Intelligent control method for ice slurry particle size uniformity

Abstract

The present application relates to an intelligent control method for the uniformity of ice particle size. By arranging an infrared thermopile sensor array on the screening conveying line, the transient thermal radiation signals generated by heat conduction of ice particles are collected. After complex space-time correlation, denoising, normalization and interpolation processing, standardized heat dissipation data are obtained. Then, the time series convolution network is used to extract the thermal decay slope and other thermodynamic characteristics, and the particle size distribution parameters are inferred in real time combined with the calibration database. The system automatically determines the particle size dispersion degree, generates dynamic aperture gradient adjustment instructions, and realizes millimeter-level aperture distribution adaptive adjustment through the electro-deformation flexible screen, achieving continuous, fine classification and particle size homogenization of ice particle flow, effectively improving the screening accuracy and intelligent level, and having the ability of dynamic environmental adaptation and closed-loop optimization.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for particle size screening, and in particular to an intelligent control method for improving the uniformity of particle size in shaken ice. Background Technology

[0002] Currently, the requirements for the uniformity of ice particle size in industrial ice making and cold chain logistics are increasingly stringent. Traditional ice particle size screening processes generally employ fixed screen structures, combined with physical control methods such as screen vibration or tilt angle. They utilize the movement of ice particles on the screen surface and the gravity-driven screening process to separate ice particles of different sizes. However, most of these methods belong to open-loop control systems, meaning they rely solely on the initially set geometric parameters and vibration amplitude and frequency of the equipment. The actual particle size distribution of the finished product during the screening process cannot be fed back to the control system in real time for adaptive parameter adjustment. Once fluctuations occur in the particle size distribution, moisture content, or ambient temperature and humidity between batches of raw ice particles, traditional solutions struggle to respond promptly, leading to amplified particle size dispersion in the screened finished product. This results in non-conforming phenomena such as "large particles passing through the screen" and "small particles being mixed in," ultimately affecting product uniformity and the quality of downstream applications.

[0003] Currently, the mainstream ice particle size screening and control technologies mainly include three approaches: First, relying on multi-stage screens and graded conveying structures, coarse particle size classification is achieved by changing screens and adjusting channels. However, this method involves frequent switching, high energy consumption, and limited automation. Second, configuring online or offline image recognition systems and laser particle size analyzers for particle size detection. The results are mostly lagging information, with insufficient real-time feedback, and factors such as high humidity, low temperature, and poor visibility limit the accuracy of image recognition. Third, indirectly estimating particle size distribution through macroscopic indicators such as material weighing and time measurement, but this method suffers from low resolution and poor environmental adaptability. Although these solutions have some versatility, none of them have established a real-time adaptive closed-loop control relationship for the screening process. In actual industrial production scenarios, they exhibit significant shortcomings such as detection lag, slow adjustment, and complex maintenance.

[0004] In summary, existing technologies generally face the following technical shortcomings and application bottlenecks: First, the screening process lacks a closed-loop adaptive feedback mechanism based on the particle size distribution of the finished product, making it difficult to correct screening deviations in a timely manner. Second, current detection methods mainly rely on lagging information such as images and weighing, resulting in poor real-time performance. Furthermore, they are subject to environmental factors such as fluctuations in temperature and humidity, and condensation on the surface of ice particles, significantly impacting the accuracy of perception. Third, the control mechanisms primarily rely on macroscopic physical quantities, lacking refined and dynamic adjustment methods for the spatial distribution of sieve openings. Fourth, multi-stage mechanical adjustment structures are bulky, complex to maintain, and difficult to automate, making them unsuitable for high-cleanliness, rapid-response scenarios such as cold chain logistics. Summary of the Invention

[0005] This application provides an intelligent control method for the uniformity of ice particle size in shaken ice, aiming to solve one of the problems or issues of the prior art mentioned in the background section above.

[0006] The intelligent control method for improving the uniformity of ice particle size in shaken ice provided in this application specifically includes: S1: Miniaturized infrared thermopile sensor arrays are deployed at the feed end, middle section and discharge end of the ice particle screening conveyor line to collect transient thermal radiation attenuation curves caused by the ambient temperature difference during the residence of ice particles on the screen surface, and generate a three-dimensional raw heat flow signal set containing spatial location coordinates, time series stamps and heat flow intensity values.

[0007] S2: Perform denoising and normalization preprocessing on the three-dimensional original heat flow signal set to eliminate background thermal noise interference and unify the dimensions, generating a standardized transient heat dissipation spatiotemporal distribution matrix as the reference data source for subsequent feature extraction.

[0008] S3: Input the standardized transient heat dissipation spatiotemporal distribution matrix into a temporal convolutional network for feature mining, and extract three core thermodynamic feature vectors for each monitoring area: heat attenuation slope, plateau duration, and hot spot spatial dispersion.

[0009] S4: Based on the standard particle size ice sample calibration mapping relationship data table, the core thermodynamic feature vector is mapped and converted into a set of statistical parameters characterizing the dispersion of the current cross-section ice particle size distribution, generating a real-time particle size distribution descriptor containing skewness coefficient, peak width index and main peak displacement.

[0010] S5: Determine whether the peak width index in the real-time particle size distribution descriptor continuously exceeds the preset particle size dispersion threshold. If the determination result is yes, trigger the pore size gradient adjustment command; otherwise, maintain the current screen shape and keep the monitoring state.

[0011] S6: Calculate the asymmetric strain distribution curve required in the length direction of the screen surface according to the aperture gradient adjustment command, and generate a gradient voltage control sequence to drive the flexible electro-deformable screen structure to produce millimeter-level aperture gradient changes.

[0012] S7: Based on the gradient voltage control sequence, drive the screen to generate asymmetric strain, and construct a dynamic aperture distribution zone with inlet tightening, middle expansion and outlet convergence.

[0013] S8: Based on the dynamic aperture distribution band, the flow of ice particles passing through the sieve surface is intercepted and diverted, and the particle size distribution is optimized in a single sieving cycle. The real-time particle size distribution descriptor is updated to enter the next round of feedback adjustment cycle.

[0014] The intelligent control method for improving the uniformity of ice particle size in shaken ice provided in this application has the following beneficial effects: (1) To address the problems of detection lag, open-loop feedback, and sluggish adjustment caused by the reliance on offline sampling, image recognition, or weighing statistics in traditional ice particle screening processes, this solution explores the non-uniform micro-melting characteristics exhibited by ice particles during transport in a pre-cooling environment after cryogenic discharge, and constructs a real-time sensing and closed-loop control mechanism based on the thermodynamic behavior of the ice particles themselves. Small-diameter ice particles are more prone to significant surface micro-melting on the vibrating screen surface due to their large specific surface area and low heat capacity, forming a thicker melt water layer and affecting local thermal radiation characteristics; while large-diameter ice particles have a slow thermal response, a thin melt water layer, and are easy to detach. The two exhibit distinguishable transient decay modes in their infrared thermal radiation responses. By deploying a miniaturized infrared thermopile sensor array at key locations on the sieve body, the system can non-contactly acquire three-dimensional data clusters of space, time, and thermal intensity during the residence of ice particles on the sieve surface. It then utilizes a temporal convolutional network to extract core dynamic features such as thermal decay slope, plateau duration, and hot spot dispersion, which are subsequently mapped to the skewness coefficient, peak width index, and main peak displacement of the current cross-section particle size distribution, achieving millisecond-level online identification of particle size dispersion. This method completely eliminates reliance on externally stimulated sensing technologies such as image recognition, laser diffraction, or acoustic emission, avoiding problems such as interference from complex ambient light, equipment contamination failure, and signal delay, significantly improving detection reliability and response speed.

[0015] (2) In terms of execution end design, this scheme abandons the traditional technical path of optimizing the screening effect by adjusting the vibration frequency, screen tilt angle or multi-level mechanical displacement, and innovatively introduces an electro-deformable composite material layer integrated into the screen substrate to realize the dynamic reconstruction of the spatial distribution of the screen holes. This functional layer is composed of conductive nanofiber fabric and PVDF-TrFE piezoelectric film. After applying gradient voltage, it generates asymmetric strain, which drives the upper conductive mesh to form controllable folds, thereby precisely controlling the geometry of the screen holes in the length direction of the screen surface, generating a dynamic aperture distribution zone of "inlet tightening - middle expansion - outlet convergence", with an accuracy of up to 0.1 mm. This structure does not change the overall open area ratio, but only adjusts the probability of ice particles of different sizes to pass through by reshaping the spatial gradient of the aperture: smaller ice particles are effectively intercepted in the inlet area to prolong the screening time, larger ice particles obtain higher permeability in the middle section, and finally achieve natural convergence and concentrated output of particle size distribution at the outlet. The entire adjustment process utilizes the latent heat of phase change of ice particles and the ambient temperature difference, eliminating the need for additional heating or cooling devices. All actions are completed within a single screening cycle (≤1.2 seconds), with a response delay less than one-fifth that of traditional PID control systems, significantly enhancing the system's real-time performance and energy efficiency. This mechanism also avoids high-energy-consuming and low-efficiency operations such as material return circulation, airflow compensation, grading weights, and screen replacement, forming a closed-loop control system with autonomous response across the entire chain from sensing to execution.

[0016] The aforementioned technical approaches collectively construct a novel closed-loop sieving system based on the intrinsic thermophysical properties of ice particles, executed by a piezoelectrically driven deformation structure, and centered on an edge intelligent model. This achieves a paradigm shift from "passive sorting" to "active adaptation." The system boasts several advantages, including no visual guidance required, no reliance on deep learning training, no mechanical moving parts, and low power consumption. It significantly improves sieving accuracy and stability, and greatly enhances adaptability and robustness in low-temperature and humid industrial environments. This provides a highly responsive, highly reliable, and self-sustaining intelligent particle size control solution for industries such as frozen food, cold chain ice making, and pharmaceutical quick-freezing. Attached Figure Description

[0017] Figure 1 This is the main flowchart of an intelligent control method for improving the uniformity of ice particle size in shaken ice.

[0018] Figure 2 This is a sub-flowchart of an intelligent control method for improving the uniformity of ice particle size in shaken ice.

[0019] Figure 3 This is another sub-flowchart of the intelligent control method for ensuring uniformity of ice particle size in shaken ice.

[0020] Figure 4 This is an application environment diagram of an intelligent control method for the uniformity of ice particle size in one embodiment.

[0021] Figure 5 An internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0022] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0023] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0024] like Figure 1 As shown, this application provides an intelligent control method for improving the uniformity of ice particle size in shaken ice, specifically including: S1: Miniaturized infrared thermopile sensor arrays are deployed at the feed end, middle section and discharge end of the ice particle screening conveyor line to collect transient thermal radiation attenuation curves caused by the ambient temperature difference during the residence of ice particles on the screen surface, and generate a three-dimensional raw heat flow signal set containing spatial location coordinates, time series stamps and heat flow intensity values.

[0025] S2: Perform denoising and normalization preprocessing on the three-dimensional original heat flow signal set to eliminate background thermal noise interference and unify the dimensions, generating a standardized transient heat dissipation spatiotemporal distribution matrix as the reference data source for subsequent feature extraction.

[0026] S3: Input the standardized transient heat dissipation spatiotemporal distribution matrix into a temporal convolutional network for feature mining, and extract three core thermodynamic feature vectors for each monitoring area: heat attenuation slope, plateau duration, and hot spot spatial dispersion.

[0027] S4: Based on the standard particle size ice sample calibration mapping relationship data table, the core thermodynamic feature vector is mapped and converted into a set of statistical parameters characterizing the dispersion of the current cross-section ice particle size distribution, generating a real-time particle size distribution descriptor containing skewness coefficient, peak width index and main peak displacement.

[0028] S5: Determine whether the peak width index in the real-time particle size distribution descriptor continuously exceeds the preset particle size dispersion threshold. If the determination result is yes, trigger the pore size gradient adjustment command; otherwise, maintain the current screen shape and keep the monitoring state.

[0029] S6: Calculate the asymmetric strain distribution curve required in the length direction of the screen surface according to the aperture gradient adjustment command, and generate a gradient voltage control sequence to drive the flexible electro-deformable screen structure to produce millimeter-level aperture gradient changes.

[0030] S7: Based on the gradient voltage control sequence, drive the screen to generate asymmetric strain, and construct a dynamic aperture distribution zone with inlet tightening, middle expansion and outlet convergence.

[0031] S8: Based on the dynamic aperture distribution band, the flow of ice particles passing through the sieve surface is intercepted and diverted, and the particle size distribution is optimized in a single sieving cycle. The real-time particle size distribution descriptor is updated to enter the next round of feedback adjustment cycle.

