A rice large-scale processing production parameter intelligent adjustment control system and method

Abstract

The application relates to the technical field of intelligent control of rice processing, and discloses a rice large-scale processing production parameter intelligent adjustment control system and method, wherein the system is provided with a transient physical property sensing module arranged on the upstream of a rice mill, physical properties of material flow are collected in real time, an edge calculation and control unit calculates a dynamic feedforward time benchmark according to the physical properties, predicts the physical property state of the material section when arriving at a milling area in the future, and solves an optimal control parameter vector, a high dynamic response execution mechanism receives the parameter and completes predictive adjustment of the processing parameters of the rice mill before the material arrives, meanwhile, a post-positioned quality detection and optimization module installed at a discharge port on-line detects the quality of finished products, and quantitative data of the module are used for periodically optimizing the mapping relationship between the material state and the control parameters in the system. Through combination of feedforward prediction and feedback optimization, the application realizes dynamic and accurate control of processing parameters, can effectively reduce the broken rice rate, and enhances the self-adaptive capacity of the system.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for rice processing, specifically to an intelligent adjustment and control system and method for large-scale rice processing production parameters. Background Technology

[0002] Rice is one of the world's main staple foods, and the rice milling process is crucial in its commercial processing. The rice milling process aims to remove the bran layer and germ from the surface of brown rice while preserving the integrity of the rice grain to the greatest extent possible. The quality of the finished product, such as the broken rice rate and the precision of milling, are key indicators for measuring its economic value.

[0003] In existing large-scale rice production lines, mechanical rice milling machines are typically used for continuous operation. The processing parameters of these machines, such as milling pressure and the gap between the rice blades and the milling rollers, are often kept constant throughout the entire batch processing, or adjusted manually and infrequently by operators based solely on experience. However, as a natural agricultural product, paddy rice or brown rice inherently possesses non-uniform physical properties. Even within the same batch of raw materials, the moisture content, hardness, brittleness, and particle size of different parts fluctuate dynamically in real time.

[0004] The uncertainty of these material properties poses a serious challenge to traditional static parameter control methods. When material segments with high hardness or low moisture content enter the milling zone, a fixed milling pressure may be insufficient to effectively remove the bran layer, resulting in uneven whitening and substandard milling precision. Conversely, when material segments with low hardness or high moisture content enter, the same pressure may be too high, easily causing rice grain breakage and significantly increasing the broken rice rate in the finished product.

[0005] While some existing technologies attempt to introduce feedback control, such as adjusting processing parameters by online detection of finished product quality, these methods inherently suffer from hysteresis. From the time an anomaly in finished product quality is detected until the control system responds and adjusts parameters, a large amount of material has already been processed under unsuitable parameters, making real-time optimization of the processing difficult. Furthermore, some pre-detection methods only provide average material characteristics and cannot capture instantaneous changes in material flow over time; therefore, their control accuracy is limited and cannot fundamentally solve the quality instability problem caused by transient fluctuations in material.

[0006] Therefore, how to perceive the transient physical characteristics of the material flow about to enter the milling zone in real time and feedforward, and thereby achieve predictive and dynamic precise adjustment of the processing parameters of the rice milling machine, has become a technical bottleneck for improving the automation level of rice processing and the uniformity of finished product quality. Summary of the Invention

[0007] The purpose of this invention is to provide an intelligent adjustment and control system and method for large-scale rice processing parameters, solving the technical problem in existing technologies where static or delayed adjustment of processing parameters due to real-time changes in the physical properties of materials leads to unstable finished product quality. The core of this invention lies in using feedforward, multimodal transient physical characteristic sensing of the material flow entering the rice milling machine, combined with accurate prediction of the material state, to achieve dynamic and predictive adjustment of key processing parameters of the rice milling machine. Furthermore, by forming a closed-loop optimization through online monitoring of finished product quality, this significantly improves the yield and quality uniformity of rice processing.

[0008] To achieve the above objectives, the first aspect of the present invention provides an intelligent adjustment and control system for large-scale rice processing parameters, the system comprising:

[0009] A transient physical characteristic sensing module is installed in the material channel upstream of the rice milling machine's feed inlet to collect real-time data on the material flow about to enter the rice milling machine and obtain raw sensing signals characterizing the material's physical properties.

[0010] A high dynamic response actuator, integrated into the rice milling machine, is used to receive control commands and physically adjust the processing parameters of the rice milling machine.

[0011] A post-processing quality inspection and optimization module is installed at the outlet of the rice milling machine to detect the quality indicators of the finished product online and output quantitative quality data.

[0012] An edge computing and control unit is connected to the aforementioned modules. This edge computing and control unit is used for:

[0013] The original sensing signal is received, and the corrected instantaneous flow velocity is calculated based on the original sensing signal, thereby determining a dynamic feedforward time reference.

[0014] Based on the original sensing signal and the dynamic feedforward time reference, the physical characteristics of the material segment when it arrives at the grinding zone in the future are predicted. Based on the predicted future state, the optimal control parameter vector is calculated through a mapping function between the material state and the control parameters.

[0015] The optimal control parameter vector is sent as a control command to the high dynamic response actuator;

[0016] The system receives quantitative quality data from the post-quality detection and optimization module and uses the quantitative quality data to periodically optimize the mapping function between the material state and control parameters.

[0017] In one specific implementation, the transient physical characteristic sensing module includes a high-frequency acoustic impact sensing array and a terahertz spectral scanning array. The high-frequency acoustic impact sensing array is used to acquire acoustic signals related to the mechanical properties of the material flow, and the terahertz spectral scanning array is used to acquire spectral data related to the internal moisture content of the material flow. The sensing areas of the two arrays are spatially aligned to achieve synchronous data acquisition of the same material segment.

[0018] In one specific implementation, the high-frequency acoustic impact sensor array consists of multiple acoustic impact sensors arranged linearly and equidistantly along the material movement direction. The edge computing and control unit determines the time delay of the signals by calculating the cross-correlation function of the two sensor signals, and calculates the original instantaneous flow velocity of the material flow based on the time delay and the spacing between the sensors.

[0019] In a specific implementation, the edge computing and control unit is logically divided into a signal preprocessing and feature extraction submodule, an instantaneous flow rate calculation and correction submodule, and a state prediction and decision-making submodule.

