A method and system for optimizing the precision milling process of a variety of rice

Abstract

The application discloses a kind of adaptation multi-species rice precision milling process optimization method and system, it is related to grain processing intelligent control technical field.The present application extracts sample rice from the rice to be processed to carry out pre-milling test, and constructs segmented milling control model based on quality variation before and after pre-milling;Adopt entropy weight-TOPSIS algorithm to calculate milling response index MRI, rice is divided into different milling process domain based on MRI, corresponding control parameters are matched, and dynamic compensation correction is carried out in combination with raw material characteristics;In the milling process, real-time acquisition operating state data, the fracture precursor prediction model is output fracture precursor index FRI and implements hierarchical closed-loop control, to reduce the risk of broken rice production;Calculate whiteness gain UBG, by analyzing the time series variation trend and attenuation characteristics of whiteness gain, the intelligent determination of optimal milling endpoint is realized.The present application improves the processing adaptability of multi-species rice, rice quality and milling process control precision, adapts to the precision milling process of multi-species rice.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for grain processing, specifically to a method and system for optimizing the precision milling process of rice that is adapted to multiple varieties. Background Technology

[0002] As my country's grain processing industry continues to develop towards intelligence, refinement, and quality, rice, as one of my country's most important grain crops, is directly affected by its processing quality, impacting product market value and enterprise economic benefits. The milling process, a crucial step in paddy processing that determines rice whiteness, integrity, broken rice rate, and milling yield, directly influences the quality of the finished rice. Currently, rice processing enterprises generally employ multi-stage milling processes, adjusting parameters such as milling pressure, roller speed, and feed rate to convert brown rice into polished rice. However, different rice varieties exhibit significant differences in moisture content, hardness, bran structure, and brittleness, leading to substantial variations in milling effects between different batches of rice under the same process parameters. In recent years, domestic and international research has gradually introduced online detection technology, intelligent control technology, and data-driven methods into the rice milling process. For example, Cheng Ke et al., in their paper "Online Detection System for Rice Processing Quality Based on Machine Vision," proposed using machine vision technology to achieve online detection of broken rice rate and bran retention, and adjusting the operating parameters of the rice milling machine based on the detection results. This indicates that online detection and automatic control technologies have begun to be applied in the rice processing process. Studies have shown that unreasonable milling parameter settings, fluctuations in raw grain characteristics, and the lack of effective online control mechanisms are important reasons for increased broken rice rate and unstable quality. Therefore, establishing an intelligent milling control method that can adapt to different raw material characteristics has become an important research direction in the field of rice processing.

[0003] Chinese patent (publication number CN111569982A) discloses a rice milling quality control device, method, and intelligent rice milling system. The method achieves rice milling quality control and automated processing by sampling and testing the rice milling machine outlet sample and adjusting the rice milling process parameters based on the test results. However, this method mainly relies on the quality test results after milling for feedback control, which is a post-event adjustment method. It is difficult to reflect the differences in processing response of different raw materials in the early stage of milling in a timely manner, and it lacks the ability to match dynamic parameters with strong adaptability to raw materials.

[0004] Chinese patent (publication number CN120508024A) discloses an intelligent control method and system for rice refining equipment. The method collects equipment operating status and processing quality information through sensors and performs intelligent control on the processing process to achieve automatic adjustment of the processing process. However, the control strategy of this method mainly focuses on adjusting the equipment operating status and does not adequately consider the processing adaptability problem caused by the difference in physical characteristics of different batches of raw grains. When the characteristics of the raw grains fluctuate greatly, the processing quality may still be unstable.

[0005] Chinese patent (publication number CN110681427A) discloses a method for milling rice using mixed fiber media. By improving the milling media conditions, the milling effect is enhanced, thereby reducing the broken rice rate and improving the quality of rice, which can improve the milling quality to a certain extent. However, this method mainly focuses on the improvement of the milling media and equipment structure, and lacks a dynamic monitoring and prediction mechanism for the real-time changes in the milling process, making it difficult to make adaptive adjustments based on the characteristics of different raw materials and the real-time operating status.

[0006] In summary, existing technologies still have shortcomings in adapting to the differences in processing various rice varieties, dynamically matching milling process parameters, predicting the risk of broken rice in advance, and accurately determining the optimal milling endpoint. There is an urgent need for a precise milling process optimization method and system that can comprehensively consider the raw material quality characteristics, processing response characteristics, and real-time operating status in order to improve the stability of rice processing quality and processing efficiency. Summary of the Invention

[0007] Based on the above-mentioned technical problems, this application discloses a method and system for optimizing the precision milling process of rice adaptable to multiple varieties, wherein the method is specifically as follows:

[0008] Samples of rice were taken from rice to be processed and pre-milled. The rice was milled under preset conditions to obtain the quality parameters of the pre-milled rice.

[0009] A segmented milling control model is constructed. The segmented milling control model calculates the milling response index MRI of the rice to be processed based on the changes in quality parameters before and after pre-milling. The milling process domain corresponding to the rice to be processed is determined based on the milling response index MRI, and segmented milling control parameters corresponding to the milling process domain are generated.

[0010] The formal grinding process is executed according to the segmented grinding control parameters, and the operating status data, whiteness change and energy consumption of the grinding equipment drive motor are collected in real time.

[0011] A fracture precursor prediction model is constructed. The fracture precursor prediction model is based on the operating status data and predicts the fracture precursor index FRI, which reflects the fracture trend of rice grains. The milling is controlled by combining the fracture precursor index FRI with a preset risk threshold.

[0012] Based on the whiteness change and energy consumption, the whiteness gain UBG, which represents the whiteness improvement capability corresponding to unit energy consumption, is calculated. By combining the whiteness gain UBG with a preset termination threshold, it is determined whether the current grinding process has reached the optimal grinding endpoint.

[0013] Preferably, the pre-milling test of the sample rice extracted from the rice to be processed specifically includes the following preset conditions: milling roller speed, milling chamber pressure, single pre-milling time, and feeding rate; the sample rice quality parameters include at least rice grain whiteness, broken rice rate, mass loss rate, rice grain integrity, and bran residue rate.

[0014] Preferably, the segmented milling control model is a multi-input hierarchical matching compensation model, which includes a six-layer structure: feature input layer, MRI calculation layer, process domain discrimination layer, segmented parameter matching layer, raw material disturbance compensation layer, and parameter output layer.

[0015] The input variables of the feature input layer are the quality parameters of the pre-milled rice sample and the auxiliary perturbation input. The auxiliary perturbation input includes the initial moisture content of rice, the impurity rate of rice, and the initial hardness of rice.

[0016] The MRI computation layer uses the entropy weight-TOPSIS algorithm to approximate the ideal solution and sort the results to calculate the grinding response index MRI.

[0017] The process domain discrimination layer adopts the interval threshold discrimination method, and divides at least three independent milling process domains based on MRI values, including low response process domain, medium response process domain, and high response process domain.

[0018] The segmented parameter matching layer is based on the process domain type and matches the corresponding basic grinding parameters, including the pressure of each grinding chamber, the speed of the grinding rollers, the feed speed, and the grinding gap.

[0019] The raw material disturbance compensation layer dynamically corrects the three basic grinding parameters based on the auxiliary disturbance input, thereby offsetting the grinding deviation caused by the differences in the physical properties of the raw materials themselves.

[0020] The parameter output layer outputs the segmented grinding control parameters after compensation and correction.

[0021] Preferably, the entropy weight-TOPSIS approximation ideal solution sorting algorithm specifically involves: constructing a quality change matrix before and after pre-milling, including changes in whiteness, broken rice rate, mass loss rate, rice grain integrity, and bran residue rate, and performing range standardization on each parameter;

[0022] The information entropy value of each indicator is obtained by using the entropy weight method. Based on the information entropy value, the weight of each indicator is determined, and the weighted standardized indicator is calculated.

[0023] The positive ideal solution for the optimal grinding condition and the negative ideal solution for the worst grinding condition are set, and the grinding response index MRI is calculated using the TOPSIS method.

[0024] Preferably, the dynamic correction of the three basic milling parameters specifically involves: based on three auxiliary disturbance inputs—initial moisture content of rice, impurity rate of rice, and initial hardness of rice—segmented linear dynamic compensation is performed on the basic process parameters of the four stages of coarse milling, medium milling, and fine milling, respectively.

