Intelligent recovery control method and system for food conveying anti-sticking powder scattering

Abstract

The present application relates to the technical field of intelligent recycling control and dynamic parameter adaptation, in particular to an intelligent recycling control method and system for preventing food conveying from sticking and powder scattering, in which a detection unit collects the proportion of powder scattering and falling off, the belt vibration frequency and the conveying speed through a high-frame-frequency industrial camera, a laser particle sensor, a motor current sensor and a belt encoder, a core processing unit synchronously calculates the vibration frequency and the proportion of powder scattering and falling off through double parallel modules, a coupling model is constructed, a control unit generates a primary suction compensation value according to the vibration frequency deviation and a secondary compensation value according to the adhesion rate deviation, the two values are superimposed to dynamically adjust the rotating speed of a variable frequency fan, and the opening duration of a valve of a recycling cover is proportionally extended according to the conveying speed, a recycling circulating unit collects dust through a negative pressure recycling cover, and the dust is filtered through a cyclone separator and then returned to a powder scattering device for recycling, so that the powder scattering recycling efficiency and the anti-sticking reliability are improved, and resource waste is reduced.

Description

Technical Field

[0001] This invention relates to the field of intelligent recycling control and dynamic parameter adaptation technology, specifically to an intelligent recycling control method and system for preventing food conveying from sticking and sprinkling powder. Background Technology

[0002] Intelligent recycling control and dynamic parameter adaptation technology is an important technology, specifically applied to the anti-sticking and dust recovery process in the conveying of food such as biscuit blanks. The core is to achieve synergistic optimization of anti-sticking effect and recovery efficiency by dynamically matching negative pressure suction and recovery time, thus meeting the core requirements of food production for anti-sticking reliability and resource utilization. During food conveying, the belt vibration frequency and the amount of powder falling off vary with the conveying speed and the state of the food blank. This dynamic change directly affects the adhesion stability of the powder on the surface of the blank, causing the fixed negative pressure recovery suction to be unable to offset the increased falling off caused by the intensified vibration and insufficient adhesion rate. This leads to insufficient powder recovery or an increased risk of powder sticking to the blank. At the same time, the fixed opening time of the recovery hood valve is difficult to adapt to the belt conveying rhythm, further affecting the dust recovery efficiency and recycling rate. To solve this technical problem, we have provided an intelligent recovery control method and system for preventing powder from sticking during food conveying. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent recycling control method and system for preventing food from sticking and sprinkling powder during transport, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, one objective of this invention is to provide an intelligent recovery control method for preventing food from sticking and sprinkling powder during transport, comprising the following steps: S1. A powder-sprinkling device is installed above the biscuit blank conveyor belt, and a negative pressure recovery hood is installed below the belt. The suction of the negative pressure recovery hood is provided by a variable frequency fan. A laser particle size sensor and a high frame rate industrial camera are installed at the inlet of the negative pressure recovery hood. S2. Real-time acquisition of the proportion of cornstarch detachment and belt vibration frequency in a designated area of ​​the conveyor belt after powdering. Continuous image capture of the biscuit blank surface after powdering is obtained using a high frame rate industrial camera. Based on the grayscale value changes in the images, the trajectory of cornstarch particles detached from the biscuit blank surface is identified to calculate the proportion of powder detachment. Simultaneously, the current of the conveyor motor is acquired through a motor current sensor. The amplitude of the fundamental component in the motor current spectrum is extracted through spectrum analysis, and the real-time vibration frequency is calculated according to the preset mapping relationship between current amplitude and vibration frequency. S3. Establish a coupled model of vibration frequency, shedding amount and suction force. With real-time vibration frequency as input, when the vibration frequency is higher than the reference value, generate a suction force compensation value according to the preset frequency. At the same time, calculate the adhesion rate based on the real-time powder shedding amount ratio. When the adhesion rate is lower than the preset adhesion rate threshold, generate an additional suction force compensation value proportionally. Send the compensated total suction force value to the variable frequency fan controller to dynamically adjust the fan output. Calculate the time for a unit length of belt to pass through the recycling hood according to the real-time conveying speed. Based on the suction force compensation value, dynamically control the opening time of the pneumatic valve of the recycling hood proportionally. S4. The dust collected by the recovery hood is separated and filtered before being returned to the dust spreading device for recycling.

[0005] The second objective of this invention is to provide a system for implementing an intelligent recycling control method for preventing food conveying and powder spillage, comprising: The detection unit consists of a high frame rate industrial camera, a laser particle size sensor, a motor current sensor, and a belt encoder. The camera captures dynamic images of the biscuit blank surface after powdering through a coaxial light source, the laser sensor synchronously corrects the particle size, the current sensor collects the three-phase current of the motor, and the encoder provides feedback on the belt linear speed. The core processing unit has two built-in parallel modules. The vibration analysis module extracts the fundamental amplitude of the current spectrum through fast Fourier transform and outputs the real-time vibration frequency by combining it with a pre-stored mapping table. The shedding amount calculation module uses the dynamic background difference method to process the image sequence and integrates laser particle size data to calculate the proportion of powder shedding and the surface adhesion rate. The control unit integrates a coupled model processor and an actuator. The model processor generates a primary suction compensation value based on the vibration frequency deviation and a secondary suction compensation value based on the adhesion rate deviation. After superposition, the total suction command is output. The actuator converts the command into a fan speed signal and calculates the unit passage time of the belt based on the linear velocity. The opening time of the recovery hood valve is dynamically extended according to the compensation value ratio. The recycling unit includes a negative pressure recycling hood, a variable frequency fan, and a cyclone separator. The recycling hood collects dust based on dynamic suction and extended opening time. The fan responds to speed commands, and the filtered dust is returned to the dust spreading device for recycling.

[0006] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention accurately captures the dynamic changes in belt vibration and powder shedding through a detection unit. A high-frame-rate camera combined with dynamic background subtraction method identifies particle trajectories, a laser particle size sensor corrects particle size errors, and a motor current sensor extracts vibration frequency through spectrum analysis, providing reliable data for subsequent control. The core processing unit uses dual parallel modules to synchronously calculate the vibration frequency and the proportion of powder shedding. The coupled model generates primary and secondary suction compensation values ​​based on this, dynamically adjusting the output of the variable frequency fan. Simultaneously, combined with the conveying speed feedback from the belt encoder, the opening time of the recovery hood valve is extended according to the compensation value ratio to adapt to the conveying rhythm. The recovery and circulation unit filters the dust and returns it to the powder spreading device for recycling. This solves the problems of insufficient recovery and blank adhesion caused by dynamic changes in working conditions, improves powder utilization and anti-sticking reliability, and reduces resource waste and production risks. Attached Figure Description

[0007] Figure 1 This is a flowchart illustrating the overall workflow of the present invention; Figure 2 This is a schematic diagram of the overall structure of the present invention; The meanings of the labels in the diagram are as follows: 1. Detection unit; 2. Core processing unit; 3. Control unit; 4. Recycling unit. Detailed Implementation

[0008] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some 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.