[0020] The signal preprocessing and feature extraction submodule is used to extract acoustic feature vectors and terahertz feature vectors from acoustic signals and spectral data, respectively, and to construct transient physical feature vectors by splicing them together.

[0021] The instantaneous velocity calculation and correction submodule is used to calculate the original instantaneous velocity and input the aforementioned transient physical feature vector into a velocity estimation correction model to obtain a correction coefficient. The corrected instantaneous velocity is obtained by multiplying the original instantaneous velocity by the correction coefficient.

[0022] The state prediction and decision-making submodule is used to input the historical transient physical feature vector sequence and the dynamic feedforward time base into a time series prediction model to predict the physical characteristics of the material segment when it arrives at the grinding zone in the future; then, the predicted future state is input into the mapping function between the material state and the control parameters to calculate the optimal control parameter vector.

[0023] In one specific implementation, the high dynamic response actuator includes a servo motor for adjusting the pressure between the grinding roller and the rice sieve, and a piezoelectric ceramic actuator for adjusting the gap between the rice knife and the grinding roller.

[0024] In one specific implementation, the post-quality inspection and optimization module includes an image acquisition unit for photographing the finished product and an image processing unit for analyzing the acquired images to calculate the broken rice rate.

[0025] A second aspect of this invention provides a method for intelligent adjustment and control of production parameters in large-scale rice processing, the method comprising the following steps:

[0026] Data acquisition: The raw sensor signals of the material flow about to enter the rice milling machine are collected in real time through a transient physical characteristic sensing module deployed upstream of the rice milling machine inlet.

[0027] Feature extraction and state prediction: In the edge computing and control unit, the original sensing signals are processed to construct transient physical feature vectors, the corrected instantaneous flow velocity is calculated and the dynamic feedforward time reference is determined, and then the time series prediction model is called to predict the physical characteristics of the material segment when it arrives at the grinding zone in the future.

[0028] Decision and Execution: Based on the predicted future state, the edge computing and control unit calculates the optimal control parameter vector through a mapping function between material state and control parameters, and sends the optimal control parameter vector as a control command to the high dynamic response actuator to adjust the processing parameters of the rice milling machine within the dynamic feedforward time reference.

[0029] Online detection and optimization: The post-quality detection and optimization module installed at the discharge port of the rice milling machine detects the quality indicators of the finished product online and generates quantitative quality data. The edge computing and control unit receives the quantitative quality data and periodically optimizes the mapping function between the material state and the control parameters accordingly.

[0030] This invention provides an intelligent adjustment and control system and method for large-scale rice processing parameters. It has the following beneficial effects:

[0031] 1. This invention utilizes a transient physical characteristic sensing module installed upstream of the rice milling machine to collect key physical characteristics such as hardness and moisture content of the material flow about to enter the milling zone. An edge computing and control unit then predicts the future state of the material segment upon arrival at the milling zone based on this data and calculates the optimal control parameters. Finally, a high-dynamic-response actuator precisely adjusts these parameters before the material arrives. This predictive control method ensures that each segment of material is processed under optimized parameters, fundamentally avoiding over-milling or under-milling caused by parameter mismatch, thereby maximizing the preservation of rice grain integrity.

[0032] 2. This invention employs a post-processing quality detection and optimization module to perform online quantitative detection of quality indicators such as the broken rice rate of the finished product, and feeds this real-time quality data back to the edge computing and control unit. This unit uses this feedback data to periodically optimize and adjust the mapping function between the internal material state and control parameters. This closed-loop optimization mechanism enables the system to autonomously learn and adapt to the effects of raw material batch changes, environmental temperature and humidity fluctuations, and even equipment wear, ensuring the long-term effectiveness of the control strategy and the continuous stability of processing quality.

[0033] 3. This invention calculates the original flow velocity of the material using the cross-correlation of signals from a high-frequency acoustic impact sensor array, and innovatively uses a transient physical feature vector constructed from acoustic and terahertz multimodal data to correct it, obtaining a corrected instantaneous flow velocity highly correlated with the actual flow characteristics of the material. Based on this precise flow velocity calculation and the resulting dynamic feedforward time reference, combined with high dynamic response actuators such as servo motors and piezoelectric ceramic actuators, it ensures that control commands can be precisely executed within a millisecond timescale, achieving precise spatiotemporal synchronization between control actions and material flow, effectively improving control accuracy and response speed. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the intelligent adjustment and control system for large-scale rice processing production parameters according to an embodiment of the present invention;

[0035] Figure 2 This is a schematic diagram illustrating the workflow of an intelligent adjustment and control method according to an embodiment of the present invention.

[0036] Among them, 40 is the rice milling machine; 41 is the feed inlet; 42 is the material channel; 43 is the grinding roller; 44 is the rice knife; and 45 is the discharge outlet.

[0037] 100. Transient physical characteristic sensing module; 110. High-frequency acoustic impact sensing array; 120. Terahertz spectral scanning array; 121. Terahertz emission source; 122. Terahertz detector; 200. Edge computing and control unit; 300. High dynamic response actuator; 310. Servo motor; 320. Piezoelectric ceramic actuator; 400. Post-processing quality detection and optimization module; 410. Image acquisition unit; 420. Image processing unit. Detailed Implementation

[0038] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] See attached document Figure 1 The system works in conjunction with a rice milling machine 40 to achieve intelligent adjustment and control of the processing parameters of the rice milling machine 40.

[0040] The system includes: a transient physical characteristic sensing module 100, an edge computing and control unit 200, a high dynamic response actuator 300, and a post-processing quality detection and optimization module 400. The edge computing and control unit 200, as the core of the system, establishes electrical or communication connections with the transient physical characteristic sensing module 100, the high dynamic response actuator 300, and the post-processing quality detection and optimization module 400 to achieve signal transmission and data interaction.

[0041] A transient physical property sensing module 100 is installed in the material channel 42 upstream of the feed inlet 41 of the rice milling machine 40. This module is used to collect real-time data on the material flow that is about to enter the milling zone of the rice milling machine 40 in order to obtain raw sensing signals characterizing the physical properties of the material.

