Anomaly monitoring method and system for operation process of non-invasive homogenizer

Abstract

The application belongs to the technical field of mixing equipment, and particularly relates to an abnormality monitoring method and system for a running process of a non-intervention homogenizer, which comprises the following steps: calculating mechanical angular acceleration according to real-time rotational speed of a motor shaft end; calculating a rheological resistance characteristic index according to motor orthogonal shaft current, real-time rotational speed of the motor shaft end and the mechanical angular acceleration based on a pre-constructed inverse dynamics model; processing the rheological resistance characteristic index by using a singular spectrum analysis algorithm, extracting a trend line of the rheological resistance characteristic index by constructing a trajectory matrix, singular value decomposition and diagonal averaging operation; calculating information entropy of a first-order difference sequence of the trend line as mixing stationarity entropy, and determining a mixing endpoint of the homogenizer running process according to convergence of the mixing stationarity entropy relative to a reference entropy, so as to realize abnormality monitoring of the non-intervention homogenizer running process. The application realizes abnormality monitoring and endpoint determination of a homogenizer mixing process.

Description

Technical Field

[0001] This invention relates to the field of mixing equipment technology. More specifically, this invention relates to a method and system for monitoring anomalies during the operation of a non-invasive homogenizer. Background Technology

[0002] Non-invasive planetary centrifugal homogenizers, also known as vacuum degassing machines, are widely used in the preparation of high-end materials in electronics, new energy, and pharmaceuticals. They are particularly effective for high-viscosity materials such as conductive silver paste, phosphors, and carbon nanotube slurries. Through the combined motion of revolution and rotation along two axes, they generate powerful shear and centrifugal forces, achieving efficient and uniform dispersion and degassing of materials without the need for contact between stirring blades and the material. In actual industrial production, the uniformity of material mixing directly determines the physicochemical properties of the final product, such as conductivity, luminescence efficiency, and slurry stability. Therefore, precise monitoring of the mixing process and accurate determination of the mixing endpoint are crucial for ensuring consistent product quality and improving production efficiency.

[0003] However, the operation and control of such equipment currently still largely rely on the traditional fixed-time method, that is, automated operation based on a pre-set empirical time formula. This open-loop control mode completely ignores the dynamic interference factors in the production process. In fact, slight differences between batches of raw materials, fluctuations in ambient temperature, and subtle changes in the amount of feed will significantly affect the rheological properties of materials and the mixing process. Under fixed-time control, two types of problems are likely to occur: first, insufficient mixing time leads to uneven material dispersion and a decrease in product qualification rate; second, excessive mixing time causes over-mixing, with a large amount of mechanical energy being converted into heat energy, leading to the denaturation and failure of heat-sensitive materials, while also causing energy waste and reduced production capacity.

[0004] To address these issues, existing technologies attempt to introduce online monitoring methods based on motor current, aiming to infer the trend of material viscosity changes by monitoring current variations. However, the rheological resistance signal caused by changes in material viscosity is extremely weak and is often submerged and masked by the large fluctuations in inertia and frictional load signals of the equipment itself, resulting in an extremely low signal-to-noise ratio. Furthermore, the periodic mechanical vibrations generated by the high-speed operation of the equipment can also cause serious interference to the signal. If conventional filtering methods are used to process the noise, a non-negligible phase lag will be introduced, causing the monitoring results to lag behind the actual state. Summary of the Invention

[0005] To address the technical problems mentioned above, where the weak material rheological resistance signal is overwhelmed by the inertial load and mechanical friction load of the equipment itself during the variable speed operation of a homogenizer, and where traditional noise reduction methods introduce phase lag, resulting in the inability to accurately and timely monitor the material mixing state, this invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a method for anomaly monitoring during the operation of a non-intrusive homogenizer, comprising: acquiring in real time the orthogonal axis current and real-time rotational speed of the homogenizer drive motor via a controller, and calculating the mechanical angular acceleration based on the real-time rotational speed of the motor shaft; calculating rheological drag characteristic indicators stripped of mechanical load torque based on a pre-constructed inverse dynamics model, according to the orthogonal axis current, real-time rotational speed of the motor shaft, and mechanical angular acceleration; processing the rheological drag characteristic indicators using a singular spectrum analysis algorithm, and extracting the trend line of the rheological drag characteristic indicators by constructing a trajectory matrix, singular value decomposition, and diagonal averaging operations; calculating the information entropy of the first-order difference sequence of the trend line as the mixing stationarity entropy, and determining the mixing endpoint of the homogenizer operation process based on the convergence of the mixing stationarity entropy relative to the reference entropy, thereby achieving anomaly monitoring during the operation of the non-intrusive homogenizer.