[0009] Please see Figure 1 As shown, one of the objectives of this embodiment is to provide an intelligent recovery control method for preventing food from sticking and sprinkling powder during transport, including the following steps: S1. A powder-sprinkling device is installed above the biscuit blank conveyor belt, and a negative pressure recovery hood is installed below the belt. The suction of the negative pressure recovery hood is provided by a variable frequency fan. A laser particle size sensor and a high frame rate industrial camera are installed at the inlet of the negative pressure recovery hood. S2. Real-time acquisition of the proportion of cornstarch detachment and belt vibration frequency in a designated area of ​​the conveyor belt after powdering. Continuous image capture of the biscuit blank surface after powdering is obtained using a high frame rate industrial camera. Based on the grayscale value changes in the images, the trajectory of cornstarch particles detached from the biscuit blank surface is identified to calculate the proportion of powder detachment. Simultaneously, the current of the conveyor motor is acquired through a motor current sensor. The amplitude of the fundamental component in the motor current spectrum is extracted through spectrum analysis, and the real-time vibration frequency is calculated according to the preset mapping relationship between current amplitude and vibration frequency. S3. Establish a coupled model of vibration frequency, shedding amount and suction force. With real-time vibration frequency as input, when the vibration frequency is higher than the reference value, generate a suction force compensation value according to the preset frequency. At the same time, calculate the adhesion rate based on the real-time powder shedding amount ratio. When the adhesion rate is lower than the preset adhesion rate threshold, generate an additional suction force compensation value proportionally. Send the compensated total suction force value to the variable frequency fan controller to dynamically adjust the fan output. Calculate the time for a unit length of belt to pass through the recycling hood according to the real-time conveying speed. Based on the suction force compensation value, dynamically control the opening time of the pneumatic valve of the recycling hood proportionally. S4. The dust collected by the recovery hood is separated and filtered before being returned to the dust spreading device for recycling.

[0010] A high frame rate industrial camera is installed at a preset distance above the belt downstream of the powder-spreading device at a vertical top-down angle. Its field of view covers the entire width of the belt and extends a set length along the conveying direction. By configuring a coaxial light source of a specific wavelength to enhance the reflective properties of corn starch particles, it continuously captures a sequence of dynamic images of the biscuit blank surface at a preset high frame rate after the powder-spreading action is triggered. The duration of the capture covers the critical adhesion stabilization period after powder spreading. At the same time, a laser particle size sensor performs real-time particle size scanning on the same area to correct image recognition errors.

[0011] When identifying particle trajectories based on changes in image grayscale values, the dynamic background subtraction method is used to segment the region of moving particles. The initial image after dusting is completed is extracted as the baseline background. Gray-level difference is performed frame by frame on subsequent image sequences. Pixel areas where the gray-level value changes exceed a set threshold are marked as detached particle groups. Then, the total pixel area of ​​the particle groups detached from the cookie dough surface in each frame is statistically analyzed through connected component analysis. Combined with the real-time average particle size data provided by the laser particle size sensor and the preset pixel area and mass conversion coefficients, the dusting mass detached from the cookie dough surface per unit time is calculated. Finally, the percentage of dusting mass detached is output as the percentage of the initial dusting mass.

[0012] The spectrum analysis specifically includes: The original current signals of each phase of the conveyor motor are collected in real time by a three-phase current sensor. After filtering out high-frequency noise components higher than half the sampling frequency using an anti-aliasing filter, the current signal is windowed to suppress spectral leakage. Then, the time-domain current signal is converted into a frequency-domain spectrum by a fast Fourier transform. The characteristic frequency band corresponding to the fundamental wave of belt vibration is located in the spectrum diagram, and the amplitude of the fundamental wave component with the largest spectral amplitude in the characteristic frequency band is extracted as the vibration characteristic quantity.

[0013] The mapping relationship between current amplitude and vibration frequency was established through laboratory calibration: When the belt is unloaded, the speed of the conveyor motor is adjusted by a preset step size to cover the working speed range. The amplitude of the fundamental component of the motor current signal and the measured frequency value of the standard vibration sensor installed on the belt support are recorded simultaneously. The linear positive correlation characteristic curve between the two is fitted by the least squares method and stored as a mapping relationship table. During actual operation, the real-time vibration frequency is calculated by interpolation based on the real-time extracted fundamental component amplitude.

[0014] The process of constructing the coupled model of vibration frequency, shedding amount, and suction force is as follows: Using real-time vibration frequency, powder shedding ratio, and initial powder amount as input parameters, the difference between the vibration frequency and the preset reference frequency is multiplied by the frequency and suction compensation coefficient to generate a primary suction compensation value. At the same time, the real-time powder shedding ratio is substituted into the preset shedding amount and adhesion rate conversion function to calculate the surface adhesion rate. When the surface adhesion rate is lower than the anti-sticking critical threshold, a secondary suction compensation value is generated based on the product of the adhesion rate deviation value, the adhesion rate, and the suction compensation coefficient. Finally, the primary compensation value and the secondary compensation value are superimposed on the reference suction value to form the total suction command.

[0015] The operation to generate suction compensation values ​​at a preset frequency is as follows: The reference frequency is set to be equal to the natural vibration frequency of the belt. When the real-time vibration frequency is higher than the reference frequency, the frequency deviation value is divided into several continuous intervals. Different intervals correspond to different frequencies and suction compensation coefficients. The corresponding compensation coefficient is selected according to the interval to which the real-time vibration frequency belongs and multiplied by the frequency deviation value to generate the primary suction compensation value.

[0016] The procedure for generating additional suction compensation value proportionally is as follows: The preset adhesion rate threshold is equal to the critical value of anti-stick failure. When the real-time surface adhesion rate is lower than this threshold, a secondary suction compensation value is generated based on the product of the absolute value of the adhesion rate deviation and the preset adhesion rate and the suction compensation coefficient. The compensation coefficient is set as a dynamic variable that is positively correlated with the surface roughness of the biscuit blank. The coefficient is adjusted by measuring the uniformity of the surface residual particle distribution in real time through a laser particle size sensor.

[0017] When dynamically adjusting the fan output, the variable frequency fan controller converts the received total suction command value into a corresponding motor speed control signal. Based on the preset suction and speed response curves, it adjusts the fan impeller speed. At the same time, it calculates the time taken per unit length based on the real-time linear velocity fed back by the belt encoder. Using this time as the base duration, it converts the compensation value component in the total suction command value into a duration expansion coefficient according to a preset ratio. Finally, the opening duration is equal to the product of the base duration and the expansion coefficient. The pneumatic valve controller controls the valve opening and closing sequence accordingly.