[0042] In one specific embodiment, the transient physical characteristic sensing module 100 includes a high-frequency acoustic impact sensor array 110 and a terahertz spectral scanning array 120. The high-frequency acoustic impact sensor array 110 consists of multiple acoustic impact sensors 111 arranged linearly and equidistantly along the direction of material movement within the material channel 42. The terahertz spectral scanning array 120 includes a terahertz emission source 121 and a terahertz detector 122, which are arranged on opposite sides of the material channel 42, forming a transmission scanning structure.

[0043] The scanning area of ​​the terahertz spectral scanning array 120 and the sensing area of ​​the high-frequency acoustic impact sensing array 110 are spatially overlapped or adjacent. This arrangement allows the two arrays to acquire data on the same material segment flowing through the region within the same time window, thereby ensuring the physical correspondence between acoustic features and terahertz spectral features in subsequent data processing.

[0044] The edge computing and control unit 200 is an industrial computer integrating a graphics processing unit (GPU) or a field-programmable gate array (FPGA). This unit receives raw sensing data from the transient physical characteristic sensing module 100, performs data processing, feature extraction, state prediction, and control decision-making algorithms according to the method provided by this invention, and finally generates control commands.

[0045] A high dynamic response actuator 300 is integrated into the key parameter adjustment section of the rice milling machine 40 to receive and precisely execute control commands from the edge computing and control unit 200. In one specific embodiment, the actuator includes a servo motor 310 for adjusting the pressure between the milling roller 43 and the rice sieve, and a piezoelectric ceramic actuator 320 for adjusting the gap between the rice knife 44 and the milling roller 43. These actuators possess millisecond-level fast response capabilities matching the system's decision-making speed.

[0046] A post-processing quality inspection and optimization module 400 is installed at the discharge port 45 of the rice milling machine 40 to detect the quality indicators of the finished product online. In one specific embodiment, this module is a machine vision-based online inspection system, including an image acquisition unit 410 that continuously or periodically captures images of the finished product at the discharge port, and an image processing unit 420 that analyzes the acquired images to calculate indicators such as broken rice rate. The image processing unit 420 sends the analyzed quality data to the edge computing and control unit 200 to provide feedback for long-term self-optimization of the control strategy.

[0047] See attached document Figure 1 and attached Figure 2 The workflow of the system of the present invention is designed to achieve predictive adjustment of the processing parameters of the rice milling machine 40.

[0048] During system operation, rice material flows continuously through the material channel 42 upstream of the feed inlet 41 of the rice milling machine 40. When the material flows through the sensing area of ​​the transient physical characteristic sensing module 100, the high-frequency acoustic impact sensing array 110 and the terahertz spectral scanning array 120 in the module synchronously acquire data of the current material segment, respectively obtaining the original multi-channel acoustic time series signal and the original terahertz spectral data.

[0049] The transient physical characteristic sensing module 100 transmits the acquired raw sensing signals to the edge computing and control unit 200 in real time. After receiving the data, the edge computing and control unit 200 executes a series of parallel processing tasks. This unit processes the sensing signals, extracts features that characterize the physical properties of the current material segment, such as hardness and moisture content, and constructs a multi-dimensional transient physical feature vector.

[0050] Simultaneously, the edge computing and control unit 200 calculates the original instantaneous flow velocity of the material flow using the signal time delay of different sensors 111 in the high-frequency acoustic impact sensor array 110. Subsequently, the unit uses the aforementioned constructed transient physical feature vector to correct the calculated original instantaneous flow velocity, thereby obtaining a corrected instantaneous flow velocity associated with the actual physical properties of the material. Based on this corrected flow velocity, the unit further calculates the dynamic feedforward time reference required for the material segment to reach the core grinding zone of the rice milling machine 40 from the sensing area.

[0051] Next, the edge computing and control unit 200 uses a time series prediction model based on the historical transient physical feature vector sequence and the dynamic feedforward time reference to predict the physical characteristics of the material segment when it reaches the core grinding zone in the future.

[0052] Based on the predicted future state, the edge computing and control unit 200 queries or calculates a set of optimal grinding control parameters, such as the specific values ​​of grinding pressure and kat gap, through a preset mapping relationship between material state and control parameters. This unit then sends these control parameters as instructions to the high dynamic response actuator 300.

[0053] Upon receiving an instruction, the high dynamic response actuator 300 immediately drives its included servo motor 310 or piezoelectric ceramic actuator 320 to adjust the corresponding physical components of the rice milling machine 40. This adjustment is completed within a time period determined by the dynamic feedforward time reference, ensuring that the rice milling machine 40 is in the optimal processing parameter setting state before the corresponding material segment reaches the core milling zone.

[0054] Furthermore, when the processed material flows through the discharge port 45, the post-processing quality detection and optimization module 400 performs online detection of the finished product quality and feeds back quantitative quality data such as the broken rice rate to the edge computing and control unit 200. The edge computing and control unit 200 uses this long-term actual quality data to periodically optimize and adjust the mapping relationship between the material state and control parameters established internally, thereby ensuring the overall control performance of the system remains robust during long-term operation.

[0055] See attached document Figure 1 The function of the transient physical characteristic sensing module 100 is to perform high-frequency synchronous data acquisition on the material flow that continuously passes through the material channel 42 in order to obtain the original sensing signal for subsequent analysis.

[0056] In one specific embodiment, the transient physical characteristic sensing module 100 includes a high-frequency acoustic impact sensing array 110 and a terahertz spectral scanning array 120. These two arrays are integrated and installed in a separate modular structure and fixed to the inner or outer wall of the material channel 42 upstream of the feed inlet 41 of the rice milling machine 40.

[0057] The high-frequency acoustic impact sensor array 110 consists of multiple independent acoustic impact sensors 111. In one embodiment, these sensors are piezoelectric ceramic acoustic impact sensors with a wide frequency response range, such as from several kilohertz to several megahertz, to effectively capture high-frequency acoustic emission signals released when materials are impacted and microcracks are generated.

[0058] These acoustic impact sensors 111 are positioned at preset fixed intervals along the direction of gravity-induced material descent within the material channel 42. The sensors are installed in a linear array. This spatial layout forms the technical basis for calculating the material flow rate using signal cross-correlation methods. During operation, rice grains or clusters of rice grains in the falling material flow physically impact the sensing surfaces of each sensor 111. The mechanical stress wave generated by the impact deforms the piezoelectric material and outputs an instantaneous voltage signal related to the impact force. Therefore, the high-frequency acoustic impact sensor array 110 outputs a set of multi-channel, time-dependent voltage waveform signals.