[0032] Step S1: Miniaturized infrared thermopile sensor arrays are deployed at the inlet, middle, and outlet ends of the ice particle screening conveyor line to collect transient thermal radiation attenuation curves caused by environmental temperature differences during the residence of ice particles on the screen surface, generating a three-dimensional raw heat flow signal set containing spatial location coordinates, time series stamps, and heat flow intensity values. Specifically, this includes: S1.1: Based on the physical layout characteristics of the ice particle screening conveyor line, spatial coordinate calibration is performed on the three monitoring areas of the feed end, middle section and discharge end to determine the three-dimensional position parameters of each area relative to the screen surface reference point, and a monitoring point configuration table containing area identifiers and spatial coordinate mapping relationships is generated as the spatial reference basis for sensor deployment.

[0033] Based on the physical layout characteristics of the ice particle screening conveyor line, a three-dimensional Cartesian coordinate system is established with the screen surface reference point as the spatial origin, and the physical structural boundaries of the inlet, middle section, and outlet are used as the initial calibration objects. Structural mapping is performed within this coordinate system, and a laser rangefinder is used to obtain the precise distance values ​​of each monitoring area relative to the screen surface reference point along the length, width, and height directions, which are then simultaneously recorded in the coordinate database.

[0034] Based on the coordinate database dataset, error calibration is performed on the three-dimensional position parameters of each monitoring area. The results of repeated measurements at multiple points are introduced for least squares fitting calculation to eliminate coordinate errors caused by rangefinder measurement noise and achieve stable spatial parameter output.

[0035] For the obtained calibration three-dimensional position parameters, perform region identifier mapping processing, assign a unique identifier to the feed end, middle section and discharge end in sequence and bind it to the corresponding three-dimensional coordinates to form a one-to-one mapping relationship table between region identifiers and spatial coordinates.

[0036] By introducing a coordinate resolution factor into the mapping table, the coordinate values ​​are uniformly normalized, so that the relative distance between each sensor deployment point conforms to the selected spatial resolution standard, which is beneficial for subsequent array deployment optimization calculations.

[0037] The normalized mapping table is encapsulated in a structured manner to generate a monitoring point configuration table that can be used by sensor deployment, including an identifier for each region, the corresponding three-dimensional position parameters, and the coordinate accuracy level identifier.

[0038] Through the above-mentioned spatial coordinate modeling, error calibration, identifier binding, normalization processing and structural encapsulation, the physical layout data constructed in the previous step is transformed into a monitoring point configuration table that can directly guide the layout of miniaturized infrared thermopile sensor arrays, thereby achieving high-precision output of sensor deployment benchmark data.

[0039] S1.2: According to the monitoring point configuration table, miniaturized infrared thermopile sensor arrays are installed at the determined feed end, middle section and discharge end positions respectively. Using the optical focusing component and thermopile detection unit inside the sensor, the transient thermal radiation energy generated by the uneven local thermal conduction on the surface of the ice particles passing through the screen is captured with high sensitivity, and the original heat flow intensity time series data stream containing the time dimension is generated.

[0040] Based on the monitoring point configuration table, at the inlet, middle, and outlet positions, mounting fixtures for miniaturized infrared thermopile sensor arrays are loaded. The sensor base is precisely embedded into the corresponding spatial coordinate position locking mechanism, and three-dimensional attitude adjustment is performed to ensure that the optical axis forms a design angle with the normal direction of the screen surface to achieve the optimal heat flow capture field of view. The focal length of the sensor's internal optical focusing components is fine-tuned, and the spot positioning is corrected. The prism tilt angle and lens group position are adjusted according to the field of view coverage parameters set based on the screen width and ice particle flow density to ensure that infrared radiation energy forms a high-energy-density focal point on the detection plane and is uniformly projected onto the sensitive area of ​​the thermopile detection unit. The thermopile detection unit undergoes response threshold calibration. The correspondence between the thermopile output voltage and radiation power is established using the standard radiation intensity of a known heat source. Gain adjustment through the internal amplification circuit ensures that the low-power radiation signal maintains a high signal-to-noise ratio during time-series sampling. In the sensor signal processing module, a high-precision sampling clock is configured and the sampling frequency is set. A time-series buffering strategy is used to continuously store the transient thermal radiation signals generated by each ice particle group passing through the screen surface along the time dimension, forming a heat flow intensity data stream containing time-series stamps. Through the above-mentioned chain processing method of installation positioning, optical focusing, detection calibration, and time-series acquisition, the spatial reference information in the monitoring point configuration table is transformed into a continuously outputting high-precision time-dimensional heat flow intensity data stream, achieving highly sensitive capture of the transient thermal radiation energy generated by the ice particle group passing through the screen surface due to local uneven heat conduction.

[0041] For example, three miniaturized infrared thermopile sensor arrays are installed at the feed end, middle section, and discharge end of the ice particle screening conveyor line. The sensor at the feed end is installed 0.5 meters from the reference point on the screen surface, the sensor at the middle section is installed 2.0 meters from the reference point, and the sensor at the discharge end is installed 4.5 meters from the reference point. The optical components of each sensor use a combination of aspherical lenses with a focal length of 5 mm, and the field of view angle is set to 15 degrees to cover 80% of the screen width. The thermopile detection unit is calibrated to obtain the transfer function of output voltage and radiated power. When the radiated power is 0.02 watts, the output voltage is 4.5 millivolts. The gain of the amplifier circuit is set to ×200 to make the signal amplitude reach 0.9 volts. The sampling clock frequency is set to 200 Hz, and the timing buffer capacity is a 2-second window length. In actual operation, the transient heat flux intensity curves captured by each set of sensors exhibit obvious peak characteristics when ice particles pass through. The peak value at the feed end appears at point 30 of the sampling sequence, the peak value in the middle section appears at point 50, and the peak value at the discharge end appears at point 80. The peak amplitude is positively correlated with the degree of micro-melting on the surface of the ice particles. After conversion, three sets of heat flux intensity data streams with timestamps are formed, which are used for subsequent S1.3 spatiotemporal correlation processing to bind the spatial position and time information of the sensors, verifying that the high-sensitivity capture mechanism can significantly improve the input accuracy of particle size distribution inversion under dynamic production conditions.

[0042] S1.3: Perform spatiotemporal correlation processing on the original heat flux intensity time series data stream, bind the heat flux intensity value at each sampling time with the corresponding sensor spatial location coordinates and high-precision time series stamp, construct a transient thermal radiation decay curve cluster data object with three-dimensional attributes, and form a three-dimensional original heat flux signal set that can completely characterize the heat dissipation behavior of ice particle groups during their residence on the sieve surface.

[0043] Step S2: Perform denoising and normalization preprocessing on the original three-dimensional heat flow signal set to eliminate background thermal noise interference and unify the dimensions, generating a standardized transient heat dissipation spatiotemporal distribution matrix as the benchmark data source for subsequent feature extraction. Specifically, this includes: S2.1: Perform filtering processing based on wavelet threshold denoising method on the three-dimensional original heat flow signal set containing spatial location coordinates, time series stamps and heat flow intensity values ​​to separate and remove high-frequency random noise components composed of environmental electromagnetic interference and equipment background noise, and obtain a cluster of pure transient thermal radiation attenuation curves after noise reduction.

[0044] For the three-dimensional raw heat flow signal set collected from sensors at the inlet, middle section, and outlet, a complete data object containing spatial location coordinates, time series stamps, and heat flow intensity values ​​is selected as the input condition. This sub-step is used in the main step to achieve high-fidelity denoising processing of multi-source signals. Based on the non-stationary characteristics of industrial heat flow signals, the multi-scale decomposition module of the wavelet threshold denoising method is called to perform discrete wavelet transform on the time series heat flow intensity data of each monitoring point, obtaining the decomposition coefficient matrix in different frequency bands. Referring to the preset threshold selection principle, a soft threshold function is used to compress the amplitude of the high-frequency component coefficients to eliminate random noise components composed of environmental electromagnetic interference and equipment background noise to the greatest extent. The coefficients of each frequency band after threshold processing are subjected to inverse wavelet transform operation to reconstruct the denoised time series heat flow intensity curve, and the original spatial location coordinate mapping relationship is maintained during the reconstruction process to ensure that the denoising result corresponds accurately with the physical layout. Energy preservation verification was used to calculate the global energy of the reconstructed curve and compared with the low-frequency energy of the curve before denoising, verifying that the denoising process did not destroy the main information of ice particle heat dissipation. The denoised time series curves were aggregated according to the sampling time sequence to generate a cluster of clean transient thermal radiation decay curves, which served as the direct input for subsequent background baseline drift correction. Wavelet threshold filtering was used to transform the original three-dimensional heat flow signal set from the previous step into a cluster of clean transient thermal radiation decay curves with a high signal-to-noise ratio, achieving the expected technical effect of removing high-frequency noise while retaining effective heat dissipation information.

[0045] S2.2: Based on the noise-reduced pure transient thermal radiation attenuation curve cluster, perform background baseline drift correction operation, dynamically calculate the background heat flux value when no ice particles pass through using the sliding window mean estimation method and subtract it from the original signal to generate a relative heat flux intensity change sequence that eliminates the influence of environmental temperature and humidity fluctuations.

[0046] The purified transient thermal radiation attenuation curve cluster, after noise reduction, is input into the background baseline drift correction module. The detection sample interval during which no ice particles pass through is designated as the environmental background reference set. The heat flux values ​​in the reference set are sorted by time series. A sliding window mean estimation method is applied to calculate the local expected value of the environmental background heat flux within each window. The window length is set to an integer multiple of the sensor sampling frequency to ensure time smoothing characteristics. The sequence of local expected values ​​output by the sliding window is used as a dynamic baseline vector, aligned point-by-point with the purified transient thermal radiation attenuation curve cluster. Baseline subtraction is performed, subtracting the baseline value at the corresponding time from the heat flux intensity value at each time sampling point in the curve cluster, generating a sequence of relative heat flux intensity changes with suppressed environmental noise. This sequence preserves the true heat dissipation fluctuation signal caused by uneven surface heat conduction during the passage of ice particles through the sieve, thus eliminating environmental temperature and humidity drift.

[0047] S2.3: Perform multi-channel gain equalization processing on the relative heat flux intensity change sequence, construct an inverse gain compensation matrix based on the factory sensitivity coefficient of each miniaturized infrared thermopile sensor array and perform matrix multiplication to eliminate the response inconsistency between different sensor units and output a uniform absolute heat dissipation rate dataset.

[0048] For the relative heat flux intensity change sequence after background baseline drift correction, the factory sensitivity coefficient data table of each miniaturized infrared thermopile sensor array stored in the edge computing unit is called, and the corresponding sensitivity coefficient is extracted according to the sensor unit identifier to form an ordered coefficient vector set.

[0049] Based on this coefficient vector set, an inverse gain compensation matrix is ​​constructed. The matrix elements are in the form of the reciprocals of each sensitivity coefficient, which are used to amplify the output of the low-sensitivity unit and suppress the over-response of the high-sensitivity unit.

[0050] The relative heat flux intensity change sequence is mapped to the corresponding matrix rows and columns according to timestamps and spatial location indices. Multi-channel gain equalization is then achieved through matrix multiplication, as shown in the following formula: Where R is the equalized absolute heat dissipation rate dataset, M is the inverse gain compensation matrix, and S is the original relative heat flux intensity change sequence matrix.

[0051] Perform inter-channel amplitude consistency verification on the matrix multiplication results, calculate the difference between the mean signal of each channel and the global mean, and perform secondary proportional correction on channels whose differences exceed the amplitude consistency tolerance to avoid local abnormal response.

[0052] The equalization results, after amplitude consistency verification and correction, are encapsulated into a dimensionally unified absolute heat dissipation rate dataset, which serves as the input benchmark for subsequent spatiotemporal alignment interpolation processing.

[0053] By using multi-channel gain equalization processing, the relative heat flux intensity change sequence from the previous step is transformed into an absolute heat dissipation rate dataset with consistent response and unified physical quantities, thus achieving complete comparability of multi-source sensor data at the dimensional level.

[0054] S2.4: Based on the dimensionally unified absolute heat dissipation rate dataset, perform spatiotemporal alignment interpolation resampling processing, use bilinear interpolation to map the non-uniformly distributed spatial location coordinates to the standard grid coordinate system and synchronize the asynchronous time series stamps to a unified clock reference, and generate a discretized heat flux spatiotemporal tensor with a regular topology.

[0055] S2.5: Perform extreme value normalization scaling on the discretized heat flux spatiotemporal tensor, select the global maximum heat flux intensity value and the minimum heat flux intensity value as scaling boundaries to compress the data numerical domain to the standard range of zero to one, and finally generate a standardized transient heat dissipation spatiotemporal distribution matrix as the direct input object for subsequent feature mining.

