multi-parameter weighted state transition matrix. Summary of the Invention

[0006] To improve the overall production efficiency and operational stability of spinning frames, the first aspect of this invention provides a method for intelligent control of spindle speed in spinning frames, comprising the following steps: A multidimensional state vector is constructed by collecting yarn tension, ambient temperature and humidity, front roller speed and current spindle speed. Tension fluctuation rate, breakage frequency and tension gradient are calculated. The grid granularity is scaled according to the historical path divergence factor of the previous cycle. The multidimensional state vector is mapped to the discrete space to obtain the initial state transition matrix. The baseline state transition matrix is ​​obtained by correcting the initial state transition matrix based on the tension fluctuation rate and the breakage frequency. The breakage margin factor is calculated, and the cross weight is calculated based on the variance fusion result of the ambient temperature and humidity and the front roller speed. The weight is applied to the baseline state transition matrix to obtain the multi-parameter weighted state transition matrix. The historical path divergence factor is calculated based on the deviation between the predicted state at the previous moment and the current multidimensional state vector. The target state transition matrix is ​​obtained by back-projecting the multi-parameter weighted state transition matrix. The step vector is adjusted using cross-weights. The candidate spindle speed adjustment amount is obtained by taking the inner product of the row vector corresponding to the current state in the target state transition matrix and the adjusted step vector. The target spindle speed is determined based on the candidate spindle speed adjustment amount. When the tension gradient exceeds the extreme value, the candidate spindle speed adjustment amount is rolled back using the breakage margin factor and the historical path divergence factor to control the spinning machine.

[0007] A second aspect of the present invention provides a system for intelligent control of spindle speed on a spinning frame, comprising the following modules: The mapping module is used to collect yarn tension, ambient temperature and humidity, front roller speed and current spindle speed to construct a multi-dimensional state vector, calculate tension fluctuation rate, breakage frequency and tension gradient, scale the grid granularity according to the historical path divergence factor of the previous cycle, and map the multi-dimensional state vector to the discrete space to obtain the initial state transition matrix. The calculation module is used to obtain the reference state transition matrix based on the dual threshold correction of the initial state transition matrix of tension fluctuation rate and head breakage frequency, calculate the head breakage margin factor, calculate the cross weight based on the variance fusion result of ambient temperature and humidity and front roller speed, and apply the weight to the reference state transition matrix to obtain the multi-parameter weighted state transition matrix. The adjustment module is used to calculate the historical path divergence factor based on the deviation between the predicted state at the previous moment and the current multi-dimensional state vector, back-project the multi-parameter weighted state transition matrix to obtain the target state transition matrix, adjust the step vector using cross weights, and take the inner product of the row vector corresponding to the current state in the target state transition matrix with the adjusted step vector to obtain the candidate spindle speed adjustment amount. The target spindle speed is determined based on the candidate spindle speed adjustment amount. When the tension gradient exceeds the extreme value, the candidate spindle speed adjustment amount is rolled back using the breakage margin factor and the historical path divergence factor to control the spinning machine to execute.

[0008] This invention comprehensively collects multi-dimensional parameters such as yarn tension, ambient temperature and humidity, front roller speed, and current spindle speed, and scales the grid granularity using a historical path divergence factor to achieve dynamic mapping of the state space. The state transition matrix is ​​corrected using dual thresholds of tension fluctuation rate and yarn breakage frequency, and cross-weights determined based on the fusion of variances of environmental and process parameters are applied, achieving coordinated constraints of multi-dimensional physical factors on the spindle speed control process. During the control execution phase, a target state transition matrix is ​​generated through back projection, and candidate spindle speed adjustment amounts are calculated using the adjusted step vector. When the tension gradient exceeds its extreme value, the rollback adjustment range is determined jointly using a yarn breakage margin factor and a historical path divergence factor, which can promptly suppress the risk of yarn breakage, improving the overall production efficiency and operational stability of the spinning machine while ensuring yarn quality and consistency. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the intelligent ingot speed control system. Figure 2 This is a flowchart of a method for intelligent control of spindle speed on a spinning frame. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0011] like Figure 1 The diagram shows the architecture of the intelligent control system for ingot speed, which is divided into three levels: the perception and acquisition layer, the intelligent decision-making layer, and the execution and output layer.

[0012] Example 1 In Embodiment 1 of the present invention, a method for intelligent control of spindle speed on a spinning frame is provided, such as... Figure 2 As shown, it includes: S1 collects yarn tension, ambient temperature and humidity, front roller speed and current spindle speed to construct a multidimensional state vector, calculates tension fluctuation rate, breakage frequency and tension gradient, scales the grid granularity according to the historical path divergence factor of the previous cycle, and maps the multidimensional state vector to the discrete space to obtain the initial state transition matrix.

[0013] Yarn tension is collected in real time at a fixed sampling frequency using a tension sensor installed at the yarn guide hook of the spinning machine. Ambient temperature and humidity are collected using temperature and humidity sensors deployed in the workshop. The front roller speed is collected using a rotary encoder at the front roller shaft end, and the current spindle speed is read by the main shaft frequency converter. The collected values ​​are spliced ​​in a fixed order to form a measured multidimensional state vector with corresponding physical dimensions. When calculating variance, bias, or model input, the corresponding sequence is normalized or standardized according to the calculation object.

