A multi-modal sensor-based rough food conveying state monitoring method and system

Abstract

The application provides a kind of based on multimodal sensing of bad grain conveying state monitoring method and system. Wherein, by arranging the multi-axis laser speed probe with dust cover on both sides of the conveying belt, eliminate the interference of dust on light path. Synchronous acquisition of surface layer particle laser doppler frequency shift signal and deep layer particle echo intensity data, establish layered control body model. Use echo intensity to compensate scattering signal, generate flow velocity distribution profile by autocorrelation processing. Further adopt optical flow method to construct particle motion vector field, calculate collision energy value through vector field energy distribution. When the energy value exceeds the preset threshold, generate positioning early warning signal, realize real-time monitoring and accurate positioning of particle collision anomaly. The technical scheme provided by the application can realize full-dimensional monitoring of particle motion state in the process of bad grain conveying in high dust environment, effectively identify collision anomaly and accurately locate the fault area, and ensure the quality of bad grain conveying.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial sensing and monitoring, and in particular to a kind of based on multi-modal sensing of can be beaten grain conveying state monitoring method and system. BACKGROUND

[0002] In the large-scale conveying scene of fermented grain, the multi-level nature of material flow state and the complexity of particle dynamic behavior directly affect the process stability and product quality. Due to the existence of flow velocity stratification and particle collision phenomenon of fermented grain in the conveying tank, it is urgent to develop a monitoring system with spatial resolution and multi-physical field perception to realize dynamic tracking of flow stratification boundary and real-time and accurate perception of particle collision state.

[0003] Currently, the current technical field adopts a fusion monitoring scheme based on acoustic signal analysis and embedded contact type sensing. By deploying an acoustic emission sensor array on the outer wall of the conveying tank, the acoustic characteristics generated by particle collision and friction during the flow of fermented grain are captured, and a flexible pressure-sensitive sensing unit is embedded in the tank body to measure the pressure distribution changes at different depths.

[0004] However, this scheme has the following core bottlenecks: acoustic signals are easily disturbed by environmental mechanical noise, especially under the vibration working condition of the conveying tank, the particle collision characteristic frequency band and the equipment noise frequency band are highly overlapped, which leads to a significant reduction in the accuracy of the bottom flow velocity inversion; the embedded contact type sensing unit is limited by physical size, and the spatial resolution is insufficient, making it difficult to accurately capture the instantaneous collision behavior of millimeter-level particles, and it cannot penetrate the high-density fermented grain flow layer to achieve full-section monitoring. SUMMARY

[0005] The present application provides a kind of based on multi-modal sensing of can be beaten grain conveying state monitoring method and system, to solve the problem that the prior art cannot penetrate the high-density fermented grain flow layer.

[0006] In a first aspect, the present application provides a kind of based on multi-modal sensing of can be beaten grain conveying state monitoring method, comprising:

[0007] Multi-axis laser speed measurement probes are arranged on both sides of the conveying belt, and a dust cover is arranged around the multi-axis laser speed measurement probes to eliminate the interference of dust on the light path reflection path during the conveying process;

[0008] The laser Doppler frequency shift signal of the surface layer particles and the echo intensity data of the deep layer particles are synchronously collected, and the multi-axis laser speed measurement probe is used to define the stratification control body of the surface layer region and the deep layer region of the conveying belt;

[0009] The scattered signals in the stratification control body are compensated in combination with the echo intensity data, and the laser Doppler frequency shift signal is autocorrelated to generate a flow velocity distribution profile containing the velocity difference between the surface layer particles and the deep layer particles.

[0010] Based on the flow velocity profile, the particle motion vector field is calculated by using the optical flow method, the collision region is identified by the sudden change characteristics of the velocity gradient in the particle motion vector field, and the energy value of the collision region is calculated according to the energy distribution of the vector field;

[0011] When the energy value exceeds the preset range, the pre-warning signal associated with the collision region is generated in combination with the corresponding beam splitting channel number of the hierarchical control body and the spatial coordinates of the multi-axis laser velocity measurement probe.

[0012] Optionally, according to the spatial distribution parameters of the multi-axis laser velocity measurement probe, the echo intensity data corresponding to each beam splitting channel in the hierarchical control body is extracted, the attenuation proportion of the scattering signal in the hierarchical control body is determined by the multi-band reflection intensity difference, and the scattering signal in the same beam splitting channel is dynamically gain compensated based on the attenuation proportion;

[0013] The compensated scattering signal and the laser Doppler frequency shift signal are matched according to the beam splitting channel, and the first frequency shift component corresponding to the surface layer particles and the second frequency shift component corresponding to the deep layer particles in the laser Doppler frequency shift signal are extracted;

[0014] The first frequency shift component and the second frequency shift component are time domain overlappingly intercepted by autocorrelation processing and phased sliding window, to generate the first signal segment associated with the surface layer particles and the second signal segment associated with the deep layer particles;

[0015] The phase difference accumulation calculation of adjacent windows is performed on the first signal segment and the second signal segment respectively, the first velocity component of the surface layer particles in the first signal segment and the second velocity component of the deep layer particles in the second signal segment are extracted according to the mapping relationship between the accumulated phase difference and the particle displacement;

[0016] According to the spatial distribution parameters of the multi-axis laser velocity measurement probe, the first velocity component and the second velocity component in the same beam splitting channel are velocity gradient superimposed in the depth direction of the hierarchical control body, to generate a hierarchical flow velocity profile containing the velocity difference between the surface layer particles and the deep layer particles.

[0017] Optionally, the first signal segment and the second signal segment are respectively divided into a plurality of equal-length time domain windows, adjacent time domain windows in each signal segment are partially overlappingly intercepted, and the length of the overlapping region between adjacent time domain windows is less than the total length of a single time domain window;

[0018] In the overlapping region of the adjacent time domain windows, the phase difference value between the end phase value of the previous time domain window and the start phase value of the next time domain window is calculated, and the phase difference values of all adjacent windows are accumulated along the time axis direction of the signal segment to obtain the accumulated phase difference value representing the particle motion trend.

[0019] According to the mapping relationship between the cumulative phase difference and the particle displacement, the cumulative phase difference value is converted into a displacement amount of the particle in a corresponding time interval of adjacent time domain windows based on a wavelength parameter of the layered control body laser, and a first speed component of a surface layer particle in the first signal segment and a second speed component of a deep layer particle in the second signal segment are extracted according to a ratio of the displacement amount to the time interval.

[0020] Optionally, when the energy value exceeds the preset range, the beam splitting channel number of the layered control body is extracted, and a position code of the collision area in the transverse direction of the conveyor belt is determined according to a preset transverse position mapping relationship of the beam splitting channel number;

[0021] The spatial coordinates of the multi-axis laser speed measurement probe are extracted, and a position code of the collision area in the longitudinal direction of the conveyor belt is determined according to a preset longitudinal position mapping relationship of the spatial coordinates;

[0022] The position code in the transverse direction and the position code in the longitudinal direction are directly spliced to form a unique location identifier of the collision area;

[0023] Based on the type mark of the energy value exceeding the preset range and the unique location identifier, a warning signal associated with the collision area is generated.

[0024] Optionally, based on the flow velocity distribution profile of adjacent time points, two groups of continuous flow velocity data with a preset sampling period as a time interval are obtained, the particle distribution in the two groups of continuous flow velocity data is calculated by using an optical flow method, and a particle motion vector field composed of a plurality of continuous regions is formed;

[0025] Each region in the particle motion vector field is traversed, the speed gradient between the current region and its directly adjacent region is calculated, the maximum value of all the speed gradient differences of the current region is extracted as a mutation feature value, and if the mutation feature value exceeds a preset threshold, the current region is marked as a collision region;

[0026] According to the energy distribution of the particle motion vector field, the motion vectors corresponding to all regions in the collision area are extracted, and energy accumulation is performed based on the square of the module length of the motion vectors to obtain an energy value of the collision area.

[0027] Optionally, for each region in the particle motion vector field, the difference absolute value of the speed gradient of each region and all directly adjacent regions is calculated based on the speed gradient, and the maximum value in the difference absolute value is selected as the mutation feature value of the current region;

[0028] The mutation feature value is compared with a preset threshold, and if the mutation feature value exceeds the preset threshold, the current region and directly adjacent regions thereof are collectively marked as an initial collision region;

[0029] When the boundary spacing of adjacent regions is less than a preset collision action radius, adjacent regions in the initial collision region that are in communication with each other are merged to form a continuous and closed final collision region.

[0030] Optionally, a plurality of multi-axis laser speed measurement probes are arranged at a preset interval on both sides of the conveyor belt, each multi-axis laser speed measurement probe includes a surface layer detection axis and a deep layer detection axis, and a spatial distribution parameter of the multi-axis laser speed measurement probe is generated;

[0031] A laser Doppler frequency shift signal reflected by surface layer particles is received by the surface layer detection axis, and echo intensity data reflected by deep layer particles is received by the deep layer detection axis;

[0032] According to the spatial distribution parameter, a surface layer penetration depth and a deep layer penetration depth corresponding to each multi-axis laser speed measurement probe are calculated;

[0033] Based on the horizontal spacing of adjacent multi-axis laser speed measurement probes and the difference between the corresponding surface layer penetration depth and deep layer penetration depth, a spatial segmentation frame composed of a rhombus grid is constructed on a cross section of the conveyor belt;

[0034] The coverage range of all surface layer penetration depths in the spatial segmentation frame is marked as a surface layer control body, and the coverage range of all deep layer penetration depths in the spatial segmentation frame is marked as a deep layer control body, which together define a layered control body of the conveyor belt.

[0035] In a second aspect, the application provides a rough grain conveying state monitoring system based on multi-modal sensing, comprising:

[0036] A configuration module is configured to set multi-axis laser speed measurement probes on both sides of the conveyor belt, and to configure a dust cover around the multi-axis laser speed measurement probes to eliminate the interference of dust on the light path reflection path during the conveying process;

[0037] A collection module is configured to synchronously collect laser Doppler frequency shift signals of surface layer particles and echo intensity data of deep layer particles, and to define a layered control body of a surface layer region and a deep layer region of the conveyor belt in combination with a spatial distribution parameter of the multi-axis laser speed measurement probes;

[0038] A processing module is configured to compensate scattering signals in the layered control body in combination with the echo intensity data, and to perform autocorrelation processing on the laser Doppler frequency shift signals to generate a flow velocity distribution profile containing the velocity difference between surface layer particles and deep layer particles;

[0039] The computing module calculates a particle motion vector field by using an optical flow method based on the flow velocity profile, identifies a collision area by a sudden change feature of a velocity gradient in the particle motion vector field, and calculates an energy value of the collision area according to an energy distribution of the vector field.

[0040] The generating module generates a pre-warning signal associated with the collision area when the energy value exceeds a preset range, in combination with a corresponding beam splitting channel number of the layered control body and a spatial coordinate of the multi-axis laser velocimeter probe.

[0041] In a third aspect, an embodiment of the present application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component, to implement the method for monitoring a conveying state of a grain according to the first aspect.

[0042] In a fourth aspect, an embodiment of the present application provides a computer storage medium, which stores a computer program; when the computer program is executed by a computer, the method for monitoring a conveying state of a grain according to the first aspect is implemented.