[0056] like Figure 2 As shown, step S3 involves inputting the standardized transient heat dissipation spatiotemporal distribution matrix into a temporal convolutional network for feature mining, extracting three core thermodynamic feature vectors for each monitoring area: the heat attenuation slope, the duration of the plateau period, and the spatial dispersion of hot spots. Specifically, this includes: S3.1: Based on the standardized transient heat dissipation spatiotemporal distribution matrix generated in the previous steps, a one-dimensional causal convolutional layer is used to perform local receptive field sliding window calculation on the heat flow intensity values ​​in the time series dimension to capture the short-term thermal radiation attenuation trend of ice particle groups during their residence on the sieve surface, generating a shallow temporal feature map containing details of high-frequency thermal fluctuations.

[0057] Based on the standardized transient heat dissipation spatiotemporal distribution matrix obtained in the preceding steps, the time series dimension of the matrix is ​​used as the input channel parameter of the one-dimensional causal convolutional layer. This ensures that the network structure performs convolution operations only within the current and historical sampling frames to avoid future information leakage. A local receptive field sliding window calculation is performed on this matrix. The window length and stride are determined based on the actual timescale of ice particles' residence on the sieve surface and the sampling frequency. The heat flux intensity values ​​contained within the window are arranged into a one-dimensional vector in chronological order for the convolution kernel to calculate. For each local window, the short-term thermal radiation decay pattern is captured through element-wise multiplication and summation of the convolution kernel weights and the heat flux intensity vector. The convolution kernel weights are initialized according to a Gaussian distribution and adaptively updated during network training to enhance sensitivity to high-frequency fluctuation details. To further quantify the decay trend, a linear regression slope calculation is performed on the local time series output by the convolution within a specified length range. Its mathematical expression is: Where, x i y is the time index value. i For the corresponding heat flux intensity value, μ is the mean of the time index sequence, and ν is the mean of the heat flux intensity sequence. The calculation result serves as the thermal decay slope feature value for this local window. All convolution window outputs and corresponding slope features are mapped back to the time axis of the original matrix according to their time position index, forming a shallow temporal feature map containing details of high-frequency thermal fluctuations. This ensures that subsequent dilated convolutions can utilize high-resolution feature inputs over short time periods, connecting with the fusion analysis of spatial location dimensions. Through the above processing method, the standardized matrix from the previous step is transformed into a feature map with short-term sensitivity and quantitative indicators of decay trends, achieving high-precision capture of changes in the thermal radiation of ice particle groups.

[0058] For example, in the monitoring of ice particle groups entering the screening module after quick-freezing, the time dimension of the standardized transient heat dissipation matrix is ​​set to 1024 frames, the sampling frequency is 500Hz, the local receptive field sliding window length is set to 128 frames, the stride is 32 frames, the convolution kernel size is 64 frames, and the initial weights follow a Gaussian distribution with a mean of 0 and a variance of 0.01. In specific execution, the first window covers the heat flux intensity values ​​from frame 0 to frame 127. After convolution calculation, the output vector performs heat attenuation slope calculation within this window. If x i Take an integer value from 0 to 127, y iTo correspond to the normalized heat flux intensity, for example, μ is 63.5 and ν is 0.48 in the first window, the slope value calculated by the formula is -0.0052, which indicates the rapid decrease in heat flux intensity over time. The above process is continuously executed along the entire time axis, ultimately forming a shallow temporal feature map. The horizontal axis represents the time frame index, and the vertical axis represents the convolutional output feature channel. The high-frequency thermal fluctuation region in the map is significantly enhanced in the subsequent dilated convolution processing. Validation results show that this shallow feature map can effectively distinguish the thermal response differences between small-diameter and large-diameter ice particles, providing reliable input for particle size inversion.

[0059] S3.2: Based on the shallow temporal feature map, dilated convolution operation is used to expand the network receptive field and multi-scale fusion processing is performed on the hot spot distribution information in the spatial location coordinate dimension to associate the heat conduction hysteresis effect of ice particles in different monitoring areas and generate a deep spatiotemporal embedding tensor that represents the medium- and long-term thermal dependence.

[0060] Based on the input state of the shallow temporal feature map, the dilated convolution operator is invoked to exponentially expand the receptive field in the time dimension in order to capture the ice particle thermal radiation change pattern across multiple sampling periods and avoid feature loss caused by repeated sliding of the convolution kernel.

[0061] Based on this expansion, the expansion rate is set as a multi-level dynamic matrix, and the convolution outputs corresponding to different expansion factors are calculated in parallel in the spatial coordinate dimension to ensure that a wide-coverage hot spot response mapping can be formed at the monitoring points at the feed end, middle section and discharge end.

[0062] The output of the above multi-expansion rate convolution is subjected to multi-scale fusion operation. A strategy combining weighted superposition and normalization is adopted to make the spatial distribution information of hot spots at different scales comparable in a unified scale domain. The ability to capture hysteretic heat conduction is enhanced by adjusting the weight coefficients.

[0063] A neighborhood feature association module based on spatial location coordinates is introduced to perform interdependence matrix calculation on the fused multi-scale hotspot data, generating a set of spatial correlation coefficients between hotspot distributions in each monitoring area, so as to quantify the transmission degree of heat conduction hysteresis effect at different locations.

[0064] The spatial correlation coefficient set and the temporal features after dilated convolution are matrix-concatenated to form a composite feature tensor that simultaneously contains long-term thermal dependence and spatial location correlation information. Local normalization is then applied within the tensor to eliminate differences in response amplitude in different regions.

[0065] By using a multi-scale dilated convolution and spatial correlation fusion processing method, the shallow temporal feature map is transformed into a deep spatiotemporal embedding tensor that can characterize medium- and long-term thermal dependence, thereby achieving a comprehensive quantification of the thermal conduction hysteresis effect of ice particles during the sieving process.

[0066] S3.3: Based on the deep spatiotemporal embedding tensor, the global average pooling operator is applied to perform dimensionality reduction and aggregation operations on the time series stamp dimension to eliminate instantaneous noise interference and retain the overall thermal response statistical regularity of each monitoring area, generating a compressed global thermal state feature vector.

[0067] Based on the deep spatiotemporal embedding tensor generated in the previous steps, the global average pooling operator is called to perform dimensionality reduction and aggregation operations in the time series stamp dimension. The heat flux intensity value at each time point is calculated globally by spatial location index to form a global statistical value sequence in the time dimension.

[0068] The global statistical value sequence is input into a smooth weight update mechanism, and the contribution of thermal response in different monitoring areas is weighted according to a fixed weight or an adaptive weight matrix to eliminate non-representative fluctuations caused by local transient thermal peaks.

[0069] An outlier removal operation is performed on the weighted global statistical value sequence. The median absolute deviation (MAD) criterion is used to remove noise points that exceed a set threshold, thereby ensuring the stability of subsequent feature aggregation.

[0070] The noise-removed sequence is recombined according to the monitoring area index by the global aggregation operator to generate a global thermal state data vector that is compressed in the time dimension and retains the overall thermal response trend.

[0071] The global thermal state data vector is normalized and mapped to adjust the numerical range to a unified interval of zero to one, so that the subsequent fully connected mapping layer can decouple nonlinear features.

[0072] By using global average pooling and dimensionality reduction aggregation, the deep spatiotemporal embedding tensor results from the previous step are transformed into a compressed global thermal state feature vector, achieving the expected technical effect of eliminating instantaneous noise interference and preserving the overall thermal response statistical regularity of each monitoring area.

[0073] S3.4: Based on the global thermal state feature vector, nonlinear transformation and feature decoupling operations are performed through a fully connected mapping layer to separate three orthogonal components that independently characterize the surface micro-melting rate of ice particles, the duration of recondensation, and the degree of hot spot dispersion, thereby generating a set of candidate thermodynamic features to be screened.

[0074] Based on the global thermal state feature vector, the input ports of the fully connected mapping layer are constructed to correspond one-to-one with each component of the feature vector, so as to ensure that the calculation operation of the nonlinear transformation can completely cover the global information of the thermal response characteristics of ice particles.

[0075] In the fully connected mapping layer, a weight matrix and a bias vector are configured for each input component. The initial values ​​of the weights and biases are set according to the pre-stored parameter initialization rules to ensure that the network has convergence stability during the iteration process.

[0076] Nonlinear activation function operations are performed on the input global thermal state feature vector. The hyperbolic tangent activation function is used to form a mapping effect that combines compression and expansion, thereby enhancing the separability of physical features such as the micro-melting rate and recondensation duration of ice particles in the feature space.

[0077] By performing feature decoupling on the activated feature vectors, the highly correlated thermodynamic parameters are separated into three independent feature components using the orthogonalization method. One component is used to characterize the surface micro-melting rate of ice particles, another component is used to characterize the recondensation duration, and the third component is used to characterize the spatial dispersion of hot spots.

[0078] The three independent components are calculated using an orthogonal projection matrix. The column vectors in the matrix are standard orthogonal bases generated by the Gram-Schmidt method, and orthogonalization is achieved using the following formula: Where F is the set of candidate thermodynamic features output, P is the orthogonal projection matrix, and V is the activated feature vector.

[0079] Through the above nonlinear transformation and orthogonalization process, the global thermal state feature vector of the previous step is transformed into a set of candidate thermodynamic features to be screened, which includes the surface micro-melting rate of ice particles, the duration of recondensation, and the spatial dispersion of hot spots, thereby realizing the separation of the physical meaning of the feature space and the optimization of the data structure.

[0080] S3.5: Based on the candidate thermodynamic feature set, feature selection and dimensional reconstruction processing are performed according to the predefined physical meaning constraint rule set to accurately lock the three core indicators corresponding to thermal decay slope, plateau duration and hot spot spatial dispersion, and finally output the core thermodynamic feature vector for particle size distribution inversion.

[0081] The physical meaning constraint rule set is a set of rules used to select core features directly related to the thermal response mechanism of ice particle phase change from a set of candidate thermodynamic features. Based on the physical laws governing the changes in three indicators during ice particle sieving—thermal decay slope, plateau duration, and hot spot spatial dispersion—with particle size (e.g., the thermal decay slope decreases with increasing particle size, while the plateau duration increases with increasing particle size), this rule set performs physical correlation checks, dimensional consistency verification, orthogonality analysis, and sign direction correction on candidate features. This ensures that the final output core thermodynamic feature vector has clear physical interpretability and meets the input requirements of the particle size distribution inversion model.

[0082] The physical constraint rule set is in the form of a rule table, and each rule contains the following fields: Rule ID: A unique identifier (e.g., "R-01").

[0083] Applicable features: Specify the feature components that this rule applies to (such as thermal decay slope, plateau duration, and hot spot spatial dispersion).

[0084] Physical correlation test conditions: Pre-set monotonicity relationship related to particle size (such as "the slope of thermal decay should be negatively correlated with particle size" and "the duration of plateau period should be positively correlated with particle size").

[0085] Dimensional reference: Standard units of measurement for each characteristic (e.g., the unit for thermal decay slope is s). - ¹, The duration of the plateau phase is measured in seconds (s), and the spatial dispersion of the hot spot is measured in mm².

[0086] Orthogonality threshold: The maximum allowed correlation coefficient between features (usually set to 0.15 or 0.2). Features exceeding this threshold are considered redundant.

[0087] Sign direction: Specifies the positive or negative correlation between the characteristic and the particle size distribution statistics (e.g., the thermal decay slope is set as the negative correlation direction, and the duration of the plateau period and the spatial dispersion of the hot spot are set as the positive correlation direction).

[0088] Construction method: (1) Induction of physical laws: Based on experimental data of the thermal response of ice particles in phase change, the changing trends of thermal decay slope, plateau duration and hot spot spatial dispersion in different particle size ranges were statistically analyzed, and the monotonic relationship with particle size was summarized (e.g., the larger the particle size, the slower the thermal decay rate, the longer the plateau duration and the more dispersed the hot spot). The above physical laws were transformed into clear judgment conditions (e.g., "the thermal decay slope decreases as the particle size increases").

[0089] (2) Dimensions and threshold settings: Determine the standard units of measurement for each feature based on the sensor sampling characteristics and physical definitions. Set orthogonality thresholds through correlation analysis of historical data (e.g., a Pearson correlation coefficient with an absolute value less than 0.15 is considered independent).

[0090] (3) Rule encapsulation and storage: The above conditions, thresholds, symbol directions, etc., are written into structured rules and stored in the configuration file of the edge computing unit, saved in JSON or XML format. A rule parsing interface is provided for the S3.5 step to call.

[0091] Based on the candidate thermodynamic feature set generated in the previous steps, a predefined set of physical meaning constraint rules is invoked to perform a physical correlation test on each feature component in the set, and redundant features that have no direct mapping relationship with the thermal response mechanism of ice particles are eliminated.

