Unattended crucible loading and unloading operation control system and method based on visual processing

Abstract

The invention relates to the technical field of operation control, in particular to an unattended operation crucible loading and unloading operation control system and method based on visual processing, and the method comprises the following steps: monitoring a static accumulation value in a graphite powder fluidization conveying process in real time through a distributed static sensor; when the static accumulation value exceeds a safety threshold value, pulse type ion wind neutralization treatment is triggered, and a static balance material flow is generated; the dynamic alignment deviation of the stock bin and the crucible opening is calculated; the stepping angle and the rotating speed of a spiral auger in the stock bin are controlled based on the dynamic alignment deviation, and an axial vibration field is synchronously applied in the discharging process; the position of an air gap in a material is predicted based on a convolutional neural network, and a vacuum negative pressure value is dynamically adjusted to form a gradient suction mode. According to the device, the risks of powder agglomeration, conveying pipeline blockage or electrostatic discharge caused by electrostatic accumulation are effectively prevented, the safety of the conveying process is ensured, and meanwhile, the flowing stability of graphite powder is optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of operation control, and particularly to an unattended crucible loading and unloading operation control system and method based on vision processing. Background Art

[0002] As a high-purity powder material, graphite powder is widely used in industries such as metallurgy and electronics. Due to its small particle size and large specific surface area, problems such as static electricity accumulation, dust suspension, and material agglomeration are likely to occur during transportation and loading, affecting the transportation stability and loading uniformity. The traditional graphite powder transportation and loading methods mainly rely on negative pressure pipeline transportation and mechanical vibration-assisted loading, but there are many deficiencies in static electricity control, precise alignment, and degassing optimization.

[0003] In terms of static electricity control, the existing technologies usually adopt single-point static electricity monitoring and fixed-frequency neutralization methods, which are difficult to perceive the dynamic changes of static electricity accumulation in the transportation process in real time, resulting in a relatively high risk of static electricity discharge. In addition, the static electricity neutralization method lacks the ability of adaptive adjustment, which may cause over-neutralization or uneven neutralization, affecting the fluidity of the powder.

[0004] In terms of loading alignment, the existing methods mostly rely on laser ranging or mechanical guidance for crucible positioning, which are difficult to adapt to high-dust environments, resulting in limited alignment accuracy. At the same time, during the unloading process, the material flow state is complex, and the traditional mechanical vibration or fixed stepping control methods are difficult to adjust in real time, easily causing material accumulation or uneven filling, affecting the consistency of loading density.

[0005] In terms of degassing control, the traditional vacuum suction method usually uses a fixed negative pressure value or a single time setting for degassing, without considering the dynamic evolution characteristics of the air gap during the unloading process, resulting in low suction efficiency and some air gap residues affecting the final loading quality. In addition, the existing system lacks a real-time closed-loop feedback mechanism and cannot adaptively adjust the negative pressure distribution and suction strategy according to the material state, resulting in poor adaptability to different batches of materials. Summary of the Invention

[0006] The present invention provides an unattended crucible loading and unloading operation control system and method based on vision processing.

[0007] The unattended crucible loading and unloading operation control method based on vision processing includes the following steps:

[0008] S1. When feeding the negative pressure pipeline to the silo, the static electricity accumulation value during the fluidized transportation of graphite powder is monitored in real time through a distributed static electricity sensor. When the static electricity accumulation value exceeds the safety threshold, pulse-type ion wind neutralization treatment is triggered to generate a static electricity balanced material flow;

[0009] S2. Adopt a polarization vision positioning method resistant to dust interference to capture the real-time pose of the crucible, generate positioning data through a multi-frame motion blur restoration algorithm, and calculate the dynamic alignment deviation between the silo and the crucible mouth;

[0010] S3. Based on the dynamic alignment deviation, control the stepping angle and rotation speed of the spiral auger in the silo, and synchronously apply an axial vibration field during the unloading process. A dynamic mapping relationship is established between the vibration frequency of the axial vibration field and the angle of repose of the graphite powder;

[0011] S4. In the degassing and feeding stage, monitor the temperature gradient distribution of the graphite powder accumulation through infrared thermal imaging, predict the position of air gaps inside the material based on a convolutional neural network, and dynamically adjust the vacuum negative pressure value to form a gradient suction mode.

[0012] Optionally, S1 includes arranging multiple groups (8 - 12 groups) of electrostatic induction rings in an annular array on the inner side of the wall of the fluidized conveying section of the negative pressure pipeline. The distance between adjacent electrostatic induction rings is 1 times the pipe diameter, and the electrostatic potential distribution data of the graphite powder flow cross-section is collected in real time.

[0013] Optionally, the calculation of the electrostatic accumulation value is performed using a sliding window algorithm, and the calculation is as follows:

[0014] where, E accumulate is the electrostatic accumulation value, ΔE i is the electrostatic potential difference between adjacent induction rings, ΔE i = E i+1 - E i , t i represents the time for the material to pass between the electrostatic induction rings i and i + 1, T window is the sliding time window, with a value range of 20 - 40 ms, and l is the number of sampling points within the sliding window;

[0015] When E accumulate > 800 V / m (i.e., the safety threshold), trigger pulsed ionic wind neutralization treatment, activate the pulsed ionic wind nozzles arranged at the pipe bends, optimize the release strategy of the ionic wind, and generate an electrostatically balanced material flow with stable electric potential.