[0059] The terahertz spectral scanning array 120 is used to acquire information on the moisture content within the material flow. In one embodiment, the array employs terahertz time-domain spectroscopy and consists of a terahertz emission source 121 and a terahertz detector 122. The terahertz emission source 121 and the terahertz detector 122 are respectively mounted on opposite sides of the material channel 42, forming a transmission-type measurement optical path, allowing the emitted terahertz beam to pass through the falling material flow.

[0060] The effective scanning area of ​​the terahertz spectral scanning array 120 is spatially aligned with the sensing area of ​​the high-frequency acoustic impact sensing array 110. This alignment ensures that the terahertz spectral data and the acoustic impact signal data represent the same material segment passing through the sensing area within the same time window. During operation, the terahertz emitter 121 emits a broadband terahertz pulse, which is received by the terahertz detector 122 after penetrating the material flow. Due to the strong absorption effect of water molecules on terahertz waves in a specific frequency band, the terahertz absorption spectrum of the material flow can be calculated by analyzing the attenuation and delay of the received signal relative to a reference signal (when no material is passing through).

[0061] The high-frequency acoustic impact sensor array 110 is a component of the transient physical characteristic sensing module 100, and its function is to collect raw signals related to the mechanical properties of the material flow.

[0062] In one specific embodiment, the high-frequency acoustic impact sensing array 110 is composed of multiple acoustic impact sensors 111. The acoustic impact sensors 111 are piezoelectric ceramic sensors with a wide frequency response range, for example, from 1 kHz to 2 MHz. This frequency range is selected to effectively capture low-frequency impact vibration signals generated when material particles or particle clusters impact the sensor's sensing surface, as well as high-frequency acoustic emission signals generated by the propagation of internal microcracks during the impact process.

[0063] These acoustic impact sensors 111 are installed on the inner wall of the material channel 42, at a pre-set, precise vertical spacing along the direction of the material's downward gravity. They are distributed in a linear array. This defined spatial geometry is a necessary prerequisite for the subsequent edge computing and control unit 200 to calculate the instantaneous velocity of the material flow.

[0064] The array operates based on the piezoelectric effect. When material particles in the material flow physically collide with the sensing surface of an acoustic impact sensor 111, the transient mechanical stress wave generated by the impact is transmitted to the piezoelectric material inside the sensor, causing it to deform. The piezoelectric material generates an electric charge due to the deformation, forming an instantaneous voltage signal related to the magnitude and frequency of the impact force.

[0065] Therefore, during system operation, the high-frequency acoustic impact sensor array 110 synchronously and continuously outputs a set of multi-channel time-series voltage signals. Each channel in this signal set corresponds to an acoustic impact sensor 111 in the array, and its signal waveform carries information about the physical properties of the material flow, such as hardness and brittleness, when it impacts the sensor location.

[0066] The terahertz spectral scanning array 120 is a component of the transient physical property sensing module 100, and its function is to acquire raw signals related to the moisture content inside the material flow.

[0067] In one specific implementation, the array employs terahertz time-domain spectroscopy and consists of a terahertz emission source 121 and a terahertz detector 122. The terahertz emission source 121 and the terahertz detector 122 are respectively installed on opposite sides of the material channel 42, forming a transmission-type measurement optical path, so that the emitted terahertz beam can pass through the falling material flow.

[0068] The array works by having a terahertz emitter 121 emit a broadband terahertz pulse, which is received by a terahertz detector 122 after penetrating the material flow. Since water molecules absorb terahertz waves at specific frequencies, the terahertz absorption spectrum of the material segment can be calculated by analyzing the time-domain delay and amplitude attenuation of the received terahertz pulse signal relative to a reference signal (i.e., the signal when no material is passing through), and then converting it to the frequency domain using a Fourier transform.

[0069] The effective scanning area of ​​the terahertz spectral scanning array 120 is spatially aligned with the sensing area of ​​the high-frequency acoustic impact sensing array 110. This alignment configuration ensures that the terahertz spectral data and the acoustic impact signal data represent the same material segment passing through the sensing area within the same time window, providing a physical basis for subsequent multimodal data fusion and feature extraction.

[0070] See attached document Figure 1The edge computing and control unit 200 is the core processing unit of the system of the present invention. Its hardware foundation is an industrial computer equipped with a graphics processing unit (GPU) for performing high-throughput parallel computing tasks.

[0071] The internal software architecture of the edge computing and control unit 200 is logically divided into multiple cooperating sub-modules. In one specific implementation, these sub-modules include: a signal preprocessing and feature extraction sub-module 210, an instantaneous flow rate calculation and correction sub-module 220, and a state prediction and decision-making sub-module 230.

[0072] In a preferred embodiment, in order for the edge computing and control unit 200 to operate effectively, the flow rate estimation correction model contained within it... Time series prediction models and the mapping function between material state and control parameters This requires an initial offline training and calibration process to generate. This process specifically includes the following steps:

[0073] First, data collection and calibration are performed. Representative rice samples covering different origins, batches, moisture contents, and other physical characteristics are selected and processed under controlled conditions using the system of this invention. During this process, the following sets of data are recorded simultaneously:

[0074] 1) Multi-channel time-series voltage signals acquired by the transient physical characteristic sensing module 100 and raw terahertz spectral data ;

[0075] 2) The actual instantaneous flow velocity of the material flow obtained through high-speed cameras or other benchmark measurement methods;

[0076] 3) A series of different combinations of control parameters for the high dynamic response actuator 300 that are manually set (e.g., multiple sets of different grinding pressures and kat gaps).

[0077] 4) The finished product quantitative quality indicators (e.g., broken rice rate) detected by the post-quality detection and optimization module 400, corresponding to each of the above control parameters. Thus, a multi-dimensional correlated dataset is constructed, containing "original sensor signals, actual material properties, control parameters, and processing results".

[0078] Secondly, model training and mapping relationship construction are performed. Using the dataset constructed above:

[0079] Correction model for flow velocity estimation Train it to extract transient physical feature vectors from the raw sensor signals. Output a correction factor This coefficient enables the corrected instantaneous flow velocity Minimize the error between the actual instantaneous flow velocity measured by the reference measuring equipment.