[0025] Among them, the moisture content disturbance is due to the significant increase in the plasticity of rice grains after the rice absorbs water. When the moisture content exceeds the standard, the milling pressure and roller speed are adaptively reduced, and the gap compensation amount is increased to avoid large-area breakage of rice grains.

[0026] Impurity rate disturbance: The higher the impurity content, the higher the risk of grinding chamber blockage. When the impurity exceeds the standard, the feed rate should be reduced to smoothly adapt to changes in equipment load.

[0027] When the hardness of rice grains is too high, the rice grains' resistance to grinding is enhanced. The model increases the grinding pressure, ensuring the peeling effect and compensating for the shortcomings of insufficient linear compensation correction.

[0028] Preferably, the fracture precursor prediction model specifically employs an improved LSTM time-series prediction algorithm. It uses four types of normalized long-term operating data—motor current, motor torque, speed fluctuation, and milling chamber vibration amplitude—as model input. The model memorizes the long-term characteristics of stress accumulation throughout the milling process using an LSTM-based gating structure. Simultaneously, an attention mechanism is introduced to adaptively allocate the weights of operating condition data at different time points, enhancing the feature extraction capability of key time nodes of milling stress mutation. The model outputs a predictive fracture precursor index (FRI) in real time. Based on the FRI and a preset risk threshold, graded closed-loop milling control is implemented.

[0029] Preferably, the improved LSTM time series prediction algorithm specifically adopts an Attention-LSTM structure, adds a time series attention layer after the output of the LSTM hidden layer, automatically increases the weight of the time series features at the moment of abnormal fluctuation of rice grain stress, and suppresses the weight of stable, empty, and invalid time series data.

[0030] Based on attention weights and combined with the output of the hidden layer, global temporal features are obtained, and the rupture precursor index is output through the output layer.

[0031] Preferably, the graded closed-loop milling control specifically involves: presetting at least two risk thresholds for graded closed-loop milling control; when the predicted time series FRI < low risk threshold FRI, it is determined that the internal stress of the rice grain is stable and there is no tendency for cracking accumulation, and all process parameters such as the current segmented milling pressure, milling roller speed, and milling gap remain constant.

[0032] When FRI≤FRI≤High-risk thresholdFRI, it is determined that the stress of the rice grains continues to rise, and there is a potential risk of breakage. A stepless smooth control method is adopted to gradually reduce the pressure in the grinding chamber to avoid secondary damage to the rice grains caused by sudden parameter changes.

[0033] When FRI > high-risk threshold FRI, it is determined that a large number of rice grains are about to break. An emergency protection strategy is immediately implemented, which simultaneously reduces the speed of the milling rollers and increases the milling gap to quickly release the accumulated stress inside the rice grains and prevent the formation of broken rice.

[0034] Preferably, determining whether the current grinding process has reached the optimal grinding endpoint specifically involves: continuously collecting multiple sets of whiteness gain (UBG) time-series data and constructing a UBG time-series array;

[0035] Calculate the attenuation slope of UBG within a continuous sampling period to identify the decreasing trend of whiteness gain;

[0036] When the UBG value continues to be lower than the preset termination threshold and the decay slope approaches 0 for three consecutive cycles, it is determined that the improvement of whiteness in milling has reached a bottleneck.

[0037] Immediately issue a stop order for the grinding process, ending this formal grinding operation.

[0038] The system in question is specifically as follows:

[0039] It includes a pre-grinding data acquisition module, a segmented grinding intelligent control module, a grinding condition real-time monitoring module, a rice grain breakage risk prediction module, and a grinding endpoint intelligent determination module, as detailed below:

[0040] The pre-milling data acquisition module extracts rice samples to be processed and conducts a pre-milling test under preset working conditions, collecting quality parameters of the rice samples before and after pre-milling.

[0041] The segmented milling intelligent control module is connected to the pre-milling data acquisition module. It calculates the milling response index MRI based on the change in quality parameters, determines the corresponding milling process domain, and generates and dynamically compensates the segmented milling control parameters.

[0042] The real-time monitoring module for the grinding condition is connected to the segmented grinding intelligent control module. During the actual grinding process, it collects the real-time data of the running status of the grinding equipment drive motor, the change in the whiteness of the rice grains, and the energy consumption of the equipment.

[0043] The rice grain breakage risk prediction module is connected to the real-time monitoring module of the milling condition. Based on the motor running sequence data, it predicts the breakage precursor index FRI through an improved LSTM time sequence prediction model, and completes the closed-loop control of milling by combining the preset risk threshold.

[0044] The intelligent milling endpoint determination module is connected to the real-time milling condition monitoring module and the rice grain breakage risk prediction module. It calculates the whiteness gain UBG based on the whiteness change and energy consumption, and determines the optimal milling endpoint by combining the preset termination threshold.

[0045] Compared with the prior art, the technical solution of this application has the following technical effects:

[0046] This invention achieves quantitative characterization of rice milling response characteristics by conducting pre-milling tests on rice samples before formal milling and constructing a milling response index (MRI) based on the changes in quality parameters before and after pre-milling. This effectively solves the problems in existing technologies where milling parameters mainly rely on experience setting and the adaptability to different varieties and batches of rice is insufficient, thus improving the versatility of multi-variety rice processing and the accuracy of process matching.

[0047] This invention constructs a segmented milling control model, which divides the rice to be processed into different milling process domains based on MRI, and matches the corresponding coarse milling, medium milling and fine milling control parameters. At the same time, it combines the raw material characteristics such as rice moisture content, impurity rate and hardness for dynamic compensation and correction, so that the milling process can be adaptively adjusted according to the changes in raw material properties, effectively reducing the milling quality fluctuation caused by raw material differences, and improving the consistency of whiteness and the stability of rice quality.

[0048] This invention constructs a fracture precursor prediction model, which extracts the temporal features from motor current, torque, speed fluctuations and vibration signals through the Attention-LSTM time-series prediction algorithm, predicts the risk of rice grain fracture in advance and generates a fracture precursor index (FRI), realizing the transformation from traditional post-event rice breakage detection to pre-event risk warning. It can adjust milling parameters in time before large-scale rice grain fracture occurs, thereby effectively reducing the rice breakage rate and improving the head rice rate.

[0049] This invention is based on whiteness gain (UBG). By comprehensively considering the relationship between whiteness enhancement effect and energy consumption, it analyzes the whiteness gain change trend in real time and combines the attenuation slope to achieve optimal milling endpoint identification. This reduces energy consumption and ineffective processing time while ensuring milling quality, thereby improving milling efficiency and energy utilization.

[0050] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the preferred embodiments of this application are described in detail below with reference to the accompanying drawings.

[0051] The above and other objects, advantages and features of this application will become more apparent to those skilled in the art from the following detailed description of specific embodiments in conjunction with the accompanying drawings. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0053] Based on the description of the figures and their corresponding technical content in the document, the titles of the figures are as follows:

[0054] Figure 1 A flowchart of a method for optimizing the precision milling process of rice that is suitable for multiple varieties;

[0055] Figure 2 This is a diagram of the architecture of the segmented grinding control model;

[0056] Figure 3 This is a diagram of the architecture of a rupture precursor prediction model;

[0057] Figure 4 An architecture diagram of a precision milling process optimization system adapted to multiple rice varieties;

[0058] Figure 5 This is a time-series data diagram of the grinding process;

[0059] Figure 6 A schematic diagram showing the processing results of different combinations of raw materials and process domains;

[0060] Figure 7 A comparison chart of compensation effects under different adjustments;

[0061] Figure 8 A comparison chart of the rupture risk early warning performance of different prediction algorithms;

[0062] Figure 9 A comparative chart showing the adaptability of different control methods to various varieties;

[0063] Figure 10 A comparative chart showing the overall performance of different intelligent control methods. Detailed Implementation

[0064] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. In the following description, specific details such as specific configurations and components are provided merely to help fully understand the embodiments of this application. Therefore, those skilled in the art should understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. In addition, for clarity and brevity, descriptions of known functions and structures are omitted in the embodiments.

[0065] It should be understood that the phrase "an embodiment" or "this embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "an embodiment" or "this embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0066] Furthermore, reference numerals and / or letters may be repeated in different examples within this application. Such repetition is for the purpose of simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or settings discussed.

[0067] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" in this article describes another type of relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the related objects before and after it are in an "or" relationship.

[0068] In this article, the term "at least one" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, "at least one of A and B" can mean: A exists alone, A and B exist simultaneously, or B exists alone.

[0069] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion.