[0092] The remaining feature components are checked for dimensional consistency. Dimensional reference vectors are constructed based on the time unit of thermal decay slope, the time scale of plateau duration, and the spatial scale of hot spot spatial dispersion. Unit conversion and numerical normalization are performed on each component to eliminate unit system differences caused by different sampling channels or calculation paths.

[0093] Orthogonality analysis of the dimensionally normalized feature components is performed using a multi-index correlation matrix. By calculating the Pearson correlation coefficient between features and applying threshold filtering, feature components with significant independence are retained, ensuring that there is no multicollinearity in the input vector when inverting the particle size distribution.

[0094] Physical quantity sign orientation correction is performed on the retained characteristic components. Based on the laws that the thermal decay slope decreases with increasing particle size, the plateau duration increases with increasing particle size, and the hot spot spatial dispersion increases with increasing particle size dispersion, the sign orientation is unified to ensure the consistency of model mapping.

[0095] A standard data structure for the core thermodynamic feature vector is constructed, and the corrected thermal decay slope, plateau duration, and hot spot spatial dispersion are encapsulated in a preset order to ensure that their index correspondence is completely consistent with the input interface of the edge computing unit calibration mapping library.

[0096] By performing feature selection and dimensional reconstruction, the candidate thermodynamic feature set from the previous step is transformed into a core thermodynamic feature vector that can be directly used for particle size distribution inversion, thereby optimizing the feature dimension and improving the physical matching degree.

[0097] like Figure 3 As shown, step S4 involves: based on the standard particle size ice sample calibration mapping relationship data table, mapping the core thermodynamic feature vector into a set of statistical parameters characterizing the dispersion of the current cross-section ice particle size distribution, and generating a real-time particle size distribution descriptor containing skewness coefficient, peak width index, and main peak displacement. Specifically, this includes: S4.1: Read the standard particle size ice sample calibration mapping relationship data table, extract the benchmark skewness coefficient, benchmark peak width index and benchmark main peak displacement of the standard particle size ice sample batch under a specific combination of thermal decay slope, plateau duration and hot spot spatial dispersion, and construct an initial lookup index library from multidimensional thermodynamic characteristics to particle size statistical parameters.

[0098] The edge computing unit online calibration library is a comprehensive database integrated into edge computing devices, used to store various calibration data and parameter sets required for online inference and closed-loop control. This library contains three core sub-data modules: a standard particle size ice sample calibration mapping relationship data table (providing a mapping benchmark from thermodynamic characteristics to particle size statistical parameters), a standard operating condition threshold parameter set (providing an acceptable upper limit for peak width exponents), and a set of physically meaningful constraint rules (providing physical rules for feature selection). The online calibration library provides a unified parameter query service for steps S3, S4, and S5, supporting long-term stable operation and adaptability of the equipment.

[0099] The online calibration library adopts a hierarchical storage structure and mainly includes the following parts: Standard particle size ice sample calibration mapping relationship data table: used for mapping thermodynamic characteristics to particle size statistical parameters, supporting multidimensional linear interpolation query.

[0100] Standard operating condition threshold parameter set: used for particle size dispersion determination, providing preset thresholds and environmental adaptation factors.

[0101] Physical meaning constraint rule set: used for feature screening and correction, providing physical correlation test, dimensional consistency, orthogonality threshold and sign direction rules.

[0102] Version management metadata table: records information such as version number, update time, and applicable device model for each sub-data table.

[0103] Indexing acceleration module: Creates multidimensional hash indexes for thermodynamic feature combination keys (thermal decay slope, plateau duration, hot spot spatial dispersion), supporting millisecond-level query response.

[0104] Construction method: (1) Initialization Construction: During the equipment delivery or field deployment phase, a standard particle size ice sample calibration mapping relationship data table is generated through offline calibration experiments. Based on the statistical analysis of historical qualified batch data, a standard operating condition threshold parameter set is generated. A set of physically meaningful constraint rules is compiled according to the physical laws of ice particle phase change thermal response.

[0105] (2) Data import and index creation: Write the above three parts of data into the local persistent storage of the edge computing unit (such as an SQLite database or an in-memory database). Create a multidimensional index for hot queries (the mapping from thermodynamic features to particle size statistics).

[0106] (3) Online update and calibration: When the equipment has been running for a long time or the environmental conditions have changed significantly, the system can automatically trigger the incremental calibration process to add or correct records in the data table. It supports manual updating of threshold parameter sets and rule sets through the configuration interface.

[0107] (4) Regular maintenance: Regularly (e.g., quarterly) expand the data table with new calibration samples to improve mapping accuracy. Record the version number of each update for easy backtracking and rollback.

[0108] Through the online calibration library of the edge computing unit, the structured centralized management of calibration data, threshold parameters, physical rules and other knowledge is realized, providing unified and reliable data support for real-time monitoring and closed-loop control of ice particle size distribution.

[0109] The standard particle size ice sample calibration mapping data table is a structured database pre-stored in the online calibration library of the edge computing unit. It is used to establish a quantitative mapping relationship between core thermodynamic feature vectors (thermal decay slope, plateau duration, hot spot spatial dispersion) and particle size distribution statistical parameters (skewness coefficient, peak width index, main peak displacement). This data table is obtained through calibration experiments on standard ice samples with known particle size distributions, providing a benchmark reference for real-time particle size distribution inversion. This allows thermodynamic features collected in actual production to be quickly and accurately converted into a set of statistical parameters describing the dispersion of ice particle size distribution.

[0110] The standard particle size ice sample calibration mapping data table adopts a relational two-dimensional table format. Each record corresponds to a set of standard ice samples with a specific particle size distribution and mainly includes the following fields: Ice Sample Identifier ID: A unique identifier for the calibrated ice sample (e.g., “S-001”).

[0111] Thermal attenuation slope: The rate of attenuation of ice particle group thermal radiation calibrated under standard environmental conditions (unit: s). - ¹).

[0112] Plateau duration: The duration (in seconds) during which the heat flux intensity curve enters a stable plateau phase.

[0113] Hot spot spatial dispersion: the variance or standard deviation of hot spot distribution on the sieve surface (unit: mm²).

[0114] Skewness coefficient: a statistical index characterizing the degree of asymmetry in particle size distribution (dimensionless, negative values ​​indicate left skewness, and positive values ​​indicate right skewness).

[0115] Peak width index: a statistical index that characterizes the discrete range of particle size distribution (dimensionless, the larger the value, the wider the distribution).

[0116] Main peak displacement: The deviation between the actual main peak particle size and the standard target particle size (unit: mm, negative value indicates smaller, positive value indicates larger).

[0117] Applicable operating condition labels: such as the temperature and humidity range of the pre-cooling room, the initial temperature range of ice particles, etc., for environmental adaptation.

[0118] This data table supports fast retrieval using a combination of three features: thermal decay slope, plateau duration, and hotspot spatial dispersion. It also supports nearest neighbor matching and interpolation queries.

[0119] Construction method: Standard ice sample preparation: Under standard environmental conditions such as constant temperature and humidity, sea level, and standard fuel oil, several batches of ice samples with different particle size distribution characteristics are prepared. The particle size distribution of each batch of ice samples is accurately determined by sieve analysis or image analysis, and the corresponding skewness coefficient, peak width index, and main peak shift are calculated. Each batch of ice samples is divided according to a uniform particle size range (e.g., 2.0 mm to 6.0 mm) to ensure that the particle size statistical parameters are representative.

[0120] Thermodynamic characteristic calibration: Each batch of standard ice samples was placed on the same conveyor sieve as the actual screening line and operated under standard environmental conditions. Transient thermal radiation decay curves were collected using miniaturized infrared thermopile sensor arrays deployed at the inlet, middle, and outlet ends. Following the procedure in step S3, the core thermodynamic characteristic vectors (thermal decay slope, plateau duration, and hot spot spatial dispersion) of the batch of ice samples were extracted. Multiple measurements were repeated and the average value was taken to eliminate random errors.

[0121] Data table population: The thermodynamic characteristic vector of each batch of ice samples and the corresponding particle size statistics parameters (skewness coefficient, peak width index, main peak shift) are recorded as a single record and written into the data table. For the same batch of ice samples, multiple sets of calibration data under different temperature and humidity environments can be recorded to support environmental compensation. After the data table is established, it is stored in the online calibration library of the edge computing unit, and a query interface is provided.

[0122] Data sheet maintenance and updates: Regularly (e.g., quarterly) recalibrate using newly prepared standard ice samples, expanding or updating the records in the data sheets. When the screening line equipment undergoes major modifications or environmental conditions exceed the original calibration range, the calibration process should be repeated to generate a new version of the data sheet.

[0123] By using a standard particle size ice sample calibration mapping data table, a quantitative and traceable mapping from thermodynamic characteristics to particle size statistical parameters was achieved, ensuring the accuracy and consistency of real-time particle size distribution inversion.

[0124] The input is the core thermodynamic feature vector output from the preceding step S3.5, containing three indices: thermal decay slope, plateau duration, and hot spot spatial dispersion. Simultaneously, a calibration mapping data table of standard particle size ice samples is stored within the edge computing unit as a reference. For this input, the online calibration library interface module of the edge computing unit is invoked to read the data table, loading calibration records of various batches of standard particle size ice samples under different thermodynamic feature combinations into the system cache. Feature combination filtering is performed on the loaded calibration records. Based on the key-value matching principle of thermal decay slope, plateau duration, and hot spot spatial dispersion, completely matching or adjacent matching thermal feature combination entries are located in the cache data. Field extraction processing is performed on the matching entries to obtain three particle size statistical parameters corresponding to the thermal feature combination: the reference skewness coefficient, the reference peak width index, and the reference main peak displacement. Data structuring processing is performed on the extracted three reference statistical parameters, storing them as an ordered vector in a temporary mapping table to construct a mapping index entry from thermodynamic features to particle size statistical parameters. An index table generation operation is performed on all batches of mapped index entries to form a multi-dimensional lookup index library containing three-dimensional feature coordinates of thermal decay slope, plateau duration, hot spot spatial dispersion, and corresponding particle size statistical parameters. This index library processing method transforms the core thermodynamic feature vectors from the previous step into initial lookup benchmarks that can be used for subsequent interpolation calculations, thus achieving the first stage of data preparation for the particle size distribution inversion process.

[0125] For example, in a certain industrial screening batch, the core thermodynamic feature vector values ​​are a thermal decay slope of 0.83, a plateau duration of 1.27 seconds, and a hot spot spatial dispersion of 0.46. In the edge computing unit's online calibration library, standard particle size ice sample records include batches with D10=2.8mm, D50=4.2mm, and D90=5.6mm, each batch possessing corresponding thermodynamic feature combinations and statistical parameters. For the input feature vector, the system reads the data table and locates the combination entries in the batch records with thermal decay slopes of 0.82~0.84, plateau durations of 1.25~1.28 seconds, and hot spot spatial dispersions of 0.45~0.47, extracting a baseline skewness coefficient of 0.12, a baseline peak width index of 1.38, and a baseline main peak shift of -0.05mm. These parameters are then used to construct a mapping index entry and stored in the generated multidimensional lookup index library, achieving a standardized mapping from thermodynamic features to particle size statistical parameters. For different batches of input feature vectors, the index library can support mapping based on approximate matching or interpolation, which improves the continuity and accuracy of particle size distribution inversion. In actual verification, the output mapping index library can significantly improve the statistical stability of particle size under environmental fluctuations.

[0126] S4.2: Based on the initial search index library, perform multidimensional linear interpolation on the input current cross-section core thermodynamic feature vector to eliminate the discrete deviation between the measured thermal decay slope, the measured plateau duration, and the measured hot spot spatial dispersion and the standard sample point, and generate an intermediate particle size distribution estimation parameter set with continuous domain adaptability.

[0127] Using the core thermodynamic feature vector as input, the values ​​of thermal decay slope, plateau duration, and hot spot spatial dispersion are multidimensionally matched with the initial search index to identify the set of neighboring sample points in the feature space of standard particle size ice samples. The coordinate dimensions of the thermodynamic feature differences between each neighboring sample point are normalized to establish a difference metric between each feature component and the standard sample point. Based on this difference metric, a piecewise linear interpolation weight matrix is ​​constructed in the three dimensions of thermal decay slope, plateau duration, and hot spot spatial dispersion to ensure scale consistency in the interpolation process across different directions of the feature space. A multidimensional linear interpolation method is used to calculate the interpolation contribution of the current feature vector at each neighboring sample point, and a three-dimensional fusion operation is performed under the constraint of the weight matrix to obtain a set of intermediate estimates reflecting the continuous-domain particle size distribution. Through the statistical parameter set obtained from the interpolation fusion, the discrete standard mapping index values ​​are transformed into continuously adaptive particle size distribution estimation parameters, achieving a smooth transition from discrete calibration data to continuous-domain prediction data.