[0014] Extract the continuous yarn tension sequence within the current time window and calculate the ratio of the standard deviation to the mean of the sequence as the tension fluctuation rate. Count the number of yarn breaks per unit time using a photoelectric yarn breakage sensor, as the yarn breakage frequency. Calculate the difference between the yarn tension at the current moment and the yarn tension at the previous sampling moment, and divide this difference by the sampling time interval to obtain the tension gradient.

[0015] The system retrieves the historical path divergence factor calculated in the previous control cycle from its storage and sets the basic grid granularity vector. The basic grid granularity is divided by this historical path divergence factor to obtain the grid granularity for the current cycle; when the historical path divergence factor is zero, it is replaced with a preset minimum positive constant. Based on this dynamic grid granularity, each data dimension of the continuous multidimensional state vector is divided into equidistant intervals, and the combination of these intervals within each dimension of the multidimensional state vector is defined as a discrete state node.

[0016] The number of times a node transitions from one discrete state node to all other possible discrete state nodes is recorded in historical data. The transition probability is obtained by dividing the number of each transition by the total number of transitions to the corresponding initial state node. The initial state transition matrix is ​​constructed from the transition probabilities among all discrete state nodes.

[0017] In some embodiments, scaling the grid granularity according to the historical path divergence factor of the previous cycle and mapping the multidimensional state vector to the discrete space to obtain the initial state transition matrix includes: Divide the grid granularity of each dimension of the basic state space by the historical path divergence factor to obtain the scaled grid granularity of each dimension of the state space. When the historical path divergence factor is zero, it is replaced with a preset minimum positive constant. Divide the element values ​​of each dimension in the multidimensional state vector by the grid granularity of the corresponding dimension in the scaled state space, and round up the result to obtain the index of each dimension in the discretized state index vector of the current time period. Map the discretized state index vector to a unique discretized state number; The number of times the discretized state number of the previous period is transferred to the discretized state number of the current period is counted within two consecutive time periods. All the occurrence counts are filled into the corresponding row and column positions of a two-dimensional matrix, and the probability is normalized for each row of the two-dimensional matrix to construct the initial state transition matrix.

[0018] The basic state space is set to include five dimensions: yarn tension, ambient temperature, ambient humidity, front roller speed, and current spindle speed. The corresponding elements of the initial grid granularity vector are 0.5 cN, 0.2℃, 1%, 5 rpm, and 10 rpm. If the historical path divergence factor calculated in the previous cycle is 1.25, then each element of the initial grid granularity vector is divided by 1.25 to obtain scaled grid granularity vector elements of 0.4 cN, 0.16℃, 0.8%, 4 rpm, and 8 rpm. If this divergence factor equals 0 due to complete system convergence, it is replaced with 0.01 to prevent division by zero errors.

[0019] When the currently acquired multidimensional state vector elements are 15.2cN, 26.5℃, 60.5%, 320rpm, and 16000rpm, each element is divided by the scaled grid granularity and rounded up to obtain the discretized state index vector elements for the current time period: 38, 166, 76, 80, and 2000. The single-dimensional index values ​​in this discretized state index vector represent the intervals within each dimension and are not directly equivalent to the total number of states in the state transition matrix.

[0020] The 5-dimensional index vector is transformed into a globally unique state key value using a base offset rule, and this state key value is mapped to a unique valid discretized state number through a valid state dictionary maintained by the system. Within a continuously running sampling period, if the state number in period (t-1) is 105 and the state number in period t is 112, then the value in the 105th row and 112th column of the two-dimensional matrix is ​​incremented by 1.

[0021] When the update cycle arrives, each row of the matrix is ​​traversed, and the value of all elements in a row is divided by the sum of the elements in that row for probability normalization. When the total number of transitions in row 105 is 50, and the number of transitions to row 112 is 10, the corresponding transition probability in the initial state transition matrix is ​​0.20, thus constructing a complete Markov initial state transition probability matrix.

[0022] S2. Based on the tension fluctuation rate and the frequency of head breakage, the initial state transition matrix is ​​corrected by a dual threshold to obtain the reference state transition matrix. The head breakage margin factor is calculated. The cross weight is calculated based on the variance fusion result of the ambient temperature and humidity and the front roller speed. The weight is applied to the reference state transition matrix to obtain the multi-parameter weighted state transition matrix.

[0023] Pre-set the tension extreme threshold for tension fluctuation rate and the decapitation extreme threshold for decapitation frequency. Iterate through each state row in the initial state transition matrix, determining whether the tension fluctuation rate of the corresponding state is not less than the tension extreme threshold and whether the corresponding decapitation frequency is not less than the decapitation extreme threshold. If both conditions are met, reset the corresponding row vector in the initial state transition matrix to a unit vector pointing to the preset low-speed safe state, so that the row only retains the transition probability pointing to the low-speed safe state, thus completing the correction and obtaining the baseline state transition matrix. If the dual conditions are not met, retain the original transition probability of the corresponding row vector for that state.

[0024] The first quotient is obtained by dividing the current tension volatility by the tension extreme threshold, and the second quotient is obtained by dividing the current detonation frequency by the detonation extreme threshold. The first and second quotients are added together to obtain the detonation margin factor. The detonation margin factor is used to characterize the degree to which the current state utilizes the safety margin; a higher value indicates that it is closer to the detonation risk boundary. The variance of the environmental temperature and humidity fusion sequence and the variance of the front roller speed sequence are calculated separately within a set time period. The two are added together to obtain the basic volatility, and the basic volatility is multiplied by the environmental compensation coefficient to obtain the cross weight.