[0007] This invention acquires real-time current and speed data of the motor through a controller and calculates angular acceleration accordingly. Then, combining this with a pre-constructed inverse dynamics model, it subtracts the inertial torque and frictional torque generated by mechanical angular acceleration and real-time speed from the motor's orthogonal axis current, which characterizes the total electromagnetic torque. This achieves mathematical decoupling of material load and equipment mechanical load, obtaining a rheological resistance characteristic index that reflects only changes in material viscosity. This effectively solves the problem that weak material rheological signals are easily overwhelmed by the large fluctuations in the equipment's mechanical load signal during variable-speed processes. Furthermore, it uses a singular spectrum analysis algorithm to process this index and extract a trend line. Combining this with information entropy quantifies the stability of the trend line and compares it with a benchmark, transforming a qualitative evaluation of mixing uniformity into a quantitative index that converges to a benchmark. This achieves high-precision, lag-free, and adaptive anomaly monitoring and endpoint determination of the homogenizer mixing process without contacting the material or relying on a fixed-time formulation.

[0008] Preferably, the pre-constructed inverse dynamics model is obtained by: controlling the motor to perform a no-load full-speed operation including acceleration, constant speed, and deceleration, and collecting multiple sets of sample data; constructing an observation vector and a regression matrix, wherein the observation vector is composed of the product of the motor orthogonal axis current and the motor torque constant in the sample data, and the regression matrix is ​​composed of the mechanical angular acceleration, the real-time speed at the motor shaft end, and the sign function value of the real-time speed at the motor shaft end in the sample data; and solving the mechanical characteristic parameters of the system using the least squares regression model, wherein the mechanical characteristic parameters include the system's equivalent moment of inertia, viscous friction coefficient, and Coulomb friction coefficient.

[0009] Preferably, the formula for calculating the rheological resistance characteristic index is: In the formula, For a moment The rheological resistance characteristic index; The torque constant of the motor; For a moment The orthogonal axis current of the motor; This is the system's equivalent moment of inertia. For a moment Mechanical angular acceleration; It is the coefficient of viscous friction; For a moment The real-time rotational speed of the motor shaft end; Coulomb friction coefficient; The sign function is used to characterize the directionality of friction. It takes the value 1 when the rotational speed is greater than zero, -1 when it is less than zero, and 0 when it is equal to zero.

[0010] This invention calculates the total output torque by multiplying the motor torque constant by the current, and then subtracts the inertial term calculated from the equivalent moment of inertia and angular acceleration, the viscous friction term calculated from the viscous friction coefficient and rotational speed, and the Coulomb friction term calculated from the Coulomb friction coefficient and the sign function of the rotational speed. This achieves an accurate solution for the rheological resistance characteristic index. In particular, the introduction of the sign function of the rotational speed to characterize the directionality of the frictional force can automatically adapt to the change in the direction of the frictional force when the motor rotates in both directions. This ensures that, in the complex revolution and rotation motion of the homogenizer, regardless of whether the equipment is in an acceleration, deceleration, or constant speed state, the calculated rheological resistance characteristic index can purely reflect the rheological properties of the material inside the tank, eliminating the nonlinear interference of changes in mechanical motion state on the monitoring results.

[0011] Preferably, constructing the trajectory matrix includes: using a sliding time window, based on the most recent Construct a trajectory matrix using all data points at each time point and their rheological drag characteristic indicators. Specifically: In the formula, Indicates the number of the current analysis window. The data point, from the first The moment and the The composition of rheological resistance characteristic indicators at each moment; , representing the number of columns in the trajectory matrix; The length of the data window currently being analyzed. The length of the embedded window. It should be taken as an integer multiple of the orbital period, with a range of 50 to 200. Need to meet To ensure the effectiveness of the decomposition.

[0012] This invention employs a sliding time window strategy to map the recent rheological drag characteristic index data sequence into a trajectory matrix, thereby transforming a one-dimensional time series signal into a two-dimensional matrix structure containing time lag information. This operation effectively embeds the dynamic evolution history of the signal into the row and column spaces of the matrix, enhancing the structured relationship between data. It provides the necessary data structure foundation for subsequently using singular value decomposition (SVD) techniques to separate noise components representing periodic mechanical vibrations and trend components representing material viscosity changes from complex mixed signals.

[0013] Preferably, the step of extracting the trend line of rheological drag characteristic indicators includes: performing singular value decomposition on the trajectory matrix, selecting the first principal component with the largest eigenvalue to calculate the reconstruction matrix; converting the reconstruction matrix into a one-dimensional time series using the diagonal averaging method to obtain the trend line; the diagonal averaging method includes calculating the mean of elements of the upper left anti-diagonal line, the middle anti-diagonal line, and the lower right anti-diagonal line of the reconstruction matrix according to the position of the elements in the reconstruction matrix.