[0043] In the embodiment of the present application, multi-axis laser velocimeter probes are arranged on both sides of a conveying belt, and dustproof covers are arranged around the multi-axis laser velocimeter probes to eliminate the interference of dust on a light path reflection path in a conveying process; laser Doppler frequency shift signals of surface layer particles and echo intensity data of deep layer particles are synchronously collected, and layered control bodies of a surface layer area and a deep layer area of the conveying belt are defined through spatial distribution parameters of the multi-axis laser velocimeter probes; scattered signals in the layered control bodies are compensated in combination with the echo intensity data, and the laser Doppler frequency shift signals are subjected to autocorrelation processing, to generate a flow velocity profile containing a velocity difference between the surface layer particles and the deep layer particles; a particle motion vector field is calculated by using an optical flow method based on the flow velocity profile, a collision area is identified through a sudden change feature of a velocity gradient in the particle motion vector field, and an energy value of the collision area is calculated according to an energy distribution of the vector field; when the energy value exceeds a preset range, a pre-warning signal associated with the collision area is generated in combination with a corresponding beam splitting channel number of the layered control body and a spatial coordinate of the multi-axis laser velocimeter probe.

[0044] The present application has the following beneficial effects:

[0045] By the cooperative setting of the multi-axis laser velocity measurement probe and the dustproof cover, the scattering interference of dust on the light signal in the conveying process is effectively inhibited, the signal-to-noise ratio of the surface layer particle laser Doppler frequency shift signal and the stability of the deep layer particle echo intensity data are improved; based on the definition of the layered control body based on the spatial distribution parameters of the probe, the dynamic division of the physical boundary of the surface layer and the deep layer area of the conveying belt is realized, and the structured spatial constraint for the layered flow velocity modeling is provided; through echo intensity compensation and autocorrelation signal processing, the influence of deep scattering signal attenuation on velocity solution is eliminated, the velocity difference characteristics of the surface layer and the deep layer particles are accurately extracted, and the flow velocity distribution profile with depth resolution is generated; combined with the gradient mutation analysis of the particle motion vector field by the optical flow method, the limitation of the traditional single velocity threshold judgment is broken through, the mechanical strength of the collision area is quantified from the energy distribution dimension, and the mapping correlation of the beam splitting channel number and the spatial coordinates is established, the real-time binding mechanism of the collision position and the warning signal is established, and finally the full-link closed-loop monitoring of "layered flow velocity-collision positioning-exception warning" is realized.

[0046] Further, by dynamic gain compensation, the interference of scattering signal attenuation on deep layer particle measurement is effectively eliminated, and the reliability of echo intensity data is improved; by using frequency shift component separation and phase difference accumulation calculation, the velocity characteristics of the surface layer and the deep layer particles are accurately distinguished and quantified, and the precision bottleneck of traditional single signal aliasing is broken through; combined with the velocity gradient superposition modeling of the spatial distribution parameters, a high-resolution dynamic layered flow velocity profile is generated, and multi-scale correlation analysis from micro-particle motion to macro-flow velocity distribution is realized, providing high-precision data support for real-time positioning of the layered boundary of the flow velocity in the conveying tank and exception condition warning.

[0047] These aspects or other aspects of the present application will be more apparent in the following description of the embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0049] Figure 1 A flow chart of a kind of based on multi-modal sensing of the present application provides a kind of state monitoring method of grain conveying is shown;

[0050] Figure 2 The structure schematic diagram of a kind of based on multi-modal sensing of the present application provides a kind of state monitoring system of grain conveying is shown;

[0051] Figure 3 The structure schematic diagram of a kind of computing device provided by the present application is shown. DETAILED DESCRIPTION

[0052] In order for those skilled in the art to better understand the scheme of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application.

[0053] In some processes described in the specification and claims of the present application and the above description, a plurality of operations appearing in a specific order are included, but it should be clearly understood that these operations can be executed in parallel or in the order in which they appear in this text. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes can include more or fewer operations, and the operations can be executed in sequence or in parallel. It should be noted that the descriptions such as "first", "second", etc. in this text are used to distinguish different messages, devices, modules, etc. and do not represent the order of precedence. Also, "first" and "second" are not of different types.

[0054] Researchers found that existing bad grain conveying state monitoring technology is difficult to synchronously obtain the accurate flow rate distribution of surface and deep layer particles due to dust interference, signal aliasing and insufficient layering perception ability, and cannot effectively identify the mechanical abnormalities of dynamic collision area, resulting in large deviation of flow rate layering boundary positioning and lagging response of collision warning. Based on this, a bad grain conveying state monitoring method based on multi-modal sensing is provided. The method can eliminate dust interference and separate surface / deep layer particle motion characteristics through spatial distribution modeling and dynamic signal compensation technology of multi-axis laser speed probe, realize millimeter-level dynamic tracking of flow rate layering boundary and real-time quantitative evaluation of particle collision energy by combining optical flow method and vector field energy analysis, and finally generate spatially correlated accurate warning signals.

[0055] The technical solutions of the present application can be applied to flow rate layering monitoring and collision abnormality diagnosis of high-dust, multi-particle fluidized conveying tanks in wine brewing, feed processing and other scenarios.

[0056] The technical solutions in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0057] Figure 1 A flowchart of a bad grain conveying state monitoring method based on multi-modal sensing is provided for the embodiments of the present application, as shown in Figure 1 The method comprises:

[0058] 101、In the two sides of the conveying belt, multi-axis laser speed measurement probes are arranged, and dustproof covers are arranged outside the multi-axis laser speed measurement probes to eliminate the interference of dust on the reflection path of the light path during the conveying process;

[0059] In step 101, the multi-axis laser speed measurement probe refers to a laser speed measurement device (such as a 905 nm wavelength semiconductor laser arranged on each axis) including three independent laser emission directions of X-axis, Y-axis and Z-axis; the dustproof cover refers to a double-layer dustproof structure composed of an outer air curtain isolation layer and an inner quartz protective window, the air curtain isolation layer refers to a ring-shaped air barrier formed by a 0.3 MPa positive pressure airflow generated by a high-pressure air pump; the two sides of the conveying belt refer to the installation positions symmetrically distributed left and right along the conveying direction of the conveying belt, and the interval is set to 1.2 times the width of the conveying belt (such as 1.2 meters for a 1-meter-wide belt).

[0060] In the embodiment of the application, first, 4 groups of multi-axis laser speed measurement probe groups are arranged on the left and right sides of the conveying belt, and each group of probes includes 3 laser emitters distributed at 120° in the circumferential direction. Secondly, a rotating air curtain isolation layer is configured, which refers to a ring-shaped air curtain nozzle rotating 15 times per minute, covering an angle of 270°. Finally, the programmable logic controller (Programmable Logic Controller) is used to link the air curtain pressure and the conveying belt motor speed, and when the conveying belt speed increases to 2 m / s, the air curtain pressure is simultaneously increased to 0.35 MPa to maintain the dust blocking efficiency.

[0061] In the apron conveying system of a large grain processing center, technicians arranged MS-8 multi-axis laser-infrared composite sensor arrays at intervals of 2 meters along the two sides of a 1.2-meter-wide U-shaped trough conveying belt. Each measurement point integrates two groups of 785 nm laser speed measurement probes and a group of 3-5 μm waveband infrared thermal imaging modules, and is wrapped with a food-grade stainless steel dust cover. The surface of the cover body is coated with a hydrophobic and oleophobic coating, and the inside is maintained at 0.15 MPa positive pressure airflow by a micro-blower. Field tests show that in an environment where 150 mg / m³ of dust is generated during rice conveying, the light transmission rate is stable at 89%, which is 62% higher than that without protection, and the temperature measurement error of the infrared thermal imaging module is controlled within ±0.5℃.

[0062] 102、Synchronously collecting the laser Doppler frequency shift signal of the surface particles and the echo intensity data of the deep particles, and defining the layered control body of the surface area and the deep area of the conveying belt through the spatial distribution parameters of the multi-axis laser speed measurement probe;

[0063] In step 102, the laser Doppler frequency shift signal of the surface layer particles refers to the laser frequency offset amount caused by the movement of the grain particles in the 0-5 cm thickness range of the surface of the conveying belt (such as a speed of 1.5 m / s corresponding to a frequency shift amount of 10.2 MHz); the echo intensity data of the deep layer particles refers to the reflected light intensity value detected by the receiving end after the laser penetrates the 5-15 cm grain layer (such as when a long-wavelength laser of 1550 nm is used, the deep layer reflection intensity decays to 12% of the surface layer value); the spatial distribution parameter refers to a geometric parameter group composed of the installation height, the inclination angle and the spacing between adjacent probes of the multi-axis laser speed measurement probe (such as a height parameter H = 30 cm, an inclination angle θ = 15°, and a spacing parameter D = 25 cm); the layered control body refers to the monitoring area of the cross section of the conveying belt divided according to the spatial distribution parameter, and the surface layer region is defined as a triangular pyramid space domain (apex angle 22°) covered by direct laser, and the deep layer region is defined as an ellipsoidal space domain (long axis 35 cm, short axis 18 cm) covered by laser scattering.

[0064] In the embodiment of the application, first, the laser Doppler frequency shift signals of the X / Y / Z three axes are synchronously collected by a field-programmable gate array (FPGA) at a sampling rate of 10 kHz, and the echo intensity data of each channel is measured by an avalanche photodiode (APD). Second, a three-dimensional coordinate system is established according to the probe installation parameters, the surface layer region is modeled as a right triangular pyramid (volume V = 0.12 m³) with the vertex at the probe position and the bottom surface at the surface of the conveying belt, and the deep layer region is modeled as a rotating ellipsoid (volume V = 0.68 m³) with the probe as the focal point through the spatial distribution parameters of the multi-axis laser speed measurement probe. Finally, the probe data is associated to the corresponding layered control body through a coordinate mapping algorithm.

[0065] The system starts multi-modal synchronous collection: the laser probe scans the surface layer 0-8 cm deep grain flow at an inclination angle of 45°, and captures the particle Doppler frequency shift through 1.5 MHz high-speed sampling; the infrared module monitors the temperature field distribution of the 10-25 cm deep layer grain pile at a frame rate of 10 Hz. Combined with the spatial coordinates of the sensor (X = 3.6 m, Y = ±0.6 m, Z = 0.2 m), the grain tank is divided into 8 layered control bodies, wherein the thickness Δ = 5 cm of the surface layer control body corresponds to the focusing depth of the laser, and the deep layer control body is layered according to the Fibonacci sequence to Δ = 12 cm, and the moisture content gradient of each layer is calibrated through the infrared radiation intensity.

[0066] 103, the scattered signals in the layered control body are compensated combined with the echo intensity data, and the laser Doppler frequency shift signals are autocorrelated to generate a flow velocity distribution profile containing the velocity difference between the surface layer particles and the deep layer particles;

[0067] In step 103, the scattering signal compensation refers to an optical attenuation compensation calculation (compensation coefficient a = exp (μ·L), where μ is the grain extinction coefficient, and L is the penetration depth) on the deep echo intensity data based on the Beer-Lambert Law; the autocorrelation processing refers to a time-domain autocorrelation function calculation (autocorrelation window length 128 sampling points, lag time τ = 0-100 μs) on the laser Doppler frequency shift signal, and the particle motion velocity is inversely calculated by extracting the peak interval of the autocorrelation curve; and the flow velocity distribution profile refers to a two-dimensional cross-sectional vector diagram (the horizontal axis is the width direction coordinate of the conveying belt, the vertical axis is the thickness direction coordinate of the grain layer, and the velocity range 0-3 m / s is represented by a blue-yellow-red color spectrum) representing the velocity difference between the surface layer and the deep layer by a color gradient.