[0025] Multiply each element in the baseline state transition matrix by the cross weight to obtain a multi-parameter weighted state transition matrix. This matrix serves as the transition weight matrix with multi-parameter perturbation intensity in subsequent calculations, and the adjustment effect of the cross weight is no longer offset by row normalization of the matrix after global scalar scaling.

[0026] In some embodiments, the step of obtaining the baseline state transition matrix based on the dual threshold correction of the tension fluctuation rate and the frequency of head loss, and calculating the head loss margin factor, includes: Determine whether the tension fluctuation rate corresponding to each state in the initial state transition matrix is ​​not less than the tension extreme value threshold and whether the corresponding decapitation frequency is not less than the decapitation extreme value threshold. For a state where the tension fluctuation rate is not less than the tension extreme threshold and the breakage frequency is not less than the breakage extreme threshold, the row vector corresponding to this state in the initial state transition matrix is ​​reset to a unit vector pointing to the preset low-speed safety state to obtain the reference state transition matrix. The first quotient is obtained by dividing the current tension fluctuation rate by the tension extreme threshold, the second quotient is obtained by dividing the current decapitation frequency by the decapitation extreme threshold, and the decapitation margin factor is obtained by adding the first quotient and the second quotient.

[0027] The pre-set extreme threshold for tension is 15%, i.e., a tension fluctuation rate of 0.15, and the extreme threshold for head breakage is 5 times per thousand spindles. Each state row in the initial state transition matrix is ​​traversed, and the average tension fluctuation rate and average head breakage frequency associated with that state in historical records are retrieved. When the tension fluctuation rate corresponding to a high-risk state is 18%, greater than 0.15, and its associated head breakage frequency is 6 times per thousand spindles, greater than 5 times, the state is determined to meet the double over-limit condition.

[0028] The corresponding row vector of this state in the initial state transition matrix is ​​reset so that the probability of all column elements is zeroed. Only the transition probability of the base steady interval corresponding to the preset low-speed safe state of 12000 rpm is set to 1.0. That is, a specific unit vector is constructed. A safe reference state transition matrix is ​​generated through this local correction operation.

[0029] After completing matrix correction, the risk approximation degree is assessed by simultaneously calculating the failure margin factor for the current real-time state of the system. If the tension volatility measured in the current real-time state is 12% (0.12), and the failure frequency is 3 times per thousand spindles, the first quotient is calculated by dividing the current tension volatility of 0.12 by the tension extreme threshold of 0.15. Simultaneously, the second quotient is calculated by dividing the current failure frequency of 3 by the failure extreme threshold of 5.

[0030] By linearly superimposing these two quotients, the breakage margin factor for the current cycle is obtained as 1.40. This factor quantifies the overall approximation of the current spinning condition relative to the extreme breakage risk. A higher value indicates a smaller safety margin, serving as the core calculation basis for the system control rollback ratio.

[0031] In some embodiments, the step of calculating cross weights based on the variance fusion result of ambient temperature and humidity and front roller speed, and applying the weights to the baseline state transition matrix to obtain a multi-parameter weighted state transition matrix includes: Calculate the first variance of the weighted fusion time series of the normalized ambient temperature time series and the normalized ambient humidity time series within the preset time window, and the second variance of the unnormalized roller speed time series. The basic volatility is obtained by adding the first variance to the second variance; The basic fluctuation amount is multiplied by the environmental compensation coefficient to obtain the cross weight; The cross weights are multiplied by the baseline state transition matrix to obtain the multi-parameter weighted state transition matrix.

[0032] A time window of 60 seconds was set. Within this window, normalized time series values ​​of ambient temperature and humidity, ranging from 0 to 1, were acquired. These values ​​were then linearly combined using a preset weighting of 0.4 for temperature and 0.6 for humidity, to generate a weighted fused time series. When the first variance of this fused time series over these 60 sampling points was 0.0025, it indicated relatively stable environmental fluctuations. Simultaneously, the time series of the roller rotation speed before normalization was acquired, and its second variance was calculated to be 0.0150, reflecting the dynamic fluctuations of the drawing mechanism.

[0033] Adding the first variance of 0.0025 to the second variance of 0.0150 yields a base fluctuation of 0.0175. Using an environmental compensation coefficient of 0.85, obtained through prior experimental calibration, the base fluctuation of 0.0175 is multiplied by 0.85 to calculate the cross-weight of 0.014875. This weight represents the combined fluctuation intensity formed by environmental parameters and mechanical drafting parameters; when fluctuations in environmental temperature and humidity or front roller speed increase, the base fluctuation increases synchronously, and the cross-weight increases accordingly.

[0034] The calculated cross weight of 0.014875 is used as a global disturbance weight and is multiplied by scalar multiplication with all elements of the corrected baseline state transition matrix. This allows the subsequent target state transition matrix and spindle speed adjustment to reflect the intensity of environmental and process fluctuations, thereby generating a multi-parameter weighted state transition matrix that incorporates the variance characteristics of external physical disturbances.

[0035] S3: Calculate the historical path divergence factor based on the deviation between the predicted state at the previous moment and the current multi-dimensional state vector. Back-project the multi-parameter weighted state transition matrix to obtain the target state transition matrix. Adjust the step vector using cross-weights. Take the inner product of the row vector corresponding to the current state in the target state transition matrix and the adjusted step vector to obtain the candidate spindle speed adjustment amount. Determine the target spindle speed based on the candidate spindle speed adjustment amount. When the tension gradient exceeds the extreme value, use the breakage margin factor and the historical path divergence factor to roll back the candidate spindle speed adjustment amount and control the spinning machine to execute.