[0080] Time series forecasting models The system is trained to accurately predict the material's feature vectors in the future based on historical transient physical feature vector sequences.

[0081] Construct a mapping function between material state and control parameters. By analyzing the relationships between the predicted material feature vectors, control parameters, and broken rice rate in the dataset, the future... Material feature vector at time step Determine the optimal control parameter vector that minimizes the broken rice rate. This establishes a mapping function between the material state and control parameters. This function can be a multidimensional lookup table, a fitted polynomial function, or another independent machine learning model.

[0082] After offline training and calibration are completed, the generated models and functions are embedded or loaded into the edge computing and control unit 200 as the initial basis for its online prediction and decision-making. During the subsequent long-term operation of the system, these models and functions can also be fine-tuned and optimized online using real-time feedback data provided by the post-processing quality inspection and optimization module 400.

[0083] The signal preprocessing and feature extraction submodule 210 receives the raw sensing signal from the transient physical characteristic sensing module 100. This submodule first performs digital filtering on the signal to remove noise, and then performs windowing processing on the continuous signal stream. For each time window, the submodule extracts feature parameters from the acoustic signal and terahertz spectral data respectively, and constructs a transient physical feature vector. .

[0084] Specifically, for the multi-channel acoustic signal output by the high-frequency acoustic impact sensor array 110, this submodule applies Fast Fourier Transform (FFT) to convert it to the frequency domain and calculates features such as energy distribution and peak frequency within a specific frequency band. These features constitute the acoustic feature sub-vector. For the spectral data output by the terahertz spectral scanning array 120, this submodule analyzes the absorption intensity and peak width at the positions of the moisture characteristic absorption peaks. These features constitute the terahertz characteristic subvector. Ultimately, the transient physical feature vector is formed by concatenating these two sub-vectors:

[0085] ,in, Indicates the current moment.

[0086] The instantaneous flow velocity calculation and correction submodule 220 and the signal preprocessing and feature extraction submodule 210 operate in parallel. This submodule selects two physically adjacent sensors in the high-frequency acoustic impact sensing array 110, for example... and The corresponding signal is and By calculating the cross-correlation function of these two signals... Time delay for determining signal mode :

[0087] ;

[0088] ;

[0089] in, This is the time shift. If the vertical distance between these two sensors is... Then the original instantaneous flow velocity for: ;

[0090] The submodule then calls a pre-trained flow rate estimation correction model. The model uses the transient physical feature vector output by the signal preprocessing and feature extraction submodule 210. As input, output a dimensionless correction coefficient. Corrected instantaneous flow rate Obtained through the following formula:

[0091] ;

[0092] ;

[0093] The state prediction and decision-making submodule 230 is responsible for generating the final control commands. This submodule first calculates and corrects the corrected instantaneous flow rate output by submodule 220 based on the instantaneous flow rate. And the already defined physical distance from the center of the sensing area to the core grinding area. Calculate the dynamic feedforward time base :

[0094] ;

[0095] Subsequently, this submodule will process the historical transient physical feature vector sequence. (in The system sampling time interval, (for sequence length) and dynamic feedforward time base They are input together into a time series prediction model (For example, in a gated recurrent unit (GRU) network, predicting the future Material feature vector at time step .

[0096] Finally, this submodule will predict the future Material feature vector at time step Input to a preset mapping function between material state and control parameters. In the process, the optimal control parameter vector is queried or calculated. For example, specific values ​​for grinding pressure and kat gap. This optimal control parameter vector. That is, as the final control command, it is sent to the high dynamic response actuator 300.

[0097] The signal preprocessing and feature extraction submodule 210 is one of the internal functional modules of the edge computing and control unit 200. The function of this submodule is to receive and process the raw sensing signals from the transient physical property sensing module 100, and extract quantitative features from them to characterize the physical properties of the material.

[0098] This submodule receives multi-channel time-series voltage signals from the high-frequency acoustic impact sensor array 110. and raw terahertz spectral data from the terahertz spectral scanning array 120. Upon receiving a continuous signal data stream, the submodule first performs digital filtering to remove interference from environmental and circuit noise. Then, the submodule uses a fixed-length time window to divide the continuous signal stream into a series of discrete data segments, each corresponding to a basic unit for subsequent processing.

[0099] For the multi-channel time-series voltage signal acquired from the high-frequency acoustic impact sensor array 110 within each time window This submodule applies a Fast Fourier Transform (FFT) to the signal of each channel, transforming it from the time domain to the frequency domain. By analyzing its spectrum, a series of characteristic parameters related to the material's hardness, brittleness, and other mechanical properties are extracted, such as the signal energy integral values ​​in the preset low, medium, and high frequency bands, and the peak value of the power spectral density. These extracted characteristic parameters are combined to form an acoustic feature vector. .

[0100] For the raw terahertz spectral data acquired from the terahertz spectral scanning array 120 within the same time window This submodule analyzes the spectral characteristics of the material at moisture-specific absorption peak frequencies (e.g., around 1.0 terahertz and 1.4 terahertz). The extracted feature parameters include the absorption intensity, full width at half maximum (FWHM), and peak area fraction of the characteristic absorption peaks. These parameters directly reflect the internal average moisture content and uniformity of moisture distribution within the material section. These extracted feature parameters are combined to form a terahertz feature subvector. .

[0101] Finally, the signal preprocessing and feature extraction submodule 210 extracts acoustic feature vectors. Terahertz eigenvectors By concatenating the data, the transient physical feature vector corresponding to that time window is constructed. .

[0102] ;in, For a moment The transient physical eigenvectors, For a moment acoustic feature vectors, For a moment The terahertz eigenvectors. This submodule will calculate the generated time. transient physical eigenvectors The output is sent to the instantaneous flow rate calculation and correction submodule 220 and the state prediction and decision-making submodule 230.

[0103] The instantaneous flow velocity calculation and correction submodule 220 is one of the internal functional modules of the edge computing and control unit 200. The function of this submodule is to calculate the instantaneous velocity of the material flow and correct the velocity value using the physical property information of the material to improve its accuracy.