[0068] In the embodiment of the application, first, dynamic compensation is performed on the deep echo intensity data: when it is detected that the echo intensity of the 5-15 cm layered layer is lower than 25 mV, a 16-stage programmable gain amplifier (Programmable Gain Amplifier) is started to amplify the signal, and the amplification multiple G = 1 + 0.15×(25-V_echo). Secondly, the laser Doppler frequency shift signal is subjected to autocorrelation processing, and an improved Gabor transform is used for time-frequency analysis of the frequency shift signal to extract the characteristic frequency component f_peak = Δf / 2π. Finally, the K-means clustering algorithm is used to divide the velocity data into a surface high-speed cluster (Cluster1: 2.1±0.4 m / s) and a deep low-speed cluster (Cluster2: 1.3±0.6 m / s), and an RGB encoded flow velocity distribution profile with a spatial resolution of 0.5 cm is generated.

[0069] In response to the detection of abnormal echoes (20% sudden increase in infrared radiation intensity) by the layered control body, the system starts multi-modal compensation: first, the deep layer grain pile moisture content is inversely calculated (from 14.5% to 17.2%) according to the infrared thermal image data, and the laser Mie scattering signal is subjected to humidity compensation correction; then, the surface Doppler signal is subjected to wavelet denoising processing to suppress the 0.8-1.2 kHz high-frequency noise caused by the conveying chain plate vibration. The compensated data shows that under the condition of chain speed 0.8 m / s, the surface particle flow velocity v_s = 0.76 m / s, while the flow velocity v_d = 0.58 m / s at a depth of 15 cm, forming a significant flow velocity retardation zone.

[0070] 104、Based on the flow velocity distribution profile, the particle motion vector field is calculated by using the optical flow method, the collision area is identified by the mutation characteristics of the velocity gradient in the particle motion vector field, and the energy value of the collision area is calculated according to the energy distribution of the vector field;

[0071] In step 104, the optical flow method calculation refers to a dense motion field estimation algorithm based on the Horn-Schunck optical flow equation (∇I·v + I_t = 0), where I is the intensity of the velocity field, and v is the velocity vector to be solved; the abrupt change of the velocity gradient refers to the spatial derivative ∂v_x / ∂x or ∂v_y / ∂y exceeding a set threshold value; and the vector field energy distribution refers to the square integral value of the length of the motion vector per unit volume.

[0072] In the embodiments of the present application, first, 10 consecutive flow velocity distribution profiles are obtained at a time interval of 100 ms, and the optical flow equation is solved by OpenCL parallel computing architecture. Second, the Sobel edge detection operator is used to calculate the amplitude of the velocity gradient, and a threshold condition is set: when the absolute value of the X-direction gradient is > 7.5 (m / s) / m and the absolute value of the Y-direction gradient is > 4.2 (m / s) / m, it is determined that the area is a collision area. Finally, the energy density of the marked area is calculated, and the energy density warning threshold is set to 200 J / m³. When the detection area E is ≥ 200 J / m³, the subsequent processing flow is triggered.

[0073] Based on the fused multi-modal data, a three-dimensional motion vector field is constructed using the Lucas-Kanade optical flow method. In the 7th section of the grain tank (X = 16.8m), an abnormal Z-axis direction velocity gradient is detected: the surface flow velocity drops from 0.75 m / s to 0.41 m / s within 0.5 seconds, and the corresponding vector field divergence value reaches 4.3 s⁻¹. Combined with the infrared temperature rise data (the local temperature rises from 25°C to 28°C), the system determines that mold clumping occurs in this area, and the potential energy E_p = ∫(ρgh)dV reaches 2.3 kJ, far exceeding the normal threshold of 0.8 kJ.

[0074] 105、When the energy value exceeds the preset range, the corresponding beam channel number of the layered control body and the spatial coordinates of the multi-axis laser velocimeter probe are combined to generate a warning signal associated with the collision area.

[0075] In step 105, the beam channel number refers to the spatial position coding of each laser beam in the multi-axis laser velocimeter probe group (e.g., X3-Y2-Z1 represents the laser beam emitted in the third column X-axis direction, the second row Y-axis direction, and the first layer Z-axis direction); the spatial coordinates refer to a meter-based coordinate system established based on the starting end of the conveyor belt (e.g., the X-axis is along the length direction of the conveyor belt, the origin is at the inlet, and the coordinate range is 0-50m); the spatial coordinates are dynamically calibrated through the synchronous measurement data of the beam channel and the installation spacing parameters of the probe; and the warning signal refers to a standardized alarm message containing position coordinates, energy value, and time stamp.

[0076] In the embodiments of the present application, firstly, a mapping relationship table of beam splitting channel space encoding and physical coordinates of the conveying belt is established (for example, X3-Y2-Z1 corresponds to the conveying belt 3.2 meters away, the coordinates X=3.2 m, Y=0.6 m, and Z=0.3 m). Secondly, when the collision area energy value exceeds the threshold value, a three-level early warning signal is published through the Message Queuing Telemetry Transport (MQTT) protocol: the first level of early warning (200-300 J / m³) triggers the sound-light alarm and the HMI interface flickers; the second level of early warning (300-500 J / m³) automatically reduces the conveying belt speed to 80% of the rated value; the third level of early warning (> 500 J / m³) executes an emergency stop instruction and sends a short message to notify the maintenance personnel. Finally, the alarm data is synchronized to the manufacturing execution system through the OPC unified architecture interface.

[0077] When the accumulated energy exceeds the preset safety value, the system automatically associates the 2# beam splitting channel (encoding GL-07-02) with the infrared module space grid (X=16.8±0.5m, Y=-0.45~+0.45m, Z=0.15-0.25m), and generates a four-level early warning instruction. Through the OPCUA protocol, the alarm information containing the blocking coordinates, moisture content and temperature rise rate is pushed to the central control system, and the high-pressure pulse air hammer (frequency 5 Hz, pressure 0.6 MPa) of the section is activated synchronously to automatically dredge. The on-site inspection by the operation and maintenance personnel found that the corn kernels were swollen due to the infiltration of rainwater, which was consistent with the system diagnosis result with a coincidence degree of 96%.

[0078] In summary, the method of real-time monitoring and abnormal early warning of grain conveying flow state based on multi-modal perception is realized through steps 101 to 105, and a flow state abnormal early warning instruction with spatial positioning accuracy is generated. This method breaks through the perception limitations of traditional single-mode monitoring, realizes multi-level dynamic perception and hidden fault prediction of grain flow conveying state under complex working conditions, and significantly improves the intelligent control level of the grain conveying system.

[0079] In order to solve the industry problems of insufficient detection accuracy of layered flow rate and coupling of multi-physical field interference in the process of distiller's grains conveying, an intelligent monitoring system for distiller's grains conveying state based on multi-modal sensor fusion is developed, which realizes millimeter-level precision dynamic monitoring of layered flow rate and three-dimensional flow field reconstruction. In some embodiments, the scattered signals in the layered control body are compensated by combining the echo intensity data in step 103, and the laser Doppler frequency shift signals are autocorrelated to generate a flow rate distribution profile containing the velocity difference between surface particles and deep particles, including:

[0080] 201. Based on the spatial distribution parameters of the multi-axis laser velocimetry probe, extract the echo intensity data corresponding to each beam channel in the layered control body, determine the attenuation ratio of the scattered signal in the layered control body through the multi-band reflection intensity difference, and perform dynamic gain compensation on the scattered signal in the same beam channel based on the attenuation ratio.

[0081] In step 201, the spatial distribution parameters of the multi-axis laser velocimetry probe refer to the set of geometric parameters consisting of the probe's installation height, laser emission tilt angle, and distance between adjacent probes; the layered control volume refers to the monitoring area of ​​the conveyor belt cross-section divided according to the spatial distribution parameters, including a surface control volume with a grain layer depth range of 0 to 5 cm and a deep control volume with a grain layer depth range of 5 to 15 cm; the beam splitting channel refers to the signal acquisition path encoding of each independent laser beam within the multi-axis laser velocimetry probe; the echo intensity data refers to the reflected light power value detected by the receiver after the laser penetrates the grain layer; the multi-band reflection intensity difference refers to the ratio of echo intensity of different wavelength lasers within the same beam splitting channel; the attenuation ratio refers to the optical signal attenuation coefficient calculated through the multi-band echo intensity difference; and the dynamic gain compensation refers to the operation of adjusting the signal amplifier gain in real time according to the attenuation ratio.

[0082] In this embodiment, a beam-splitting channel spatial mapping table is first generated based on the probe's installation height, tilt angle, and spacing parameters. For example, the X1-Y1-Z1 channel corresponds to a conveyor belt with a horizontal coordinate of 0.3 meters, a vertical coordinate of 0.2 meters, and a vertical coordinate of 0.1 meters. Next, frequency band differential processing is performed on the dual-band laser echo intensity data. When the echo intensity of the deep control body is detected to be lower than a preset threshold, the echo intensity ratio of different bands is calculated, and the optical signal attenuation ratio is derived by combining the wavelength parameters. Then, the gain value of the signal link is adjusted using a digital potentiometer. For example, when the attenuation ratio is 0.83, the gain coefficient of the corresponding beam-splitting channel is increased to 1.34 times to compensate for signal attenuation. Finally, the compensated echo intensity data is associated with and stored in conjunction with the beam-splitting channel encoding.

[0083] 202. Match the compensated scattering signal with the laser Doppler frequency shift signal according to the beam splitting channel, and extract the first frequency shift component corresponding to the surface particles and the second frequency shift component corresponding to the deep particles in the laser Doppler frequency shift signal;

[0084] In step 202, the compensated echo intensity data refers to the reflected light signal after dynamic gain compensation processing; the laser Doppler frequency shift signal refers to the light wave frequency shift caused by particle motion; beam splitting channel matching refers to aligning the compensated echo intensity data with the corresponding frequency shift signal according to spatial coordinates; the first frequency shift component refers to the frequency shift signal component generated by the surface particle motion; and the second frequency shift component refers to the frequency shift signal component generated by the deep particle motion.

[0085] In the embodiments of the present application, firstly, the compensated echo intensity data and the laser Doppler frequency shift signal are aligned according to the three-dimensional coordinates through the beam splitting channel spatial mapping table, for example, the X1-Y1-Z1 channel data pair of the coordinate (0.3 meters, 0.2 meters, 0.1 meters). Secondly, two groups of finite-length unit impulse response (FIR) digital filters are designed, the passband of the first filter is set to 6-12 megahertz to extract the surface layer frequency shift component (corresponding to the speed of 0.9-1.8 meters per second), and the passband of the second filter is set to 2-5 megahertz to extract the deep layer frequency shift component (corresponding to the speed of 0.3-0.75 meters per second). Then, the cross-correlation algorithm is used to verify the data alignment accuracy, and the correlation coefficient of the echo intensity data and the frequency shift signal is calculated. When the correlation coefficient exceeds 0.85, it is determined that the matching is successful. Finally, the first frequency shift component and the second frequency shift component separated are respectively stored in the double-channel first-in-first-out (FIFO) storage queue.