[0018] Further explanation is needed: to accurately capture the trajectory of cornstarch granules after dusting, and to ensure the accuracy of particle size data to support subsequent calculations of the amount of detachment, the installation positions, operating parameters, and collaborative logic of the high-frame-rate industrial camera and the laser particle size sensor need to be meticulously designed. The specific implementation method is as follows: A high-frame-rate industrial camera is mounted vertically at a preset distance above the conveyor belt downstream of the powder-spreading device. This vertical angle, where the camera lens optical axis is at a 90-degree angle to the conveyor belt plane, avoids image distortion caused by oblique projection and ensures spatial accuracy for particle trajectory recognition. The preset distance is determined based on the camera resolution and belt width. For example, if a 1920×1080 pixel camera is used and the belt width is 800mm, to ensure the field of view covers the entire width of the belt and each pixel corresponds to an actual size less than or equal to 0.5mm, optical calculations determine the installation distance to be 500mm. This distance also avoids image blurring caused by belt vibration. The field of view must cover the entire width of the belt and extend a preset length along the conveying direction. The preset length is calculated based on the belt conveyor speed and the particle adhesion stabilization time. If the belt speed is 0... With a speed of 0.5 m / s and a particle adhesion stabilization time of 2 seconds, the extension length is set to 1 m (0.5 m / s × 2 s) to ensure complete capture of particle dynamics in the designated area of ​​the belt during the critical time period. The reflective properties of corn starch particles are enhanced by configuring a coaxial light source with a specific wavelength. The coaxial light source is a light source arranged coaxially with the camera's optical axis, which can reduce shadow interference. Its specific wavelength is selected as 550-600 nm yellow-green light, which matches the peak of the reflection spectrum of corn starch, enabling the particles to form obvious reflective differences on the surface of the biscuit blank, facilitating subsequent grayscale value recognition. The brightness of the light source is dynamically adjusted according to the ambient light intensity. The ambient brightness is detected in real time by the camera's built-in light sensor. When the brightness is below 500 lux, the light source output is automatically enhanced, and when it is above 1000 lux, it is appropriately weakened to ensure stable image grayscale contrast.

[0019] After the powder-sprinkling action is triggered, a series of dynamic images of the biscuit dough surface are continuously captured at a preset high frame rate. The powder-sprinkling action is triggered synchronously via the solenoid valve switch signal of the powder-sprinkling device, ensuring seamless connection between the shooting and powder-sprinkling actions. The preset high frame rate is set to 200fps, which can capture the instantaneous motion trajectory of the cornstarch granules falling off. The particle falling speed is approximately 0.1m / s, and 200fps can record the position every 5ms to avoid missing the trajectory. The shooting duration covers the critical adhesion stabilization period after powder-sprinkling. The critical adhesion stabilization period refers to the time from particle falling off to stable adhesion after powder-sprinkling, which is experimentally determined to be 2 seconds. Therefore, the shooting duration is set to 2 seconds, and a total of 400 frames of images are collected. At the same time, a laser particle size sensor performs real-time particle size analysis on the same area. To correct image recognition errors, a laser particle size sensor is installed next to the camera, completely overlapping the camera's shooting area. The scanning frequency is set to 100Hz, and the average particle size data of corn starch particles in the scanning area is output in real time. This corrects image recognition errors because image recognition can only obtain the pixel area of ​​particles and cannot directly correspond to the actual mass. The average particle size provided by the laser particle size sensor can establish a relationship between pixel area and actual volume, and then convert it into mass. For example, when the image recognition of a particle group has a pixel area of ​​100 pixels, the laser sensor measures an average particle size of 50μm. Combined with the corn starch density, the actual mass of the particle group can be accurately calculated through a preset conversion factor, avoiding the calculation deviation of the amount of detachment caused by the difference in particle size.

[0020] Based on dynamic image sequences acquired by high-frame-rate industrial cameras, precise image processing algorithms are needed to identify the detachment trajectory of corn starch granules, and then calculate the percentage of starch detachment. The core method employs dynamic background subtraction to segment the moving particle region, ensuring effective separation of detached particles from the background. The specific implementation method is as follows: When identifying particle trajectories based on image grayscale value changes, the core method employed is dynamic background subtraction to segment the moving particle region. Dynamic background subtraction is an image processing method that extracts moving targets by comparing the grayscale differences between a sequence of images and a reference background. This effectively eliminates interference from static textures on the cookie dough surface and accurately locates detached moving particles. The execution flow of this method first extracts the initial image after the powdering process is completed as the reference background. The initial image after powdering is completed refers to the image captured in the first frame after the powdering action ends. At this time, the particles have not yet begun to detach, and the image only contains the cookie dough surface and stably attached particles. After extraction, it is stored in the buffer of the image processing module as a reference for subsequent comparisons. To avoid background drift, every 50 frames... The baseline background is updated once per frame to ensure consistency with the actual scene. Then, gray-level difference operations are performed frame-by-frame on subsequent image sequences. This involves subtracting the gray-level value of each pixel in the current frame from the corresponding pixel in the baseline background to obtain the gray-level difference. Pixels with gray-level changes exceeding a set threshold are marked as detached particle groups. This threshold, determined experimentally to be 30, means that when the gray-level difference is greater than 30, the corresponding pixel is identified as a moving detached particle and marked as white (gray-level value 255). Other pixels are marked as black (gray-level value 0), forming a binarized image that separates the detached particle group from the background. Finally, connected component analysis is used to statistically analyze the number of particles detached from the cookie dough surface in each frame. The total pixel area of ​​the particle cluster is calculated using connected component analysis. This involves clustering adjacent white pixels in a binarized image, with each cluster forming a connected component representing a detached particle or particle cluster. The sum of the number of pixels in all connected components is the total pixel area of ​​the particle cluster in that frame. To avoid noise interference, tiny connected components with an area less than 5 pixels are removed as image noise, ensuring that the statistical results only reflect real detached particles. Combining real-time average particle size data provided by a laser particle size sensor with preset pixel area and mass conversion factors, the mass of powder detached from the biscuit dough surface per unit time is calculated. The pixel area and mass conversion factors are preset based on the density of corn starch, particle sphericity, and camera imaging ratio. A constant, for example, when the average particle size is 50μm, the particle mass corresponding to 1000 pixels is 0.01mg. This coefficient is stored in the parameter library of the processing module and can be dynamically adjusted according to the real-time average particle size feedback from the laser sensor. During calculation, the total pixel area of ​​each frame image is first multiplied by the conversion coefficient to obtain the particle mass corresponding to a single frame image. Then, the particle masses of 400 frames are summed to obtain the total detachment mass within 2 seconds. Dividing by 2 seconds gives the detachment mass per unit time. Finally, the percentage of detached powder mass is output as the percentage of the initial powder detachment mass. The initial powder detachment mass is the set mass of the powder detachment device for a single powder detachment and is stored in the system parameters; for example, the detachment mass per unit time is 0.At a speed of 5g / s, the total mass of powder falling off within 2 seconds is 1g. If the initial powder mass is 10g, then the powder falling off represents 10%. This value is output to the core processing unit in real time, providing crucial information for subsequent suction compensation control.