[0014] This invention effectively separates the main trend components in a signal mathematically by performing singular value decomposition on the trajectory matrix and selecting the first principal component with the highest energy for reconstruction. It automatically filters out mechanical vibration noise and random interference corresponding to smaller eigenvalues. Subsequently, the reconstructed matrix is ​​inversely transformed into a one-dimensional time series using the diagonal averaging method. This ensures that the generated trend line retains the original abrupt change characteristics of the signal while achieving smooth processing. Moreover, this process is entirely based on the structural characteristics of the data itself. Compared with traditional low-pass filters, it effectively avoids phase lag and ensures the real-time response of the mixed endpoint determination to changes in the material state.

[0015] Preferably, the formula for calculating the reconstruction matrix is: In the formula, To reconstruct the matrix; It is the largest eigenvalue; This is the first left singular vector; The first right singular vector is denoted as ; the eigenvalue, the first left singular vector, and the first right singular vector are all results of singular value decomposition.

[0016] Preferably, calculating the information entropy of the first-order difference sequence of the trend line as the mixed stationarity entropy includes: calculating the first-order difference value of each point on the trend line, and calculating the proportion of the absolute value of the first-order difference value of each point in the sum of the absolute values ​​of the first-order difference values ​​of all points as the normalized probability; calculating the negative of the sum of the products of the normalized probabilities of all points and the natural logarithm of the normalized probabilities to obtain the mixed stationarity entropy.

[0017] This invention transforms the fluctuation amplitude of a trend line into a probability distribution by calculating the first-order difference value of each point on the trend line and statistically analyzing its normalized probability. Then, it uses the natural logarithm to calculate the information entropy of this distribution as the mixing stationarity entropy, thus quantifying the disorder of the system using a single scalar value. This method utilizes the physical property that the rate of change of resistance during mixing tends to zero and the distribution gradually concentrates as uniformity increases. This allows the mixing stationarity entropy to serve as a sensitive indicator of whether the system has reached a steady state; that is, as mixing becomes more uniform, the entropy value decreases significantly, providing a robust quantitative basis for endpoint determination.

[0018] Preferably, determining the mixing endpoint of the homogenizer operation process based on the convergence of the mixing stability entropy relative to the reference entropy includes: after the real-time speed of the motor shaft enters the constant speed segment, calculating the average value of the mixing stability entropy at all times before the start of the constant speed segment as the reference entropy; calculating the product of the reference entropy and a preset convergence coefficient to obtain a convergence threshold, wherein the value of the convergence coefficient ranges from 0.1 to 0.3; if the mixing stability entropy at the current time is less than the convergence threshold and the duration exceeds a preset holding time, then the mixing is determined to be complete, wherein the value of the holding time ranges from 3 seconds to 7 seconds.

[0019] Preferably, the constant speed range is determined as follows: the absolute value of the deviation between the real-time rotational speed of the motor shaft and the target rotational speed is less than a preset error band, and the duration of this state exceeds a preset stabilization time; wherein, the target rotational speed is included in the process formula, and the error band is set to two percent of the target rotational speed; the stabilization time ranges from 0.5 seconds to 3 seconds.

[0020] This invention transforms the fluctuation amplitude of a trend line into a probability distribution by calculating the first-order difference value of each point on the trend line and statistically analyzing its normalized probability. Then, it uses the natural logarithm to calculate the information entropy of this distribution as the mixing stationarity entropy, thus quantifying the disorder of the system using a single scalar value. This method utilizes the physical property that the rate of change of resistance during mixing tends to zero and the distribution gradually concentrates as uniformity increases. This allows the mixing stationarity entropy to serve as a sensitive indicator of whether the system has reached a steady state; that is, as mixing becomes more uniform, the entropy value decreases significantly, providing a robust quantitative basis for endpoint determination.

[0021] In a second aspect, the present invention provides an anomaly monitoring system for the operation of a non-invasive homogenizer, comprising a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the aforementioned anomaly monitoring method for the operation of a non-invasive homogenizer is implemented.

[0022] By adopting the above technical solution, a computer program is generated from the above-mentioned method for abnormal monitoring of the operation process of a non-invasive homogenizer, and stored in a memory for loading and execution by a processor. Terminal devices are then made based on the memory and processor for convenient use.