[0036] Based on the row vector corresponding to the current state in the multi-parameter weighted state transition matrix of the previous cycle, select the next discrete state with the largest transition probability or transition weight, or sum the center values ​​of each candidate state according to the row vector to obtain the predicted state vector of the previous time step. Calculate the element-wise differences between the predicted state vector of the previous time step and the actual collected current multi-dimensional state vector in each dimension. Divide the element-wise differences of each dimension by the normalization scaling factor of the corresponding dimension to obtain the standard deviation of each dimension. Sum the squares of the standard deviations of each dimension, take the square root, and add it to a minimal positive constant to prevent division by zero to obtain the historical path divergence factor.

[0037] Based on the row and column dimensions of the multi-parameter weighted state transition matrix, a diagonal matrix is ​​generated where all diagonal elements are historical path divergence factors. The target state transition matrix is ​​obtained by multiplying the multi-parameter weighted state transition matrix with this diagonal matrix. This back projection process does not transpose the multi-parameter weighted state transition matrix, nor does it use the reciprocals of historical path divergence factors as projection coefficients. This ensures that the greater the prediction bias, the higher the magnitude of the weights in the target state transition matrix involved in subsequent adjustments.

[0038] A standard step vector containing multiple selectable spindle speed increment / decrement values ​​is pre-defined. A constant safety bias is added to the cross weights to obtain a composite weight. This composite weight is then multiplied by the standard step vector to generate an adjusted step vector. A row of weight distributions corresponding to the discrete nodes of the current multidimensional state vector is extracted from the target state transition matrix and used as a row vector. This row vector is multiplied by the corresponding elements of the adjusted step vector, and the sum is performed to complete the inner product operation, yielding the candidate spindle speed adjustment amount.

[0039] A preset extreme value for the tension gradient is established. The system then checks if the absolute value of the current tension gradient exceeds this extreme value. If the absolute value is not greater than the extreme value, the current spindle speed is added to the candidate spindle speed adjustment amount to obtain the actual target spindle speed. If the absolute value is greater than the extreme value, the break-off margin factor is multiplied by the historical path divergence factor to calculate the rollback ratio coefficient. The sum of the absolute value of the candidate spindle speed adjustment amount and the preset safe rollback bias is multiplied by the rollback ratio coefficient to generate the rollback speed width. When the tension gradient is greater than the extreme value, the current spindle speed is subtracted from the rollback speed width; when the tension gradient is less than a negative extreme value, the current spindle speed is added to the rollback speed width to obtain the actual target spindle speed.

[0040] The determined target spindle speed is converted into a frequency control command and sent to the spindle frequency converter of the spinning frame. The spindle frequency converter changes the output AC frequency according to the received command, driving the spindle motor to rotate, thereby realizing closed-loop control and physical execution of the spinning frame spindle speed.

[0041] In some embodiments, the step of calculating the historical path divergence factor based on the deviation between the predicted state at the previous moment and the current multidimensional state vector, and back-projecting the multi-parameter weighted state transition matrix to obtain the target state transition matrix includes: Calculate the element differences between the predicted state vector at the previous moment and the current multidimensional state vector in each dimension. Divide the element differences in each dimension by the normalization scaling factor of the corresponding dimension to obtain the standard deviation of each dimension. Sum the squares of the standard deviations in each dimension, take the square root, and add it to the smallest positive constant to prevent division by zero to obtain the historical path divergence factor. Generate a diagonal matrix whose main diagonal elements are all the historical path divergence factors and whose number of rows and columns is exactly the same as the multi-parameter weighted state transition matrix. Perform matrix multiplication operation between the multi-parameter weighted state transition matrix and the diagonal matrix to generate the target state transition matrix.

[0042] The predicted state vector at the previous moment is determined by the multi-parameter weighted state transition matrix or the target state transition matrix of the previous cycle. Specifically, the system reads the matrix row vector corresponding to the actual state of the previous cycle, selects the target state number with the largest weight, and obtains the predicted state vector by looking up the mapping relationship between the pre-stored state number and the state center value; or, the system sums the weights of each candidate state in the row vector with the corresponding state center value to obtain the predicted state vector.

[0043] Taking the output predicted state vector elements as 15.0cN, 26.0℃, 60.0%, 315rpm, and 15900rpm, and the actual measured multidimensional state vector elements at the current moment as 15.2cN, 26.5℃, 60.5%, 320rpm, and 16000rpm as an example, the absolute element differences between these two vectors in each dimension are calculated, resulting in a difference vector element of 0.2cN, 0.5℃, 0.5%, 5rpm, and 100rpm.

[0044] Preset normalized scaling factors for each dimension, namely tension scaling 2.0 cN, temperature scaling 5.0℃, humidity scaling 10.0%, front roller speed scaling 50 rpm, and spindle speed scaling 1000 rpm. Divide the element differences by the scaling factors respectively to obtain the dimensionless standard deviation vector elements 0.1, 0.1, 0.05, 0.1, and 0.1.

[0045] The squares of each element in the standard deviation vector are 0.01, 0.01, 0.0025, 0.01, and 0.01. The sum of all squared values ​​is 0.0425. Taking the square root of 0.0425 yields the Euclidean distance of 0.206. Adding this to a small positive constant of 0.0001 to prevent division by zero gives the historical path divergence factor of 0.2061, representing the deviation between the model-predicted trajectory and the actual physical trajectory.