[0021] While collecting the percentage of powder fallout in real time, the belt vibration frequency needs to be obtained through spectrum analysis of the conveyor motor current. Since belt vibration directly changes the motor load resistance, the current signal fluctuates periodically with the vibration period. Therefore, the belt vibration information can be extracted by analyzing the characteristic components in the current spectrum. The specific implementation method is as follows: The core of spectrum analysis is to separate the fundamental component corresponding to belt vibration from the raw motor current signal. First, the raw current signals of each phase of the conveyor motor are acquired in real time using three-phase current sensors. Switch-type Hall current sensors are selected and installed on the A, B, and C phase power supply lines of the conveyor motor, respectively. The acquisition frequency is set to 10kHz. The sensors convert the current signals into 0-5V analog voltage signals, which are then converted into digital signals by a 16-bit ADC and transmitted to the vibration analysis module of the core processing unit. Simultaneously, the average value of the three-phase current signals is taken to obtain the single-phase equivalent current signal, simplifying the subsequent processing flow. Then, anti-vibration analysis is performed... The aliasing filter removes high-frequency noise components above half the sampling frequency. The anti-aliasing filter is a low-pass filter with a cutoff frequency equal to half the sampling frequency. Its function is to prevent high-frequency noise from mixing into the low-frequency band during Fourier transform, causing spectral distortion. An RC active low-pass filter is selected, with a passband ripple of less than or equal to 1dB and a stopband attenuation of greater than or equal to 40dB, ensuring that high-frequency noise above 5kHz is effectively filtered out, outputting a stable low-frequency current signal. After noise filtering, the current signal is windowed to suppress spectral leakage. Spectral leakage refers to the diffusion of spectral energy into adjacent frequencies after Fourier transform when the signal period does not match the sampling period. The phenomenon of segmentation makes it difficult to locate the fundamental wave component of vibration. A Hanning window is used for windowing, as this window function effectively reduces spectral leakage while preserving the amplitude accuracy of the signal. Specifically, the 1024 continuously acquired current sampling points are multiplied point-by-point by the 1024 coefficients of the Hanning window function to obtain the windowed current signal sequence. Then, a Fast Fourier Transform (FFT) is used to convert the time-domain current signal into a frequency-domain spectrum. The FFT is an efficient algorithm for calculating the Discrete Fourier Transform, which can convert the time-domain current signal into a frequency-domain amplitude and frequency distribution spectrum. Here, a 1024-point FFT is used, dividing the 0-5kHz frequency range into 10... 24 frequency points, each spaced approximately 4.88 Hz apart, are used to transform and output the spectral amplitude corresponding to each frequency point, forming a complete frequency domain spectrum. Finally, the characteristic frequency band corresponding to the fundamental wave of belt vibration is located in the spectrum. The characteristic frequency band is determined based on the inherent characteristics of the belt conveyor system. Through preliminary no-load testing, it was found that the fundamental frequency of belt vibration is concentrated in the range of 5-20 Hz. Therefore, this range is set as the characteristic frequency band. The frequency point with the largest spectral amplitude is searched within the characteristic frequency band. The spectral amplitude corresponding to this point is the amplitude of the fundamental wave component of belt vibration, which is stored as a vibration characteristic quantity to provide data support for subsequent calculation of real-time vibration frequency.

[0022] After obtaining the vibration characteristic quantity (fundamental component amplitude) through spectrum analysis, it is necessary to establish its correspondence with the actual vibration frequency of the belt. Since the correlation between current amplitude and vibration frequency differs under different belt tensions and motor loads, a mapping relationship needs to be constructed through laboratory calibration to ensure accurate conversion during actual operation. The specific implementation method is as follows: Establishing the mapping relationship between current amplitude and vibration frequency needs to be done in a laboratory environment. First, the belt is placed in an unloaded state (i.e., no biscuit blanks are placed on the belt surface), with only the motor driving the belt to run unloaded. This is to eliminate the influence of the weight of the biscuit blanks on the belt vibration, ensuring that the calibration results only reflect the correlation between the belt's own vibration and the motor current. Simultaneously, the belt tension is checked, and the tensioning pulley is adjusted to achieve the design tension, avoiding calibration deviations caused by abnormal tension. The conveyor motor speed is adjusted using a preset step size of 0.1 m / s to cover the working speed range, which is determined by actual production conditions. The required speed is set at 0.2-1.0 m / s (covering the common speed of biscuit blank conveying). The speed is adjusted successively by the motor frequency converter. After each adjustment, the belt is run for 30 seconds to ensure stable belt vibration. The amplitude of the fundamental component of the motor current signal and the measured frequency value of the standard vibration sensor installed on the belt support are recorded simultaneously. The standard vibration sensor is a piezoelectric accelerometer, which is installed in the middle of the belt support. The vibration frequency signal output by the sensor is synchronously transmitted to the computer via the data acquisition card. It corresponds one-to-one with the amplitude of the fundamental component of the motor current to form a three-dimensional dataset of speed, current amplitude and vibration frequency.

[0023] The linear positive correlation curve between the two components is fitted using the least squares method. The least squares method is a mathematical approach that finds the best-fitting straight line by minimizing the sum of squared errors. Here, it is used to fit the relationship between the amplitude of the fundamental current component (independent variable) and the vibration frequency (dependent variable). Fifty sets of current amplitude and vibration frequency data recorded during calibration are input into MATLAB software, and the `polyfit` function is called to perform a first-order linear fit. The slope and intercept of the fitted line are obtained, which represents the linear positive correlation curve between the two components. The goodness of fit R0 is [value missing]. 2 The correlation coefficient must be greater than or equal to 0.95 to ensure reliable correlation. The fitted characteristic curve is stored as a mapping table. The mapping table is divided into intervals of 0.1A based on the current amplitude. Each interval corresponds to a vibration frequency value. For example, a current amplitude of 10A corresponds to a vibration frequency of 3.5Hz, and 10.1A corresponds to 3.52Hz. The table is stored in the Flash memory of the core processing unit and supports fast reading. During actual operation, the vibration analysis module finds the two closest current amplitude intervals in the mapping table based on the real-time extracted fundamental component amplitude. Linear interpolation is used to calculate the real-time vibration frequency. For example, if the real-time current amplitude is 10.05A, which is between 10A (3.5Hz) and 10.1A (3.52Hz), the vibration frequency is calculated to be 3.51Hz through interpolation, ensuring that the conversion accuracy meets the control requirements.