[0070] Example 1 mainly describes an optimization method for precision milling of rice suitable for multiple varieties, such as... Figure 1 As shown, it specifically includes:

[0071] Samples of rice were taken from rice to be processed and pre-milled. The rice was milled under preset conditions to obtain the quality parameters of the pre-milled rice.

[0072] A segmented milling control model is constructed. The segmented milling control model calculates the milling response index MRI of the rice to be processed based on the changes in quality parameters before and after pre-milling. The milling process domain corresponding to the rice to be processed is determined based on the milling response index MRI, and segmented milling control parameters corresponding to the milling process domain are generated.

[0073] The formal grinding process is executed according to the segmented grinding control parameters, and the operating status data, whiteness change and energy consumption of the grinding equipment drive motor are collected in real time.

[0074] A fracture precursor prediction model is constructed. The fracture precursor prediction model is based on the operating status data and predicts the fracture precursor index FRI, which reflects the fracture trend of rice grains. The milling is controlled by combining the fracture precursor index FRI with a preset risk threshold.

[0075] Based on the whiteness change and energy consumption, the whiteness gain UBG, which represents the whiteness improvement capability corresponding to unit energy consumption, is calculated. By combining the whiteness gain UBG with a preset termination threshold, it is determined whether the current grinding process has reached the optimal grinding endpoint.

[0076] Furthermore, samples of rice were taken from the rice to be processed for a pre-milling test. Specifically, the preset conditions included the milling roller speed, milling chamber pressure, single pre-milling time, and feeding rate. The quality parameters of the sample rice included at least the whiteness of the rice grains, the broken rice rate, the mass loss rate, the integrity of the rice grains, and the bran residue rate.

[0077] Furthermore, the preset conditions are as follows: the grinding roller speed is set to 1200r / min-1400r / min, the grinding chamber pressure is set to 0.2MPa-0.35MPa, the single pre-grinding time is set to 15s-25s, and the feeding rate is set to 150kg / h-200kg / h.

[0078] Furthermore, the whiteness of rice grains was measured using an intelligent rice whiteness meter. Uniform rice samples were randomly selected, and the average whiteness value was measured under light-proof conditions. The broken rice rate was determined by sieving rice samples using a rice broken rice sorting sieve, weighing the broken rice that passed through the sieve, and the ratio of the broken rice mass to the total rice mass was the broken rice rate. The mass loss rate was calculated by weighing the net weight of rice samples of the same mass before and after pre-milling, and the ratio of the mass difference to the initial rice mass. The rice grain integrity was determined using an image vision inspection device to identify cracks and damage in each sample rice grain, and the percentage of intact rice grains was statistically analyzed. The bran residue rate was determined by detecting the bran residue content on the surface of the rice grains using a near-infrared spectroscopy detector to obtain the bran residue percentage.

[0079] Furthermore, such as Figure 2 The diagram shown is an architecture diagram of the segmented milling control model. Specifically, the segmented milling control model is a multi-input hierarchical matching compensation model, which includes a six-layer structure: feature input layer, MRI calculation layer, process domain discrimination layer, segmented parameter matching layer, raw material disturbance compensation layer, and parameter output layer.

[0080] The input variables of the feature input layer are the quality parameters of the pre-milled rice sample and the auxiliary perturbation input. The auxiliary perturbation input includes the initial moisture content of rice, the impurity rate of rice, and the initial hardness of rice.

[0081] The MRI computation layer uses the entropy weight-TOPSIS algorithm to approximate the ideal solution and sort the results to calculate the grinding response index MRI.

[0082] The process domain discrimination layer adopts the interval threshold discrimination method, and divides at least three independent milling process domains based on MRI values, including low response process domain, medium response process domain, and high response process domain.

[0083] The segmented parameter matching layer is based on the process domain type and matches the corresponding basic grinding parameters, including the pressure of each grinding chamber, the speed of the grinding rollers, the feed speed, and the grinding gap.

[0084] The raw material disturbance compensation layer dynamically corrects the three basic grinding parameters based on the auxiliary disturbance input, thereby offsetting the grinding deviation caused by the differences in the physical properties of the raw materials themselves.

[0085] The parameter output layer outputs the segmented grinding control parameters after compensation and correction.

[0086] Furthermore, the segmented milling process is divided into three consecutive stages: coarse milling, medium milling, and fine milling. Each stage is matched with an independent risk threshold for early signs of breakage and a whiteness gain termination threshold. The coarse milling stage focuses on rapid peeling of the outer layer and relaxes the risk threshold for rice grain breakage. The fine milling stage focuses on protecting the integrity of rice grains and tightens the risk threshold for rice grain breakage, thus achieving multi-stage differentiated and precise risk control and endpoint determination.

[0087] Furthermore, the low-response process domain MRI∈[0,0.3], the medium-response process domain MRI∈[0.3,0.7], and the high-response process domain MRI∈(0.7,1]; the low-response process domain corresponds to rice with hard texture and difficult-to-peel outer layer, the medium-response process domain corresponds to conventional rice, and the high-response process domain corresponds to rice with soft texture and easy to break.

[0088] Furthermore, the entropy-weighted TOPSIS algorithm for approximating the ideal solution ranking is as follows: A quality change matrix is ​​constructed before and after pre-milling, including changes in whiteness, broken rice rate, mass loss rate, grain integrity, and bran residue rate. Each parameter is then standardized using its range. The change in whiteness is a positive indicator, and the standardization formula is:

[0089]

[0090] Other parameters are negative indicators, and the standardized formula is:

[0091]

[0092] in, For the first Standardized values ​​of rice quality indicators For the first The measured changes in several rice quality indicators. , These represent the maximum and minimum values ​​of the corresponding quality index changes in multiple pre-tests, respectively.

[0093] The information entropy value of each indicator is obtained using the entropy weight method, with the following formula:

[0094]

[0095]

[0096] in, For the first The percentage of each quality indicator in the total five indicators. For the first Information entropy of quality indicators This represents the total number of indicator types.

[0097] Based on the information entropy value, the weight of each indicator is determined, and the weighted standardized index is calculated using the following formula:

[0098]

[0099]

[0100] in, For the first The objective weights of various rice quality indicators Weighted standardized indicators;

[0101] The optimal grinding condition (positive ideal solution) and the worst grinding condition (negative ideal solution) are defined. The grinding response index (MRI) is calculated using the TOPSIS method. The formula is as follows:

[0102]

[0103]

[0104] in, This represents the Euclidean distance between the current sample index and the optimal grinding condition. For the positive ideal solution, The Euclidean distance between the current sample index and the worst-case grinding condition. For a negative ideal solution, The grinding response index.

[0105] Furthermore, the specific setting method for positive ideal solution and negative ideal solution is as follows: all indicators of positive ideal solution (optimal grinding condition) take the ideal optimal value, positive indicators take the maximum value of 1, and negative indicators take the minimum value of 0;

[0106] All indices of the negative ideal solution Z (worst grinding condition) are taken as the worst ideal values, the positive indices are taken as the minimum value of 0, and the negative indices are taken as the maximum value of 1.

[0107] Furthermore, the segmented parameter matching layer matches the corresponding basic grinding parameters based on the process domain type, specifically as follows:

[0108] In the low-response process domain (MRI∈[0,0.3), for hard and difficult-to-remove bran rice), the coarse grinding section has a pressure of 0.32MPa, a roller speed of 1350r / min, a feed rate of 190kg / h, and a grinding gap of 0.8mm; the medium grinding section has a pressure of 0.28MPa, a roller speed of 1300r / min, a feed rate of 170kg / h, and a grinding gap of 1.0mm; and the fine grinding section has a pressure of 0.22MPa, a roller speed of 1200r / min, a feed rate of 150kg / h, and a grinding gap of 1.2mm. The entire process utilizes high pressure and high speed to achieve forced peeling of the bran from hard rice.

[0109] The parameters for the coarse grinding stage in the medium response process domain (MRI∈[0.3,0.7], for conventional rice) are: pressure 0.28MPa, roller speed 1300r / min, feed rate 170kg / h, and grinding gap 0.9mm; pressure 0.25MPa, roller speed 1250r / min, feed rate 160kg / h, and grinding gap 1.1mm; and pressure 0.20MPa, roller speed 1150r / min, feed rate 140kg / h, and grinding gap 1.3mm. Standard medium parameters are used to balance grinding whiteness and rice grain integrity.