[0128] S4.3: Based on the drift of the thermal conductivity coefficient caused by the temperature and humidity fluctuations in the pre-cooling room, perform dynamic compensation and correction processing on the values ​​of the intermediate particle size distribution estimation parameter set to offset the influence of environmental noise on the analysis of heat dissipation rate, and output the corrected particle size distribution statistics after environmental robustness optimization.

[0129] An environmental fluctuation detection interface is established for the intermediate particle size distribution estimation parameter set. Real-time temperature and humidity data of the precooling room are extracted as input conditions, and the coupling relationship between the surface heat conduction behavior of ice particles and environmental changes is analyzed.

[0130] Real-time temperature and humidity data are converted into the drift of thermal conductivity coefficient. The theoretical heat dissipation correction value under the current operating conditions is calculated based on the relationship model between the latent heat of ice phase change and thermal conductivity. This relationship model is constructed by mapping physical constants and quantified environmental parameters.

[0131] A dynamic compensation factor is applied to each parameter set of the intermediate particle size distribution estimation parameters by means of the correction of the phase change thermal response characteristics of ice particles. This compensation factor is calculated from the deviation between the thermal conductivity drift and the actual thermal attenuation slope.

[0132] The calculated compensation factor is adjusted proportionally to the skewness coefficient, peak width index, and main peak displacement in the estimated parameter set to offset the deviation caused by environmental noise.

[0133] By smoothing the compensated parameter curves using polynomial fitting, overcorrection caused by instantaneous environmental fluctuations can be avoided, thus improving the stability of the correction results.

[0134] A corrected particle size distribution statistic optimized for environmental robustness is constructed and used as the direct input for subsequent probability density function reconstruction, thereby achieving an adaptive offsetting effect against temperature and humidity fluctuations.

[0135] S4.4: Based on the corrected particle size distribution statistics, perform probability density function reconstruction calculation to transform the discrete statistics into a continuous probability distribution curve, and accurately calculate the real-time skewness coefficient characterizing the degree of asymmetry in the current ice particle size distribution, the real-time peak width index characterizing the particle size discrete range, and the real-time main peak displacement characterizing the position of the mainstream particle size.

[0136] Using the environmentally robust optimized corrected particle size distribution statistic as input, the probability density function reconstruction module is invoked to establish a continuous-domain particle size distribution curve. Discrete data points in this statistic are arranged in order of particle size, and a local weighting function is applied around each data point using a kernel density estimation method to form smoothed probability estimates for particle size intervals. A Gaussian kernel is used as the weighting function, and the bandwidth parameter for different particle size intervals is adaptively adjusted to eliminate curve distortion caused by uneven sampling density. Normalization is performed on all smoothed probability values ​​to maintain the integral sum within a unit interval, thus ensuring the mathematical completeness of the probability curve. Using the reconstructed probability distribution curve, the skewness coefficient calculation formula is invoked to obtain the real-time skewness coefficient. Based on the same probability distribution curve, the peak width index calculation formula is invoked to reflect the particle size distribution width. By performing extreme value location calculations on the probability distribution curve, the particle size value corresponding to the maximum probability point is calculated, and this value is compared with a standard benchmark to obtain the real-time main peak displacement. The skewness coefficient, peak width index, and main peak displacement are output sequentially to form three real-time evaluation indicators characterizing the asymmetry, discrete range, and mainstream particle size position of the current ice particle group's particle size distribution. Through probability density function reconstruction, normalization, and precise calculation of the three statistics, the corrected particle size distribution statistics from the previous step are transformed into real-time descriptive parameters that can be used for closed-loop screening control decisions, achieving accurate characterization of particle size distribution based on thermodynamic characteristics and probability statistics.

[0137] S4.5: Integrate the real-time skewness coefficient, real-time peak width index, and real-time main peak displacement, and encapsulate them to generate a standardized real-time particle size distribution descriptor data structure. This data structure serves as the final statistical parameter set characterizing the dispersion of the current cross-section ice particle size distribution, which is then used for threshold comparison and closed-loop control decision-making in subsequent steps.

[0138] Step S5: Determine whether the peak width index in the real-time particle size distribution descriptor continuously exceeds the preset particle size dispersion threshold. If the determination result is yes, trigger the pore size gradient adjustment command; otherwise, maintain the current screen shape and continue monitoring. Specifically, this includes: S5.1: Based on the real-time particle size distribution descriptor generated in the previous steps, which includes skewness coefficient, peak width index and main peak displacement, the core indicator peak width index, which characterizes the dispersion of particle size distribution, is extracted and used as the quantitative evaluation input of the uniformity of ice particle group screening at the current moment, providing a standardized data benchmark for subsequent threshold comparison.

[0139] Based on the skewness coefficient, peak width index, and main peak displacement data structures generated in the previous steps and encapsulated in the real-time particle size distribution descriptor, the field retrieval interface inside the descriptor is called to lock the peak width index value, which characterizes the dispersion of particle size distribution, as the core processing object of this sub-step.

[0140] The extracted peak width index value is input into the standardized data parsing module to perform a unit consistency check, ensuring that the value uses the same statistical unit and scale as the preset threshold, so as to avoid judgment deviations caused by unit mismatch in subsequent comparison stages.

[0141] Based on the peak width index value output by the analysis module, the particle size uniformity quantitative evaluation function is called. Combined with the time series index of batch production, the index is located to the time marker of the current screening cycle, so as to extract a static and comparable uniformity evaluation quantity from the dynamic production flow.

[0142] The evaluation function generates a double-key mapping record containing the peak width index value and the current time identifier in the internal buffer, and submits the mapping to the input buffer of the threshold comparison control unit in a standardized format to establish the data benchmark for this round of judgment.

[0143] Through the above chain derivation process, the real-time particle size distribution descriptor output in the previous step is transformed into a standardized quantitative evaluation input for particle size screening uniformity, thus providing an accurate and directly callable data foundation for the subsequent threshold comparison process.

[0144] S5.2: Using the standard operating condition threshold parameter set stored in the online calibration library of the edge computing unit, the preset particle size dispersion threshold corresponding to the current production batch is called. The preset particle size dispersion threshold is used as a static reference benchmark for judging the stability of the screening process, so as to eliminate the risk of misjudgment caused by environmental fluctuations.

[0145] The standard operating condition threshold parameter set is a set of parameters stored in the online calibration library of the edge computing unit. It is used to define the acceptable upper limit of particle size distribution dispersion during the ice particle screening process (i.e., the preset particle size dispersion threshold). This parameter set is set according to different production batches, different ice particle types, and environmental conditions (such as temperature and humidity in the pre-cooling room). It serves as a static reference benchmark for judging whether the peak width index exceeds the standard, so as to eliminate the risk of misjudgment caused by environmental fluctuations.

[0146] The standard operating condition threshold parameter set is in the form of a key-value table. Each record corresponds to a production operating condition and includes the following fields: Operating condition label: A label that uniquely identifies the current production conditions.

[0147] Preset particle size dispersion threshold: the maximum allowable value of peak width index (dimensionless). If the value is exceeded, the particle size distribution is considered to be too wide.

[0148] Environmental adaptation factor: A coefficient used to compensate for the environmental impact of the threshold (e.g., the influence coefficient of temperature and humidity on particle size dispersion).

[0149] Applicable temperature and humidity range: This threshold is effective for the temperature range and relative humidity range of the pre-cooling room.

[0150] Version number: The version identifier of the parameter set, which is easy to trace.

[0151] Construction method: (1) Historical data statistics: Collect historical peak width index data of multiple normal production batches (finished products with qualified quality), and calculate their mean μ and standard deviation σ. Set the threshold as μ + k·σ, where k is usually taken as 2 to 3 (corresponding to 95% to 99% confidence interval). Statistically analyze batches under different environmental conditions to obtain thresholds for different temperature and humidity ranges.

[0152] (2) Environmental adaptation factor calibration: Analyze the impact of temperature and humidity changes on the peak width index through experimental or field data, and fit the compensation coefficient C_env (for example, the threshold increases by 0.01 for every 1℃ increase in temperature; the threshold increases by 0.02 for every 10% increase in humidity). Store the coefficient in the parameter set.

[0153] (3) Parameter set encapsulation and deployment: The above thresholds, environmental adaptation factors, applicable ranges, etc. are written into the configuration file and stored in the online calibration library of the edge computing unit. A query interface is provided for searching by operating condition label and temperature and humidity range.

[0154] Based on the peak width index value and the corresponding real-time particle size distribution descriptor output from the previous steps, the standard operating condition threshold parameter set pre-stored in the online calibration library of the edge computing unit is called, and the particle size dispersion threshold item matching the current production batch is obtained through index retrieval as a static reference benchmark data source.

[0155] The peak width index value is made consistent with the particle size dispersion threshold of the current batch by performing data type consistency processing, unit conversion and dimension unification operations to ensure that there is no impact from unit system differences during the comparison process.

[0156] Based on the operating condition labels of the pre-cooling room and the screening module, the environmental adaptation factors in the threshold parameter set are read and bound to the threshold values ​​to form a composite reference matrix containing the threshold value and the environmental compensation coefficient, which provides environmental robustness assurance for subsequent anomaly judgment.

[0157] The instantaneous difference between the current measured value and the reference threshold is calculated using the threshold deviation.

[0158] The instantaneous difference is input into the environmental fluctuation compensation function. in As an environmental adaptation factor, we obtain differential data after environmental compensation to eliminate the risk of misjudgment caused by temperature and humidity fluctuations.

[0159] By using the above processing method, the peak width index data from the previous step is transformed into a comparable difference quantity that is strictly consistent with the static threshold benchmark, thereby realizing the environmental robustness judgment input for the stability of the screening process.

[0160] S5.3: Perform continuous monitoring processing of the peak width index using a sliding time window. By setting a requirement for the number of continuous sampling periods, determine whether the peak width index continuously exceeds the preset particle size dispersion threshold within a specified time series. This filters out instantaneous noise interference and confirms the authenticity of the particle size distribution anomaly, generating an out-of-tolerance judgment result with time-series confidence.

[0161] To perform continuous monitoring of the peak width index in the input real-time particle size distribution descriptor using a sliding time window, the absolute value of the peak width index for the current sampling period needs to be retrieved from the core evaluation module of the edge computing unit as the basic data source for the time series. A fixed-length sliding window buffer is established on the basic data source for the time series, and the length of this window is set to the number of continuous sampling periods pre-calibrated according to the production cycle and process requirements to ensure that the data within the window can fully cover the dynamic changes of a screening cycle. For each peak width index data sample within the sliding window, a numerical difference comparison operation is performed with a preset particle size dispersion threshold to generate a corresponding binary judgment result sequence, where a judgment value of true indicates that the peak width index of the sample exceeds the threshold, and a judgment value of false indicates that it does not exceed the threshold. The binary judgment result sequence is cumulatively statistically analyzed to calculate the length of the continuous segment where true values ​​appear within the window, and compared with the specified number of consecutive out-of-tolerance periods to confirm the continuity of the out-of-tolerance phenomenon. The continuous out-of-tolerance ratio is calculated using the following formula: Where C represents the length of the continuous out-of-tolerance segment within the window, and N represents the sliding window length. If the calculated result exceeds the preset critical value for the continuous out-of-tolerance ratio, the entire window is assigned a "continuous out-of-tolerance" status in the judgment logic. Based on this continuous out-of-tolerance status, combined with the temporal consistency check of similar monitoring results outside the window, an out-of-tolerance judgment result with temporal confidence is finally generated, serving as the direct input for subsequent logical branch decisions. Through continuous monitoring processing via a sliding time window, the peak width index calculated in the previous step is transformed into continuous out-of-tolerance judgment data with resistance to transient noise interference, achieving the expected technical effect of confirming abnormal states in the screening process.

[0162] S5.4: Execute a logical branch selection operation based on the out-of-tolerance judgment result with time-series confidence. If the judgment result is yes, instantiate the aperture gradient adjustment command signal. If the judgment result is no, otherwise generate a holding command to maintain the current screen shape, thereby outputting a deterministic control strategy decision vector to drive the subsequent actuator action.

[0163] Based on the error determination result input with time-series confidence, the logic branch parser of the control decision module is called, and the result is loaded into the status register of the branch determination unit as a Boolean state quantity.

[0164] In the branch decision unit, a binary threshold matching operation is performed. By comparing the value bit by bit with the truth constant table, the corresponding execution path index code is generated to determine the target type of subsequent control signal instantiation.

[0165] For execution paths where the matching result is true, the aperture gradient adjustment instruction constructor is called. The morphological parameter templates for inlet tightening, middle expansion, and outlet convergence are loaded into the constructor memory. Combined with the screen morphological parameters of the current production batch, the aperture gradient adjustment instruction signal is instantiated and assigned a unique identifier for subsequent link tracking.