[0023] The beneficial effects of this invention are as follows:

[0024] This invention combines a pre-constructed inverse dynamics model to subtract the inertial torque and frictional torque generated by mechanical angular acceleration and real-time rotational speed from the orthogonal axis current of the motor, which characterizes the total electromagnetic torque. This achieves mathematical decoupling between material load and equipment mechanical load, obtaining a rheological resistance characteristic index that reflects only the change in material viscosity. This effectively solves the problem that weak material rheological signals are easily drowned out by the large fluctuations in the mechanical load signal of the equipment itself during variable speed processes. Furthermore, the singular spectrum analysis algorithm is used to process this index and extract the trend line. The stability of the trend line is quantified by combining information entropy and compared with a benchmark. This transforms the qualitative evaluation of mixing uniformity into a quantitative index that converges to the benchmark. Thus, it achieves high-precision, lag-free, and adaptive anomaly monitoring and endpoint determination of the homogenizer mixing process without contacting the material and without relying on a fixed-time formulation. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating an anomaly monitoring method for the operation of a non-invasive homogenizer according to the present invention;

[0026] Figure 2 This is a schematic diagram illustrating the comparison of the decoupling effects of dynamic characteristics;

[0027] Figure 3 This is an illustrative comparison diagram of the trend extraction algorithm of the present invention and existing algorithms;

[0028] Figure 4 This is a schematic diagram illustrating the evolution of the mixed stationarity entropy of the trend line and the adaptive endpoint determination process. Detailed Implementation

[0029] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0030] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0031] This invention discloses a method for monitoring anomalies during the operation of a non-invasive homogenizer, referring to... Figure 1 This includes steps S1-S4:

[0032] S1: The controller acquires the orthogonal shaft current and the real-time rotational speed of the homogenizer drive motor in real time, and calculates the mechanical angular acceleration based on the real-time rotational speed of the motor shaft.

[0033] It should be noted that the dynamic state of the homogenizer changes rapidly during operation, especially during acceleration and deceleration, when the load characteristics of the motor fluctuate drastically. If the sampling frequency is too low, key transient dynamic characteristics will be lost, making it impossible to accurately distinguish between inertial load and material load. Therefore, this embodiment adopts a high-frequency synchronous acquisition method to capture microsecond-level dynamic changes, providing a high-precision data foundation for dynamic decoupling.

[0034] Specifically, the controller reads the following data from the servo driver in real time via a high-speed industrial bus: (1) the motor quadrature shaft current, which represents the electromagnetic torque; (2) the real-time rotational speed of the motor shaft end, in radians per second; wherein the sampling frequency ranges from 1000 Hz to 5000 Hz, and in this embodiment, the sampling frequency is set to 2000 Hz; in other embodiments, the implementer can select a suitable sampling frequency according to the controller's processing capability and bus bandwidth.

[0035] Furthermore, based on the collected real-time rotational speed of the motor shaft, the mechanical angular acceleration is calculated. The mechanical angular acceleration is equal to the difference between the current real-time rotational speed of the motor shaft and the real-time rotational speed of the motor shaft at the previous sampling time, divided by the sampling period. Here, the mechanical angular acceleration characterizes the rate of velocity change, and this value is significantly non-zero when the equipment is in an acceleration or deceleration state. The larger the real-time rotational speed of the motor shaft, the greater the centrifugal force. The sampling period is the reciprocal of the sampling frequency, and the smaller the value, the higher the time resolution. This calculation process uses the difference method to approximate the derivative, which can reflect the acceleration state of the equipment rotor in real time.

[0036] S2: Based on the pre-built inverse dynamics model, the rheological drag characteristic index, after removing the mechanical load torque, is calculated according to the motor orthogonal shaft current, the real-time speed of the motor shaft end, and the mechanical angular acceleration; the rheological drag characteristic index is processed using the singular spectrum analysis algorithm.

[0037] It should be noted that since the total torque output by the motor is mainly used to overcome the three aspects of resistance: the rotor's own rotational inertia, the friction of the mechanical structure, and the rheological resistance of the material inside the tank, in variable speed processes, inertial torque and frictional torque dominate. Directly monitoring the total current cannot reflect the viscosity change of the material, leading to monitoring failure. Therefore, this embodiment establishes a physical inverse dynamics model to separate the mechanical load of the equipment itself from the total load, decoupling the pure resistance characteristics that are only related to the rheological properties of the material.

[0038] Specifically, after the equipment leaves the factory or is maintained, an unloaded calibration procedure is performed. Without placing a material tank, the motor is controlled to perform a full-speed operation including acceleration, constant speed, and deceleration, and several sets of data are collected. The sample size needs to be large enough to eliminate the influence of random noise on the regression results. Therefore, in this embodiment, the sample size is set to 2000. In other embodiments, the implementer can set the sample size according to the duration of the calibration process.

[0039] Furthermore, based on the principle of dynamic equilibrium, a least squares regression model is constructed to solve for the parameters.