[0110] In the high-response processing domain (MRI∈(0.7,1], soft and easily broken rice), the coarse grinding section has a pressure of 0.24MPa, a roller speed of 1250r / min, a feed rate of 160kg / h, and a grinding gap of 1.0mm; the medium grinding section has a pressure of 0.21MPa, a roller speed of 1200r / min, a feed rate of 150kg / h, and a grinding gap of 1.2mm; and the fine grinding section has a pressure of 0.18MPa, a roller speed of 1100r / min, a feed rate of 130kg / h, and a grinding gap of 1.4mm. Overall, the grinding intensity is reduced and the grinding gap is increased to avoid a large amount of broken rice in soft rice.

[0111] Furthermore, the three basic milling parameters are dynamically corrected. Specifically, based on the three auxiliary disturbance inputs of initial rice moisture content, rice impurity rate, and initial rice hardness, the basic process parameters of the four stages of coarse milling, medium milling, and fine milling are respectively subjected to piecewise linear dynamic compensation.

[0112] Among them, the moisture content disturbance is due to the significant increase in the plasticity of rice grains after the rice absorbs water. When the moisture content exceeds the standard, the milling pressure and roller speed are adaptively reduced, and the gap compensation amount is increased to avoid large-area breakage of rice grains.

[0113] Impurity rate disturbance: The higher the impurity content, the higher the risk of grinding chamber blockage. When the impurity exceeds the standard, the feed rate should be reduced to smoothly adapt to changes in equipment load.

[0114] When the rice grain hardness is too high, the rice grain's resistance to grinding is enhanced. The model increases the grinding pressure to ensure the peeling effect and compensates for the shortcomings of linear compensation correction. The formula is:

[0115]

[0116] in, To compensate for the post-grinding process parameters, The original basic grinding parameters output by the segmented parameter matching layer. , , These are the measured rice moisture content, rice impurity rate, and rice hardness, respectively. , , These are the process baseline values, namely, baseline moisture content 13.5%, baseline impurity rate 0.8%, and baseline rice grain hardness 420g. , , These are the compensation coefficients corresponding to moisture content, impurity rate, and hardness, respectively, with the compensation coefficients for the three grinding sections assigned independently.

[0117] Furthermore, the operating status data of the grinding equipment drive motor includes four types of time-series data: real-time motor current, real-time motor torque, motor speed fluctuation value, and real-time vibration amplitude of the grinding chamber.

[0118] Furthermore, such as Figure 3 The diagram shows the architecture of the fracture precursor prediction model. Specifically, the fracture precursor prediction model adopts an improved LSTM time series prediction algorithm. It uses four types of normalized long-term operating data of the grinding equipment motor current, motor torque, speed fluctuation value, and grinding chamber vibration amplitude as model input. The model memorizes the long-term characteristics of stress accumulation throughout the grinding process through the LSTM basic gating structure. At the same time, it introduces an attention mechanism to adaptively allocate the weight of the operating data at different time points, enhances the feature extraction capability of key time nodes of grinding stress change, and outputs the fracture precursor index FRI with predictive power in real time. The fracture precursor index FRI is combined with a preset risk threshold to carry out graded closed-loop grinding control.

[0119] Furthermore, the improved LSTM time series prediction algorithm specifically employs an Attention-LSTM structure, adding a time series attention layer after the LSTM hidden layer output. This automatically increases the weight of time series features at times of abnormal fluctuations in grain stress and suppresses the weight of stationary, idle, and invalid time series data. The formula is as follows:

[0120]

[0121] in, For the first Attention weights corresponding to each time step For the output of the LSTM hidden layer, For the attention layer query vector, The attention layer feature mapping weight matrix, For attention layer bias terms, Input the total time step size into the model. The tanh activation function;

[0122] Based on attention weights and combined with the hidden layer output, global temporal features are obtained. The rupture precursor index is then output through the output layer, with the following formula:

[0123]

[0124]

[0125] in, For global time series features, As a precursor index for rupture, , These are the output layer weight matrix and the bias term, respectively.

[0126] Furthermore, the improved LSTM time series prediction algorithm has an input layer with a dimension of 4, corresponding to four time series signals: motor current, motor torque, speed fluctuation, and grinding chamber vibration. The first LSTM layer has 64 hidden layer neurons, with tanh as the activation function and dropout=0.1 set to suppress time series overfitting. The second LSTM layer has 32 hidden layer neurons, inheriting the time series features from the previous layer and retaining long-cycle stress accumulation information. After embedding the time series attention layer into the two-layer LSTM, it completes the adaptive allocation of global time series feature weights. The fully connected output layer has an output dimension of 1, outputting the rupture precursor index FRI at the current moment.

[0127] Furthermore, the graded closed-loop milling control is specifically as follows: at least two risk thresholds are preset for graded closed-loop milling control. When the predicted time series FRI < low risk threshold FRI, it is determined that the internal stress of the rice grain is stable and there is no tendency for cracking accumulation. All process parameters such as the current segmented milling pressure, milling roller speed, and milling gap remain constant.

[0128] When FRI≤FRI≤High-risk thresholdFRI, it is determined that the stress of the rice grains continues to rise, and there is a potential risk of breakage. A stepless smooth control method is adopted to gradually reduce the pressure in the grinding chamber to avoid secondary damage to the rice grains caused by sudden parameter changes.

[0129] When FRI > high-risk threshold FRI, it is determined that a large number of rice grains are about to break. An emergency protection strategy is immediately implemented, which simultaneously reduces the speed of the milling rollers and increases the milling gap to quickly release the accumulated stress inside the rice grains and prevent the formation of broken rice.

[0130] Furthermore, the calculation of whiteness gain UBG is as follows:

[0131]

[0132] in, For whiteness gain, This represents the real-time change in the whiteness of rice grains. The UBG value represents the cumulative energy consumption of the grinding equipment within the corresponding time period. The higher the UBG value, the higher the energy utilization efficiency under the current grinding conditions, and the better the cost-effectiveness of whiteness improvement.

[0133] Furthermore, determining whether the current grinding process has reached the optimal grinding endpoint involves: continuously collecting multiple sets of whiteness gain (UBG) time-series data and constructing a UBG time-series array;

[0134] Calculate the attenuation slope of UBG within a continuous sampling period to identify the decreasing trend of whiteness gain;

[0135] When the UBG value continues to be lower than the preset termination threshold and the decay slope approaches 0 for three consecutive cycles, it is determined that the improvement of whiteness in milling has reached a bottleneck.

[0136] Immediately issue a stop order for the grinding process, ending this formal grinding operation.

[0137] This embodiment details a method for optimizing the precision milling process of rice suitable for multiple varieties. Sample rice is extracted from the rice to be processed for pre-milling experiments, and a segmented milling control model is constructed based on the quality changes before and after pre-milling. The entropy weight-TOPSIS algorithm is used to calculate the milling response index MRI. Based on the MRI, the rice is divided into different milling process domains, and corresponding control parameters are matched. Simultaneously, dynamic compensation and correction are performed in conjunction with raw material characteristics. During the milling process, real-time operating status data is collected, and the fracture precursor index FRI is output through a fracture precursor prediction model and graded closed-loop control is implemented to reduce the risk of broken rice. The whiteness gain UBG is calculated, and by analyzing the temporal change trend and attenuation characteristics of the whiteness gain, intelligent determination of the optimal milling endpoint is achieved.

[0138] Example 2, based on Example 1, details a precision milling process optimization system for multiple rice varieties, including a pre-milling data acquisition module, a segmented milling intelligent control module, a milling condition real-time monitoring module, a rice grain breakage risk prediction module, and a milling endpoint intelligent determination module, as follows:

[0139] The pre-milling data acquisition module extracts rice samples to be processed and conducts a pre-milling test under preset working conditions, collecting quality parameters of the rice samples before and after pre-milling.

[0140] The segmented milling intelligent control module communicates with the pre-milling data acquisition module, calculates the milling response index MRI based on the change in quality parameters, determines the corresponding milling process domain, and generates and dynamically compensates the segmented milling control parameters.

[0141] The real-time monitoring module for milling conditions is connected to the segmented milling intelligent control module. During the actual milling process, it collects real-time data on the operating status of the milling equipment drive motor, the change in rice grain whiteness, and the energy consumption of the equipment.

[0142] The rice grain breakage risk prediction module is connected to the milling condition real-time monitoring module. Based on the motor running sequence data, it predicts the breakage precursor index FRI through an improved LSTM time sequence prediction model, and completes the milling closed-loop control by combining the preset risk threshold.