[0046] Based on the row and column dimensions of the aforementioned multi-parameter weighted state transition matrix, when the number of valid discrete states is N and the matrix size is N×N, a diagonal matrix of the same order is constructed; for example, when N=1000, a 1000×1000 diagonal matrix is ​​constructed, with all elements on the diagonal assigned the value 0.2061, and all other off-diagonal elements set to 0. Here, N represents the number of valid discrete states in the state dictionary, which does not conflict with the discretized state index value of a single dimension.

[0047] The multi-parameter weighted state transition matrix is ​​right-multiplied by the diagonal matrix. Matrix multiplication is used to perform a proportional scaling and back projection of the column vectors of the original matrix, so that the target state transition matrix carries the divergence intensity corresponding to the current prediction bias, and the target state transition matrix is ​​output.

[0048] In some embodiments, the step of adjusting the step vector using cross-weights, and taking the inner product of the row vector corresponding to the current state in the target state transition matrix and the adjusted step vector to obtain the candidate spindle speed adjustment amount, includes: Adding a constant safety bias to the cross weights yields the composite weights; The adjusted step vector is obtained by multiplying the composite weight with a preset step vector whose dimension is the same as the total number of states. Extract the row vector corresponding to the current state from the target state transition matrix, and perform a dot product operation between the row vector and the adjusted step vector to obtain the candidate ingot speed adjustment amount.

[0049] The cross-weight calculated in the previous stage is 0.014875. A constant safety bias of 1.500 is preset to ensure the basic response margin of the motor speed regulation. The cross-weight 0.014875 is added to the constant safety bias 1.500 to calculate the composite weight 1.514875. At the same time, a step vector with the same dimension as the total number of discrete states N is defined in the initialization stage, for example, N=1000. Each element of this vector represents the expected basic speed increment when the system transitions to that state, that is, the initial value of an element in a certain state is +10 rpm, and the initial value of an element in a certain low-speed state is -50 rpm.

[0050] By performing a scalar multiplication operation, the calculated composite weight 1.514875 is multiplied one by one with each element of the step vector to perform a global magnitude adjustment, generating an adjusted step vector, that is, the original +10rpm is updated to +15.14875rpm, and the original -50rpm is updated to -75.74375rpm.

[0051] After completing the step vector update, based on the current discretized state number obtained from real-time monitoring as state index number 350, the 1×N-dimensional row vector corresponding to the 350th row is extracted from the generated target state transition matrix. This row vector contains the weighted prediction weight distribution of the transition from the current state to all other discrete states.

[0052] The row vector is multiplied by the aforementioned adjusted N×1 dimensional step vector, which involves multiplying the transition weight of each state by its corresponding expected speed increment and summing all the results. When the sum of the inner products is 12.56, the candidate spindle speed adjustment for the current control cycle is determined to be +12.56 rpm, thus transforming the state transition weights in the high-dimensional space into a one-dimensional motor speed control command with physical execution meaning.

[0053] In some embodiments, determining the target spindle speed based on the candidate spindle speed adjustment amount, and when the tension gradient exceeds the extreme value, using the breakage margin factor and the historical path divergence factor to roll back the candidate spindle speed adjustment amount to control the spinning machine to execute, includes: Determine whether the absolute value of the tension gradient in the current cycle is greater than the set extreme value; When the absolute value of the tension gradient is not greater than the extreme value, the candidate ingot speed adjustment amount is added to the current ingot speed to obtain the actual target ingot speed. When the absolute value of the tension gradient is greater than the extreme value, the head break margin factor is multiplied by the historical path divergence factor to obtain the rollback ratio coefficient. The absolute value of the candidate spindle speed adjustment is added to the preset safe rollback bias, and the result is multiplied by the rollback ratio coefficient to generate the rollback speed width. When the tension gradient is greater than the extreme value, the actual target spindle speed is obtained by subtracting the rollback speed width from the current spindle speed. When the tension gradient is less than the negative extreme value, the current spindle speed is added to the rollback speed width to obtain the actual target spindle speed. The actual target spindle speed command is sent to the variable frequency servo motor driver of the spinning machine to modify the current motor speed.

[0054] The candidate spindle speed adjustment is calculated to be +12.56 rpm in the current control cycle, and the current real-time operating spindle speed of the motor is 16000 rpm. The absolute value of the yarn tension gradient calculated in real time by the sensor is extracted as 0.8 cN / s, and compared with the preset tension gradient extreme value of 1.5 cN / s.

[0055] If the absolute value of the tension gradient (0.8) is less than the extreme value (1.5), it indicates that the current spinning tension change is stable with no risk of abrupt change. The candidate spindle speed adjustment amount +12.56 rpm is added to the current spindle speed of 16000 rpm to obtain the actual target spindle speed of 16012.56 rpm. Conversely, if the system detects tension contraction and the absolute value of the tension gradient reaches 2.5 cN / s, which is greater than the extreme value of 1.5, the emergency rollback adjustment mechanism is triggered. In this case, the previously calculated yarn breakage margin factor is extracted as 1.40. During this abnormal cycle, due to the increased deviation between the predicted state and the current state, the historical path divergence factor is updated to 2.061. Multiplying 1.40 by 2.061 yields a rollback ratio coefficient of 2.8854.

[0056] The absolute value of the candidate spindle speed adjustment amount, 12.56 rpm, is extracted and added to the preset safe rollback bias amount of 50 rpm to obtain a sum of 62.56 rpm. This sum is then multiplied by the rollback ratio coefficient of 2.8854 to generate a rollback speed width of approximately 180.5 rpm.