[0040] First, construct the observation vector. It is a column vector composed of the motor orthogonal axis currents at all sampling times multiplied by the motor torque constant; where the motor torque constant is a known physical quantity used to convert current into torque; the value of the motor orthogonal axis current increases with the increase of load.

[0041] Secondly, construct the regression matrix. It is a matrix with the number of rows equal to the number of samples and the number of columns 3. The first column of the matrix is ​​the mechanical angular acceleration at all sampling times, the second column is the real-time rotational speed of the motor shaft at all sampling times, and the third column is the sign function value of the real-time rotational speed of the motor shaft at all sampling times. The sign function value is used to characterize the directionality of the friction force. It takes a value of 1 when the rotational speed is greater than zero, a value of -1 when it is less than zero, and a value of 0 when it is equal to zero.

[0042] Finally, the mechanical characteristic parameter vector is solved using pseudo-inverse matrix operations. The specific calculation expression is as follows:

[0043]

[0044] In the formula, The vector of mechanical characteristic parameters to be solved contains the mechanical characteristics of the system; This is the regression matrix; For observation vectors; This represents finding the inverse of a matrix.

[0045] Solving this expression yields ,in, The equivalent moment of inertia of the system characterizes the rotor's ability to resist changes in velocity; is the coefficient of viscous friction, which characterizes the resistance proportional to speed; Let be the Coulomb friction coefficient, which characterizes the inherent static friction resistance; store these parameters in the system constant library.

[0046] Specifically, the rheological drag characteristic index is calculated in real time using the inverse dynamics model. The specific calculation expression is as follows:

[0047]

[0048] In the formula, For a moment The rheological resistance characteristic index; the larger the value, the higher the viscosity of the material. The torque constant of the motor; For a moment The orthogonal axis current of the motor; This is the system's equivalent moment of inertia. For a moment Mechanical angular acceleration; It is the coefficient of viscous friction; For a moment The real-time rotational speed of the motor shaft end; Coulomb friction coefficient; The sign function is used to characterize the directionality of friction. It takes the value 1 when the rotational speed is greater than zero, -1 when it is less than zero, and 0 when it is equal to zero.

[0049] In this calculation expression, Indicates the time when the motor is The theoretical torque required to overcome rotor inertia and mechanical friction; derived from the total electromagnetic torque. The residual obtained by subtracting the theoretical torque from the middle This refers to the resistance torque generated solely by the flow of material within the tank, thereby enabling independent monitoring of the material's rheological properties during speed changes.

[0050] For example, such as Figure 2 As shown, a comparison diagram of the decoupling effect of dynamic characteristics is presented. Regarding the curve corresponding to the orthogonal axis current of the original motor in the diagram, it rises sharply during acceleration due to inertial impact and drops sharply during deceleration due to reverse braking. This large fluctuation masks the true change in material resistance. The rheological resistance characteristic index calculated by the method of this invention, after eliminating inertial and frictional loads, maintains a stable trend throughout the process, intuitively demonstrating the effective suppression of speed variation interference and proving the feasibility of monitoring material rheological properties under variable speed processes.

[0051] S3: By constructing a trajectory matrix, performing singular value decomposition and diagonal averaging operations, the trend line of the rheological drag characteristic index is extracted.

[0052] It should be noted that, due to the superposition of revolution and rotation during the operation of the planetary centrifuge homogenizer, the decoupled rheological resistance characteristic index still contains high-frequency mechanical periodic fluctuation noise. If a traditional low-pass filter is used to filter out this noise, it will introduce significant phase lag, resulting in a delay in the determination of the mixing endpoint and thus causing over-mixing. Therefore, this embodiment uses a singular spectrum analysis algorithm, which utilizes its zero-phase-shift data-driven characteristics to accurately extract the smooth trend representing the change in the mixing state from the noisy signal.

[0053] Specifically, a sliding time window method is used, based on the most recent Construct a trajectory matrix using all data points at each time point and their rheological drag characteristic indicators. Specifically:

[0054]

[0055] In the formula, Indicates the number of the current analysis window. The data point, from the first The moment and the The composition of rheological resistance characteristic indicators at each moment; , where represents the number of columns in the trajectory matrix; The current data window length represents the total number of historical data points participating in this analysis; its value determines the width of the time horizon. The embedded window length... This determines the fineness and frequency resolution of the algorithm's decomposition.

[0056] Among them, the length of the embedded window The orbital period should be an integer multiple of the orbital period to effectively separate periodic components; if If it is too small, it cannot effectively separate noise from the trend; if If the embedding window length is too large, the computational load increases dramatically and the algorithm's ability to capture transient changes decreases, leading to a deterioration in real-time performance; therefore, the embedding window length should be adjusted accordingly. The value range is from 50 to 200. In this embodiment, the value will be... Set to 100; in other embodiments, the implementer can set it according to the device's orbital frequency. The value of .