[0143] The intelligent milling endpoint determination module is connected to the real-time milling condition monitoring module and the rice grain breakage risk prediction module. It calculates the whiteness gain UBG based on the whiteness change and energy consumption, and determines the optimal milling endpoint by combining the preset termination threshold.

[0144] Furthermore, the pre-grinding data acquisition module includes a pre-grinding execution unit and a multi-dimensional quality detection unit;

[0145] Among them, the pre-grinding execution unit is preset with fixed grinding conditions, the grinding roller speed is 1200r / min-1400r / min, the grinding chamber pressure is 0.2MPa-0.35MPa, the single pre-grinding time is 15s-25s, and the feeding rate is 150kg / h-200kg / h.

[0146] The multi-dimensional quality testing unit can complete the testing of five rice quality parameters, namely rice grain whiteness, broken rice rate, quality loss rate, rice grain integrity, and bran residue rate. The supporting testing equipment includes an intelligent rice whiteness meter, broken rice sorting sieve, high-precision electronic scale, image vision inspection equipment, and near-infrared spectroscopy detector.

[0147] This embodiment describes in detail a precision rice milling process optimization system adapted to multiple varieties. It consists of a milling data acquisition module, a segmented milling intelligent control module, a milling condition real-time monitoring module, a rice grain breakage risk prediction module, and a milling endpoint intelligent determination module. The modules work together to optimize the precision rice milling process.

[0148] Example 3, based on Example 1 or 2, details the implementation and verification of the optimized rice milling process for multiple varieties described in this invention, using japonica rice as the research object. The specific details are as follows:

[0149] The raw materials used were cleaned and purified japonica rice, with a total weight of 2000 kg. Before formal processing, the basic characteristics of the raw materials were tested. The test results showed that the moisture content of this batch of rice was 13.8%, the impurity rate was 0.7%, the grain hardness was 438 g, the original whiteness was 8.6, and the original broken rice rate was 0.9%.

[0150] 100g of rice was randomly selected from the raw materials to be processed for a pre-milling test. The pre-milling test was conducted in a laboratory rice milling apparatus. The milling roller speed was set to 1300 r / min, the milling chamber pressure was set to 0.28 MPa, the single pre-milling time was set to 20s, and the feeding speed was controlled at 180 kg / h. After the pre-milling was completed, the quality parameters of the rice sample were tested. The changes in whiteness, broken rice rate, mass loss rate, rice grain integrity, and bran residue rate before and after pre-milling were obtained. The data are shown in Table 1 below.

[0151] Table 1 Results of changes in quality parameters before and after pre-grinding

[0152] Whiteness 8.6 24.8 +16.2 Broken rice rate 0.9% 1.7% +0.8% Quality loss rate 0% 1.6% +1.6% Rice grain integrity 99.1% 98.2% -0.9%

[0153] The above quality change data were input into the segmented milling control model. The model standardized each quality index, determined the objective weight of each index using the entropy weight method, and calculated the milling response index MRI of this batch of rice using the TOPSIS approximation ideal solution sorting algorithm. The calculated MRI value was 0.56. According to the preset process domain division rules, when the MRI is in the range of 0.3 to 0.7, it corresponds to the medium response process domain. The system determined that this batch of rice belongs to the medium response process domain.

[0154] After completing the process domain identification, the system automatically calls the basic parameter library corresponding to the process domain to obtain the basic grinding parameters for the three stages of coarse grinding, medium grinding, and fine grinding. The system further detects the auxiliary disturbance parameters of the current raw material and inputs the measured moisture content of 13.8%, impurity rate of 0.7%, and hardness of 438g into the raw material disturbance compensation layer. The model uses the baseline moisture content of 13.5%, the baseline impurity rate of 0.8%, and the baseline hardness of 420g as reference values ​​to dynamically compensate and correct the basic parameters of each stage, obtaining the segmented grinding control parameters for formal production, as shown in Table 2 below:

[0155] Table 2 Control parameters for segmented milling

[0156] coarse mill 0.287MPa 1315r / min 172kg / h 0.89mm medium mill 0.258MPa 1263r / min 162kg / h 1.08mm Fine grinding 0.206MPa 1162 r / min 142kg / h 1.28mm

[0157] The formal grinding process was carried out according to the parameters in Table 2. During the entire grinding process, the system collected real-time operating status data such as motor current, motor torque, speed fluctuation value, and grinding chamber vibration amplitude, and continuously input the data into the fracture precursor prediction model for analysis with a sampling period of 1 second. The fracture precursor prediction model adopts the Attention-LSTM structure, which learns the correlation between the equipment operating status and the change of force on the rice grains, and outputs the fracture precursor index FRI in real time.

[0158] The process data from monitoring the grinding process are shown in Table 3 below:

[0159] Table 3 Time-series monitoring data of the grinding process

[0160] 1min coarse mill 0.89mm 0.18 18.6 2.4kWh — 2min coarse mill 0.89mm 0.23 22.1 4.8kWh 1.46 3min coarse mill 0.89mm 0.29 25.6 7.2kWh 1.46 4min coarse mill 0.89mm 0.36 28.7 9.7kWh 1.24 5min coarse mill 0.89mm 0.41 31.4 12.3kWh 1.04 6min medium mill 1.08mm 0.46 34.0 14.8kWh 1.04 7min medium mill 1.08mm 0.54 36.1 17.4kWh 0.81 8min medium mill 1.08mm 0.61 37.8 20.1kWh 0.63 9min medium mill 1.08mm 0.68 39.1 22.9kWh 0.46 10min medium mill 1.08mm 0.73 40.2 25.8kWh 0.38 11min Fine grinding (risk control) 1.35mm 0.60 41.3 28.8kWh 0.37 12min Fine grinding 1.28mm 0.53 41.8 31.8kWh 0.17 13min Fine grinding 1.28mm 0.47 42.1 34.9kWh 0.10 14min Fine grinding 1.28mm 0.42 42.3 38.0kWh 0.06 15min Fine grinding 1.28mm 0.39 42.42 41.0kWh 0.04 16min Fine grinding 1.28mm 0.36 42.53 44.0kWh 0.037 17min Fine grinding 1.28mm 0.34 42.63 47.1kWh 0.032

[0161] like Figure 5 The time-series data diagram of the milling process shown indicates that as the milling process continues, the model detects that the internal stress of the rice grains gradually accumulates. When the milling process reaches about 10 minutes, the FRI reaches 0.73, exceeding the preset high-risk threshold. Based on this, the system determines that there is a risk of mass breakage of rice grains and automatically activates the high-risk protection strategy, appropriately reducing the milling pressure and the speed of the milling rollers, while increasing the milling gap, thereby releasing the accumulated stress inside the rice grains and preventing the broken rice rate from increasing further.

[0162] The system continuously monitors online whiteness changes and equipment energy consumption, and calculates the whiteness gain (UBG) based on real-time data. As the grinding process progresses, the rate of whiteness improvement gradually slows down, while the whiteness improvement capacity per unit of energy consumption continues to decline. When grinding reaches the 14th minute, the UBG for multiple consecutive sampling cycles is below the preset termination threshold of 0.05, and its decay slope gradually stabilizes. Based on this, the system determines that the current whiteness improvement has entered a plateau period, and continuing grinding will not yield significant quality gains. Therefore, it automatically issues a shutdown command to end the current grinding process.

[0163] After milling, the finished rice was subjected to quality testing, and the test results are shown in Table 3:

[0164] Table 4. Results of Quality Inspection of Finished Rice

[0165] Whiteness 43.2 Broken rice rate 3.8% Head rice yield 68.4% Rice grain integrity 96.2% Unit energy consumption 18.0 kWh / t

[0166] As shown in Table 4, this invention establishes a raw material milling response evaluation mechanism through pre-milling experiments. Combined with MRI process domain matching, raw material disturbance compensation, crack precursor prediction, and whiteness gain endpoint determination, it achieves adaptive optimization control of the entire milling process. For this batch of rice, while ensuring a high whiteness level, it effectively suppresses broken rice production, improves the head rice rate, and avoids increased energy consumption caused by ineffective over-milling. Experimental results show that this invention can automatically optimize milling parameter configuration according to different raw material characteristics, improve processing quality stability and energy utilization efficiency, and has good engineering application value.