[0024] After obtaining the real-time vibration frequency and the percentage of powder detachment through spectrum analysis and image recognition, a coupling model of these three parameters with the suction force of the negative pressure recovery hood needs to be constructed. Increased vibration frequency will exacerbate particle detachment, and decreased powder adhesion rate means insufficient anti-sticking effect. Both need to be improved by adjusting suction compensation to improve recovery efficiency. Therefore, the model needs to respond to these two parameters simultaneously and generate coordinated suction control commands. The specific implementation method is as follows: The construction of the coupled model first clarifies the input parameters, with real-time vibration frequency, powder detachment ratio, and initial powder application amount as the core inputs. Real-time vibration frequency, calculated through the mapping relationship between current amplitude and vibration frequency, reflects the influence of belt vibration on particle detachment. The powder detachment ratio, obtained after image recognition, is the percentage of detached particle mass relative to the initial powder application mass, directly reflecting the particle adhesion state. The initial powder application amount is the set mass of powder applied by the powder application device in a single application, used to assist in calibrating the practical significance of the detachment ratio and avoid compensation deviations caused by different powder application amounts. All three are synchronously input into the model at a frequency of 100ms / time to ensure data timeliness. The first stage of the model generates the primary absorption... The suction compensation value is specifically calculated by multiplying the difference between the vibration frequency and the preset reference frequency by a frequency and a suction compensation coefficient. The preset reference frequency is the natural vibration frequency measured when the belt is unloaded, calibrated in the laboratory using a standard vibration sensor. At this point, the amount of particle shedding is minimal, and no additional suction compensation is needed. This value is pre-stored in the model parameter library and can be recalibrated quarterly based on belt wear. The frequency and suction compensation coefficients are constants reflecting the relationship between the vibration frequency deviation and the required suction compensation. Through experimental measurements, the suction increment required to return the amount of particle shedding to the normal range was recorded under different vibration frequency deviations. The fitted coefficient is 0.3 kPa / Hz, meaning that for every 1 Hz increase in deviation, an additional 0.3 kPa of suction compensation is required. For example, if the real-time vibration frequency is 10Hz and deviates from the reference frequency by 2Hz, then the primary suction compensation value = 2Hz × 0.3kPa / Hz = 0.6kPa. This value is used to offset the increased shedding caused by the intensified vibration. At the same time, the second stage of the model calculates the surface adhesion rate and determines whether a secondary suction compensation value needs to be generated: the proportion of real-time powder shedding is substituted into the preset shedding amount and adhesion rate conversion function. The conversion function is constructed based on the logic of adhesion rate = 1 - proportion of powder shedding. Since the particles that do not fall off are attached to the surface of the biscuit dough, for example, if the shedding amount is 8%, then the surface adhesion rate = 92%. This function also needs to be calibrated in conjunction with the initial powder shedding amount. If the initial powder shedding amount is insufficient (below 3g / m), the function will be affected. Even if the percentage of detachment is low, the actual adhesion may still be insufficient. In this case, the adhesion rate needs to be reduced by 5%-10% to ensure that the calculation matches the actual anti-sticking requirements. Then, the surface adhesion rate is compared with the critical anti-sticking threshold. The critical anti-sticking threshold is the minimum adhesion rate to ensure that the biscuit dough does not stick to the belt. If the real-time adhesion rate is higher than the threshold, no secondary compensation is needed. If it is lower than the threshold, the adhesion rate deviation value is calculated and then multiplied by the adhesion rate and the suction compensation coefficient to generate the secondary suction compensation value. The adhesion rate and the suction compensation coefficient are also calibrated experimentally and set to 0.2 kPa / %, that is, for every 1% decrease in adhesion rate, an additional 0.2 kPa of suction is required for compensation. For example, when the deviation is 3%, the secondary compensation value = 3% × 0.2 kPa / % = 0.6 kPa is used to improve the anti-sticking effect and reduce particle shedding. In the final stage of the model, the compensation values ​​are integrated, and the primary compensation value and secondary compensation value are superimposed on the baseline suction value to form the total suction command. The baseline suction value is the basic suction force that can stably recover a small amount of detached particles when the belt is unloaded and the vibration frequency is the baseline frequency. 2 kPa is set experimentally and is directly added arithmetically during superposition. For example, baseline suction force 2 kPa + primary compensation 0.6 kPa + secondary compensation 0.6 kPa = 3.2 kPa. This total suction value is the final control command sent to the variable frequency fan controller, ensuring that the suction force can simultaneously cope with vibration and adhesion issues, balancing recovery efficiency and biscuit blank integrity.

[0025] In the process of generating the initial suction compensation value in the coupled model, it is necessary to refine the matching logic between the frequency deviation and the compensation coefficient. Since the particle shedding increase is different for different vibration frequency deviations, using a fixed compensation coefficient may easily lead to insufficient or excessive compensation. Therefore, it is necessary to dynamically select the coefficient according to the deviation range. The specific implementation method is as follows: First, the reference frequency is set to be equal to the belt's natural vibration frequency. The belt's natural vibration frequency is the natural vibration frequency of the belt under no-load and standard tension, obtained through laboratory calibration. The belt is adjusted to the design tension (sag less than or equal to 5mm / 1m), the motor is run under no-load, and the vibration frequency of the support is collected using a standard vibration sensor. The average of three measurements is taken as the natural vibration frequency, which is stored in the model parameter library. It is recalibrated every six months due to belt aging. The reason for setting the reference frequency to be equal to the natural frequency is that the belt vibration is most stable at this time, the amount of particle shedding is at a natural level, and no additional suction compensation is required. When the real-time vibration frequency is higher than the reference frequency (the amount of particle shedding is small when it is lower than the reference frequency, so no compensation is needed), first calculate the frequency deviation value (real-time vibration frequency - reference frequency, e.g., 10Hz - 8Hz = 2Hz), then divide the deviation value into several continuous intervals. The interval division is based on experimental data. The increase in particle shedding amount under different deviations is statistically analyzed. It is found that the increase in shedding amount is slow when the deviation is 0-2Hz (5% increase per Hz), the increase is faster when the deviation is 2-4Hz (8% increase per Hz), and the increase is drastic above 4Hz (12% increase per Hz). Therefore, the deviation value is divided into [0,2)Hz, [2,4 ... The 0Hz, 2Hz, and [4, +∞)Hz ranges correspond to different frequencies and suction compensation coefficients. These coefficients are calculated based on the required suction compensation / frequency deviation. The coefficient for the [0, 2)Hz range is 0.3 kPa / Hz (a 1Hz deviation requires 0.3 kPa compensation), the coefficient for the [2, 4)Hz range is 0.5 kPa / Hz (a 3Hz deviation requires 1.5 kPa compensation), and the coefficient for the [4, +∞)Hz range is 0.8 kPa / Hz (a 5Hz deviation requires 4 kPa compensation). These coefficient values ​​are pre-stored in a range-coefficient mapping table and can be fine-tuned according to the cookie dough thickness. (The billet thickness coefficient is reduced by 10% to avoid excessive suction damaging the billet.) Finally, the corresponding compensation coefficient is selected based on the real-time vibration frequency range and multiplied by the frequency deviation value to generate the primary suction compensation value. For example, if the real-time vibration frequency is 11Hz, the reference frequency is 8Hz, and the deviation is 3Hz, it belongs to the [2,4)Hz range, with a corresponding coefficient of 0.5kPa / Hz. The primary compensation value is 3Hz × 0.5kPa / Hz = 1.5kPa. If the deviation is 5Hz (real-time 13Hz), it belongs to the [4,+∞)Hz range, with a coefficient of 0.8kPa / Hz. The compensation value is 5 × 0.8 = 4kPa. The generated primary compensation value will be labeled with its range to facilitate subsequent model diagnosis and ensure that the compensation is effective without excessively consuming wind turbine energy.