[0057] in, Need to meet To ensure the effectiveness of the decomposition; in this embodiment, The value is 400, meaning that the trajectory matrix is ​​constructed each time using the latest 400 data points.

[0058] The obtained trajectory matrix It has a Hankel structure, meaning that the elements on the antidiagonal are equal, which allows subsequent SVD decomposition to capture the dynamic features in the time series.

[0059] Specifically, a singular spectral analysis algorithm is used to perform singular value decomposition on the trajectory matrix containing the time delay information of the original time series. The specific calculation expression is as follows:

[0060]

[0061] In the formula, This is the trajectory matrix, which contains the time delay information of the original time series; For the first The eigenvalues ​​are arranged in descending order, with larger values ​​representing stronger energy of the corresponding components. For the first One left singular vector; For the first One right singular vector; This indicates the matrix transpose.

[0062] Furthermore, select the first principal component with the largest eigenvalue, that is, let Because during the mixing process, the slowly changing trend component has the highest energy, and the remaining components are considered as vibrational noise; the reconstruction matrix is ​​calculated, and the specific calculation expression is as follows:

[0063]

[0064] In the formula, To reconstruct the matrix, only the main trend information in the original signal was retained, while high-frequency noise was removed; It is the largest eigenvalue; This is the first left singular vector; It is the first right singular vector.

[0065] Specifically, based on the elements in the reconstructed matrix, a trend line is constructed, and the first trend line is... Points The calculation is divided into three cases:

[0066] Scenario 1, when When calculating the mean of the anti-diagonal line at the top left corner of the reconstructed matrix, the specific calculation expression is as follows:

[0067]

[0068] Scenario 2, when When calculating the mean of the anti-diagonal lines in the middle region of the reconstructed matrix, the specific calculation expression is as follows:

[0069]

[0070] Scenario 3, when When calculating the mean of the anti-diagonal line in the lower right corner of the reconstructed matrix, the specific calculation expression is as follows:

[0071]

[0072] In the formula, The first trend line The value of each point represents the rheological resistance after denoising at that moment; To reconstruct the first in the matrix Line 1 Column elements; Get the length of the embedded window and The smaller value in; Get the length of the embedded window and The larger value in the range.

[0073] By performing the diagonal averaging operation described above, the two-dimensional matrix is ​​transformed back into a one-dimensional time series, resulting in a smooth trend line that retains the characteristics of abrupt changes.

[0074] For example, such as Figure 3 As shown, a comparison diagram of the existing algorithm and the trend extraction algorithm of the present invention is presented. Among them, for the decoupled noisy rheological drag characteristic index, the trend line obtained by the traditional moving average filter is significantly lagging behind the changes in the original data. In contrast, the trend line extracted by the present invention using singular spectrum analysis closely follows the central trend of the data with almost no phase lag. This intuitively demonstrates that the present invention achieves zero-delay monitoring while ensuring the filtering effect, avoiding the judgment delay caused by signal lag.

[0075] S4: Calculate the information entropy of the first-order difference sequence of the trend line as the mixing stationarity entropy, and determine the mixing endpoint of the homogenizer operation process based on the convergence of the mixing stationarity entropy relative to the reference entropy, so as to realize the abnormal monitoring of the non-intrusive homogenizer operation process.

[0076] It should be noted that, since the essence of the mixing process is that the internal structure of the material tends to become ordered from disorder, the macroscopic manifestation is that the rate of change of rheological resistance gradually approaches zero and the fluctuation decreases. However, the absolute resistance values ​​of different batches of materials are different, and it is difficult to adapt using a fixed threshold. Therefore, this embodiment introduces information entropy to characterize the stability of the trend line and adopts a relative threshold judgment logic based on the initial state to achieve adaptive endpoint control without manual calibration.

[0077] First, to avoid misjudgments due to system instability during sudden speed changes, the system is configured with detection enable logic: only when the absolute value of the deviation between the real-time speed and the target speed is less than a preset error band, and the duration of this state exceeds a preset stabilization period, will subsequent entropy calculation be initiated, and this moment will be taken as the start time of the constant speed segment; where the target speed is included in the process formula, and the error band is set according to the control accuracy, and in this embodiment it is set to two percent of the target speed.