[0167] This embodiment details the implementation and verification of the precision milling process optimization method for multiple rice varieties described in this invention, using japonica rice as the research object. During the process, raw material properties are detected, and samples are taken for pre-milling to obtain quality change data. The milling response index is calculated using the entropy weight-TOPSIS algorithm to divide the process domain, matching corresponding segmented milling parameters and compensating for parameters based on raw material disturbances. Segmented milling is then carried out based on the optimized parameters. During processing, time-series data from the collected equipment is input into an improved LSTM model to predict breakage risk and dynamically adjust the milling conditions. Simultaneously, whiteness and energy consumption data are collected to calculate whiteness gain and automatically determine the milling endpoint. Finished product testing results demonstrate that this invention can adaptively control the processing process according to raw material characteristics, balancing finished product quality, rice grain integrity, and processing energy consumption.

[0168] Example 4, based on Example 3, uses the same milling equipment, testing equipment, and environmental conditions as Example 2, only changing the type of rice raw material to be processed. Hard japonica rice, conventional japonica rice, and soft indica rice were selected as research objects. 100g of rice samples were randomly selected from each group for pre-milling. The pre-milling conditions were consistent with Example 2: milling roller speed 1300r / min, milling chamber pressure 0.28MPa, pre-milling time 20s, and feeding speed 180kg / h. Subsequently, the changes in quality parameters before and after pre-milling were detected, and the results were input into the segmented milling control model to calculate MRI. The MRI calculation results are shown in Table 5 below.

[0169] Table 5. MRI calculation results for different rice raw materials

[0170] Hard japonica rice 13.4% 482g 0.24 conventional japonica rice 13.8% 438g 0.56 Soft Indica Rice 14.1% 371g 0.83

[0171] The process domain was partitioned according to the process domain partitioning rules, and the results are shown in Table 6.

[0172] Table 6. MRI Process Domain Division Results

[0173] Hard japonica rice 0.24 Low-response process domain conventional japonica rice 0.56 Mid-response process domain Soft Indica Rice 0.83 High-response process domain

[0174] After determining the process domain, the system calls the corresponding parameter library to perform formal grinding. The low-response process domain employs a high-pressure, high-speed intensive grinding strategy; the medium-response process domain employs a standard grinding strategy; and the high-response process domain employs a low-pressure, large-gap flexible grinding strategy. Furthermore, different raw materials are combined with other process domains to compare the effectiveness of this method. After grinding, the quality indicators of each group of samples are measured, and the results are shown in Table 7 below.

[0175] Table 7 Processing results of different combinations of raw materials and process domains

[0176] Hard japonica rice Low-response process domain 42.8 4.2% Hard japonica rice Mid-response process domain 40.6 4.8% Hard japonica rice High-response process domain 36.8 3.9% conventional japonica rice Low-response process domain 44.0 6.5% conventional japonica rice Mid-response process domain 43.2 3.8% conventional japonica rice High-response process domain 39.5 3.4% Soft Indica Rice Low-response process domain 43.1 8.7% Soft Indica Rice Mid-response process domain 42.7 5.6% Soft Indica Rice High-response process domain 42.5 3.5%

[0177] According to Table 7 and Figure 6 The schematic diagrams showing the processing results of different raw materials and process domain combinations indicate that different raw materials achieve the best overall processing effect when corresponding to their respective MRI-matched process domains. Among them, hard japonica rice achieves a whiteness of 42.8 and a head rice rate of 67.6% under the low-response process domain; conventional japonica rice achieves a whiteness of 43.2 and a head rice rate of 68.4% under the medium-response process domain; and soft indica rice achieves a head rice rate of 69.1% and a minimum broken rice rate of 3.5% under the high-response process domain.

[0178] This embodiment describes in detail the milling process optimization process for different types of processed rice raw materials using this method. The results show that the MRI process domain division mechanism established by this invention can effectively characterize the milling response characteristics of different raw materials and achieve precise matching between process parameters and raw material characteristics, thereby obtaining better comprehensive indicators of whiteness, broken rice rate and head rice rate.

[0179] Example 5, based on Example 4, verifies the impact of the dynamic compensation mechanism of the present invention on the stability of processing quality when moisture content, impurity rate, and rice grain hardness change, as follows:

[0180] Conventional japonica rice with an MRI value of 0.56 was uniformly selected as the experimental subject. The basic parameters of the medium response process domain determined in Example 2 were used as the initial parameters, namely, coarse mill pressure 0.287 MPa, medium mill pressure 0.258 MPa, and fine mill pressure 0.206 MPa, coarse mill roller speed 1315 r / min, medium mill roller speed 1263 r / min, and fine mill roller speed 1162 r / min. During the experiment, the equipment model, ambient temperature, total feed amount, and sampling method were kept consistent, and only the raw material physical properties were changed.

[0181] Four groups of rice with moisture contents of 12.5%, 13.5%, 14.5%, and 15.5% were selected for the experiment. One group was processed directly using the original fixed process parameters, while the other group was dynamically corrected using a raw material disturbance compensation layer. The quality indicators were tested after milling, as shown in Table 8 below:

[0182] Table 8 Comparison of compensation effects under different moisture contents

[0183] 12.5% no 42.9 5.8% 65.7% 12.5% yes 42.8 4.2% 68.1% 13.5% no 43.0 4.0% 68.2% 13.5% yes 43.1 3.9% 68.4% 14.5% no 43.3 5.4% 66.7% 14.5% yes 43.2 4.0% 68.3% 15.5% no 43.5 7.1% 63.9% 15.5% yes 43.1 4.5% 67.8%

[0184] To verify the effectiveness of impurity rate disturbance compensation, experimental groups with impurity rates of 0.3%, 0.8%, 1.5%, and 2.2% were constructed, with 0.8% serving as the baseline. Equipment load changes and quality results were recorded during the experiments, and the results are shown in Table 9 below.

[0185] Table 9 Comparison of compensation effects under different impurity rates

[0186] 0.3% no 43.1 3.9% 68.4% 0.3% yes 43.1 3.8% 68.5% 0.8% no 43.0 4.0% 68.2% 0.8% yes 43.1 3.9% 68.4% 1.5% no 42.7 5.6% 66.1% 1.5% yes 42.9 4.3% 67.8% 2.2% no 42.2 7.4% 63.4% 2.2% yes 42.7 4.8% 67.1%

[0187] According to Tables 8 and 9, Figure 7 The comparison chart showing the compensation effects under different adjustments demonstrates... Figure 7(a) The comparison of compensation effects under different moisture contents shows that when the moisture content deviates from the benchmark value by 13.5%, the broken rice rate of the uncompensated processing scheme increases significantly, especially under the high moisture content condition of 15.5%, where the broken rice rate reaches 7.1%; while after compensation is enabled, the system automatically reduces the milling pressure and roller speed, and increases the milling gap, reducing the broken rice rate to 4.5% and increasing the head rice rate to 67.8%; Figure 7 (b) The comparison of compensation effects under different impurity rates shows that as the impurity rate increases, the equipment load fluctuation intensifies; the broken rice rate increases significantly when compensation is not enabled, and the system balances the equipment load changes by reducing the feeding speed after compensation is enabled, so that the processing quality remains stable.

[0188] This embodiment details the experimental verification of the impact of the dynamic compensation mechanism of the present invention on the stability of processing quality when the moisture content, impurity rate, and rice grain hardness change. The results verify that the raw material disturbance compensation mechanism established by the present invention can dynamically correct the basic milling parameters according to the changes in moisture content and impurity rate. Even if the physical properties of the raw materials fluctuate greatly, the system can still maintain a stable whiteness level and a low broken rice rate. The experimental results prove that the present invention has strong raw material adaptability and process robustness.

[0189] Example 6, based on Example 5, details the experimental implementation process for verifying the early warning accuracy of the Attention-LSTM rupture precursor prediction model and the improvement effect on processing quality after closed-loop regulation, as follows:

[0190] The conventional japonica rice raw material in Example 2 was used uniformly, with an MRI of 0.56, corresponding to the middle response process domain. The moisture content of the raw material was controlled at 13.8%, the impurity rate was controlled at 0.7%, and the grain hardness was controlled at 438g. The milling equipment parameters were executed according to the segmented milling control parameters determined in Example 2. During the experiment, the batch of raw materials, equipment status, environmental conditions, and total feed amount were kept consistent.

[0191] Six algorithms were used to construct rupture precursor prediction models: BP neural network, random forest (RF), XGBoost, GRU, LSTM, and Attention-LSTM of this invention. All models used the same training and testing samples. The output of each model was the rupture precursor index FRI. When the predicted FRI exceeded the high-risk threshold, process control was triggered.