[0057] Since the current tension gradient is +2.5cN / s, which is greater than the positive value, it indicates that the tension is too tight. The speed reduction logic is executed, and the current spindle speed of 16000rpm is subtracted from the rollback speed width of 180.5rpm to obtain the actual target spindle speed of 15819.5rpm.

[0058] The calculated actual target spindle speed is converted into a standard control message command and sent to the variable frequency servo motor driver at the bottom of the spinning machine. The stator current frequency is changed to control the motor speed reduction, realizing closed-loop adaptive safety avoidance under the dangerous condition of spindle breakage.

[0059] Example 2 Embodiment 2 of the present invention proposes a system for intelligent control of spindle speed of a spinning frame, comprising the following modules: The mapping module is used to collect yarn tension, ambient temperature and humidity, front roller speed and current spindle speed to construct a multi-dimensional state vector, calculate tension fluctuation rate, breakage frequency and tension gradient, scale the grid granularity according to the historical path divergence factor of the previous cycle, and map the multi-dimensional state vector to the discrete space to obtain the initial state transition matrix. The calculation module is used to obtain the reference state transition matrix based on the dual threshold correction of the initial state transition matrix of tension fluctuation rate and head breakage frequency, calculate the head breakage margin factor, calculate the cross weight based on the variance fusion result of ambient temperature and humidity and front roller speed, and apply the weight to the reference state transition matrix to obtain the multi-parameter weighted state transition matrix. The adjustment module is used to calculate the historical path divergence factor based on the deviation between the predicted state at the previous moment and the current multi-dimensional state vector, back-project the multi-parameter weighted state transition matrix to obtain the target state transition matrix, adjust the step vector using cross weights, and take the inner product of the row vector corresponding to the current state in the target state transition matrix with the adjusted step vector to obtain the candidate spindle speed adjustment amount. The target spindle speed is determined based on the candidate spindle speed adjustment amount. When the tension gradient exceeds the extreme value, the candidate spindle speed adjustment amount is rolled back using the breakage margin factor and the historical path divergence factor to control the spinning machine to execute.

[0060] In some embodiments, scaling the grid granularity according to the historical path divergence factor of the previous cycle and mapping the multidimensional state vector to the discrete space to obtain the initial state transition matrix includes: Divide the grid granularity of each dimension of the basic state space by the historical path divergence factor to obtain the scaled grid granularity of each dimension of the state space. When the historical path divergence factor is zero, it is replaced with a preset minimum positive constant. Divide the element values ​​of each dimension in the multidimensional state vector by the grid granularity of the corresponding dimension in the scaled state space, and round up the result to obtain the index of each dimension in the discretized state index vector of the current time period. Map the discretized state index vector to a unique discretized state number; The number of times the discretized state number of the previous period is transferred to the discretized state number of the current period is counted within two consecutive time periods. All the occurrence counts are filled into the corresponding row and column positions of a two-dimensional matrix, and the probability is normalized for each row of the two-dimensional matrix to construct the initial state transition matrix.

[0061] In some embodiments, the step of obtaining the baseline state transition matrix based on the dual threshold correction of the tension fluctuation rate and the frequency of head loss, and calculating the head loss margin factor, includes: Determine whether the tension fluctuation rate corresponding to each state in the initial state transition matrix is ​​not less than the tension extreme value threshold and whether the corresponding decapitation frequency is not less than the decapitation extreme value threshold. For a state where the tension fluctuation rate is not less than the tension extreme threshold and the breakage frequency is not less than the breakage extreme threshold, the row vector corresponding to this state in the initial state transition matrix is ​​reset to a unit vector pointing to the preset low-speed safety state to obtain the reference state transition matrix. The first quotient is obtained by dividing the current tension fluctuation rate by the tension extreme threshold, the second quotient is obtained by dividing the current decapitation frequency by the decapitation extreme threshold, and the decapitation margin factor is obtained by adding the first quotient and the second quotient.

[0062] In some embodiments, the step of calculating cross weights based on the variance fusion result of ambient temperature and humidity and front roller speed, and applying the weights to the baseline state transition matrix to obtain a multi-parameter weighted state transition matrix includes: Calculate the first variance of the weighted fusion time series of the normalized ambient temperature time series and the normalized ambient humidity time series within the preset time window, and the second variance of the unnormalized roller speed time series. The basic volatility is obtained by adding the first variance to the second variance; The basic fluctuation amount is multiplied by the environmental compensation coefficient to obtain the cross weight; The cross weights are multiplied by the baseline state transition matrix to obtain the multi-parameter weighted state transition matrix.

[0063] In some embodiments, the step of calculating the historical path divergence factor based on the deviation between the predicted state at the previous moment and the current multidimensional state vector, and back-projecting the multi-parameter weighted state transition matrix to obtain the target state transition matrix includes: Calculate the element differences between the predicted state vector at the previous moment and the current multidimensional state vector in each dimension. Divide the element differences in each dimension by the normalization scaling factor of the corresponding dimension to obtain the standard deviation of each dimension. Sum the squares of the standard deviations in each dimension, take the square root, and add it to the smallest positive constant to prevent division by zero to obtain the historical path divergence factor. Generate a diagonal matrix whose main diagonal elements are all the historical path divergence factors and whose number of rows and columns is exactly the same as the multi-parameter weighted state transition matrix. Perform matrix multiplication operation between the multi-parameter weighted state transition matrix and the diagonal matrix to generate the target state transition matrix.