[0166] For execution paths where the matching result is false, the screen holding instruction generator is called to copy the existing screen aperture distribution matrix to the instruction buffer, instantiate the screen holding instruction, and mark the holding mode flag to solidify the current aperture state.

[0167] The two types of control instructions that have been instantiated above are uniformly connected to the decision vector encapsulator. The instruction type, target parameter set, execution priority and other elements are arranged in a standardized field order to generate a deterministic control strategy decision vector to ensure the safety and predictability of the actions of the driving actuator.

[0168] By unifying the decision vector, the judgment result of the previous step is transformed into standardized control data that can be directly parsed by the control interface, thereby achieving the expected technical effect of real-time adjustment or maintenance of the flexible electro-deformable screen structure.

[0169] For example, on a quick-freezing screening line designed to produce 1200 ice particles per minute, the measured peak width index over 15 consecutive sampling cycles (80 milliseconds per cycle) is 0.62, exceeding the preset particle size dispersion threshold of 0.55, and the time-series confidence level is determined to be true. The control decision module reads the target pore size template with the inlet tightening to 2.0 mm, the middle expanding to 4.8 mm, and the outlet converging to 3.5 mm, and instantiates the pore size gradient adjustment command signal. If the peak width index continuously drops to 0.50 and is determined to be false, the screen holding command generator copies the current pore size matrix and instantiates the holding command. The encapsulator encapsulates the adjustment or holding command along with the execution priority (adjustment mode priority is set to 1, holding mode priority is set to 2) into a unified control strategy decision vector, and transmits it to the driver interface through a standardized communication protocol. In adjustment mode, the flexible electro-deformable screen structure completes the asymmetric strain distribution adjustment. After one screening cycle, the peak width index drops to 0.54, and the particle size uniformity is significantly improved. In maintenance mode, the screen maintains its original geometric shape, and the system enters a continuous monitoring state with no performance fluctuations.

[0170] S5.5: The deterministic control strategy decision vector is encapsulated into a standardized communication protocol data packet and sent to the drive control interface of the flexible electro-deformable screen structure to complete the construction of a closed-loop feedback link from data analysis to execution triggering, ensuring that the system can respond in real time to the dynamic changes in ice particle size distribution.

[0171] Step S6: Calculate the required asymmetric strain distribution curve along the length of the screen surface according to the aperture gradient adjustment command, and generate a gradient voltage control sequence to drive the flexible electro-deformable screen structure to produce millimeter-level aperture gradient changes. Specifically, this includes: S6.1: Based on the target morphological parameters of inlet tightening, middle expansion and outlet convergence defined in the aperture gradient adjustment command, a nonlinear mapping function between the screen surface length coordinate axis and the target screen aperture change is constructed, and the ideal asymmetric strain distribution curve continuously distributed along the conveying direction is calculated to establish the deformation reference profile of the flexible electro-deformable screen structure in the spatial dimension.

[0172] Based on the target morphological parameters of inlet tightening, middle expansion, and outlet convergence contained in the received aperture gradient adjustment command, the screen surface length reference coordinate system data stored in the edge computing unit is invoked to quantize the physical length corresponding to the screen surface conveying direction into continuous coordinate axes, establishing a mapping data source between morphological parameters and screen aperture change. Using this mapping data source, a set of node parameters capable of characterizing the aperture change trends in the feed end, middle, and discharge end regions is selected as the control nodes for function construction. A cubic spline interpolation method is used to generate continuous target aperture change curves between nodes, ensuring the continuity of the derivative of the curve fitting to avoid abrupt deformation. The target aperture change curve is transformed into a corresponding strain demand curve. A physical mapping relationship is established between the screen aperture change and the linear strain of the screen substrate through material geometry formulas. The ideal strain amplitude at each coordinate position is calculated by invoking the elastic modulus and structural geometric parameters of the screen material. The calculated strain amplitude is adjusted using a sinusoidal modulation superposition of asymmetric components. A high-amplitude tightening waveform is introduced in the inlet region, a low-amplitude expanding waveform in the middle region, and a medium-amplitude converging waveform in the outlet region to form a strain distribution curve with asymmetric characteristics along the conveying direction. The synthesized strain distribution curve is used as the deformation reference profile input to the subsequent discretization module. Millimeter-level aperture gradient adjustment of the flexible electro-deformable screen structure is achieved through continuous spatial strain control. By constructing a nonlinear mapping function between the screen surface length coordinate axis and the target screen aperture change, and calculating the ideal asymmetric strain distribution curve, the aperture gradient adjustment command from the previous step is transformed into spatial deformation reference data, achieving precise deformation control of the flexible electro-deformable screen structure in the conveying direction.

[0173] S6.2: Discretize the ideal asymmetric strain distribution curve into a mesh, decompose the continuous strain reference profile into the local strain demand of several independent control units, and generate a discrete strain demand vector set containing the location index of each control unit and its corresponding target strain value, which serves as the basic data source for voltage calculation.

[0174] Based on the ideal asymmetric strain distribution curve and its continuous spatial function expression obtained in the previous steps, the standard coordinate axis along the length of the sieve surface is selected as the discretization reference, and the center point position index of each control unit is defined to form a position index set. Using the curvature change of the ideal asymmetric strain distribution curve at each index position as the calculation basis, numerical integration based on equidistant sampling is performed to map the strain values ​​on the continuous curve to local strain requirements. To ensure the calculation accuracy of each local strain requirement, a piecewise polynomial fitting method is used to approximate the original curve locally, and the strain value at each position is obtained on the fitted model. After calculating all local strain requirements, a discrete strain requirement vector set containing the control unit position index and the corresponding target strain value is constructed, and boundary condition verification is performed on this vector set to eliminate outliers exceeding the material's physical limits. The verified discrete strain requirement vector set is used as the basic data source for voltage calculation input, realizing the transformation from a spatially continuous curve to local strain parameters.

[0175] By using discretization meshing and numerical approximation processing, the ideal asymmetric strain distribution curve obtained in the previous step is transformed into a discrete strain demand vector that can be directly called by the voltage calculation module, thereby achieving precise zoning control of the flexible screen.

[0176] For example, within a range of 0 to 300 mm along the length of the sieve surface, a discretization grid spacing of 10 mm is set based on the characteristics of the asymmetric strain distribution curve, forming a total of 30 control units. The curve is approximated using cubic spline fitting, and strain values ​​are extracted at the center of each control unit. For instance, at the 12th unit position (120 mm), the strain value is 0.0035, and at the 20th unit position (200 mm), the strain value is 0.0052. These strain values ​​are bound to position indices to form a discrete strain demand vector set. Boundary checks are performed to remove unit values ​​exceeding 0.0080, ultimately generating a dataset containing 28 effective control units as the input for voltage calculation. In this embodiment, the discretized strain demand data maintains a stable physical match during subsequent electric field strength calculations, significantly improving the spatial resolution and execution accuracy of the control sequence.

[0177] S6.3: Using the pre-stored calibration matrix of the electromechanical coupling characteristics of the polyvinylidene fluoride-trifluoroethylene copolymer piezoelectric thin film, the local strain demand quantities in the discrete strain demand vector set are inversely solved into the corresponding electric field intensity demand values, generating an initial electric field intensity distribution sequence that reflects the intrinsic response characteristics of the material, so as to establish the physical quantity conversion relationship from macroscopic deformation to microscopic electric field.

[0178] Based on the target local strain values ​​in the discrete strain demand vector set, a pre-stored calibration matrix of the electromechanical coupling characteristics of a polyvinylidene fluoride-trifluoroethylene copolymer piezoelectric thin film is used to perform inverse calculations to obtain the electric field intensity values ​​that drive the corresponding local region to reach the target strain. During the calculation process, each local strain demand is input into the inverse calculation model of the material property matrix, and the material response coefficients characterized by the matrix elements are used to achieve the mapping from macroscopic deformation to microscopic electric field intensity. Numerical stability control is introduced during matrix inversion, and singular values ​​smaller than a set threshold are filtered out through singular value decomposition to ensure that the calculation process is not affected by ill-conditioned matrices. The calculated electric field intensity values ​​are rearranged according to the control unit position index to form an initial electric field intensity distribution sequence. This sequence completely describes the physical quantity transformation relationship from macroscopic aperture gradient deformation to microscopic electric field driving conditions. Through this inverse mapping method based on the intrinsic response characteristics of the material, the discrete strain demand from the previous step is transformed into physically feasible electric field distribution data, achieving precise matching between the material driving layer and the target deformation parameters.

[0179] For example, in an embodiment where a section of the sieve surface is divided into 16 independent control units along its length, the target strain requirement for each control unit ranges from 0.15% to 0.32%. The pre-stored piezoelectric thin film electromechanical coupling characteristic matrix is ​​16×16 dimensional, with its diagonal elements representing the self-response coefficients of each unit, averaging 3.8×10⁻⁶. - ³ m / V, where off-diagonal elements represent the coupling coefficients of adjacent units, with an average value of 1.2 × 10⁻⁶. -4 m / V. Using the above formula for inverse calculation, assuming the target strain of the 5th element in the strain demand vector is 0.27%, the corresponding electric field strength solution after matrix inversion is 71.05 V / m, which is within the safe driving range. Arranging the solution electric field strengths of all 16 elements by index forms the initial electric field strength distribution sequence, which spatially exhibits a gradient characteristic of high inlet region, low middle section, and medium outlet region. This sequence is directly used for subsequent potential difference integration calculations. Verification showed that in the screen deformation response test, the aperture shrinkage amplitude in the inlet region was significantly improved, the aperture expansion in the middle section was stable, and the aperture convergence in the outlet region was uniform, achieving the expected effect of significantly improved particle size distribution.

[0180] S6.4: Based on the initial electric field intensity distribution sequence and the geometric spacing parameters of the conductive nanofiber fabric electrode layer, perform potential difference integration calculation to convert the microscopic electric field intensity requirement value into the driving potential difference value required by each independent control unit, and generate a preliminary gradient voltage control original sequence to match the actual power supply interface characteristics of the electro-deformable composite material layer.

[0181] Based on the initial electric field intensity distribution sequence obtained by inverse calculation of the polyvinylidene fluoride-trifluoroethylene copolymer piezoelectric film and the geometric spacing parameters of the conductive nanofiber fabric electrode layer, the electric field intensity requirement value corresponding to each control unit is selected as the input quantity for the integration operation, and a spatial voltage distribution integration path along the length of the sieve surface is established.

[0182] A geometric spacing parameter is introduced as a position variable on the integration path. The cumulative value of the potential difference between each control unit is calculated by interval integration, forming a potential difference sequence that corresponds one-to-one with the spatial coordinates.

[0183] The integral calculation results are combined with the dielectric constant of the electrode layer and the effective length factor to perform physical quantity unit conversion, ensuring that the potential difference value conforms to the actual power supply interface standard.

[0184] The potential difference sequence after unit conversion is reconstructed into the initial gradient voltage control original sequence according to the control unit arrangement order. This sequence is then compared with the interface matching parameters of the composite material layer to filter out incompatible voltage values ​​to avoid interface failure.

[0185] Through the above integration and reconstruction process, the initial electric field intensity distribution sequence generated in the previous step is transformed into the original gradient voltage control sequence that can be directly applied to the flexible electro-deformable screen structure, thereby achieving a matching effect between the voltage spatial distribution and the material response characteristics.

[0186] S6.5: Apply amplitude limiting constraints and timing smoothing filtering to the original gradient voltage control sequence to eliminate step abrupt changes caused by discretization calculation and ensure that the output voltage is within the safe threshold range, thereby generating a standardized gradient voltage control sequence that is ultimately used to drive the flexible electro-deformable screen structure to generate millimeter-level aperture gradient changes, thus completing the complete closed-loop transformation from control command to execution signal.

[0187] Step S7: Based on the gradient voltage control sequence, the screen is driven to generate asymmetric strain, constructing a dynamic aperture distribution zone that tightens at the inlet, expands in the middle, and converges at the outlet. Specifically, this includes: S7.1: Perform digital-to-analog conversion and power amplification on the gradient voltage control sequence to generate a high-fidelity analog voltage signal set that can directly drive the polyvinylidene fluoride-trifluoroethylene copolymer piezoelectric film in the electrodeformable composite material layer, ensuring that the voltage amplitude accuracy meets the millimeter-level aperture adjustment requirements.