[0086] 203. The first signal segment associated with the surface layer particles and the second signal segment associated with the deep layer particles are generated by time domain overlapping truncation of the first frequency shift component and the second frequency shift component through autocorrelation processing and a phased sliding window.

[0087] In step 203, autocorrelation processing refers to time domain autocorrelation function calculation of the frequency shift signal; phased sliding window refers to a time domain truncation method using a variable window length; first signal segment refers to a time domain data segment of the surface layer frequency shift component after truncation; and second signal segment refers to a time domain data segment of the deep layer frequency shift component after truncation.

[0088] In the embodiments of the present application, firstly, the first frequency shift component is truncated using a fixed window (window length 20 milliseconds, sampling rate 128 kilohertz, containing 2560 sampling points), and each time the sliding step is 10 milliseconds, to generate a time domain signal segment with an overlap rate of 50%. Secondly, the second frequency shift component is truncated using an adaptive window truncation algorithm: when the signal variance exceeds 0.1 volts square, the window length is shortened to 30 milliseconds (3840 sampling points), otherwise the default 50 milliseconds (6400 sampling points) is maintained. Then, autocorrelation function calculation is performed on each signal segment, the window length is 128 sampling points, the lag time range is 0 to 100 microseconds, and the main period is extracted by the peak detection algorithm (for example, the main period of the surface layer signal segment is 0.15 milliseconds, corresponding to the speed of 1.2 meters per second). Finally, a time stamp label is added to each signal segment, and stored in a time sequence database.

[0089] 204. The first velocity component of the surface layer particles in the first signal segment and the second velocity component of the deep layer particles in the second signal segment are extracted according to the mapping relationship between the cumulative phase difference and the particle displacement by performing phase difference accumulation calculation on adjacent windows of the first signal segment and the second signal segment respectively.

[0090] In step 204, the phase difference accumulation calculation of adjacent windows refers to the integration operation of the phase change of continuous signal segments to eliminate random noise; the accumulated phase difference refers to the sum of the phase differences of multiple windows; the particle displacement refers to the movement calculated according to the physical relationship between the phase difference and the wavelength of laser; the first velocity component refers to the instantaneous velocity value of the surface layer particles; and the second velocity component refers to the instantaneous velocity value of the deep layer particles.

[0091] In the embodiment of the present application, first, the phase unwrapping algorithm is performed on the first signal segment to extract the phase difference sequence of adjacent windows (for example, the first window phase difference is 0.12 radians, and the second window phase difference is 0.15 radians). Second, the accumulated phase difference is calculated, and when the accumulated phase difference is greater than or equal to 2π, the modulo 2π correction is performed (that is, subtracting an integer multiple of 2π). Then, according to the wavelength of laser 905 nanometers, the particle displacement is calculated to be equal to the wavelength multiplied by the accumulated phase difference and then divided by 4π, for example, the accumulated phase difference of 2.7 radians corresponds to a displacement of 0.194 millimeters. Finally, the velocity component is calculated by difference, and the formula is displacement divided by time interval. When the time interval is 20 milliseconds, the velocity is calculated to be 0.485 meters per second, and the result is written into the velocity buffer.

[0092] 205. According to the spatial distribution parameters of the multi-axis laser velocity measurement probe, the first velocity component and the second velocity component in the same beam splitting channel are superimposed in the depth direction of the layered control body to generate a layered flow velocity distribution profile containing the velocity difference between the surface layer particles and the deep layer particles.

[0093] In step 205, the velocity gradient superposition refers to the spatial integration of the velocity component in the depth direction of the layered control body to quantify the flow heterogeneity; and the layered flow velocity distribution profile refers to a two-dimensional heat map representing the velocity difference at different depths in pseudo-color coding.

[0094] In the embodiment of the present application, first, a velocity-depth mapping table is established according to the vertical coordinates of the beam splitting channel (for example, a vertical 0.3 meters corresponds to a grain layer depth of 5 centimeters, and a vertical 0.8 meters corresponds to 12 centimeters). Second, the velocity gradient value of each depth layer is calculated, and the formula is the integral of the difference between the surface layer velocity and the deep layer velocity along the depth direction, for example, at a depth of 5 centimeters, the surface layer velocity is 1.2 meters per second, the deep layer velocity is 0.8 meters per second, and the gradient value is 0.4 meters per second per centimeter. Then, a continuous velocity field is generated by using a cubic spline interpolation algorithm, and the velocity gradient is mapped to a pseudo-color image (the velocity difference range of negative 0.5 to positive 1.2 meters per second is represented by blue to red gradient) through a graphics rendering engine. Finally, the area where the gradient value exceeds the threshold value of 0.3 meters per second per centimeter is marked as a red warning area.

[0095] The following is a specific example:

[0096] In the dynamic monitoring scenario of the grain conveying system, when the conveying belt transports broken corn kernels with a moisture content of 17% at a speed of 2.5 m / s, the multi-modal sensing system starts layered flow rate monitoring. Five groups of multi-axis laser velocimeter probes (wavelength 1550 nm, detection angle 60°) are arranged transversely along the conveying belt to synchronously emit 32 beams of probe light. According to the spatial distribution parameters, the echo intensity data of each beam channel in the layered control body is extracted. For the high attenuation phenomenon detected by the No. 3 beam channel (20 kHz band attenuation up to 43 dB), the system dynamically activates the gain compensation algorithm to increase the gain of the 2.8 GHz high-frequency scattering signal by 12 dB to the effective detection threshold. After compensation, the scattering signal is time-frequency matched with the Doppler frequency shift signal, and the fast frequency shift component of the surface layer particles and the slow component of the deep layer particles are separated in the 58-62 MHz frequency domain interval. The surface layer signal segment S1 and the deep layer signal segment S2 are obtained by autocorrelation function window interception, and the cumulative calculation of the phase difference between adjacent windows shows that S1 segment generates a 37π phase difference within 200 ms, corresponding to a surface layer particle displacement of 41 cm; S2 segment generates a 28π phase difference, corresponding to a deep layer displacement of 32 cm. Finally, the v1 and v2 of each beam channel are gradient superimposed in the conveying depth direction, and the generated layered flow rate profile shows that the surface layer 0-15 cm flow rate is 2.42±0.15 m / s, and the deep layer 15-30 cm flow rate decreases to 1.82±0.21 m / s. After the edge computing node processes the profile data, it successfully warns of the abnormal stagnation phenomenon in the left 15% area of the conveying belt, triggering the automatic shaking device to clear the accumulated material within 3 seconds, which is 8 minutes earlier than the traditional single-point monitoring method to identify the blockage risk.

[0097] In summary, the three-dimensional dynamic monitoring method of material flow state based on multi-spectral layered analysis is realized through steps 201 to 205. The layered control body is constructed by the spatial distribution parameters of the multi-axis laser velocimeter probe, the scattering signal attenuation is dynamically compensated by combining the multi-band reflection intensity difference, and the interference of the dust environment on the echo signal of the deep layer particles is eliminated. The layered signal decoupling technology is used to separate the high-frequency Doppler frequency shift component of the surface layer particles and the low-frequency vibration characteristics of the deep layer particles, the time and space continuity parameters of particle motion are extracted by autocorrelation processing and phase difference accumulation calculation, and the independent velocity components of the surface layer and deep layer particles are reconstructed. Based on the spatial gradient superposition algorithm, the layered flow rate distribution profile containing velocity difference is generated, breaking through the dimensional limitation of traditional single-point flow rate detection, realizing the three-dimensional dynamic analysis of the particle flow state on the cross section of the conveying belt, accurately identifying the hidden velocity gradient anomaly of surface layer slip and deep layer stagnation, and providing high-precision layered velocity field data support for the flow state imbalance warning of the material conveying process.

[0098] To address the technical challenge of accurately calculating the velocity of stratified particles during grain transport, a system for accurately calculating stratified flow velocity based on phase difference accumulation was developed, achieving millimeter-level precision in separating and measuring the velocities of surface and deep particles. In some embodiments, step 204 involves performing phase difference accumulation calculations on adjacent windows of the first and second signal segments, and extracting the first velocity component of surface particles in the first signal segment and the second velocity component of deep particles in the second signal segment based on the mapping relationship between the accumulated phase difference and particle displacement. This includes:

[0099] 301. Divide the first signal segment and the second signal segment into multiple time-domain windows of equal length, and partially overlap and truncate adjacent time-domain windows in each signal segment, wherein the length of the overlapping area between adjacent time-domain windows is less than the total length of a single time-domain window.

[0100] In step 301, the time-domain window refers to the time interval unit divided into signal segments according to a fixed duration; partial overlap truncation refers to the truncation method of retaining some repeated data segments between adjacent windows; the first signal segment refers to the data segment after time-domain truncation of the frequency shift signal corresponding to the surface particle movement; the second signal segment refers to the data segment after time-domain truncation of the frequency shift signal corresponding to the deep particle movement.

[0101] In this embodiment, the first and second signal segments are first divided into multiple time-domain windows of equal length. The window length is set according to the signal sampling rate; for example, when the sampling rate is 128 kHz, the window length is set to 20 milliseconds, containing 2560 sampling points. A sliding window algorithm is used to achieve partial overlap truncation. The overlap region length of adjacent windows is set to 40% of the total window length, i.e., the overlap region of a 20-millisecond window is 8 milliseconds, ensuring that the overlap region is less than the total length of a single window. For the surface first signal segment, a fixed window division is used, while for the deeper second signal segment, the window length is dynamically adjusted according to the real-time signal-to-noise ratio: when the signal-to-noise ratio is lower than 15 dB, the window length is shortened to 15 milliseconds to improve signal stability.

[0102] 302. Within the overlapping area of ​​the adjacent time domain windows, calculate the phase difference between the end phase value of the previous time domain window and the beginning phase value of the next time domain window, and accumulate the phase difference values ​​of all adjacent windows along the time axis of the signal segment to obtain the cumulative phase difference value characterizing the particle motion trend.

[0103] In step 302, the overlapping region refers to the data segment repeatedly extracted between adjacent time domain windows; the phase difference refers to the start and end phase offset of adjacent windows within the overlapping region; and the cumulative phase difference refers to the sum of phase differences accumulated along the time axis of the signal segment.

[0104] In the embodiment of the present application, first, the phase alignment operation is performed on each signal segment in the overlapping area, and the least square phase estimation algorithm is used to calculate the difference between the end phase value of the previous time domain window and the start phase value of the next time domain window. For example, the end phase value of the previous window is 1.2 radians, and the start phase value of the next window is 1.5 radians, and the phase difference is 0.3 radians. Secondly, the phase difference values of all adjacent windows are accumulated along the time axis direction, and the calculation formula is that the cumulative phase difference is equal to the sum of the phase differences of all adjacent windows. In order to suppress noise interference, the cumulative phase difference is subjected to a sliding average filter, and the filter window length is 5 data points, eliminating random phase jumps. Finally, the cumulative phase difference values of the surface signal segment and the deep signal segment are recorded, for example, the cumulative phase difference of the surface 10 windows is 2.7 radians, and the cumulative phase difference of the deep 6 windows is 1.2 radians.