[0078] The preset stabilization time is used to confirm whether the servo motor has completely overcome mechanical inertia and entered a steady state, in order to filter out overshoot oscillations or transient jitters that may occur during speed adjustment. If the stabilization time is set too short, the system may mistakenly start monitoring before the motor speed has truly stabilized, causing incompletely decoupled inertial components to be mixed into the subsequently calculated rheological resistance characteristics, affecting the accuracy of the base entropy calculation. If the stabilization time is set too long, it will cause the monitoring system to intervene late, miss the key rheological change characteristics in the early stage of mixing, and reduce the time resolution of the endpoint determination. Therefore, the value range of the stabilization time is 0.5 seconds to 3 seconds. In this embodiment, the stabilization time is set to 1 second. In other embodiments, the implementer can adjust it according to the response bandwidth and mechanical rigidity of the homogenizer servo system.

[0079] Then, calculate the first-order difference sequence of the trend line, including the first-order difference value of each point on the trend line; then calculate the normalized probability of each point, which is equal to the absolute value of the first-order difference value of each point on the trend line, divided by the sum of the absolute values ​​of the first-order difference values ​​of all points on the trend line; this probability value represents the proportion of the change in that point in the overall change, reflecting how fast the resistance changes at that moment.

[0080] Furthermore, the Shannon entropy principle is used to calculate the mixing stability entropy, which is equal to the negative of the sum of the products of the normalized probabilities of all points on the trend line and the natural logarithm of those probabilities. The smaller this value, the more ordered the system is. In this calculation process, when mixing is incomplete, the resistance changes drastically and irregularly, resulting in a relatively uniform distribution of the difference values ​​and a higher calculated entropy value. When mixing is close to uniform, the resistance tends to be constant, the difference values ​​approach zero noise, the distribution is concentrated, and the calculated entropy value is significantly reduced.

[0081] Furthermore, the average value of the mixing stability entropy at all moments before the start of the constant velocity phase is calculated as the baseline entropy; at this point, the material is not yet uniformly mixed, and the entropy value is at a relatively high level. This baseline entropy reflects the initial disorder of the material at this time.

[0082] Furthermore, a convergence threshold is set, which is equal to the convergence coefficient multiplied by the baseline entropy. The convergence coefficient is used to determine the rate of decrease relative to the initial state. If the convergence coefficient is too large, it may lead to premature shutdown and insufficient mixing. If the convergence coefficient is too small, it may lead to excessively long decision time and reduced efficiency. Therefore, the value range of the convergence coefficient is 0.1 to 0.3. In this embodiment, the convergence coefficient is set to 0.2. In other embodiments, the implementer can adjust the coefficient according to the requirements of mixing uniformity.

[0083] Furthermore, it is determined whether the current mixing stability entropy is less than the convergence threshold and whether the duration of this state exceeds the preset holding time; wherein, the holding time is used to prevent accidental fluctuations from causing misjudgment, and its value ranges from 3 seconds to 7 seconds. In this embodiment, the holding time is set to 4 seconds.

[0084] If the above conditions are met, the mixing is considered complete, the system automatically issues a stop command, and records the current duration as the optimal process parameters for this batch of materials.

[0085] For example, such as Figure 4 As shown, a schematic diagram illustrates the evolution of the mixing stability entropy and the adaptive endpoint determination process of the trend line. The curve corresponding to the mixing stability entropy remains at a high level and exhibits large fluctuations in the early stage of mixing due to the drastic changes in resistance. As the mixing process progresses, the material gradually becomes uniform, and the entropy value shows a significant downward trend. When the entropy value falls below the convergence threshold and persists for a certain period of time, the system marks the accurate trigger signal for the mixing endpoint, realizing data-driven adaptive intelligent control.

[0086] This invention also discloses an anomaly monitoring system for the operation of a non-invasive homogenizer, comprising a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement an anomaly monitoring method for the operation of a non-invasive homogenizer according to the present invention.