[0026] The specific implementation method for generating additional suction compensation values ​​proportionally: After the coupled model addresses the particle shedding problem caused by belt vibration with primary suction compensation values, if the real-time surface adhesion rate still does not meet the minimum requirement for ensuring the anti-stick effect, additional secondary suction compensation values ​​need to be generated proportionally. The core logic of this compensation value is constructed around the adhesion rate threshold and the dynamic compensation coefficient. It must accurately offset the anti-stick risk caused by insufficient adhesion rate, and also adapt to the individual differences in the surface roughness of the biscuit blank. The specific implementation method is as follows: First, the preset adhesion rate threshold is equal to the anti-sticking failure threshold. The anti-sticking failure threshold refers to the minimum surface adhesion rate required to ensure that the biscuit blank does not stick to the conveyor belt during transport. This value was calibrated through numerous clinical trials. One hundred groups of biscuit blanks with different formulations were selected, and belt adhesion was tested at different adhesion rates. Statistical analysis showed that when the adhesion rate was greater than or equal to 85%, all types of biscuit blanks showed no adhesion. Below 85%, the adhesion rate of soft biscuit blanks surged to over 30%. Therefore, the anti-sticking failure threshold was uniformly set to 85% and pre-stored in the control unit's parameter library. Considering the differences in biscuit blank types, the system supports manual fine-tuning of the threshold through the user interface to ensure that the threshold adapts to the anti-sticking requirements of different products. When the real-time surface adhesion rate calculated by the core processing unit is lower than this threshold, the absolute value of the adhesion rate deviation is first calculated, i.e., the anti-sticking failure threshold and the real-time surface adhesion rate. This value directly reflects the degree of insufficient adhesion rate. For example, with a threshold of 85% and a real-time adhesion rate of 82%, the absolute value of the deviation is... The larger the deviation (3%), the higher the risk of adhesion and the greater the additional suction compensation required. A secondary suction compensation value is then generated by multiplying the absolute value of this deviation by the preset adhesion rate and suction compensation coefficient. The initial values ​​of the adhesion rate and suction compensation coefficient are determined experimentally. Within the adhesion rate deviation range of 1%-5%, the suction increment required to restore the adhesion rate to the critical value is recorded. The initial coefficient is fitted to be 0.2 kPa / %, meaning that for every 1% decrease in adhesion rate, an additional 0.2 kPa of suction is required. For example, when the absolute deviation is 3%, the initial secondary compensation value = 3% × 0.2 kPa / % = 0.6 kPa. It should be noted that this compensation coefficient is not a fixed value but a dynamic variable positively correlated with the surface roughness of the biscuit dough. The surface roughness of the biscuit dough is a key factor affecting particle adhesion. The rougher the surface, the more difficult it is for cornstarch particles to adhere stably, requiring greater suction compensation to reduce detachment. The smoother the surface, the stronger the particle adhesion, and the compensation coefficient can be appropriately reduced.Surface roughness cannot be directly measured and must be indirectly reflected by the uniformity of residual particle distribution on the surface, measured in real time by a laser particle size sensor. The laser particle size sensor scans the biscuit blank surface at a frequency of 100 ms / time, statistically analyzes the distribution density of residual particles per unit area, and calculates the distribution uniformity, which is the ratio of the area where 90% of the residual particles are located to the total scanned area. The higher the distribution uniformity, the more regular the particle adhesion and the smoother the biscuit blank surface; the lower the uniformity, the more scattered the particle adhesion and the rougher the surface. Based on this distribution uniformity, the compensation coefficient is dynamically adjusted. The system pre-stores uniformity and coefficient adjustment tables. For example, when the distribution uniformity is greater than or equal to 90%, the surface is considered smooth. The compensation coefficient is reduced by 10% (from 0.2 kPa / % to 0.18 kPa / %). When the uniformity is between 70% and 90%, the initial coefficient of 0.2 kPa / % is maintained. When the uniformity is less than 70%, the surface is considered rough, and the coefficient is increased by 10%. For example, in a scenario with an absolute deviation of 3%, if the laser particle size sensor measures a distribution uniformity of 65% (less than 70%), the adjusted coefficient is 0.22 kPa / %. The final secondary suction compensation value is 3% × 0.22 kPa / % = 0.66 kPa. This value not only compensates for the problem of insufficient adhesion rate but also adapts to the adhesion problem caused by surface roughness, ensuring that the secondary compensation is accurate and fits the actual working conditions.