[0104] This submodule receives multi-channel time-series voltage signals from the high-frequency acoustic impact sensor array 110. To calculate the flow rate, this submodule selects two acoustic impact sensors 111 that are physically adjacent along the material movement direction in the array, for example, sensor... and The corresponding signals are respectively and By calculating the cross-correlation function of these two signals... The time delay between these two sensors determines the material flow pattern. .

[0105] ;

[0106] ;

[0107] in, It is a signal and The cross-correlation function, It is a time shift. This is the time shift that maximizes the cross-correlation function, i.e., the time delay. If the vertical distance between the two acoustic impact sensors 111 is... Then the original instantaneous flow velocity The calculation is as follows:

[0108] ;in, The vertical distance between the two acoustic impact sensors 111 This represents the original instantaneous flow velocity.

[0109] Considering that the physical properties of materials (such as moisture content and particle viscosity) can affect their flow behavior and thus impact velocity calculations based on signal pattern matching, this submodule further performs a correction step. This submodule calls a velocity estimation correction model pre-trained using experimental data. The model receives the timing data from the signal preprocessing and feature extraction submodule 210. transient physical eigenvectors As input, it outputs a dimensionless correction coefficient. .

[0110] ;

[0111] Corrected instantaneous flow rate By using the original instantaneous flow rate With correction factor Multiplication yields: ,in, For a moment The transient physical eigenvectors, The flow velocity estimation model was modified. For correction factor, This is the corrected instantaneous flow velocity. This submodule will ultimately calculate the corrected instantaneous flow velocity. Output to the State Prediction and Decision Submodule 230.

[0112] The state prediction and decision-making submodule 230 is one of the internal functional modules of the edge computing and control unit 200. The function of this submodule is to predict the future state of the material when it reaches the grinding zone based on the current state and flow rate of the material, and calculate the optimal control command accordingly.

[0113] This submodule receives the corrected instantaneous flow rate output from the instantaneous flow rate calculation and correction submodule 220. Based on the flow rate and a pre-calibrated physical distance DD from the center of the sensing area of ​​the transient physical characteristic sensing module 100 to the core grinding zone of the rice milling machine 40, this submodule calculates the time required for the current material segment to reach the grinding zone, i.e., the dynamic feedforward time reference. .

[0114] ;in, It is the physical distance from the center of the sensing area to the core area of ​​the grinding mill. For the corrected instantaneous flow rate, It serves as a dynamic feedforward time reference.

[0115] Subsequently, the state prediction and decision-making submodule 230 calls a pre-trained time series prediction model. In one specific implementation, the model is a gated recurrent unit (GRU) network. The model receives a sequence of historical transient physical feature vectors generated by the signal preprocessing and feature extraction submodule 210. and dynamic feedforward time base As input, the future of this material segment is predicted. The physical characteristics of the object when it reaches the grinding zone, which is predicted by the future... Material feature vector at time step To characterize.

[0116] ;in, It is a sequence of transient physical feature vectors in history. It is the system sampling time interval. It is the sequence length. It is a time series forecasting model. In the future Material feature vector at time t.

[0117] Finally, this submodule will predict the future Material feature vector at time step Input to a preset mapping function between material state and control parameters. In the diagram, the mapping function between the material state and the control parameters... The correspondence between different material states and optimal grinding parameters is defined. This function allows the submodule to calculate the optimal control parameter vector. .

[0118] ;in, It is a mapping function between material state and control parameters. This is the optimal control parameter vector. This vector contains specific control command values ​​for the high dynamic response actuator 300, such as the target torque of the servo motor 310 and the target drive voltage of the piezoelectric ceramic actuator 320. This submodule will use this optimal control parameter vector... The final instruction is sent to the high dynamic response actuator 300.

[0119] See attached document Figure 1 The high dynamic response actuator 300 is integrated inside the rice milling machine 40 and is used to receive control commands and physically adjust the milling process. The response speed of this actuator is matched with the decision frequency of the edge computing and control unit 200 to achieve segmented and precise control of the material flow.

[0120] In one specific embodiment, the high dynamic response actuator 300 includes a servo motor 310 for adjusting the grinding pressure and a piezoelectric ceramic actuator 320 for adjusting the gap between the rice knives.

[0121] The servo motor 310 is connected to the pressure mechanism of the grinding roller 43 of the rice milling machine 40 via a precision transmission device. The servo motor 310 has high torque density and millisecond-level response capability, enabling it to quickly and accurately change the pressure applied to the grinding roller 43.

[0122] A piezoelectric ceramic actuator 320 is mounted on the adjustment mechanism of the rice knife 44. This actuator utilizes the inverse piezoelectric effect to generate a micron-level precise displacement when a driving voltage is applied, thereby achieving rapid and hysteresis-free adjustment of the gap between the rice knife 44 and the grinding roller 43.

[0123] The workflow of the high dynamic response actuator 300 begins with receiving the optimal control parameter vector from the state prediction and decision submodule 230. The drive controller within the actuator parses this vector, converting its component values ​​into physical setpoints for the corresponding actuators. For example, the pressure setpoint in the vector is converted into a target torque or position command for the servo motor 310, and the gap setpoint is converted into a drive voltage applied to the piezoelectric ceramic actuator 320.

[0124] The entire operation of this mechanism, from receiving the optimal control parameter vector to completing the adjustment action of the physical components, is designed to have a total time consumption that is less than the dynamic feedforward time base calculated by the state prediction and decision submodule 230. This timing matching ensures that the processing parameters of the rice mill 40 are adjusted to be optimized for a specific material segment before the segment arrives at the core milling zone, thus achieving predictive control.

[0125] See attached document Figure 1The post-processing quality detection and optimization module 400 is installed at the discharge port 45 of the rice milling machine 40. Its function is to perform online quality quantitative detection of the finished product and provide long-term optimization feedback data for the system's control strategy.

[0126] In one specific implementation, the module is a machine vision-based online inspection system, which mainly includes an image acquisition unit 410 and an image processing unit 420. The image acquisition unit 410 is positioned at the discharge port 45, so that its field of view can cover the falling finished product material flow. The image processing unit 420 is electrically connected to the image acquisition unit 410 and establishes a communication connection with the edge computing and control unit 200.

[0127] During system operation, the image acquisition unit 410, such as a high-speed industrial camera, continuously or periodically captures images of the finished product flowing through the discharge port 45, obtaining digital images containing a large number of rice grains. This unit transmits the acquired images to the image processing unit 420 in real time.