[0105] 303、According to the mapping relationship between the cumulative phase difference and the particle displacement, the cumulative phase difference value is converted into the displacement amount of the particle in the time interval corresponding to the adjacent time domain window based on the wavelength parameter of the laser in the layered control body, and the first speed component of the surface particle in the first signal segment and the second speed component of the deep particle in the second signal segment are extracted according to the ratio of the displacement amount to the time interval.

[0106] In step 303, the displacement amount refers to the physical moving distance of the particle in the time interval corresponding to the adjacent time domain window; the first speed component refers to the instantaneous speed value of the surface particle; the second speed component refers to the instantaneous speed value of the deep particle; and the wavelength parameter refers to the wavelength value of the laser in the layered control body.

[0107] In the embodiment of the present application, first, the cumulative phase difference value is converted into the displacement amount according to the laser wavelength parameter, and the calculation formula is that the displacement amount is equal to the laser wavelength multiplied by the cumulative phase difference and then divided by 4π. For example, when the laser wavelength is 905 nanometers and the cumulative phase difference is 2.7 radians, the displacement amount is calculated as 905 nanometers multiplied by 2.7 divided by 4π, which is 0.194 millimeters. Secondly, the speed component is calculated according to the time interval, and the speed component is equal to the displacement amount divided by the window time interval. For example, when the time interval is 20 milliseconds, the speed component is 0.194 millimeters divided by 20 milliseconds, and the result is 0.485 meters per second. For the first signal segment of the surface, the speed components of all windows are extracted and the median is taken as the first speed component; for the second signal segment of the deep, the mean value is taken as the second speed component after removing outliers. Finally, the speed data is associated with the depth coordinate of the layered control body, and written into the flow rate database.

[0108] The following is a specific example:

[0109] In the multi-modal monitoring scenario of the grain conveying system, when the conveying belt transports the broken grain with high moisture content at medium speed, the system starts the layered particle motion analysis process. For the surface high-frequency signal segment and the deep low-frequency signal segment, first, the surface signal is cut according to a 50 millisecond time window length, and a 30% overlap rate is set to form a continuous analysis window. The deep signal uses an 80 millisecond time window length and retains a 25% overlap area. In the overlap area of adjacent windows, the system automatically captures the continuity of the phase value, for example, in a certain surface signal segment, the phase transition of three consecutive windows is detected, and a significant phase difference is accumulated. Based on the 1550 nanometer laser wavelength parameter, the system maps the accumulated phase change to the actual displacement data of the particles through the phase difference-displacement conversion model, and automatically calculates the surface particle motion speed combined with the corresponding time window length. The synchronous processing of the deep signal analysis shows that the deep displacement of a specific beam channel area is only 43% of the surface value, triggering the layered flow rate abnormality alarm. Through multi-modal data fusion verification, it is confirmed that the abnormality is caused by the material accumulation caused by the deformation of the conveying belt side baffle, and the system immediately links the malting mechanism to implement directional vibration dredging, and restores the flow rate balance state within 12 seconds, which is 15 times more efficient than manual inspection, and avoids about 3.5 tons of material from being hardened.

[0110] In summary, the particle layered velocity analysis method based on time domain phase chain accumulation is realized through steps 301 to 303, the particle motion phase continuity tracking mechanism is established through the equal-length time domain window overlap interception technology, the phase difference value is extracted in the overlapping area of adjacent windows and is chain accumulated along the time axis, and the cumulative phase feature quantity of particle displacement is constructed; combined with the laser wavelength parameter, the cumulative phase difference value is converted into the actual displacement of the layered particles, and the independent velocity components of the surface and deep particles are analyzed through the dynamic correlation of displacement and time interval. This method uses the window overlap phase inheritance strategy to eliminate the phase jump error caused by traditional signal truncation, uses the phase chain accumulation to enhance the spatiotemporal continuity representation ability of the particle motion trend, realizes the high-precision dynamic detection of the layered particle velocity, and provides the transient velocity field data support with sub-millimeter level displacement resolution for the abnormal conveying flow state.

[0111] In order to solve the accurate positioning problem of collision energy abnormality in the conveying process of malting grains, a collision warning system based on multi-axis laser positioning is developed, which realizes real-time monitoring and accurate alarm of the collision area with millimeter level spatial resolution. In some embodiments, when the energy value exceeds the preset range in step 105, the warning signal associated with the collision area is generated in combination with the corresponding beam channel number of the layered control body and the spatial coordinates of the multi-axis laser velocity probe, including:

[0112] 401. When the energy value exceeds the preset range, extract the beam channel number of the layered control body, and determine the position code of the collision area in the transverse direction of the conveyor belt according to the beam channel number and the preset transverse position mapping relationship.

[0113] In step 401, the beam splitting channel number refers to the unique identifier of each laser beam in the multi-axis laser velocimetry probe, which is composed of an axial code (X / Y / Z) and a serial number; the lateral position mapping relationship refers to the mapping rule for determining the lateral position of the collision area on the conveyor belt according to the axial code in the beam splitting channel number and the preset correspondence table of lateral coordinate intervals of the conveyor belt; the collision area refers to the abnormal particle flow area determined by energy value detection; the lateral position code of the conveyor belt refers to the digital code used to identify the position of the collision area in the width direction of the conveyor belt.

[0114] In this embodiment, firstly, the beam channel number (e.g., X3-Y2-Z1) corresponding to the layered control body that triggered the energy value exceeding the limit is extracted. Secondly, coordinate transformation is performed according to the lateral position mapping table: the X-axis code in the beam channel number corresponds to the lateral coordinate range of the conveyor belt (e.g., X3 corresponds to 3.0-3.5 meters laterally), and the lateral position code "H3" is generated by looking up the table. Finally, the lateral position code is stored in the cache queue, and timestamp information is recorded.

[0115] 402. Extract the spatial coordinates of the multi-axis laser velocity measuring probe, and determine the position code of the collision area in the longitudinal direction of the conveyor belt according to the spatial coordinates and the preset longitudinal position mapping relationship;

[0116] In step 402, spatial coordinates refer to the installation position parameters of the multi-axis laser velocity measuring probe in the three-dimensional coordinate system of the conveyor belt, including longitudinal coordinates (Y-axis), transverse coordinates (X-axis), and vertical coordinates (Z-axis); longitudinal position mapping relationship refers to the mapping rule for determining the position of the collision area in the length direction of the conveyor belt based on the correspondence table between the probe's longitudinal coordinate parameters and the segmented intervals of the conveyor belt length direction; the conveyor belt longitudinal position code refers to the digital code used to identify the position of the collision area in the length direction of the conveyor belt.

[0117] In this embodiment, the spatial coordinate parameters of the triggering warning probe (e.g., longitudinal coordinate Y = 5.2 meters) are first read from the device configuration library. Next, according to the longitudinal position mapping rules, the longitudinal coordinate is segmented at 0.5-meter intervals (e.g., the 5.0-5.5 meter interval is coded as "V5"), and a linear interpolation algorithm is used to determine the longitudinal position code. Finally, the longitudinal position code "V5" is associated with the lateral position code "H3" and stored.

[0118] 403. The horizontal position code and the vertical position code are directly concatenated to form a unique location identifier for the collision area;

[0119] In step 403, the unique location identifier refers to a globally unique location identifier formed by concatenating the lateral position code and the longitudinal position code in a fixed order; direct concatenation refers to the operation of sequentially connecting the string type position code to generate a composite identifier.

[0120] In the embodiment of the application, the lateral position code "H3" and the longitudinal position code "V5" are first extracted from the cache queue. Secondly, the string concatenation operation is performed according to the "lateral code + longitudinal code" rule to generate the unique location identifier "H3V5". Finally, the uniqueness of the identifier is verified by the hash check algorithm: calculate the SHA-256 hash value of the identifier, compare with the historical database, if not repeated, it is determined as a valid identifier.

[0121] 404、Based on the type mark of the energy value exceeding the preset range and the unique location identifier, a pre-warning signal associated with the collision area is generated.

[0122] In step 404, the type mark refers to a pre-warning classification mark according to the level of energy value exceeding the preset range; the unique location identifier refers to a globally unique location identifier formed by concatenating the lateral position code and the longitudinal position code; the pre-warning signal refers to a standardized alarm message containing the type mark, the unique location identifier and the timestamp.

[0123] In the embodiment of the application, three levels of pre-warning types are defined according to the energy value exceeding the range: the energy value in the range of 200 to 300 joules per cubic meter is marked as a first level pre-warning "W1", the energy value in the range of 300 to 500 joules per cubic meter is marked as a second level pre-warning "W2", and the energy value exceeding 500 joules per cubic meter is marked as a third level pre-warning "W3". Secondly, the standardized alarm message is constructed in JSON format, and the key fields include the type mark (alert_type), the unique location identifier (location_code) and the timestamp in ISO 8601 format (timestamp).

[0124] The following is a specific example:

[0125] In the multi-modal monitoring scenario of the grain conveying system, when the system detects that the energy value of the No. 5 beam splitting channel increases to 3.7 times the normal threshold, the collision positioning mechanism is triggered. The system first analyzes the beam splitting channel number CH_005, determines the corresponding channel according to the lateral mapping table, determines that the channel corresponds to the 1.2-1.5 meter area of the conveying belt from the left, generates the lateral position code "L1.3", and synchronously calls the spatial coordinate data of the No. 3 multi-axis laser speed probe (longitudinal reference point offset 8.6 meters). Combined with the conveying belt running speed 1.2 m / s and the event occurrence timestamp, the longitudinal position of the collision point is calculated as 10.3 meters from the starting point of the conveying belt, and the longitudinal code "Y10.3" is generated. The two codes are spliced to form a unique collision identifier "L1.3-Y10.3", and marked as a "mechanical component shedding" type warning based on the energy waveform characteristics. The system drives the inspection robot to arrive at the target area within 0.8 seconds, and the infrared scan confirms that there is a 17 cm long broken beating plate residue at this position, and the local high-pressure air blowing of the can-be-bad mechanism is started. This mechanism shortens the foreign matter removal response time from 25 minutes of manual processing to 92 seconds, avoids the scratching of the conveying belt and the cross-contamination of the subsequent 12 tons of wheat, and controls the processing range to 0.8 square meters by precise positioning, reducing the production capacity loss by 87% compared with the traditional full-segment shutdown inspection method.

[0126] In summary, the collision area precise positioning and dynamic warning method based on multi-dimensional coordinate coding is realized by steps 401 to 404. The collision area lateral positioning code is generated by mapping the beam splitting channel number and the lateral position, and the longitudinal positioning code is analyzed by combining the spatial coordinates of the multi-axis laser speed probe and the longitudinal position mapping model. The unique spatial identifier of the collision area is constructed by directly splicing the lateral and longitudinal codes. The warning signal with spatial coordinate directionality is generated by fusing the energy abnormal type label and the positioning identifier, forming a collaborative warning mechanism for collision area type identification and position labeling. This method constructs a dynamic correlation system of conveying belt lateral and longitudinal position coding, realizes cm-level spatial positioning and multi-dimensional state labeling of energy abnormal area, and the warning signal can synchronously feedback the collision type, energy level and precise coordinate information, significantly improving the dynamic perception ability of hidden collision risk and the targeted processing efficiency of equipment maintenance in the material conveying process.