[0192] The time difference between the time when the statistical model issues a risk warning and the actual time when the rice fragments rapidly increase during the experiment is defined as the warning lead time. The prediction errors of each algorithm model are statistically analyzed, and the results are shown in Table 10.

[0193] Table 10 Comparison of the performance of different prediction algorithms for rupture risk early warning

[0194] BP neural network 24.6s 0.089 0.115 Random Forest (RF) 13.3s 0.051 0.072 XGBoost 28.7s 0.046 0.068 GRU 41.6s 0.058 0.070 LSTM 67.9s 0.053 0.064 This invention 76.2s 0.037 0.049

[0195] According to Table 10 and Figure 8 As shown in the comparison chart of the rupture risk warning performance of different prediction algorithms, the Attention-LSTM of this invention introduces an attention mechanism on the basis of LSTM, which further improves the early warning capability while maintaining a low prediction error, and achieves simultaneous optimization of prediction accuracy and prediction timeliness.

[0196] Each model was combined with the same hierarchical closed-loop control strategy to perform formal milling on the same batch of raw materials. After processing, the quality indicators of the finished product were tested, and the results are shown in Table 11.

[0197] Table 11 Comparison of processing results after closed-loop control using different prediction algorithms

[0198] BP neural network 42.7 5.8% 65.9% 18.7kWh / t Random Forest (RF) 43.1 4.7% 67.4% 18.4 kWh / t XGBoost 43.3 4.5% 67.6% 18.6 kWh / t GRU 43.0 4.3% 67.8% 18.2 kWh / t LSTM 43.2 4.1% 68.0% 18.3 kWh / t This invention 43.2 3.8% 68.4% 18.0 kWh / t

[0199] As shown in Table 11, the Attention-LSTM algorithm of this invention can achieve a lower broken rice rate, a higher whole rice rate, and a lower unit energy consumption, while maintaining the same excellent level of whiteness index. Its overall milling control effect is better than other comparative algorithms.

[0200] This embodiment verifies the early warning accuracy of the Attention-LSTM fracture precursor prediction model and the effect of closed-loop control on improving processing quality. The experimental results show that the model of the present invention exhibits superior comprehensive performance in terms of prediction accuracy, early warning timeliness, and processing quality improvement. It can effectively improve the risk control capability of the grinding process and provide a reliable decision basis for subsequent closed-loop control.

[0201] Example 7, based on Example 6, details a comprehensive performance comparison experiment between the existing mainstream milling control method and the proposed method, as follows:

[0202] Three types of typical rice—hard japonica rice, conventional japonica rice, and soft indica rice—were selected as experimental subjects. All experiments used the same type of rice milling equipment, the same processing environment, and the same batch of raw materials. Each group of experiments was repeated 10 times and the average value was taken.

[0203] The comparison methods include WFC (Whiteness Feedback Control), ECOC (Energy Optimization Control), FLC (Fuzzy Logic Control), BPNN (BP Neural Network Control), LSTM (Long Short-Term Memory Network Control) and the method of this invention. Among them, WFC (Whiteness Feedback Control) collects rice grain whiteness data in real time during the milling process, dynamically adjusts the milling chamber pressure and milling roller speed according to the deviation between the current whiteness and the target whiteness, and continuously executes the whiteness detection and parameter adjustment process. When the whiteness reaches the set target value, the milling process ends.

[0204] ECOC (Energy Optimization Control) collects real-time data on the operating power and cumulative energy consumption of the grinding equipment, constructs an energy consumption evaluation index per unit output, and dynamically adjusts the grinding pressure, roller speed and feeding speed according to the energy consumption change trend. The grinding process ends when the processing quality meets the set requirements and the unit energy consumption reaches the optimization target.

[0205] FLC (Fuzzy Logic Control) acquires operating parameters such as whiteness, equipment power and output in real time, inputs each parameter into the fuzzy inference system, calculates the control output according to the preset fuzzy rule base, and dynamically adjusts the grinding pressure, grinding roller speed and feeding speed according to the inference results to complete the online control of the whole process.

[0206] BPNN (BP Neural Network Control) collects equipment operating parameters and quality inspection data, constructs a BP neural network quality prediction model, uses the model to predict the current processing quality status, and dynamically adjusts the grinding process parameters based on the prediction results to achieve intelligent control of the processing process.

[0207] LSTM (Long Short-Term Memory Network Control) collects time-series operating data such as motor current, motor torque, equipment vibration, and speed changes, constructs an LSTM time-series prediction model, uses the model to predict the quality change trend during the grinding process, and dynamically adjusts the grinding process parameters based on the prediction results to achieve intelligent time-series control throughout the entire process.

[0208] The method of this invention obtains information on the quality changes of rice samples through pre-milling tests, uses MRI to complete the analysis of the milling response characteristics of raw materials and the identification of process domains, and completes segmented parameter compensation by combining the moisture content, impurity rate and hardness of raw materials. During the formal milling process, the Attention-LSTM model is used to predict the fracture precursor index (FRI) in real time and perform graded closed-loop control. At the same time, the milling endpoint is determined based on the whiteness gain (UBG), thus completing the precise milling control of multiple rice varieties.

[0209] The head rice yield obtained under three raw material conditions was statistically analyzed, and the standard deviation between different varieties was calculated as an adaptability evaluation index. The results are shown in Table 12.

[0210] Table 12 Comparison of multi-variety adaptability of different control methods

[0211] WFC 66.3% 66.9% 64.5% 1.23 ECOC 65.8% 67.4% 65.8% 0.77 FLC 67.0% 67.2% 66.1% 0.61 BPNN 66.7% 67.7% 66.4% 0.56 LSTM 67.6% 67.5% 67.1% 0.38 Method of the present invention 68.0% 68.4% 68.7% 0.29

[0212] According to Table 12 and Figure 9 As shown in the comparison chart of the multi-variety adaptability of different control methods, the method of the present invention maintains a high head rice yield and has the smallest standard deviation under three typical raw material conditions. This indicates that the present invention can automatically match appropriate process parameters according to the milling response characteristics of raw materials and has stronger cross-variety adaptability.

[0213] After processing, the whiteness, broken rice rate, head rice rate, and unit energy consumption were tested. The results are shown in Table 13.

[0214] Table 13 Comparison of the overall performance of different intelligent control methods

[0215] WFC 43.1 5.2% 66.9% 18.9kWh / t ECOC 42.5 4.8% 67.2% 17.8 kWh / t FLC 43.3 4.6% 67.5% 18.4 kWh / t BPNN 43.2 4.5% 67.7% 18.6 kWh / t LSTM 43.1 4.1% 68.0% 18.3 kWh / t Method of the present invention 43.4 3.8% 68.4% 18.0 kWh / t

[0216] According to Table 13 and Figure 10 As shown in the comparison chart of the comprehensive performance of different intelligent control methods, the method of the present invention achieves the lowest broken rice rate and the highest head rice rate while maintaining a high whiteness level, thus realizing the synergistic optimization of processing quality and energy consumption level.

[0217] This embodiment describes in detail the selection of three types of rice and the comparative test conducted under unified experimental conditions between the present invention and five existing milling control schemes. The multi-raw material adaptability and comprehensive processing indicators of each scheme were compared. The test results show that the present invention has a higher head rice rate, smaller fluctuations in indicators between varieties, and better cross-raw material adaptability in the processing of various types of rice. At the same time, it can maintain ideal whiteness, lower broken rice rate, higher head rice rate and lower processing energy consumption. The comprehensive control performance is superior to all the comparative schemes.

[0218] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any changes, modifications, substitutions, integrations, and parameter changes made to these embodiments within the spirit and principles of the present invention, without departing from the principles and spirit of the present invention, through conventional substitutions or to achieve the same function, fall within the scope of protection of the present invention.