[0128] The image processing unit 420 executes preset image processing algorithms, such as image segmentation, morphological analysis, and feature extraction, to identify each grain of rice in the image and determine whether it is broken rice. Based on the analysis results of continuous images, the image processing unit 420 calculates the quantity or area ratio of whole rice to broken rice, thereby calculating a quantitative quality indicator such as the broken rice rate of the finished product within the current time period. This unit periodically sends the calculated quality indicator data to the edge computing and control unit 200.

[0129] After receiving this actual quality feedback data, the edge computing and control unit 200 compares it with the predicted material feature vectors recorded during that time period. and the optimal control parameter vector adopted accordingly. The association is established. Through an online learning or offline retraining process, the unit maps the internal material states to control parameters using a function. Adjustments and optimizations will be made.

[0130] This optimization process ensures that the system's control strategy can be adaptively adjusted based on long-term actual processing results, thereby maintaining robust and high-quality processing performance even when material properties drift slowly or undergo unforeseen changes.

[0131] See attached document Figure 2 The figure illustrates in detail the specific implementation process of the intelligent adjustment and control method of the present invention. The following will refer to the attached figure. Figure 1 The process is explained in detail.

[0132] In a specific operational instance, the implementation steps of the method of the present invention are as follows:

[0133] Step S100: Data Acquisition. During the operation of the rice milling machine 40, the rice material flow continuously passes through the transient physical characteristic sensing module 100 installed upstream of the inlet 41 due to gravity. When the material flow passes through the sensing area, the high-frequency acoustic impact sensing array 110 and the terahertz spectral scanning array 120 work synchronously to acquire multi-channel time-series voltage signals in real time. Compared with the original terahertz spectral data These raw data streams are continuously sent to the edge computing and control unit 200.

[0134] Step S200: Feature Extraction and Flow Rate Calculation. After receiving the data, the edge computing and control unit 200 initiates parallel processing with its internal signal preprocessing and feature extraction submodule 210 and instantaneous flow rate calculation and correction submodule 220. The signal preprocessing and feature extraction submodule 210 processes the received signal, extracts acoustic and terahertz features, and constructs a time-varying flow rate. transient physical eigenvectors Meanwhile, the instantaneous flow rate calculation and correction submodule 220 utilizes multi-channel time-series voltage signals. Calculate the original instantaneous flow velocity And call the flow rate estimation correction model. Utilize time transient physical eigenvectors The flow velocity is corrected to obtain the corrected instantaneous flow velocity. .

[0135] Step S300: State Prediction. The state prediction and decision-making submodule 230 inside the edge computing and control unit 200 receives the corrected instantaneous flow rate. Then, based on the known physical distance from the center of the sensing area to the grinding core area... Calculate the dynamic feedforward time base Next, the submodule calls the time series prediction model. Based on the historical transient physical feature vector sequence and this dynamic feedforward time base Predicting the future of this material segment Material feature vector at time step .

[0136] Step S400: Decision and Execution. The State Prediction and Decision Submodule 230 will predict the future... Material feature vector at time step Input to the preset mapping function between material state and control parameters In this way, a set of optimal control parameter vectors can be calculated. The optimal control parameter vector This command is then sent as a control instruction to the high dynamic response actuator 300. Upon receiving the instruction, the high dynamic response actuator 300 immediately drives its internal servo motor 310 and piezoelectric ceramic actuator 320, based on the dynamic feedforward time reference. Within a limited time, complete the adjustment of the grinding pressure and the gap between the rice knives.

[0137] Step S500: Online Detection and Optimization. While the aforementioned predictive control process continues, the post-processing quality detection and optimization module 400, installed at the discharge port 45, performs online quality detection on the finished product, such as calculating the broken rice rate. This module feeds back the quantified quality data to the edge computing and control unit 200. The edge computing and control unit 200 uses this long-term, real-time quality data to periodically adjust the mapping function between the material state and control parameters. Iterative optimization is performed to achieve adaptive adjustment of the system control strategy and performance improvement, forming a complete closed-loop learning and optimization process.

[0138] The present invention has been disclosed with reference to the above embodiments; however, the scope of protection of the present invention is not limited thereto. Those skilled in the art should understand that various modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the invention.

[0139] In an alternative implementation, the sensing method used to acquire material moisture information in the transient physical property sensing module 100 can be replaced. For example, near-infrared spectroscopy (NIR) analysis technology can be used instead of the aforementioned terahertz spectral scanning array 120. Near-infrared spectroscopy also has characteristic absorption peaks for moisture, enabling non-contact online measurement of the moisture content of the material flow.

[0140] In another alternative implementation, the acoustic impact sensor 111 in the high-frequency acoustic impact sensor array 110 can be a microelectromechanical system (MEMS) accelerometer or a fiber Bragg grating (FBG) vibration sensor, in addition to a piezoelectric ceramic sensor. These sensors also have the characteristics of high sensitivity and wide frequency response, and can effectively collect vibration signals generated by material impact.

[0141] Regarding the hardware foundation of the edge computing and control unit 200, in addition to using an industrial computer equipped with a graphics processing unit, for applications requiring extremely low latency, a field-programmable gate array (FPGA) or a dedicated AI acceleration chip can also be used. These hardware solutions can embed algorithms into the hardware logic, achieving faster signal processing and decision response speeds.

[0142] At the algorithm level, the time series prediction model used in the state prediction and decision submodule 230 Besides using gated recurrent unit (GRU) networks, long short-term memory (LSTM) networks, Transformer networks, or other deep learning models suitable for time series forecasting can also be employed. Furthermore, the mapping function between material states and control parameters... In addition to using preset functions or lookup tables, this can also be achieved through a reinforcement learning agent. This agent takes the predicted material state as input, directly outputs control actions, and continuously learns the optimal control strategy by interacting with feedback from the post-quality detection and optimization module 400.

[0143] Regarding the high dynamic response actuator 300, the servo motor 310 used to adjust the grinding pressure can be replaced by a voice coil motor or a high-speed electro-hydraulic proportional valve to obtain different force control characteristics in specific applications. The piezoelectric ceramic actuator 320 used to adjust the gap between the milling knives can also be replaced by a magnetostrictive actuator, which can also provide micron-level, high-frequency displacement control.