[0127] In order to solve the technical problem of dynamic identification of collision area in the conveying process of grain, an intelligent collision detection system based on optical flow field analysis is developed, which realizes sub-second response of automatic identification of collision area and energy quantification evaluation. In some embodiments, based on the flow velocity distribution profile, the particle motion vector field is calculated by using the optical flow method in step 104, the collision area is identified by the sudden change of the velocity gradient in the particle motion vector field, and the energy value of the collision area is calculated according to the vector field energy distribution, including:

[0128] 501. Based on the velocity distribution profile at adjacent time points, two sets of continuous velocity data with a preset sampling period are obtained, and the particle distribution in the two sets of continuous velocity data is calculated using the optical flow method to form a particle motion vector field composed of multiple continuous regions.

[0129] In step 501, adjacent time points refer to the time interval between two sets of continuous data collected according to a preset sampling period; the velocity distribution profile refers to image data representing the velocity distribution of particles at different positions on the conveyor belt cross section in the form of a two-dimensional heat map; the optical flow method refers to a motion field estimation algorithm based on the constraint relationship between the spatial gradient of the velocity field and the time derivative; the particle motion vector field refers to a two-dimensional vector field composed of velocity vector data from multiple continuous regions.

[0130] In this embodiment, two sets of velocity distribution profiles at adjacent time points are first acquired with a preset sampling period of 100 milliseconds. The motion field of the two sets of image data is calculated using the optical flow method: an energy function containing velocity field intensity, velocity vector components and smoothing coefficient is constructed, and the velocity vector field is solved using the Gauss-Newton iteration method. A particle motion vector field composed of 128 by 64 grid units is generated, with each grid unit corresponding to a 1 cm by 1 cm area on the conveyor belt surface. The magnitude and direction data of the velocity vector are stored.

[0131] 502. Traverse each region in the particle motion vector field, calculate the velocity gradient between the current region and its directly adjacent regions, extract the maximum value among all the velocity gradient differences in the current region as the mutation feature value, and mark it as a collision region if the mutation feature value exceeds a preset threshold.

[0132] In step 502, the velocity gradient refers to the spatial rate of change of the velocity vector of adjacent regions; the abrupt change characteristic value refers to the maximum velocity gradient difference between the current region and its directly adjacent regions; the collision region refers to the set of continuous grid cells where the velocity gradient abrupt change exceeds a preset threshold.

[0133] In this embodiment, each grid cell of the particle motion vector field is first traversed, and the horizontal and vertical velocity gradient components between the current cell and its 8 neighboring cells are calculated using the central difference method. The maximum absolute value of the gradient in each direction is extracted as the mutation feature value. When the mutation feature value exceeds the threshold of 5 meters per second per meter, the current cell is marked as a candidate collision cell. Next, adjacent candidate cells are merged using a region growing algorithm to form a continuous collision region with a minimum area of ​​4 square centimeters.

[0134] 503. Based on the energy distribution of the particle motion vector field, extract the motion vectors corresponding to all regions within the collision region, and accumulate the energy based on the square of the magnitude of the motion vectors to obtain the energy value of the collision region.

[0135] In step 503, the energy distribution refers to the kinetic energy density distribution of the particle motion vector; the square of the module length of the motion vector refers to the sum of the square of the horizontal component and the square of the vertical component of the velocity vector; and the energy accumulation refers to the integral sum of the energy density of all grid cells in the collision region.

[0136] In the embodiment of the present application, first, the velocity vector data of each grid cell in the collision region is extracted, and the square of the module length thereof is calculated (for example, when the horizontal component of the velocity vector is 1.2 meters per second and the vertical component is 0.8 meters per second, the square of the module length is 1.2 square plus 0.8 square, which equals 2.08 square meters per second square). Second, the square of the module length of all cells in the region is accumulated to obtain the total energy value of the collision region (for example, if the region contains 10 cells, the total energy value is 2.08 multiplied by 10, which equals 20.8 joules per cubic meter). Finally, the energy value is compared with the preset threshold value to trigger the corresponding level of warning signal.

[0137] The following is a specific example:

[0138] Under the high-speed running condition of the grain storage and conveying system, the multi-modal monitoring system starts the collision detection mechanism when the particle flow state is abnormal. When the conveying belt transports a mixture of sorghum and corn at a speed of 2.4 m / s, the system acquires continuous flow profile data with a sampling period of 500 milliseconds. Through the improved pyramid optical flow algorithm, the particle distribution of adjacent time windows is tracked at the sub-pixel level, and a motion field composed of 128 vector units is generated at monitoring section No. 6, in which the 23rd vector unit detects a direction mutation (the motion angle is deflected from 15° to 82°). Using the eight-neighborhood gradient analysis method to traverse the motion field, a speed gradient mutation feature value of 1.8 m / s² is found in the area 2.1 meters to the right of the conveying belt, which exceeds the preset threshold value of 0.6 m / s² by 300%. The system extracts the square of the module length of 12 associated vector units in the collision region to perform energy integration, and measures an energy peak value of 4780 au (the reference energy level is 1200±300 au), triggering a level 3 warning. The three-dimensional point cloud scanning device is linked to the target area for millisecond-level review, accurately locating the 38 cm long metal foreign object embedding point, and driving the beatable mechanism to perform a 23 kPa directional air flushing removal. The measured data shows that this mechanism realizes a 9.8 second closed-loop response from the occurrence to the disposal of the collision event under the full-speed running state of the conveying belt, which is 46 times more efficient than traditional manual inspection, successfully avoids a 150,000 yuan screen tearing accident, and suppresses the false positive rate to below 0.3% through energy threshold control.

[0139] In summary, the particle collision dynamic detection and energy quantification method based on spatiotemporal correlation vector field analysis is realized through steps 501 to 503. This method breaks through the limitations of traditional static energy threshold detection, realizes dynamic capture and energy evolution tracking of transient collision events in the conveying process, improves the robustness of collision detection through the double verification mechanism of vector field gradient mutation and energy field spatial superposition, and provides a multi-dimensional dynamic criterion with spatial positioning accuracy and energy intensity calibration for collision risk warning of the material conveying system.

[0140] To solve the technical problems of collision area boundary ambiguity and scattered detection in the conveying process of distiller's grains, an intelligent collision recognition system based on gradient mutation clustering is developed to realize high-precision collision area boundary extraction and region fusion. In some embodiments, in step 502, each region in the particle motion vector field is traversed, the velocity gradient between the current region and its directly adjacent regions is calculated, the maximum value of all the velocity gradient differences of the current region is extracted as the mutation feature value, and if the mutation feature value exceeds the preset threshold, the current region and its directly adjacent regions are marked as an initial collision region, including:

[0141] 601. For each region in the particle motion vector field, based on the velocity gradient of each region and all directly adjacent regions, the difference absolute value of the velocity gradient is calculated, and the maximum value of the difference absolute value is selected as the mutation feature value of the current region;

[0142] In step 601, the velocity gradient difference absolute value refers to the absolute value of the difference between the velocity gradient of the current region and each of its directly adjacent regions; the mutation feature value refers to the maximum value of all directly adjacent region velocity gradient difference absolute values.

[0143] In the embodiments of the present application, first, each region (such as a 1 cm by 1 cm grid cell) of the particle motion vector field is traversed, and the following operations are performed on each region: the central difference method is used to calculate the velocity gradient component difference between the current region and eight adjacent regions (including upper, lower, left, right, and diagonal adjacent regions). Specifically, the horizontal gradient difference is equal to the absolute value of the difference between the horizontal gradient of the current region and the horizontal gradient of the adjacent region, and the vertical gradient difference is equal to the absolute value of the difference between the vertical gradient of the current region and the vertical gradient of the adjacent region. The maximum value of all adjacent gradient differences is extracted as the mutation feature value of the current region. For example, the horizontal gradient of the current region is 5 meters per second per meter, and the horizontal gradient of the adjacent region is 8 meters per second per meter. The horizontal gradient difference is 3 meters per second per meter, and if this value is the maximum value of all adjacent differences, the mutation feature value is recorded as 3 meters per second per meter.

[0144] 602. Compare the mutation feature value with the preset threshold value, and if it exceeds the preset threshold value, mark the current region and its directly adjacent regions as an initial collision region;

[0145] In step 602, the preset threshold refers to a critical value for determining the speed gradient mutation; and the initial collision region refers to a set of the region marked when the mutation characteristic value exceeds the threshold and its adjacent regions.

[0146] In the embodiment of the present application, first, the mutation characteristic value is compared with the preset threshold (e.g., 5 meters per second per meter): when the mutation characteristic value exceeds the threshold, the current region and its eight directly adjacent regions are collectively marked as the initial collision region. For example, if the mutation characteristic value of a region is 6 meters per second per meter, the region and its surrounding neighborhood are marked as the initial collision region. Then, the coordinate positions of all the initial collision regions are recorded and stored in a temporary buffer.

[0147] 603、When the boundary spacing of adjacent regions is less than the preset collision action radius, the adjacent regions in the initial collision region that are connected to each other are merged to form a continuous and closed final collision region.

[0148] In step 603, the collision action radius refers to the maximum spacing between adjacent regions allowed to be merged; and the final collision region refers to a closed and continuous region formed by merging the connected initial regions.

[0149] In the embodiment of the present application, first, the collision action radius is set to 2 centimeters (i.e., the spacing of two grid cells). Then, all the initial collision regions are traversed: if the boundary spacing of two regions is less than or equal to 2 centimeters, the two regions are determined to be connected regions. The region growing algorithm is used for merging: each initial region is taken as a seed point, and the adjacent regions that satisfy the spacing condition are expanded outward until no further expansion is possible. Finally, a continuous and closed final collision region is generated, for example, three initial regions with a spacing of 1.5 centimeters are merged into a final collision region with an area of 6 square centimeters.

[0150] The following is a specific example:

[0151] In the multi-modal monitoring scenario of the grain conveying system, when the conveying belt transports high-impurity rough rice at a speed of 1.6 m / s, the system triggers the collision positioning mechanism based on the speed field anomaly detection. The multi-axis laser speed probe generates a particle motion vector field composed of 256 hexagonal cells at monitoring section No. 7, where the 89th cell detects a sudden change in the velocity gradient, with the lateral velocity difference from adjacent cells increasing from the normal value of 0.15 m / s to 1.2 m / s, and the longitudinal velocity difference reaching 0.8 m / s. The system performs eight-neighbor gradient analysis to calculate the absolute value of the velocity gradient difference between the cell and its eight adjacent cells, and measures the maximum mutation feature value as 2.3 m / s², which is 360% higher than the preset threshold of 0.5 m / s². After triggering the warning, the 89th cell and its adjacent five cells are marked as the initial collision area (covering a diameter of 0.45 m). After the collision action radius verification (preset value of 0.3 m), it is found that the boundary distance of the left adjacent 87th cell is only 0.18 m, so the region merging is performed to form a continuous closed region, finally covering a trapezoidal region of 1.8-2.3 meters right and Y12.7-Y13.5 in the longitudinal coordinate. The system links the X-ray foreign matter detection module to perform millisecond-level scanning on the region, confirms the existence of a 22 cm long metal screen fragment embedded in the material flow, and drives the beatable rough mechanism to perform 18 kPa directional high-pressure air flushing to remove it. The actual operation data shows that from the speed gradient anomaly detection to the completion of foreign matter removal, only 6.7 seconds are consumed, which is 53 times higher than the efficiency of traditional manual disposal, and avoids the risk of scratching the 7.5-meter-long conveying belt and the subsequent pollution loss of 8 tons of rough rice. Through the region merging algorithm, the system reduces the false positive rate of discrete detection points from 12% to 0.7%, and the positioning accuracy is controlled within ±5 cm.