[0027] After the coupled model outputs the total suction command, the variable frequency fan controller and the pneumatic valve controller need to work together to convert the abstract suction command into the actual speed of the fan and the precise opening time of the valve. The former directly determines the suction strength of the negative pressure recovery hood, while the latter matches the recovery time according to the belt conveyor speed to avoid insufficient recovery or energy waste caused by the mismatch between suction and time. The specific implementation method is as follows: When dynamically adjusting the fan output, the variable frequency fan controller first receives the total suction command value sent by the coupled model and converts it into a corresponding motor speed control signal. The variable frequency fan controller uses a dedicated control module based on PLC. Its core is to establish the correspondence between suction and speed response curves through pre-stored data. This curve is generated through laboratory calibration: under fan no-load conditions, the target suction is adjusted from 1.8 kPa to 4.0 kPa in steps of 0.2 kPa. After each step stabilizes, the corresponding fan motor speed is recorded. For example, 2.0 kPa corresponds to 1000 rpm, 2.5 kPa corresponds to 1300 rpm, 3.0 kPa corresponds to 1600 rpm, and 3.5 kPa corresponds to 1900 rpm. These suction and speed data are fitted into a quadratic curve using the least squares method. The fan suction and speed have a square relationship, and the quadratic curve accurately matches this characteristic. The data is stored in the controller's EEPROM memory. During conversion, the controller first verifies the range of the total suction command value, then calculates the target speed through curve interpolation (e.g., 3.2 kPa corresponds to 1720 rpm). Subsequently, it generates a 0-10V analog speed control signal (0V corresponds to 0 rpm, 10V to 3000 rpm, and 1720 rpm to 5.73V), which is sent to the fan motor's frequency converter. The frequency converter then adjusts the fan impeller speed according to the preset suction and speed response curves. The driver uses a vector control algorithm to receive... After the controller sends the analog signal, it converts it into the motor's output frequency. Simultaneously, the actual rotational speed is collected in real-time by the motor's built-in encoder, forming a closed-loop control: if the actual speed is lower than the target speed, the driver automatically increases the output frequency; if it is higher than the target speed, the output frequency decreases, ensuring that the deviation between the actual and target speeds is always less than or equal to 5 rpm to avoid suction fluctuations. After the impeller speed is adjusted, the actual suction force of the negative pressure recovery hood is monitored in real-time by a pressure sensor installed inside the hood. If the deviation between the actual suction force and the command value is greater than 0.05 kPa, the controller will readjust the speed control signal to further calibrate the suction force and ensure accuracy. While adjusting the fan speed, the system also uses the real-time linear velocity meter feedback from the belt encoder... The unit length passage time is calculated using an incremental photoelectric encoder mounted on the drive roller of the belt. For each rotation of the roller, the encoder outputs 1000 pulses. The controller collects the encoder pulse signals via a high-speed counter module, counting the pulses at 100ms intervals. Combined with the circumference of the drive roller, the real-time linear velocity is calculated. For example, if 50 pulses are collected within 100ms, the linear velocity = (50 pulses / 1000 pulses / revolution) × 0.5m / revolution ÷ 0.1s = 0.25m / s. The unit length passage time refers to the time it takes for a unit length of belt to completely pass through the negative pressure recovery hood. The calculation logic is unit length ÷ real-time linear velocity. For example, when the linear velocity is 0.25m / s, the unit length passage time = 1m ÷ 0.1s.25 m / s = 4 s. This time is the baseline duration, representing the theoretical time required for a 1 m conveyor belt to travel from entering the recovery hood to leaving. It forms the basis for calculating the subsequent valve opening time. Then, based on this baseline duration, the compensation component of the total suction command value is converted into a duration extension coefficient according to a preset ratio. The compensation component is the difference between the total suction command value and the baseline suction value; its magnitude reflects the additional suction strength required. This additional suction requires a longer recovery time to fully recover the detached particles. The preset ratio was calibrated experimentally. Testing the recovery efficiency under different compensation values ​​revealed that for every 0.1 kPa increase in compensation value, the baseline duration needs to be extended by 0.05 times to maintain a recovery efficiency above 95%. Therefore, the ratio was set as extension coefficient = compensation component × 0.5 (1 / kPa). For example, when the compensation component is 1.2 kPa, the extension coefficient = 1.2 × 0.5 = 0.6. This coefficient represents the additional extension required beyond the baseline duration. Ultimately, the opening time equals the baseline... The product of the duration and (1 + expansion coefficient) (ensuring the duration is the base duration plus the expansion portion), for example, a base duration of 4s × (1 + 0.6) = 6.4s. The pneumatic valve controller uses this to control the valve opening and closing sequence. The pneumatic valve controller is a dedicated control module for solenoid valves. After receiving the opening duration signal, it first determines the belt position through the pulse signal from the belt encoder. When the front end of a certain belt segment reaches the inlet of the recovery hood, the controller sends a power-on command to the solenoid valve, the valve opens, and negative pressure begins to recover dust. When the power-on duration reaches 6.4s, i.e., the rear end of the belt leaves the recovery hood, the controller sends a power-off command, and the valve closes. Simultaneously, the controller dynamically updates the base duration and expansion coefficient based on changes in real-time linear velocity. For example, when the linear velocity increases to 0.3m / s, the base duration becomes 3.33s. With the compensation value unchanged, the opening duration becomes 3.33s × 1.6 = 5.33s, ensuring the valve opening duration is always synchronized with the belt conveyor rhythm, avoiding suction wastage and fully recovering particles detached from each belt segment.

[0028] The second objective of this invention is to provide a system for implementing an intelligent recovery control method for preventing food conveying and powder spillage, including any of the above-mentioned features, comprising: The detection unit 1 consists of a high frame rate industrial camera, a laser particle size sensor, a motor current sensor, and a belt encoder. The camera captures dynamic images of the biscuit blank surface after powdering through a coaxial light source, the laser sensor synchronously corrects the particle size, the current sensor collects the three-phase current of the motor, and the encoder feeds back the belt linear speed. The core processing unit 2 has two built-in parallel modules. The vibration analysis module extracts the fundamental amplitude of the current spectrum through fast Fourier transform and outputs the real-time vibration frequency by combining it with a pre-stored mapping table. The shedding amount calculation module uses the dynamic background difference method to process the image sequence and integrates laser particle size data to calculate the proportion of powder shedding and the surface adhesion rate. The control unit 3 integrates a coupled model processor and an actuator. The model processor generates a primary suction compensation value based on the vibration frequency deviation and a secondary suction compensation value based on the adhesion rate deviation. After superposition, the total suction command is output. The actuator converts the command into a fan speed signal and calculates the unit passage time of the belt based on the linear velocity. The opening time of the recovery hood valve is dynamically extended according to the compensation value ratio. The recycling unit 4 includes a negative pressure recycling hood, a variable frequency fan, and a cyclone separator. The recycling hood collects dust according to the dynamic suction and the extended opening time. The fan responds to the speed command, and the filtered dust is returned to the dust spreading device for recycling.

[0029] In this invention, the detection unit uses a high-frame-rate industrial camera, a laser particle size sensor, a motor current sensor, and a belt encoder to collect data on the percentage of powder detachment, belt vibration frequency, and conveying speed. The core processing unit uses two parallel modules to simultaneously calculate the vibration frequency and the percentage of detachment, constructing a coupled model. The control unit generates a primary suction compensation value based on the vibration frequency deviation and a secondary compensation value based on the adhesion rate deviation. After superposition, the variable frequency fan speed is dynamically adjusted. Simultaneously, the opening time of the recovery hood valve is proportionally extended in conjunction with the conveying speed. The recovery and circulation unit collects dust through a negative pressure recovery hood, filters it through a cyclone separator, and returns it to the powder spreading device for recycling, improving powder recovery efficiency and anti-sticking reliability, and reducing resource waste.

[0030] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent recovery control of anti-stick dusting of food delivery, characterized in that: Includes the following steps: S1. A powder-sprinkling device is installed above the biscuit blank conveyor belt, and a negative pressure recovery hood is installed below the belt. The suction of the negative pressure recovery hood is provided by a variable frequency fan. A laser particle size sensor and a high frame rate industrial camera are installed at the inlet of the negative pressure recovery hood. S2. Real-time acquisition of the proportion of cornstarch detachment and belt vibration frequency in a designated area of ​​the conveyor belt after cornstarch scattering. Continuous image capture of the biscuit blank surface after cornstarch scattering is obtained using a high frame rate industrial camera. Based on the grayscale value changes in the images, the trajectory of cornstarch particles detached from the biscuit blank surface is identified to calculate the proportion of cornstarch detachment. Simultaneously, the current of the conveyor motor is acquired through a motor current sensor. The amplitude of the fundamental component in the motor current spectrum is extracted through spectrum analysis, and the real-time vibration frequency is calculated based on the preset mapping relationship between the current amplitude and the real-time vibration frequency. S3. Establish a coupled model of real-time vibration frequency, powder fall-off ratio, and suction force, with real-time vibration frequency, powder fall-off ratio, and initial powder amount as input parameters; When the real-time vibration frequency is higher than the reference frequency, the difference between the real-time vibration frequency and the preset reference frequency is multiplied by the frequency and suction compensation coefficient to generate the primary suction compensation value; the frequency and suction compensation coefficient are constants that reflect the relationship between the vibration frequency deviation and the required suction compensation. Meanwhile, the adhesion rate is calculated based on the proportion of real-time powder fallout. When the adhesion rate is lower than the preset adhesion rate threshold, a secondary suction compensation value is generated by multiplying the absolute value of the adhesion rate deviation with the preset adhesion rate and the suction compensation coefficient. The preset adhesion rate and suction compensation coefficient are set as dynamic variables that are positively correlated with the surface roughness of the biscuit blank. The coefficients are adjusted by measuring the uniformity of the surface residual particle distribution in real time through a laser particle size sensor. The primary compensation value and the secondary compensation value are superimposed on the reference suction value to form the total suction command. The total suction command is sent to the variable frequency fan controller to dynamically adjust the fan output and calculate the time for a unit length of belt to pass through the recovery hood based on the real-time conveying speed. The compensation value component in the total suction command value is converted into a duration extension coefficient according to a preset ratio. The final opening duration is equal to the product of the base duration and the extension coefficient. The pneumatic valve controller controls the valve opening and closing sequence accordingly. S4. The dust collected by the recovery hood is separated and filtered before being returned to the dust spreading device for recycling.