[0064] In some embodiments, the step of adjusting the step vector using cross-weights, and taking the inner product of the row vector corresponding to the current state in the target state transition matrix and the adjusted step vector to obtain the candidate spindle speed adjustment amount, includes: Adding a constant safety bias to the cross weights yields the composite weights; The adjusted step vector is obtained by multiplying the composite weight with a preset step vector whose dimension is the same as the total number of states. Extract the row vector corresponding to the current state from the target state transition matrix, and perform a dot product operation between the row vector and the adjusted step vector to obtain the candidate ingot speed adjustment amount.

[0065] In some embodiments, determining the target spindle speed based on the candidate spindle speed adjustment amount, and when the tension gradient exceeds the extreme value, using the breakage margin factor and the historical path divergence factor to roll back the candidate spindle speed adjustment amount to control the spinning machine to execute, includes: Determine whether the absolute value of the tension gradient in the current cycle is greater than the set extreme value; When the absolute value of the tension gradient is not greater than the extreme value, the candidate ingot speed adjustment amount is added to the current ingot speed to obtain the actual target ingot speed. When the absolute value of the tension gradient is greater than the extreme value, the head break margin factor is multiplied by the historical path divergence factor to obtain the rollback ratio coefficient. The absolute value of the candidate spindle speed adjustment is added to the preset safe rollback bias, and the result is multiplied by the rollback ratio coefficient to generate the rollback speed width. When the tension gradient is greater than the extreme value, the actual target spindle speed is obtained by subtracting the rollback speed width from the current spindle speed. When the tension gradient is less than the negative extreme value, the current spindle speed is added to the rollback speed width to obtain the actual target spindle speed. The actual target spindle speed command is sent to the variable frequency servo motor driver of the spinning machine to modify the current motor speed.

[0066] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for intelligent control of spindle speed on a spinning frame, characterized in that, include: A multidimensional state vector is constructed by collecting yarn tension, ambient temperature and humidity, front roller speed and current spindle speed. Tension fluctuation rate, breakage frequency and tension gradient are calculated. The grid granularity is scaled according to the historical path divergence factor of the previous cycle. The multidimensional state vector is mapped to the discrete space to obtain the initial state transition matrix. The baseline state transition matrix is ​​obtained by correcting the initial state transition matrix based on the tension fluctuation rate and the breakage frequency. The breakage margin factor is calculated, and the cross weight is calculated based on the variance fusion result of the ambient temperature and humidity and the front roller speed. The weight is applied to the baseline state transition matrix to obtain the multi-parameter weighted state transition matrix. The historical path divergence factor is calculated based on the deviation between the predicted state at the previous moment and the current multidimensional state vector. The target state transition matrix is ​​obtained by back-projecting the multi-parameter weighted state transition matrix. The step vector is adjusted using cross-weights. The candidate spindle speed adjustment amount is obtained by taking the inner product of the row vector corresponding to the current state in the target state transition matrix and the adjusted step vector. The target spindle speed is determined based on the candidate spindle speed adjustment amount. When the tension gradient exceeds the extreme value, the candidate spindle speed adjustment amount is rolled back using the breakage margin factor and the historical path divergence factor to control the spinning machine.

2. The method according to claim 1, characterized in that, The step of scaling the grid granularity according to the historical path divergence factor of the previous cycle and mapping the multidimensional state vector to the discrete space to obtain the initial state transition matrix includes: Divide the grid granularity of each dimension of the basic state space by the historical path divergence factor to obtain the scaled grid granularity of each dimension of the state space. When the historical path divergence factor is zero, it is replaced with a preset minimum positive constant. Divide the element values ​​of each dimension in the multidimensional state vector by the grid granularity of the corresponding dimension in the scaled state space, and round up the result to obtain the index of each dimension in the discretized state index vector of the current time period. Map the discretized state index vector to a unique discretized state number; The number of times the discretized state number of the previous period is transferred to the discretized state number of the current period is counted within two consecutive time periods. All the occurrence counts are filled into the corresponding row and column positions of a two-dimensional matrix, and the probability is normalized for each row of the two-dimensional matrix to construct the initial state transition matrix.

3. The method according to claim 2, characterized in that, The baseline state transition matrix is ​​obtained by correcting the initial state transition matrix based on the tension fluctuation rate and the frequency of head loss, and the head loss margin factor is calculated, including: Determine whether the tension fluctuation rate corresponding to each state in the initial state transition matrix is ​​not less than the tension extreme value threshold and whether the corresponding decapitation frequency is not less than the decapitation extreme value threshold. For a state where the tension fluctuation rate is not less than the tension extreme threshold and the breakage frequency is not less than the breakage extreme threshold, the row vector corresponding to this state in the initial state transition matrix is ​​reset to a unit vector pointing to the preset low-speed safety state to obtain the reference state transition matrix. The first quotient is obtained by dividing the current tension fluctuation rate by the tension extreme threshold, the second quotient is obtained by dividing the current decapitation frequency by the decapitation extreme threshold, and the decapitation margin factor is obtained by adding the first quotient and the second quotient.

4. The method according to claim 1, characterized in that, The calculation of cross weights based on the variance fusion results of ambient temperature and humidity and front roller speed, and the application of these weights to the baseline state transition matrix to obtain a multi-parameter weighted state transition matrix, includes: Calculate the first variance of the weighted fusion time series of the normalized ambient temperature time series and the normalized ambient humidity time series within the preset time window, and the second variance of the unnormalized roller speed time series. The basic volatility is obtained by adding the first variance to the second variance; The basic fluctuation amount is multiplied by the environmental compensation coefficient to obtain the cross weight; The cross weights are multiplied by the baseline state transition matrix to obtain the multi-parameter weighted state transition matrix.