[0152] In summary, the collision area dynamic identification and spatial aggregation method based on gradient mutation propagation is realized through steps 601 to 603, the local mutation feature value is extracted by calculating the absolute value of the velocity gradient difference between regions, and the dynamic threshold comparison mechanism is used to mark the initial collision area and inclusively include the adjacent associated area; the spatial proximity of the initial collision area is analyzed by combining the collision action radius criterion, and a closed continuous collision scope is generated by region merging. This method builds a dual detection mechanism of velocity gradient mutation propagation model and spatial continuity verification, breaks through the spatial fragmentation limitation of traditional fixed threshold detection on the collision range, realizes the integrity capture and boundary accurate fitting of the collision action area in the dynamic conveying scene, and significantly reduces the region misjudgment rate caused by complex fluctuations of particle flow state, providing spatial coherent and morphologically complete abnormal area detection capability for collision damage warning of conveying equipment.

[0153] In order to solve the technical problem of difficult accurate division of the layered monitoring area in the process of transporting distilled sorghum, an intelligent layered control system based on diamond grid segmentation is developed, which realizes the layered monitoring of the flow field of the conveying belt with millimeter-level precision. In some embodiments, the laser Doppler frequency shift signal of the surface layer particles and the echo intensity data of the deep layer particles are synchronously collected in step 102, and the layered control body of the surface layer area and the deep layer area of the conveying belt is defined through the spatial distribution parameters of the multi-axis laser speed measurement probe, including:

[0154] 701. Arranging multi-axis laser speed measurement probes at a predetermined interval on both sides of the conveying belt, each multi-axis laser speed measurement probe comprising a surface layer detection axis and a deep layer detection axis, and generating spatial distribution parameters of the multi-axis laser speed measurement probe;

[0155] In step 701, the multi-axis laser speed measurement probe refers to a composite sensor array integrating a surface layer detection axis and a deep layer detection axis; the surface layer detection axis refers to an optical channel emitting short-wavelength laser (such as 650 nanometer red light) for surface layer particle speed measurement; the deep layer detection axis refers to an optical channel emitting long-wavelength laser (such as 1550 nanometer infrared light) for deep layer particle detection; the spatial distribution parameters refer to a group of geometric parameters composed of installation interval, inclination angle and three-dimensional coordinates of the probe array.

[0156] In the embodiments of the present application, first, the multi-axis laser speed measurement probes are symmetrically deployed at an interval of 0.5 meters on both sides of the conveying belt, 8 groups of probes are deployed on each side. Each group of probes comprises two independent optical axes: the surface layer detection axis is installed at an inclination angle of 15 degrees, and the deep layer detection axis is installed at an inclination angle of 30 degrees. Secondly, the spatial parameters of each group of probes (such as the coordinates of the third group of probes X=1.5 meters, Y=0 meters, Z=0.3 meters) are recorded by a three-dimensional coordinate measuring instrument to generate a spatial distribution parameter table.

[0157] 702. Receiving the laser Doppler frequency shift signal reflected by the surface layer particles through the surface layer detection axis, and receiving the echo intensity data reflected by the deep layer particles through the deep layer detection axis;

[0158] In step 702, the laser Doppler frequency shift signal refers to the frequency shift amount of the laser caused by the movement of the surface layer particles (1 m / s speed corresponds to 6.7 MHz frequency shift); the echo intensity data refers to the energy attenuation value of the laser reflected by the deep layer particles (such as energy attenuation to 12% of the initial value after penetrating 5 cm of grain layer).

[0159] In the embodiments of the present application, first, the frequency shift signal is collected by the surface layer detection axis at a sampling rate of 100 kHz, for example, a frequency shift of 12.3 MHz corresponds to a surface layer particle speed of 1.84 m / s. At the same time, the deep layer detection axis collects the echo intensity at a sampling rate of 1 kHz, for example, the receiving end detects an echo voltage value of 18 mV. After all data are time-stamped synchronously, they are stored in a ring buffer.

[0160] 703、According to the spatial distribution parameters, the surface penetration depth and the deep penetration depth corresponding to each multi-axis laser speed measurement probe are calculated;

[0161] In step 703, the surface penetration depth refers to the effective detection depth of short-wavelength laser in the grain layer, and the deep penetration depth refers to the effective detection depth of long-wavelength laser in the grain layer, both of which are calculated based on the Beer-Lambert law.

[0162] In the embodiments of the present application, for the surface penetration depth, the calculation formula is the natural logarithm (initial emission intensity divided by received intensity) divided by the surface extinction coefficient, for example, when the emission intensity is 50 milliwatts, the received intensity is 45 milliwatts, and the extinction coefficient is 0.03 per centimeter, the surface penetration depth is the natural logarithm (50 / 45) divided by 0.03, and the calculation result is about 3.2 centimeters; for the deep penetration depth, the same formula is used but the extinction coefficient is adjusted to 0.07 per centimeter, for example, when the received intensity is 8 milliwatts, the calculated depth is about 12.1 centimeters.

[0163] 704、Based on the horizontal spacing of adjacent multi-axis laser speed measurement probes and the difference between the corresponding surface penetration depth and deep penetration depth, a spatial segmentation frame composed of a rhombus grid is constructed on the cross section of the conveyor belt;

[0164] In step 704, the horizontal spacing refers to the installation interval of adjacent probes in the length direction of the conveyor belt, and the rhombus grid refers to a quadrilateral space unit composed of the penetration depth difference and the horizontal spacing of adjacent probes.

[0165] In the embodiments of the present application, taking two adjacent groups of probes as an example, the horizontal spacing is 0.5 meters, the surface penetration depth difference is 0.3 centimeters, and the deep penetration depth difference is 0.3 centimeters. The rhombus grid is constructed through geometric modeling, the horizontal side length of the grid is 0.5 meters, the vertical side length is obtained by subtracting the surface penetration depth from the deep penetration depth (for example, 12.1 centimeters minus 3.2 centimeters is 8.9 centimeters), and the included angle of the diagonal line is calculated by the inverse tangent function (2 times the penetration depth difference divided by the horizontal spacing). Finally, a rhombus grid unit with a size of 0.5 meters by 0.089 meters and an included angle of the diagonal line of 0.69 degrees is generated.

[0166] 705、Mark the coverage range of all surface penetration depths in the spatial segmentation frame as a surface control body, and mark the coverage range of all deep penetration depths in the spatial segmentation frame as a deep control body, both of which define the layered control body of the conveyor belt.

[0167] In step 705, the surface control body refers to the three-dimensional region covered by short-wavelength laser in the rhombus grid, and the deep control body refers to the three-dimensional region covered by long-wavelength laser in the rhombus grid.

[0168] In the embodiments of the present application, the rhombic grid is layered along the vertical direction: the surface layer control body covers a depth range of 0 to 3.2 cm, corresponding to the short-wavelength laser detection area; the deep layer control body covers a depth range of 3.2 to 12.1 cm, corresponding to the long-wavelength laser detection area. A three-dimensional modeling software is used to generate the vector boundary of the layered control body, for example, the volume of the surface layer control body corresponding to a single rhombic grid is 0.5 meters by 0.5 meters by 0.032 meters, and the volume of the deep layer control body is 0.5 meters by 0.5 meters by 0.089 meters.

[0169] The following is a specific example:

[0170] In the rough grain storage conveying system, multi-axis laser speed measurement probes (including surface layer 45° detection axis and deep layer 60° detection axis) are arranged every 0.6 meters along the two sides of the conveying belt to generate spatial distribution parameters. By calculating the penetration depth of the probes (0-15 cm for the surface layer and 15-40 cm for the deep layer), combined with the spacing, a rhombic grid layered control body is constructed. In actual operation, the system detects that the deep layer flow rate of a certain grid is abnormal (sudden drop of 18 dB), and the surface layer flow rate suddenly increases at the same time, and is located to the right 1.2 meter area. After the infrared thermal imager confirms that the metal friction is overheated, it triggers directional high-frequency vibration for 5.3 seconds to remove the foreign matter and restore the normal flow rate. This scheme improves the foreign matter detection rate to 96% and reduces the maintenance range by 62%.

[0171] In summary, the conveying belt layered control body dynamic construction method based on multi-axis laser detection and rhombic spatial segmentation is realized through steps 701 to 705. By arranging multi-axis laser speed measurement probe arrays integrated with surface layer detection axes and deep layer detection axes at a predetermined interval on both sides of the conveying belt, spatial distribution parameters of the probes are generated, and laser Doppler frequency shift signals of surface layer particles and deep layer particle echo intensity data are synchronously collected. Based on the installation angle of the probe and the laser wavelength parameters, the surface layer penetration depth and the deep layer penetration depth are calculated, and the rhombic grid spatial segmentation frame on the cross section of the conveying belt is constructed by combining the horizontal spacing of adjacent probes with the penetration depth difference. The surface layer penetration depth coverage range in the rhombic grid is defined as the surface layer control body, and the deep layer penetration depth coverage range is marked as the deep layer control body, forming the spatial constraint boundary of layered material flow state monitoring, realizing independent collection and dynamic spatial mapping of surface layer and deep layer particle motion data, providing a geometric self-adaptive control body division mechanism for layered flow rate analysis, effectively improving the spatial accuracy and data reliability of particle flow state layered monitoring, and supporting real-time adjustment of the layered control body range as the material accumulation thickness changes.

[0172] Figure 2 A structure diagram of a rough grain conveying state monitoring system based on multi-modal sensing is provided for the embodiments of the present application, as shown in Figure 2 The system includes:

[0173] The configuration module 21 is configured to set a multi-axis laser speed measurement probe on both sides of the conveying belt and set a dustproof cover around the multi-axis laser speed measurement probe to eliminate the interference of dust on the light path reflection path during the conveying process.

[0174] The acquisition module 22 is configured to synchronously acquire the laser Doppler frequency shift signal of the surface layer particles and the echo intensity data of the deep layer particles, and define a layered control body of the surface layer region and the deep layer region of the conveying belt in combination with the spatial distribution parameters of the multi-axis laser speed measurement probe.

[0175] The processing module 23 is configured to compensate the scattering signal in the layered control body in combination with the echo intensity data, and perform autocorrelation processing on the laser Doppler frequency shift signal to generate a flow velocity distribution profile containing the velocity difference between the surface layer particles and the deep layer particles.

[0176] The calculation module 24 is configured to calculate a particle motion vector field by using an optical flow method based on the flow velocity distribution profile, identify a collision region through the mutation characteristics of the velocity gradient in the particle motion vector field, and calculate an energy value of the collision region according to the energy distribution of the vector field.