Claims

1. A method for optimizing the precision milling process of rice adaptable to multiple varieties, characterized in that, include: Samples of rice were taken from rice to be processed and pre-milled. The rice was milled under preset conditions to obtain the quality parameters of the pre-milled rice. A segmented milling control model is constructed. The segmented milling control model calculates the milling response index MRI of the rice to be processed based on the changes in quality parameters before and after pre-milling. The milling process domain corresponding to the rice to be processed is determined based on the milling response index MRI, and segmented milling control parameters corresponding to the milling process domain are generated. The formal grinding process is executed according to the segmented grinding control parameters, and the operating status data, whiteness change and energy consumption of the grinding equipment drive motor are collected in real time. A fracture precursor prediction model is constructed. The fracture precursor prediction model is based on the operating status data and predicts the fracture precursor index FRI, which reflects the fracture trend of rice grains. The milling is controlled by combining the fracture precursor index FRI with a preset risk threshold. Based on the whiteness change and energy consumption, the whiteness gain UBG, which represents the whiteness improvement capability corresponding to unit energy consumption, is calculated. By combining the whiteness gain UBG with a preset termination threshold, it is determined whether the current grinding process has reached the optimal grinding endpoint.

2. The method for optimizing the precision milling process of rice adaptable to multiple varieties as described in claim 1, characterized in that, The pre-milling test of rice samples taken from the rice to be processed specifically includes the following preset conditions: milling roller speed, milling chamber pressure, single pre-milling time, and feeding rate; the quality parameters of the rice samples include at least rice grain whiteness, broken rice rate, mass loss rate, rice grain integrity, and bran residue rate.

3. The method for optimizing the precision milling process of rice adaptable to multiple varieties as described in claim 1, characterized in that, The segmented milling control model is specifically a multi-input hierarchical matching compensation model, which includes a six-layer structure: feature input layer, MRI calculation layer, process domain discrimination layer, segmented parameter matching layer, raw material disturbance compensation layer, and parameter output layer. The input variables of the feature input layer are the quality parameters of the pre-milled rice sample and the auxiliary perturbation input. The auxiliary perturbation input includes the initial moisture content of rice, the impurity rate of rice, and the initial hardness of rice. The MRI computation layer uses the entropy weight-TOPSIS algorithm to approximate the ideal solution and sort the results to calculate the grinding response index MRI. The process domain discrimination layer adopts the interval threshold discrimination method, and divides at least three independent milling process domains based on MRI values, including low response process domain, medium response process domain, and high response process domain. The segmented parameter matching layer is based on the process domain type and matches the corresponding basic grinding parameters, including the pressure of each grinding chamber, the speed of the grinding rollers, the feed speed, and the grinding gap. The raw material disturbance compensation layer dynamically corrects the three basic grinding parameters based on the auxiliary disturbance input, thereby offsetting the grinding deviation caused by the differences in the physical properties of the raw materials themselves. The parameter output layer outputs the segmented grinding control parameters after compensation and correction.

4. The method for optimizing the precision milling process of rice adaptable to multiple varieties as described in claim 3, characterized in that, The entropy weight-TOPSIS approximation ideal solution sorting algorithm is as follows: construct a quality change matrix before and after pre-milling, including changes in whiteness, broken rice rate, mass loss rate, rice grain integrity, and bran residue rate, and perform range standardization on each parameter. The information entropy value of each indicator is obtained by using the entropy weight method. Based on the information entropy value, the weight of each indicator is determined, and the weighted standardized indicator is calculated. The positive ideal solution for the optimal grinding condition and the negative ideal solution for the worst grinding condition are set, and the grinding response index MRI is calculated using the TOPSIS method.

5. The method for optimizing the precision milling process of rice adaptable to multiple varieties according to claim 4, characterized in that, The dynamic correction of the three basic milling parameters is specifically as follows: based on the three auxiliary disturbance inputs of initial rice moisture content, rice impurity rate and initial rice hardness, the basic process parameters of coarse milling, medium milling and fine milling are respectively linearly and dynamically compensated in segments. Among them, the moisture content disturbance is due to the significant increase in the plasticity of rice grains after the rice absorbs water. When the moisture content exceeds the standard, the milling pressure and roller speed are adaptively reduced, and the gap compensation amount is increased to avoid large-area breakage of rice grains. Impurity rate disturbance: The higher the impurity content, the higher the risk of grinding chamber blockage. When the impurity exceeds the standard, the feed rate should be reduced to smoothly adapt to changes in equipment load. When the hardness of rice grains is too high, the rice grains' resistance to grinding is enhanced. The model increases the grinding pressure, ensuring the peeling effect and compensating for the shortcomings of insufficient linear compensation correction.

6. The method for optimizing the precision milling process of rice adaptable to multiple varieties as described in claim 5, characterized in that, The aforementioned fracture precursor prediction model specifically employs an improved LSTM time-series prediction algorithm. It uses four types of normalized long-term operating data—motor current, motor torque, speed fluctuation, and milling chamber vibration amplitude—as model input. The model memorizes the long-term characteristics of stress accumulation throughout the milling process using an LSTM-based gating structure. Simultaneously, an attention mechanism is introduced to adaptively allocate weights to operating data at different time points, enhancing the feature extraction capability of key time nodes in milling stress mutation. The model outputs a predictive fracture precursor index (FRI) in real time. Based on the FRI and a preset risk threshold, graded closed-loop milling control is implemented.

7. The method for optimizing the precision milling process of rice adaptable to multiple varieties as described in claim 6, characterized in that, The improved LSTM time series prediction algorithm specifically adopts an Attention-LSTM structure and adds a time series attention layer after the output of the LSTM hidden layer to automatically increase the weight of time series features at the moment of abnormal fluctuation of rice grain stress and suppress the weight of stationary empty invalid time series data. Based on attention weights and combined with the output of the hidden layer, global temporal features are obtained, and the rupture precursor index is output through the output layer.

8. The method for optimizing the precision milling process of rice adaptable to multiple varieties according to claim 7, characterized in that, The graded closed-loop milling control is specifically as follows: at least two risk thresholds are preset for graded closed-loop milling control. When the predicted time series FRI < the low risk threshold FRI, it is determined that the internal stress of the rice grain is stable and there is no tendency for cracking accumulation. All process parameters such as the current segmented milling pressure, milling roller speed, and milling gap are kept constant. When FRI≤FRI≤High-risk thresholdFRI, it is determined that the stress of the rice grains continues to rise, and there is a potential risk of breakage. A stepless smooth control method is adopted to gradually reduce the pressure in the grinding chamber to avoid secondary damage to the rice grains caused by sudden parameter changes. When FRI > high-risk threshold FRI, it is determined that a large number of rice grains are about to break. An emergency protection strategy is immediately implemented, which simultaneously reduces the speed of the milling rollers and increases the milling gap to quickly release the accumulated stress inside the rice grains and prevent the formation of broken rice.

9. The method for optimizing the precision milling process of rice adaptable to multiple varieties as described in claim 8, characterized in that, The determination of whether the current grinding process has reached the optimal grinding endpoint is specifically as follows: continuously collect multiple sets of whiteness gain UBG time series data and construct a UBG time series array; Calculate the attenuation slope of UBG within a continuous sampling period to identify the decreasing trend of whiteness gain; When the UBG value continues to be lower than the preset termination threshold and the decay slope approaches 0 for three consecutive cycles, it is determined that the improvement of whiteness in milling has reached a bottleneck. Immediately issue a stop order for the grinding process, ending this formal grinding operation.

10. A precision milling process optimization system for multiple rice varieties, used to implement the precision milling process optimization method for multiple rice varieties as described in any one of claims 1-9, comprising a pre-milling data acquisition module, a segmented milling intelligent control module, a real-time milling condition monitoring module, a rice grain breakage risk prediction module, and a milling endpoint intelligent determination module, as detailed below: The pre-milling data acquisition module extracts rice samples to be processed and conducts a pre-milling test under preset working conditions, collecting quality parameters of the rice samples before and after pre-milling. The segmented milling intelligent control module is connected to the pre-milling data acquisition module. It calculates the milling response index MRI based on the change in quality parameters, determines the corresponding milling process domain, and generates and dynamically compensates the segmented milling control parameters. The real-time monitoring module for the grinding condition is connected to the segmented grinding intelligent control module. During the actual grinding process, it collects the real-time data of the running status of the grinding equipment drive motor, the change in the whiteness of the rice grains, and the energy consumption of the equipment. The rice grain breakage risk prediction module is connected to the real-time monitoring module of the milling condition. Based on the motor running sequence data, it predicts the breakage precursor index FRI through an improved LSTM time sequence prediction model, and completes the closed-loop control of milling by combining the preset risk threshold. The intelligent milling endpoint determination module is connected to the real-time milling condition monitoring module and the rice grain breakage risk prediction module. It calculates the whiteness gain UBG based on the whiteness change and energy consumption, and determines the optimal milling endpoint by combining the preset termination threshold.

Images