[0188] The standardized gradient voltage control sequence output from step S6.5 is used as the input signal and loaded into the parallel buffer register array of the digital-to-analog converter unit to ensure the timing consistency of the gradient voltage during multi-channel synchronous conversion. The discrete voltage values ​​in the buffer register array are input to a high-precision multi-resolution digital-to-analog converter (DAC). Voltage quantization error is minimized through bit-by-bit weighted current integration, resulting in a continuous analog voltage waveform. The analog voltage waveform output from the DAC is introduced into a linear power amplification stage. A high-bandwidth amplifier with fixed gain parameters is selected based on the piezoelectric film's driving power requirements, and amplitude drift caused by load impedance fluctuations is eliminated through dynamic bias adjustment. Band-limited filtering is performed during amplification, using a third-order Butterworth low-pass filter to suppress high-frequency harmonics, ensuring a pure driving signal spectrum and avoiding unwanted lattice vibrations within the material. The analog signal, after power amplification and band-limited filtering, is distributed to the electrode connections of the electrodeformable composite layer by channel, and impedance matching is performed to achieve optimal energy transfer efficiency between the signal source and the input of the piezoelectric film within the operating frequency band. Through the aforementioned digital-to-analog conversion and power amplification link, the digital control sequence of the previous step is accurately converted into a high-fidelity analog voltage signal set that can directly drive the polyvinylidene fluoride-trifluoroethylene copolymer piezoelectric film, thereby achieving the amplitude accuracy and waveform integrity required for millimeter-level aperture adjustment.

[0189] S7.2: Based on the high-fidelity analog voltage signal set, a gradient electric field is applied to the polyvinylidene fluoride-trifluoroethylene copolymer piezoelectric film sandwiched between two layers of conductive nanofiber fabric to excite the internal lattice of the piezoelectric film to undergo an asymmetric electromechanical coupling response and generate a micro-strain field that varies along the length of the sieve surface.

[0190] Based on the high-fidelity analog voltage signal set, the voltage amplitude corresponding to each independent control unit output by digital-to-analog conversion and power amplification is selected as the input quantity and applied to the electrode port of the polyvinylidene fluoride-trifluoroethylene copolymer piezoelectric film sandwiched between two layers of conductive nanofiber fabric to form a gradient electric field varying along the length of the sieve surface. By controlling the potential difference between each electrode segment, the electric field intensity distribution inside the film is made asymmetrical, and the strain response value of the local lattice is calculated based on the electromechanical coupling coefficient matrix of the piezoelectric material. According to the designed gradient electric field distribution, the strain values ​​of each segment of the film lattice are combined to form a continuous strain gradient curve varying along the coordinate axis of the sieve surface. The anisotropic characteristics of the material guide the strain field to generate micro-deformation modes with inconsistent amplitudes and directions in different regions. The gradient electric field-strain field mapping relationship is converted into displacement field distribution data of the film surface using a finite element solver, and this displacement field is mapped and combined with the geometric reference of the sieve surface to drive the upper conductive nanofiber fabric to generate a non-uniform mechanical traction force in the next sub-step. By using the above-mentioned gradient electric field excitation and strain gradient generation processing method, the high-fidelity voltage signal from the previous step is converted into a microscopic strain field that varies along the sieve surface inside the polyvinylidene fluoride-trifluoroethylene copolymer piezoelectric film, thus realizing the physical basis for constructing the dynamic pore size distribution zone.

[0191] S7.3: The micro-strain field is used to apply a non-uniform mechanical traction force to the upper conductive nanofiber fabric to overcome the tension constraint of the fabric itself and induce it to undergo controlled buckling deformation, thereby transforming the micro-linear strain of the piezoelectric film into a grid wrinkle morphology at the macro level.

[0192] S7.4: Based on the spatial distribution characteristics of the grid fold morphology, the geometric parameters of the conductive grid openings integrated in the screen substrate are reconstructed in real time to form a shrinking aperture structure at the feed end of the screen surface, an expanding aperture structure in the middle section, and a converging aperture structure at the discharge end.

[0193] Fine geometric feature extraction is performed on the spatial distribution data of the grid fold morphology obtained after processing by S7.3 to obtain a complete parameter set of fold peak and valley positions, amplitudes and local opening ratio variation curves.

[0194] The sequence of fold peak and valley positions is regionalized along the length of the screen surface. The feed end, middle section and discharge end are identified as independent structural control domains, and a mapping index table of fold spatial distribution to opening geometric parameters is established.

[0195] Based on the fold amplitude values ​​in the mapping index table, the aperture change is calculated. A nonlinear geometric transformation model is used to convert the peak height and valley depth of the folds into the aperture reduction and expansion of the current region, respectively, to obtain the aperture change parameters of the sub-region.

[0196] Spatial smoothing is applied to the regional aperture variation parameters, and the abrupt aperture changes caused by uneven local strain of the folds are eliminated by the neighborhood weighted average method, forming a smooth and continuous aperture variation curve to meet the gradual separation requirements of ice particle flow.

[0197] The smooth and continuous aperture change curve is applied to the conductive mesh structure model of the screen substrate, and the opening geometry parameters at the feed end are reconstructed in real time as a shrinking aperture, the opening geometry parameters in the middle section as an expanding aperture, and the opening geometry parameters at the discharge end as a converging aperture.

[0198] Through the above real-time geometric reconstruction process, the microscopic fold morphology of the previous step is transformed into a macroscopically controllable pore size distribution structure, thereby realizing the differentiated flow path design of the ice particle screening area.

[0199] S7.5: Integrate the shrinking pore structure, the expanding pore structure and the converging pore structure to form a continuously changing dynamic pore size distribution zone, which serves as an adaptive interception and diversion control medium when ice particles pass through the sieve surface, thus completing the preparation for closed-loop optimization of particle size distribution within a single sieve cycle.

[0200] Step S8: Based on the dynamic pore size distribution band, the flow of ice particles passing through the sieve surface is intercepted and diverted for regulation. Within a single sieving cycle, the particle size distribution is optimized in a closed-loop manner, and the real-time particle size distribution descriptor is updated to enter the next feedback regulation cycle. Specifically, this includes: S8.1: Based on the spatial geometric topology of the dynamic aperture distribution zone, the mixed ice particle flow entering the feed end of the screen is subjected to initial contact and size pre-screening. The physical blocking effect generated by the inlet tightening aperture structure is used to intercept small ice particles smaller than the target lower limit and prolong their residence time, thereby generating a group of ice particles retained at the feed end and a penetrating ice particle flow with preliminary particle size separation characteristics.

[0201] S8.2: Based on the motion state of the penetrating ice particle flow in the middle section of the screen surface, a differentiated throughput control process is performed on the large-diameter ice particles and medium-diameter ice particles contained therein. The increased aperture probability provided by the central extended aperture structure is used to improve the penetration efficiency of large-diameter ice particles while allowing medium-diameter ice particles to continue to be conveyed forward, generating a middle section diverted ice particle set stratified according to particle size.

[0202] Based on the through-flow of ice particles generated by pre-screening at the feed end, differentiated throughput control is applied to large-diameter and medium-diameter ice particles. The input conditions are the real-time motion parameters of the through-flow ice particles in the middle section of the screen surface, including particle size classification labels, velocity vectors, pore size geometric distribution data, and the value of the expanded pore opening in the middle section. A particle size distribution-pore size matching probability matrix is ​​constructed. The particle size value of large-diameter ice particles is compared with the pore opening in the middle section to form a particle size-pore size coupling factor, and the throughput probability weight is adjusted based on this factor. For medium-diameter ice particles, conveying path tracking is performed, collecting their residence time and peak vibration acceleration within the pore area. The suitability for continued forward conveying is evaluated using a time-dynamic weighting function, and the evaluation results are mapped to the corresponding positions in the pore throughput probability matrix. The target passage probabilities of large-diameter and medium-diameter ice particles are extracted from the aforementioned matrix. The target probability of large-diameter ice particles is amplified by leveraging the increased aperture probability provided by the extended aperture structure, while the original probability of medium-diameter ice particles is retained or slightly adjusted to ensure the difference in flow between the two types of ice particles. The adjusted probability matrix is ​​transformed into a mid-section particle size stratified aperture control sequence, driving the corresponding aperture adjustment actions in the mid-section aperture region of the screen surface to improve the transmission efficiency of large-diameter ice particles and achieve steady-state transport of medium-diameter ice particles. Through this processing method, the penetrating ice particle flow from the previous step is transformed into a mid-section diverted ice particle set stratified by particle size, achieving enhanced spatial separation of particle size distribution and optimized input for subsequent outlet convergence screening.

[0203] S8.3: Based on the spatiotemporal distribution characteristics of the mid-section diversion ice particle collection reaching the discharge end, the remaining unpassed ice particles are subjected to final particle size convergence and homogenization output processing. The secondary screening effect of the outlet convergent aperture structure is used to remove abnormally large particles and force the remaining ice particle group to pass through the standard aperture area, generating a finished uniform ice particle flow with significantly reduced particle size dispersion.

[0204] Based on the spatiotemporal distribution characteristics of the mid-section diversion ice particle set at the discharge end, the preceding mid-section diversion results are used as input, and the geometric parameters and spatial layout description of the outlet convergent aperture structure are loaded to establish a candidate set for the remaining impermeable ice particles. Threshold segmentation is performed on the particle size data of this candidate set, identifying abnormal particles exceeding the target upper limit as rejection targets. The transmission probability is calculated based on the particle moment of inertia and sieve aperture matching condition; particles below the threshold enter the rejection channel. A secondary particle size filter is applied spatially using the physical screening effect of the aperture convergence structure. The local contraction effect at the aperture edge guides abnormal particles away from the main flow, while the vibration inertia of the sieve surface propels the ice particle group conforming to the standard particle size into the standard aperture zone. The remaining ice particle group after secondary screening undergoes homogenization output processing. By controlling the passage length and vibration amplitude of the standard aperture zone, the transmission consistency of ice particles of different sizes is enhanced, forming a uniform ice particle flow with reduced particle size dispersion. The particle size distribution results of the homogenized ice particle flow are encapsulated into an optimized output set and marked as the final particle size convergence result entering the finished product output section.

[0205] By removing abnormally large particles and forcing the remaining ice particle group through the standard aperture zone, the mid-section diversion result is transformed into a finished uniform ice particle flow with significantly reduced particle size dispersion, thus achieving the goal of particle size convergence and quality consistency at the final output.

[0206] S8.4: Based on the deviation relationship between the actual output effect of the finished uniform ice particle flow and the preset target quality, perform closed-loop evaluation and data update processing on the control efficiency of the current screening cycle, and recalculate and refresh the skewness coefficient, peak width index and main peak displacement values ​​in the real-time particle size distribution descriptor using the latest collected heat flow signal, and generate an updated real-time particle size distribution descriptor containing the latest operating condition information.

[0207] Based on the output state of the uniform ice particle flow of the finished product, the real-time particle size distribution descriptor data after screening in the previous sub-step is used as the evaluation input source for this sub-step. For the three statistical indicators in this input source—skewness coefficient, peak width index, and main peak displacement—difference calculation processing is performed with the preset target quality parameter set to form a quantified quality deviation vector, which characterizes the control performance of the current screening cycle. This quality deviation vector is matched temporally with the infrared heat flow signal recently acquired after screening through the dynamic aperture distribution band to ensure that the thermodynamic feature analysis is based on the actual finished product state rather than historical samples. After matching, the heat flow signal is input into the existing thermal decay slope, plateau duration, and hot spot spatial dispersion inversion to recalculate the three-dimensional thermodynamic feature vector of the ice particle group at the current cross-section. Based on this, parameter reconstruction processing is performed using the standard particle size mapping relationship to obtain new values ​​for the skewness coefficient, peak width index, and main peak displacement. For the three newly calculated particle size statistical indicators, an environmental drift correction model is applied to dynamically compensate the values ​​to offset the impact of temperature and humidity changes in the pre-cooling room on heat dissipation analysis, thereby ensuring the environmental robustness of the descriptor. After compensation, the latest values ​​replace the corresponding fields in the original real-time particle size distribution descriptor, generating an updated real-time particle size distribution descriptor containing complete data of the current operating conditions, and outputting it to the subsequent closed-loop decision module through a structured data interface. Through difference calculation, heat flow signal recalculation, and environmental drift compensation processing, the results of the previous step are transformed into an updated real-time particle size distribution descriptor that accurately reflects the control effectiveness of the current screening cycle, achieving the expected technical effects of closed-loop evaluation and data status refresh.

[0208] S8.5: Based on the status identifier of the updated real-time particle size distribution descriptor, the decision logic of the next round of screening control cycle is triggered and iteratively initialized. The updated statistical parameters are fed back to the threshold judgment module as a new input benchmark to start a new round of aperture gradient adjustment command calculation process, generating a continuously rolling adaptive closed-loop control timing sequence.

[0209] The intelligent control method for improving the uniformity of ice particle size in shaken ice provided in this application can be applied to, for example... Figure 4 In the application environment shown, terminal 102 communicates with server 104 via a network. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices, and server 104 can be a standalone server or a server cluster consisting of multiple servers.