2. The intelligent recovery control method for preventing food conveying and powder spillage according to claim 1, characterized in that: The high frame rate industrial camera is installed at a preset distance above the belt downstream of the powder-spreading device at a vertical top-down angle. Its field of view covers the entire width of the belt and extends a set length along the conveying direction. By configuring a coaxial light source of a specific wavelength to enhance the reflective properties of corn starch particles, it continuously captures a sequence of dynamic images of the biscuit blank surface at a preset high frame rate after the powder-spreading action is triggered. The duration of the capture covers the critical adhesion stabilization period after powder spreading. At the same time, a laser particle size sensor performs real-time particle size scanning on the same area to correct image recognition errors.

3. The intelligent recovery control method for preventing food conveying and powder spillage according to claim 2, characterized in that: When identifying particle trajectories based on changes in image grayscale values, the dynamic background subtraction method is used to segment the region of moving particles. The initial image after dusting is completed is extracted as the baseline background. Gray-level difference is performed frame by frame on subsequent image sequences. Pixel areas where the gray-level value changes exceed a set threshold are marked as detached particle groups. Then, the total pixel area of ​​the particle groups detached from the cookie dough surface in each frame is statistically analyzed through connected component analysis. Combined with the real-time average particle size data provided by the laser particle size sensor and the preset pixel area and mass conversion coefficients, the dusting mass detached from the cookie dough surface per unit time is calculated. Finally, the percentage of dusting mass detached from the cookie dough surface per unit time is output as the percentage of the initial dusting mass.

4. The intelligent recovery control method for preventing food conveying and powder spillage according to claim 1, characterized in that: The spectrum analysis specifically refers to: The original current signals of each phase of the conveyor motor are collected in real time by a three-phase current sensor. After filtering out high-frequency noise components higher than half the sampling frequency using an anti-aliasing filter, the current signal is windowed to suppress spectral leakage. Then, the time-domain current signal is converted into a frequency-domain spectrum by a fast Fourier transform. The characteristic frequency band corresponding to the fundamental wave of belt vibration is located in the spectrum diagram, and the amplitude of the fundamental wave component with the largest spectral amplitude in the characteristic frequency band is extracted as the vibration characteristic quantity.

5. The intelligent recovery control method for preventing food conveying and powder spillage according to claim 4, characterized in that: The mapping relationship between the current amplitude and the real-time vibration frequency was established through laboratory calibration: When the belt is unloaded, the speed of the conveyor motor is adjusted by a preset step size to cover the working speed range. The amplitude of the fundamental component of the motor current signal and the measured frequency value of the standard vibration sensor installed on the belt support are recorded simultaneously. The linear positive correlation characteristic curve between the two is fitted by the least squares method and stored as a mapping relationship table. During actual operation, the real-time vibration frequency is calculated by interpolation based on the real-time extracted fundamental component amplitude.

6. The intelligent recovery control method for preventing food conveying and powder spillage according to claim 5, characterized in that: The operation to generate the initial suction compensation value is as follows: The reference frequency is set to be equal to the natural vibration frequency of the belt. When the real-time vibration frequency is higher than the reference frequency, the frequency deviation value is divided into several continuous intervals. Different intervals correspond to different frequencies and suction compensation coefficients. The corresponding compensation coefficient is selected according to the interval to which the real-time vibration frequency belongs and multiplied by the frequency deviation value to generate the primary suction compensation value.

7. The intelligent recovery control method for preventing food sticking and powder scattering during food conveying according to claim 5, characterized in that: The operation to generate the secondary suction compensation value is as follows: The preset adhesion rate threshold is equal to the critical value of anti-stick failure. When the real-time surface adhesion rate is lower than this threshold, a secondary suction compensation value is generated based on the product of the absolute value of the adhesion rate deviation and the preset adhesion rate and suction compensation coefficient.

8. The intelligent recovery control method for preventing food sticking and powder scattering during food conveying according to claim 7, characterized in that: When dynamically adjusting the fan output, the variable frequency fan controller converts the received total suction command value into a corresponding motor speed control signal. Based on the preset suction and speed response curves, it adjusts the fan impeller speed. At the same time, it calculates the time taken per unit length based on the real-time linear velocity fed back by the belt encoder. Using this time as the base duration, it converts the compensation value component in the total suction command value into a duration expansion coefficient according to a preset ratio. Finally, the opening duration is equal to the product of the base duration and the expansion coefficient. The pneumatic valve controller controls the valve opening and closing sequence accordingly.

9. A system for implementing an intelligent recycling control method for preventing food conveying and powder spillage as described in any one of claims 1-8, characterized in that, include: The detection unit (1) consists of a high frame rate industrial camera, a laser particle size sensor, a motor current sensor and a belt encoder. The camera captures dynamic images of the biscuit blank surface after powdering through a coaxial light source. The laser sensor synchronously corrects the particle size. The current sensor collects the three-phase current of the motor and the encoder feeds back the belt linear speed. The core processing unit (2) has a built-in dual parallel module. The vibration analysis module extracts the fundamental amplitude of the current spectrum through fast Fourier transform and outputs the real-time vibration frequency by combining the pre-stored mapping table. The shedding amount calculation module uses the dynamic background difference method to process the image sequence and integrates the laser particle size data to calculate the proportion of powder shedding and the surface adhesion rate. The control unit (3) integrates the coupled model processor and actuator. The model processor generates a primary suction compensation value based on the real-time vibration frequency deviation and a secondary suction compensation value based on the adhesion rate deviation. After superposition, the total suction command is output. The actuator converts the command into a fan speed signal and calculates the unit passing time of the belt based on the linear velocity. The opening time of the recovery hood valve is dynamically extended according to the compensation value ratio. The recycling unit (4) includes a negative pressure recycling hood, a variable frequency fan and a cyclone separator. The recycling hood recycles dust according to the dynamic suction and extended opening time. The fan responds to the speed command. After filtration by the separator, the dust is returned to the dust spreading device for recycling.

Images