5. The method according to claim 1 or 4, characterized in that, The step of calculating the historical path divergence factor based on the deviation between the predicted state at the previous moment and the current multidimensional state vector, and then back-projecting the multi-parameter weighted state transition matrix to obtain the target state transition matrix includes: Calculate the element differences between the predicted state vector at the previous moment and the current multidimensional state vector in each dimension. Divide the element differences in each dimension by the normalization scaling factor of the corresponding dimension to obtain the standard deviation of each dimension. Sum the squares of the standard deviations in each dimension, take the square root, and add it to the smallest positive constant to prevent division by zero to obtain the historical path divergence factor. Generate a diagonal matrix whose main diagonal elements are all the historical path divergence factors and whose number of rows and columns is exactly the same as the multi-parameter weighted state transition matrix. Perform matrix multiplication operation between the multi-parameter weighted state transition matrix and the diagonal matrix to generate the target state transition matrix.

6. The method according to claim 1, characterized in that, The step of adjusting the step vector using cross-weights, which involves taking the inner product of the row vector corresponding to the current state in the target state transition matrix and the adjusted step vector to obtain the candidate spindle speed adjustment amount, includes: Adding a constant safety bias to the cross weights yields the composite weights; The adjusted step vector is obtained by multiplying the composite weight with a preset step vector whose dimension is the same as the total number of states. Extract the row vector corresponding to the current state from the target state transition matrix, and perform a dot product operation between the row vector and the adjusted step vector to obtain the candidate ingot speed adjustment amount.

7. The method according to claim 1, characterized in that, The process of determining the target spindle speed based on candidate spindle speed adjustment amounts, and adjusting the candidate spindle speed adjustment amounts by using a breakage margin factor and a historical path divergence factor when the tension gradient exceeds an extreme value to control the spinning machine execution, includes: Determine whether the absolute value of the tension gradient in the current cycle is greater than the set extreme value; When the absolute value of the tension gradient is not greater than the extreme value, the candidate ingot speed adjustment amount is added to the current ingot speed to obtain the actual target ingot speed. When the absolute value of the tension gradient is greater than the extreme value, the head break margin factor is multiplied by the historical path divergence factor to obtain the rollback ratio coefficient. The absolute value of the candidate spindle speed adjustment is added to the preset safe rollback bias, and the result is multiplied by the rollback ratio coefficient to generate the rollback speed width. When the tension gradient is greater than the extreme value, the actual target spindle speed is obtained by subtracting the rollback speed width from the current spindle speed. When the tension gradient is less than the negative extreme value, the current spindle speed is added to the rollback speed width to obtain the actual target spindle speed. The actual target spindle speed command is sent to the variable frequency servo motor driver of the spinning machine to modify the current motor speed.

8. A system for intelligent control of spindle speed on a spinning frame, characterized in that, include: The mapping module is used to collect yarn tension, ambient temperature and humidity, front roller speed and current spindle speed to construct a multi-dimensional state vector, calculate tension fluctuation rate, breakage frequency and tension gradient, scale the grid granularity according to the historical path divergence factor of the previous cycle, and map the multi-dimensional state vector to the discrete space to obtain the initial state transition matrix. The calculation module is used to obtain the reference state transition matrix based on the dual threshold correction of the initial state transition matrix of tension fluctuation rate and head breakage frequency, calculate the head breakage margin factor, calculate the cross weight based on the variance fusion result of ambient temperature and humidity and front roller speed, and apply the weight to the reference state transition matrix to obtain the multi-parameter weighted state transition matrix. The adjustment module is used to calculate the historical path divergence factor based on the deviation between the predicted state at the previous moment and the current multi-dimensional state vector, back-project the multi-parameter weighted state transition matrix to obtain the target state transition matrix, adjust the step vector using cross weights, and take the inner product of the row vector corresponding to the current state in the target state transition matrix with the adjusted step vector to obtain the candidate spindle speed adjustment amount. The target spindle speed is determined based on the candidate spindle speed adjustment amount. When the tension gradient exceeds the extreme value, the candidate spindle speed adjustment amount is rolled back using the breakage margin factor and the historical path divergence factor to control the spinning machine to execute.

9. The system according to claim 8, characterized in that, The step of scaling the grid granularity according to the historical path divergence factor of the previous cycle and mapping the multidimensional state vector to the discrete space to obtain the initial state transition matrix includes: Divide the grid granularity of each dimension of the basic state space by the historical path divergence factor to obtain the scaled grid granularity of each dimension of the state space. When the historical path divergence factor is zero, it is replaced with a preset minimum positive constant. Divide the element values ​​of each dimension in the multidimensional state vector by the grid granularity of the corresponding dimension in the scaled state space, and round up the result to obtain the index of each dimension in the discretized state index vector of the current time period. Map the discretized state index vector to a unique discretized state number; The number of times the discretized state number of the previous period is transferred to the discretized state number of the current period is counted within two consecutive time periods. All the occurrence counts are filled into the corresponding row and column positions of a two-dimensional matrix, and the probability is normalized for each row of the two-dimensional matrix to construct the initial state transition matrix.

10. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program, when executed by a processor, implements the method as described in any one of claims 1-7.

Images