[0144] Regarding the post-processing quality inspection and optimization module 400, in addition to using a conventional machine vision system to detect broken rice rate, a hyperspectral imaging system can also be used. The hyperspectral imaging system can not only identify the geometry of rice grains but also provide richer information on material composition, such as detecting chalkiness or micro-cracks in the rice grains, thus providing a mapping function between material state and control parameters. The optimization provides more dimensions of quality feedback data.

Claims

1. A smart adjustment and control system for large-scale rice processing parameters, characterized in that, include: A transient physical characteristic sensing module is installed in the material channel upstream of the rice milling machine inlet to collect real-time data on the material flow about to enter the rice milling machine and obtain the original sensing signals characterizing the physical characteristics of the material. A high dynamic response actuator, integrated into the rice milling machine, is used to receive control commands and physically adjust the processing parameters of the rice milling machine; A post-processing quality inspection and optimization module is installed at the discharge port of the rice milling machine to detect the quality indicators of the finished product online and output quantitative quality data. An edge computing and control unit, which is used for: The original sensing signal is received, and the corrected instantaneous flow velocity is calculated based on the original sensing signal, thereby determining a dynamic feedforward time reference. Based on the original sensing signal and the dynamic feedforward time reference, the physical characteristics of the material segment when it arrives at the grinding zone in the future are predicted. Based on the predicted future state, the optimal control parameter vector is calculated through a mapping function between the material state and the control parameters. The optimal control parameter vector is sent as a control command to the high dynamic response actuator; The system receives quantitative quality data from the post-quality detection and optimization module and uses the quantitative quality data to periodically optimize the mapping function between the material state and control parameters.

2. The intelligent adjustment and control system for large-scale rice processing parameters according to claim 1, characterized in that, The transient physical characteristic sensing module includes: A high-frequency acoustic impact sensor array is used to acquire acoustic signals related to the mechanical properties of material flow; A terahertz spectral scanning array is used to acquire spectral data related to the internal moisture content of the material flow; The sensing areas of the high-frequency acoustic impact sensing array and the terahertz spectral scanning array are spatially aligned to achieve synchronous data acquisition of the same material segment.

3. The intelligent adjustment and control system for large-scale rice processing parameters according to claim 2, characterized in that, The high-frequency acoustic impact sensor array consists of multiple acoustic impact sensors arranged linearly and equidistantly along the direction of material movement within the material channel. The edge computing and control unit determines the time delay of the signal by calculating the cross-correlation function of the signals from two sensors in the acoustic impact sensor, and calculates the original instantaneous flow velocity of the material flow based on the time delay and the spacing between the sensors.

4. The intelligent adjustment and control system for large-scale rice processing parameters according to claim 1, characterized in that, The high dynamic response actuator includes: A servo motor for adjusting the pressure between the grinding roller and the rice sieve; A piezoelectric ceramic actuator for adjusting the gap between the rice cutter and the grinding roller.

5. The intelligent adjustment and control system for large-scale rice processing parameters according to claim 1, characterized in that, The post-processing quality detection and optimization module includes: An image acquisition unit for photographing the finished product, and an image processing unit for analyzing the acquired images and calculating the broken rice rate.

6. The intelligent adjustment and control system for large-scale rice processing parameters according to claim 3, characterized in that, The edge computing and control unit is logically divided into: A signal preprocessing and feature extraction submodule is used to extract features from the original sensing signal and construct a transient physical feature vector. An instantaneous flow velocity calculation and correction submodule is used to determine the time delay by calculating the cross-correlation function of two sensor signals from the high-frequency acoustic impact sensing array, thereby calculating the original instantaneous flow velocity, and using the transient physical feature vector to correct the original instantaneous flow velocity to obtain the corrected instantaneous flow velocity; A state prediction and decision-making submodule is used to perform state prediction, decision-making, and optimization functions.

7. The intelligent adjustment and control system for large-scale rice processing parameters according to claim 6, characterized in that, The signal preprocessing and feature extraction submodule is used to: extract acoustic feature vectors and terahertz feature vectors from the acoustic signal output by the high-frequency acoustic impact sensing array and the spectral data output by the terahertz spectral scanning array, respectively, and construct the transient physical feature vector by concatenating the acoustic feature vectors and the terahertz feature vectors.

8. The intelligent adjustment and control system for large-scale rice processing parameters according to claim 6, characterized in that, The instantaneous flow velocity calculation and correction submodule is used for: The transient physical feature vector is input into a flow velocity estimation correction model to obtain a correction coefficient; The original instantaneous flow velocity is multiplied by the correction coefficient to obtain the corrected instantaneous flow velocity.

9. The intelligent adjustment and control system for large-scale rice processing parameters according to claim 6, characterized in that, The state prediction and decision-making submodule is used for: The historical transient physical feature vector sequence and the dynamic feedforward time base are input into a time series prediction model to predict the physical characteristics of the material segment when it arrives at the grinding zone in the future. The predicted future state is input into the mapping function between the material state and the control parameters to calculate the optimal control parameter vector.

10. A method for intelligent adjustment and control of production parameters in large-scale rice processing, characterized in that, Includes the following steps: Data acquisition: The raw sensor signals of the material flow about to enter the rice milling machine are collected in real time through a transient physical characteristic sensing module deployed upstream of the rice milling machine inlet. Feature extraction and state prediction: In the edge computing and control unit, the original sensing signals are processed to construct transient physical feature vectors, calculate the corrected instantaneous flow velocity and determine the dynamic feedforward time reference, and then call the time series prediction model to predict the physical characteristics of the material segment when it arrives at the grinding zone in the future. Decision and Execution: Based on the predicted future state, the edge computing and control unit calculates the optimal control parameter vector through a mapping function between material state and control parameters, and sends the optimal control parameter vector as a control command to the high dynamic response actuator to adjust the processing parameters of the rice milling machine within the dynamic feedforward time reference. Online detection and optimization: The post-quality detection and optimization module installed at the discharge port of the rice milling machine detects the quality indicators of the finished product online and generates quantitative quality data. The edge computing and control unit receives the quantitative quality data and periodically optimizes the mapping function between the material state and the control parameters accordingly.

Images