[0210] In one embodiment, an intelligent control device for determining the uniformity of ice particle size in shaken ice is provided, comprising: a data acquisition module, a preprocessing module, a feature extraction module, a parameter mapping module, a decision-making module, a control generation module, a drive module, and a closed-loop update module, wherein: Data acquisition module: Miniaturized infrared thermopile sensor arrays are deployed at the feed end, middle section and discharge end of the ice particle screening conveyor line to collect transient thermal radiation attenuation curves caused by the ambient temperature difference during the residence of ice particles on the screen surface, and generate a three-dimensional raw heat flow signal set containing spatial location coordinates, time series stamps and heat flow intensity values.

[0211] Preprocessing module: Performs denoising and normalization preprocessing on the three-dimensional original heat flow signal set to eliminate background thermal noise interference and unify the dimensions, generating a standardized transient heat dissipation spatiotemporal distribution matrix as the reference data source for subsequent feature extraction.

[0212] Feature extraction module: The standardized transient heat dissipation spatiotemporal distribution matrix is ​​input into a temporal convolutional network for feature mining, and three core thermodynamic feature vectors are extracted for each monitoring area: heat attenuation slope, plateau duration, and hot spot spatial dispersion.

[0213] Parameter mapping module: Based on the standard particle size ice sample calibration mapping relationship data table, the core thermodynamic feature vector is mapped and converted into a set of statistical parameters characterizing the dispersion of the current cross-section ice particle size distribution, generating a real-time particle size distribution descriptor containing skewness coefficient, peak width index and main peak displacement.

[0214] Decision module: Determines whether the peak width index in the real-time particle size distribution descriptor continuously exceeds the preset particle size dispersion threshold. If the determination result is yes, a pore size gradient adjustment command is triggered; otherwise, the current screen shape is maintained and the monitoring state is kept.

[0215] Control generation module: Calculates the asymmetric strain distribution curve required in the length direction of the screen surface according to the aperture gradient adjustment command, and generates a gradient voltage control sequence to drive the flexible electro-deformable screen structure to produce millimeter-level aperture gradient changes.

[0216] Drive module: Based on the gradient voltage control sequence, the screen is driven to generate asymmetric strain, constructing a dynamic aperture distribution zone with inlet tightening, middle expansion and outlet convergence.

[0217] Closed-loop update module: Based on the dynamic aperture distribution band, the ice particle flow passing through the sieve surface is intercepted and diverted for regulation. Within a single sieving cycle, the particle size distribution is optimized in a closed loop and the real-time particle size distribution descriptor is updated to enter the next round of feedback regulation cycle.

[0218] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 5As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements an intelligent control method for determining the uniformity of ice particle size in shaken ice.

[0219] In one embodiment, a computer device is provided, the computer device including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of any of the methods in the above method embodiments.

[0220] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of any of the methods described in the above method embodiments.

[0221] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0222] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0223] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The “multiple” mentioned in the embodiments of this application refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0224] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and such modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An intelligent control method for improving the uniformity of ice particle size in shaken ice, specifically including: S1: Collect the transient thermal radiation attenuation curve clusters caused by the ambient temperature difference during the residence of ice particles on the sieve surface, and generate a three-dimensional original heat flow signal set; S2: Perform denoising and normalization preprocessing on the original three-dimensional heat flow signal set to eliminate background thermal noise interference and unify the dimensions, and generate the transient heat dissipation spatiotemporal distribution matrix. S3: Input the transient heat dissipation spatiotemporal distribution matrix into a temporal convolutional network for feature mining, and extract three core thermodynamic feature vectors for each monitoring area: heat attenuation slope, plateau duration, and hot spot spatial dispersion. S4: Based on the standard particle size ice sample calibration mapping relationship data table, the core thermodynamic feature vector is mapped and converted into a set of statistical parameters characterizing the dispersion of the current cross-section ice particle size distribution, generating a real-time particle size distribution descriptor containing skewness coefficient, peak width index and main peak displacement. S5: Determine whether the peak width index in the real-time particle size distribution descriptor continuously exceeds the particle size dispersion threshold. If so, trigger the pore size gradient adjustment command; otherwise, maintain the current screen shape and keep the monitoring status. S6: Calculate the asymmetric strain distribution curve required in the length direction of the screen surface according to the aperture gradient adjustment command, and generate a gradient voltage control sequence to drive the flexible electro-deformable screen structure to produce millimeter-level aperture gradient changes. S7: Based on gradient voltage control sequence, the screen is driven to generate asymmetric strain, and a dynamic aperture distribution zone is constructed with inlet tightening, middle expansion and outlet convergence.

2. The intelligent control method for improving the uniformity of ice particle size in shaken ice according to claim 1, characterized in that, The three-dimensional raw heat flow signal set includes spatial location coordinates, time series stamps, and heat flow intensity values.

3. The intelligent control method for improving the uniformity of ice particle size in shaken ice according to claim 1, characterized in that, The standard particle size ice sample calibration mapping relationship data table includes ice sample ID, thermal attenuation slope, plateau duration, hot spot spatial dispersion, skewness coefficient, peak width index, main peak displacement, and applicable working condition label.

4. The intelligent control method for improving the uniformity of ice particle size in shaken ice according to claim 1, characterized in that, Following S7, the following also includes: S8: Based on the dynamic aperture distribution band, the flow of ice particles passing through the sieve surface is intercepted and diverted, and the particle size distribution is optimized in a single sieving cycle. The real-time particle size distribution descriptor is updated to enter the next round of feedback adjustment cycle.

5. The intelligent control method for improving the uniformity of ice particle size in shaken ice according to claim 2, characterized in that, S3 specifically includes: Based on the standardized transient heat dissipation spatiotemporal distribution matrix, a one-dimensional causal convolutional layer is used to perform local receptive field sliding window calculation on the heat flow intensity values ​​in the time series dimension to capture the short-term thermal radiation attenuation trend of ice particle groups during their residence on the sieve surface, generating a shallow time series feature map containing details of high-frequency thermal fluctuations. Based on shallow temporal feature maps, dilated convolution operations are used to expand the network's receptive field and multi-scale fusion processing is performed on the hot spot distribution information in the spatial location coordinate dimension to associate the heat conduction hysteresis effect of ice particles in different monitoring areas, thereby generating a deep spatiotemporal embedding tensor that represents the medium- and long-term thermal dependence. Based on the deep spatiotemporal embedding tensor, the global average pooling operator is applied to perform dimensionality reduction and aggregation operations on the time series stamp dimension, eliminating instantaneous noise interference and preserving the overall thermal response statistical regularity of each monitoring area, generating a compressed global thermal state feature vector. Based on the global thermal state feature vector, nonlinear transformation and feature decoupling operations are performed through a fully connected mapping layer to separate three orthogonal components that independently characterize the surface micro-melting rate of ice particles, the recondensation duration, and the degree of hot spot dispersion, thereby generating a set of candidate thermodynamic features to be screened. Based on the candidate thermodynamic feature set, feature selection and dimensional reconstruction are performed according to the physical meaning constraint rule set to accurately lock the three core indicators corresponding to thermal decay slope, plateau duration and hot spot spatial dispersion, and finally output the core thermodynamic feature vector for particle size distribution inversion.

6. The intelligent control method for improving the uniformity of ice particle size in shaken ice according to claim 1, characterized in that, S4 specifically includes: Read the calibration mapping data table of standard particle size ice samples, extract the benchmark skewness coefficient, benchmark peak width index and benchmark main peak displacement of standard particle size ice sample batches under specific combinations of thermal decay slope, plateau duration and hot spot spatial dispersion, and construct an initial lookup index library from multidimensional thermodynamic characteristics to particle size statistical parameters. Based on the initial search index library, multidimensional linear interpolation is performed on the input current cross-section core thermodynamic feature vector to eliminate the discrete deviation between the measured thermal decay slope, the measured plateau duration and the measured hot spot spatial dispersion and the standard sample points, and to generate an intermediate particle size distribution estimation parameter set with continuous domain adaptability. Based on the drift of the thermal conductivity coefficient caused by the temperature and humidity fluctuations in the pre-cooling room, dynamic compensation and correction are performed on the values ​​of the intermediate particle size distribution estimation parameter set to offset the influence of environmental noise on the analysis of heat dissipation rate, and the corrected particle size distribution statistics after environmental robustness optimization are output. Based on the corrected particle size distribution statistics, a probability density function reconstruction calculation is performed to transform the discrete statistics into a continuous probability distribution curve, and from this curve, the real-time skewness coefficient characterizing the degree of asymmetry in the current ice particle size distribution, the real-time peak width index characterizing the particle size discrete range, and the real-time main peak displacement characterizing the position of the mainstream particle size are accurately calculated. By integrating real-time skewness coefficient, real-time peak width index, and real-time main peak displacement, a standardized real-time particle size distribution descriptor data structure is generated. This structure serves as the final set of statistical parameters characterizing the dispersion of ice particle size distribution at the current cross-section, which can then be used for threshold comparison and closed-loop control decision-making in subsequent steps.

7. The intelligent control method for uniformity of ice particle size in shaken ice according to claim 1, characterized in that, S5 specifically includes: Based on the real-time particle size distribution descriptor, the peak width index, a core indicator representing the dispersion of particle size distribution, is extracted and used as a quantitative evaluation input for the uniformity of ice particle group screening at the current moment, providing a standardized data benchmark for subsequent threshold comparison. Using the standard operating condition threshold parameter set, the preset particle size dispersion threshold corresponding to the current production batch is called. This preset particle size dispersion threshold is used as a static reference benchmark for judging the stability of the screening process, so as to eliminate the risk of misjudgment caused by environmental fluctuations. The peak width index is continuously monitored by a sliding time window. By setting the number of consecutive sampling periods, it is determined whether the peak width index continuously exceeds the preset particle size dispersion threshold within a specified time series. This filters out instantaneous noise interference and confirms the authenticity of the particle size distribution anomaly, generating an out-of-tolerance judgment result with time-series confidence. Based on the out-of-tolerance judgment result with time-series confidence, a logical branch selection operation is executed. If the judgment result is yes, the aperture gradient adjustment command signal is instantiated. If the judgment result is no, otherwise a holding command to maintain the current screen shape is generated, thereby outputting a deterministic control strategy decision vector to drive the subsequent actuator action. The deterministic control strategy decision vector is encapsulated into a standardized communication protocol data packet and sent to the drive control interface of the flexible electro-deformable screen structure. This completes the construction of a closed-loop feedback link from data analysis to execution triggering, ensuring that the system can respond instantly to the dynamic changes in ice particle size distribution.

8. An intelligent control device for improving the uniformity of ice particle size in shaken ice, characterized in that, The device includes: Data acquisition module: Miniaturized infrared thermopile sensor arrays are deployed at the feed end, middle section and discharge end of the ice particle screening conveyor line to collect transient thermal radiation attenuation curves caused by the ambient temperature difference during the residence of ice particles on the screen surface, and generate a three-dimensional raw heat flow signal set containing spatial location coordinates, time series stamps and heat flow intensity values. Preprocessing module: Performs denoising and normalization preprocessing on the original 3D heat flow signal set, eliminates background thermal noise interference and unifies the dimensions, and generates a standardized transient heat dissipation spatiotemporal distribution matrix as the benchmark data source for subsequent feature extraction. Feature extraction module: Input the standardized transient heat dissipation spatiotemporal distribution matrix into the temporal convolutional network for feature mining, and extract three core thermodynamic feature vectors for each monitoring area: heat attenuation slope, plateau duration, and hot spot spatial dispersion. Parameter mapping module: Based on the standard particle size ice sample calibration mapping relationship data table, the core thermodynamic feature vector is mapped and converted into a set of statistical parameters characterizing the dispersion of the current cross-section ice particle size distribution, generating a real-time particle size distribution descriptor containing skewness coefficient, peak width index and main peak displacement. Decision module: Determines whether the peak width index in the real-time particle size distribution descriptor continuously exceeds the preset particle size dispersion threshold. If the determination result is yes, it triggers the pore size gradient adjustment command; otherwise, it maintains the current screen shape and keeps the monitoring status. Control generation module: Calculates the asymmetric strain distribution curve required in the length direction of the screen surface according to the aperture gradient adjustment command, and generates a gradient voltage control sequence that drives the flexible electro-deformable screen structure to produce millimeter-level aperture gradient changes; Drive module: Based on gradient voltage control sequence, the screen is driven to generate asymmetric strain, and a dynamic aperture distribution zone is constructed with inlet tightening, middle expansion and outlet convergence.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that... When the computer program is executed by a processor, it implements the steps of the method described in any one of claims 1-7.

Images