[0177] The generation module 25 is configured to generate a pre-warning signal associated with the collision region in combination with the corresponding beam splitting channel number of the layered control body and the spatial coordinates of the multi-axis laser speed measurement probe when the energy value exceeds a preset range. Figure 2 The monitoring system can perform the method Figure 1 The implementation principle and technical effects of the monitoring method are not described again. The specific operation modes of each module and unit of the monitoring system are described in the method embodiment, and will not be described in detail here.

[0178] In one possible design, Figure 2 The monitoring system can be implemented as a computing device, such as a computer Figure 3 The computing device can include a storage component 31 and a processing component 32.

[0179] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32.

[0180] The processing component 32 is configured to perform the above Figure 1 The monitoring method of the embodiment based on multi-modal sensing.

[0181] The processing component 32 can include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component can also be one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components, for executing the above method.

[0182] The storage component 31 is configured to store various types of data to support the operation of the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0183] Of course, the computing device can also include other components, such as an input / output interface, a display component, a communication component, etc.

[0184] The input / output interface provides an interface between the processing component and peripheral interface modules, which can be output devices, input devices, etc.

[0185] The communication component is configured to facilitate wired or wireless communication between the computing device and other devices, etc.

[0186] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform, and the computing device can be a cloud server, and the processing component, the storage component, etc. can be a basic server resource rented or purchased from the cloud computing platform.

[0187] The embodiment of the present application also provides a computer storage medium, which stores a computer program, and the computer program can implement the above Figure 1 A food conveying state monitoring method based on multi-modal sensing is provided.

[0188] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0189] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0190] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and the necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of software products, and the computer software products can be stored in a computer readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and include a plurality of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0191] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A multi-modal sensor based monitoring method of a food conveying condition, characterized in that, The application relates to a multi-axis laser speed measurement probe and a dustproof cover arranged outside the multi-axis laser speed measurement probe to eliminate the interference of dust on the reflection path of the light path during conveying. Synchronously collecting laser Doppler frequency shift signals of surface layer particles and echo intensity data of deep layer particles, and defining layered control bodies of surface layer regions and deep layer regions of the conveying belt through spatial distribution parameters of the multi-axis laser speed measurement probe; Compensating scattering signals in the layered control bodies in combination with the echo intensity data, and performing autocorrelation processing on the laser Doppler frequency shift signals to generate a flow velocity distribution profile containing velocity differences of the surface layer particles and the deep layer particles; Based on the flow velocity distribution profile, calculating a particle motion vector field by using an optical flow method, identifying a collision region through a sudden change characteristic of a velocity gradient in the particle motion vector field, and calculating an energy value of the collision region according to energy distribution of the vector field; When the energy value exceeds a preset range, generating a pre-warning signal associated with the collision region in combination with a corresponding beam splitting channel number of the layered control bodies and spatial coordinates of the multi-axis laser speed measurement probe. Compensating scattering signals in the layered control bodies in combination with the echo intensity data, and performing autocorrelation processing on the laser Doppler frequency shift signals to generate a flow velocity distribution profile containing velocity differences of the surface layer particles and the deep layer particles, including:

2. The method of claim 1, wherein, According to the spatial distribution parameters of the multi-axis laser speed measurement probe, extracting echo intensity data corresponding to each beam splitting channel in the layered control bodies, determining an attenuation ratio of the scattering signals in the layered control bodies through multi-frequency band reflection intensity differences, and dynamically compensating the scattering signals in the same beam splitting channel based on the attenuation ratio; Matching the compensated scattering signals and the laser Doppler frequency shift signals according to the beam splitting channels, extracting a first frequency shift component corresponding to the surface layer particles and a second frequency shift component corresponding to the deep layer particles in the laser Doppler frequency shift signals; Through autocorrelation processing and a phase sliding window, time domain overlapping interception is performed on the first frequency shift component and the second frequency shift component to generate a first signal segment associated with the surface layer particles and a second signal segment associated with the deep layer particles; According to a mapping relationship between the accumulated phase difference and particle displacement, a first velocity component of the surface layer particles in the first signal segment and a second velocity component of the deep layer particles in the second signal segment are extracted; According to the spatial distribution parameters of the multi-axis laser speed measurement probe, velocity gradient superposition is performed on the first velocity component and the second velocity component in the same beam splitting channel in the depth direction of the layered control bodies to generate a layered flow velocity distribution profile containing velocity differences of the surface layer particles and the deep layer particles. According to a mapping relationship between the accumulated phase difference and particle displacement, a first velocity component of the surface layer particles in the first signal segment and a second velocity component of the deep layer particles in the second signal segment are extracted, including:

3. The method of claim 2, wherein, ​ The first signal segment and the second signal segment are respectively divided into a plurality of equal-length time domain windows, adjacent time domain windows in each signal segment are partially overlapped, and the length of the overlapping area between adjacent time domain windows is less than the total length of a single time domain window; In the overlapping area of the adjacent time domain windows, the phase difference value between the end phase value of the previous time domain window and the start phase value of the next time domain window is calculated, and the phase difference values of all adjacent windows are accumulated along the time axis direction of the signal segment to obtain a cumulative phase difference value representing the movement trend of the particles; According to the mapping relationship between the cumulative phase difference and the particle displacement, the cumulative phase difference value is converted into the displacement amount of the particles in the time interval corresponding to the adjacent time domain windows based on the wavelength parameter of the laser in the hierarchical control body, and the first speed component of the surface layer particles in the first signal segment and the second speed component of the deep layer particles in the second signal segment are extracted according to the ratio of the displacement amount to the time interval.

4. The method of claim 1, wherein, When the energy value exceeds the preset range, in combination with the corresponding beam channel number of the hierarchical control body and the spatial coordinates of the multi-axis laser velocity measurement probe, a warning signal associated with the collision area is generated, including: When the energy value exceeds the preset range, the beam channel number of the hierarchical control body is extracted, and the position code of the collision area in the transverse direction of the conveyor belt is determined according to the mapping relationship between the beam channel number and the preset transverse position; The spatial coordinates of the multi-axis laser velocity measurement probe are extracted, and the position code of the collision area in the longitudinal direction of the conveyor belt is determined according to the mapping relationship between the spatial coordinates and the preset longitudinal position; The position code in the transverse direction and the position code in the longitudinal direction are directly spliced to form a unique positioning identifier of the collision area; Based on the type mark of the energy value exceeding the preset range and the unique positioning identifier, a warning signal associated with the collision area is generated.

5. The method of claim 1, wherein, Based on the flow velocity distribution profile, the optical flow method is used to calculate the particle motion vector field, the collision area is identified by the mutation characteristics of the velocity gradient in the particle motion vector field, and the energy value of the collision area is calculated according to the energy distribution of the vector field, including: Based on the flow velocity distribution profile of adjacent time points, two groups of continuous flow velocity data with a preset sampling period as the time interval are obtained, the particle distribution in the two groups of continuous flow velocity data is calculated using the optical flow method, and a particle motion vector field composed of a plurality of continuous regions is formed; Each region in the particle motion vector field is traversed, the velocity gradient between the current region and its directly adjacent region is calculated, the maximum value of all the velocity gradient differences of the current region is extracted as a mutation characteristic value, and if the mutation characteristic value exceeds a preset threshold, it is marked as a collision area; According to the energy distribution of the particle motion vector field, the motion vectors corresponding to all regions in the collision area are extracted, the energy is accumulated based on the square of the module length of the motion vectors, and the energy value of the collision area is obtained.

6. The method of claim 5, wherein, The velocity gradient between the current region and its directly adjacent regions is calculated by traversing each region in the particle motion vector field, the maximum value of all the velocity gradient differences of the current region is extracted as a mutation feature value, and if the mutation feature value exceeds a preset threshold, the current region is marked as a collision region, comprising: For each region in the particle motion vector field, based on the velocity gradient of each region and all directly adjacent regions, the absolute value of the difference of the velocity gradient is calculated, and the maximum value of the absolute value of the difference is selected as the mutation feature value of the current region; The mutation feature value is compared with a preset threshold, and if it exceeds the preset threshold, the current region and its directly adjacent regions are collectively marked as an initial collision region; When the boundary distance between adjacent regions is less than a preset collision action radius, the adjacent regions in the initial collision region that are connected to each other are merged to form a continuous and closed final collision region.

7. The method of claim 1, wherein, Synchronously collect the laser Doppler frequency shift signal of the surface layer particles and the echo intensity data of the deep layer particles, and define the layered control body of the surface layer and deep layer regions of the conveying belt through the spatial distribution parameters of the multi-axis laser velocity measurement probe, comprising: Arranging multi-axis laser velocity measurement probes at a preset interval on both sides of the conveying belt, each multi-axis laser velocity measurement probe including a surface layer detection axis and a deep layer detection axis, and generating the spatial distribution parameters of the multi-axis laser velocity measurement probe; Receiving the laser Doppler frequency shift signal reflected by the surface layer particles through the surface layer detection axis, and receiving the echo intensity data reflected by the deep layer particles through the deep layer detection axis; According to the spatial distribution parameters, calculating the surface layer penetration depth and deep layer penetration depth corresponding to each multi-axis laser velocity measurement probe; Based on the horizontal distance between adjacent multi-axis laser velocity measurement probes and the difference between the corresponding surface layer penetration depth and deep layer penetration depth, constructing a spatial segmentation frame composed of a rhombus grid on the cross section of the conveying belt; Marking the coverage range of all surface layer penetration depths in the spatial segmentation frame as a surface layer control body, and marking the coverage range of all deep layer penetration depths in the spatial segmentation frame as a deep layer control body, which together define the layered control body of the conveying belt.

8. A multi-modal sensor based spoilage food conveying condition monitoring system, characterized in that, Comprising: The configuration module is arranged on both sides of the conveying belt, and a dust cover is arranged around the multi-axis laser velocity measurement probe to eliminate the interference of dust on the light path reflection path during the conveying process; The acquisition module synchronously collects the laser Doppler frequency shift signal of the surface layer particles and the echo intensity data of the deep layer particles, and defines the layered control body of the surface layer and deep layer regions of the conveying belt through the spatial distribution parameters of the multi-axis laser velocity measurement probe; The processing module combines the echo intensity data to compensate the scattering signal in the layered control body, and performs autocorrelation processing on the laser Doppler frequency shift signal to generate a flow velocity distribution profile containing the velocity difference between the surface layer particles and the deep layer particles; The calculation module calculates the particle motion vector field using the optical flow method based on the flow velocity distribution profile, identifies the collision region through the mutation feature of the velocity gradient in the particle motion vector field, and calculates the energy value of the collision region according to the energy distribution of the vector field. The generating module generates a pre-warning signal associated with the collision area in combination with the corresponding beam channel number of the hierarchical control body and the spatial coordinates of the multi-axis laser velocity measurement probe when the energy value exceeds the preset range.

9. A computing device, comprising: The method comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to realize the method for monitoring the conveying state of the bad grain based on the multi-modal sensing according to any one of claims 1-7.

10. A computer storage medium, characterized in that, The computer program is stored and executed by a computer to realize the method for monitoring the conveying state of the bad grain based on the multi-modal sensing according to any one of claims 1-7.

Images