[0087] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A method of monitoring the operation of a non-invasive homogenizer for abnormalities, characterized in that, The application relates to a non-invasive homogenizer operation process abnormality monitoring method. The orthogonal shaft current of a homogenizer driving motor and the real-time rotating speed of a motor shaft end are acquired in real time through a controller, and mechanical angular acceleration is calculated according to the real-time rotating speed of the motor shaft end; Based on a pre-constructed inverse dynamics model, a rheological resistance characteristic index stripped of mechanical load torque is calculated according to the motor orthogonal shaft current, the real-time rotating speed of the motor shaft end and the mechanical angular acceleration; The rheological resistance characteristic index is processed by using a singular spectrum analysis algorithm, a trajectory matrix is constructed, singular value decomposition and diagonal averaging operation are performed, and a trend line of the rheological resistance characteristic index is extracted, including: performing singular value decomposition on the trajectory matrix, selecting a first principal component with the largest eigenvalue to calculate a reconstruction matrix; converting the reconstruction matrix into a one-dimensional time series by using a diagonal averaging method to obtain the trend line; the diagonal averaging method includes calculating the element mean values of the upper left anti-diagonal line, the middle region anti-diagonal line and the lower right anti-diagonal line of the reconstruction matrix respectively according to the positions of the elements in the reconstruction matrix; the calculation formula of the reconstruction matrix is: , is the reconstruction matrix, is the largest eigenvalue, is the first left singular vector, is the first right singular vector; the eigenvalue, the first left singular vector and the first right singular vector are all results of singular value decomposition; The information entropy of the first-order difference sequence of the trend line is calculated as the mixed stationarity entropy, and the mixed end point of the homogenizer operation process is determined according to the convergence of the mixed stationarity entropy relative to the reference entropy, so as to realize abnormality monitoring of the non-invasive homogenizer operation process; The mixed end point of the homogenizer operation process is determined according to the convergence of the mixed stationarity entropy relative to the reference entropy, which comprises the following steps: when the real-time rotating speed of the motor shaft end enters a constant speed section, the average value of the mixed stationarity entropy at all time points before the start time of the constant speed section is calculated as the reference entropy; the product of the reference entropy and a preset convergence coefficient is calculated to obtain a convergence threshold; the value range of the convergence coefficient is 0.1-0.3; if the mixed stationarity entropy at the current time point is less than the convergence threshold and the duration exceeds a preset holding time, it is determined that the mixing is completed, and the value range of the holding time is 3-7 seconds.

2. The method of claim 1, wherein, The pre-constructed inverse dynamics model is obtained by the following method: The motor is controlled to perform a no-load full-speed section operation action containing acceleration, uniform speed and deceleration, and multiple groups of sample data are collected; An observation vector and a regression matrix are constructed, the observation vector is composed of the product of the motor orthogonal shaft current and the motor torque constant in the sample data, and the regression matrix is composed of the mechanical angular acceleration, the real-time rotating speed of the motor shaft end and the sign function value of the real-time rotating speed of the motor shaft end in the sample data; The mechanical characteristic parameters of the system, including the equivalent rotational inertia, the viscous friction coefficient and the Coulomb friction coefficient of the system, are solved by using a least square method regression model.

3. A method of monitoring the operation of a non-invasive homogenizer according to claim 2, characterized in that The calculation formula of the rheological resistance characteristic index is: ； In the formula, is the rheological resistance characteristic index of the moment of time ; is the motor torque constant; is the motor orthogonal axis current of the moment of time ; is the equivalent moment of inertia of the system; is the mechanical angular acceleration of the moment of time ; is the viscous friction coefficient; is the real-time rotational speed of the motor shaft end of the moment of time ; is the Coulomb friction coefficient; is the sign function, and the sign function value is used to represent the directionality of the friction. When the rotational speed is greater than zero, the value is 1; when the rotational speed is less than zero, the value is -1; and when the rotational speed is equal to zero, the value is 0.

4. The method of claim 1, wherein, The trajectory matrix is constructed, which comprises: Using a sliding time window, based on the most recent Construct a trajectory matrix using all data points representing each moment and its rheological drag characteristic indicators. Specifically: ； In the formula, represents the first data point in the current analysis window, composed of the flow resistance characteristic index of the first time point and the second time point. represents the number of columns of the trajectory matrix; is the length of the current analysis data window, is the embedding window length, and the embedding window length should be an integer multiple of the revolution period, and the value range is 50 to 200, needs to meet to ensure the effectiveness of decomposition.​​​ 5. The method of claim 1, wherein, The information entropy of the first-order difference sequence of the trend line is calculated as the mixed stationarity entropy, which comprises: The first-order difference value of each point on the trend line is calculated, and the proportion of the absolute value of the first-order difference value of each point in the sum of the absolute values of all points is calculated as a normalized probability; The reciprocal of the sum of the product of the normalized probability of all points and the natural logarithm of the normalized probability is calculated as the mixed stationarity entropy.

6. The method of claim 1, wherein, The determination condition of the constant speed section is that the absolute value of the deviation of the real-time rotating speed of the motor shaft end and a target rotating speed is less than a preset error band, and the state duration exceeds a preset stable time length; wherein the target rotating speed is contained in a process formula, the error band is set as two percent of the target rotating speed; and the value range of the stable time length is 0.5-3 seconds.

7. A system for monitoring the operation of a non-invasive homogenizer for abnormalities, comprising: The application further provides a non-invasive homogenizer operation process abnormality monitoring device. The device comprises a processor and a memory, and the memory stores computer program instructions which are executed by the processor to realize the non-invasive homogenizer operation process abnormality monitoring method.

Images