[0016] Optionally, S2 specifically includes:

[0017] S21. Polarization imaging unit arrangement: Arrange three groups of polarization imaging units below the silo. Each group includes linearly polarized filters distributed in a ring, and cooperate with an 850 nm near-infrared polarization light source to irradiate the moving crucible. The angles of the linearly polarized filters distributed in a ring are 0°, 60°, and 120°;

[0018] S22. Stokes vector calculation: Generate a polarization feature map free of dust interference based on the Stokes vector calculation, expressed as:

[0019] Among them, S is the polarization feature vector, and I θ is the image gray value corresponding to the polarization angles (0°, 60°, 120°);

[0020] S23, motion blur restoration: Input the polarization feature maps of multiple consecutive frames (5 frames can be selected) into the motion blur restoration neural network. The motion blur restoration neural network adopts an encoder-decoder structure and embeds temporal optical flow constraints to output a high-definition positioning image after displacement compensation;

[0021] S24, three-dimensional template matching and dynamic alignment deviation calculation: Solve the three-dimensional pose of the crucible mouth through the three-dimensional template matching method, and calculate the dynamic alignment deviation, expressed as:

[0022] ΔP = K·∥T actual -T target ∥+(1 - K)·arccos(q actual ·q target ), where

[0023] ΔP is the dynamic alignment deviation, T actual , T target are the translation vectors of the actual position of the current crucible mouth and the target crucible mouth respectively, q actual , q target are the rotation quaternion representations of the actual attitude of the current crucible mouth and the rotation quaternion of the target crucible mouth respectively, K is the weight coefficient, and its value range is 0.7 -

[0024] 0.9, which is used to balance the contributions of translation error and rotation error to the alignment deviation, and is adaptively adjusted according to the material loading amount. arccos(q actual ·q target ) represents the calculation of the quaternion rotation error.

[0025] Optionally, the S3 specifically includes:

[0026] S31, establish the mapping relationship between the dynamic alignment deviation ΔP and the control parameters of the screw auger, adjust the step angle and rotation speed of the screw auger. The step angle is non-linearly adjusted based on the reference angle to avoid overshoot or oscillation, and the rotation speed adopts an exponential decay method and is adaptively reduced according to the deviation magnitude;

[0027] S32, adaptive frequency control of the axial vibration field: Embed a piezoelectric actuator array at the axis of the screw auger to generate an axial vibration field, and the vibration frequency of the axial vibration field is adjusted according to the change of the repose angle of the graphite powder;

[0028] S33, Vibration - flow coupling control strategy: Construct a vibration - flow coupling control model. When the repose angle of graphite powder exceeds the set threshold, activate the high - frequency pulse vibration mode based on the vibration - flow coupling control model. The frequency of the high - frequency pulse adopts a stepped activation method, which is increased in stages according to the increment of the repose angle, and compensation is considered according to the changing trend of the rotational speed of the spiral auger.

[0029] Optionally, in S31, the mapping relationship between the dynamic alignment deviation ΔP and the control parameters of the spiral auger is expressed as:

[0030]

[0031] where θ is the step angle of the spiral auger, θ0 is the reference angle, k θ is the adjustment gain, with a value range of 5° - 8°, ΔP is the dynamic alignment deviation, ΔP max is the maximum allowable deviation, with a value of 5 mm, n is the real - time rotational speed of the spiral auger, n base is the reference rotational speed, with a value range of 30 - 50 rpm, γ is the rotational speed attenuation coefficient, with a value range of 0.2 - 0.3.

[0032] Optionally, the vibration frequency f of the axial vibration field in S32 satisfies: ·Δf, where f is the real - time frequency of the axial vibration field, f0 is the reference frequency, with a value of 80 Hz, α is the repose angle of graphite powder measured in real - time, α c is the critical repose angle, with a value of 35°, α max ,α min is the range of repose angle change, Δf is the adjustment bandwidth, with a value of 120 Hz, A is the vibration amplitude, with a value range of 0.1 - 0.3 mm, which is inversely proportional to the feeding speed.

[0033] Optionally, S4 specifically includes:

[0034] S41, Axial temperature gradient calculation: Arrange multiple groups of infrared thermal imaging sensors circumferentially at the discharge port of the silo to form an annular monitoring array, synchronously collect the temperature field data of the graphite powder accumulation body during the discharging process, and calculate the axial temperature gradient distribution: where, is the axial temperature gradient, T top ,T bottom are the average temperatures at 20 cm and 80 cm away from the discharge port respectively, H is the monitoring spacing, with a value of 60 cm, and t is the discharging duration;

[0035] S42, Three - dimensional air gap probability distribution calculation: Construct a dual - channel convolutional neural network (DC - CNN), and the input layer simultaneously receives:

[0036] Channel 1: Axial temperature gradient distribution matrix (20×20 grid data);

[0037] Channel 2: X-ray backscattering signal (density distribution map, material density information of the corresponding spatial region);

[0038] The network output is the air gap probability distribution map P(x, y, z). When P>0.7, mark this area as an effective air gap;

[0039] S43, Generation of gradient negative pressure curve: Dynamically generate the gradient negative pressure curve according to the air gap distribution:

[0040] where P vac (r) is the vacuum negative pressure at the radial coordinate r, P base is the reference negative pressure, r is the radial coordinate of the vacuum pipeline, r i is the center position of the i-th air gap, w i is the air gap influence weight, with a value range of 0.3 - 0.6, σ is the air gap diffusion influence range, with a value of 0.2R, where R is the pipeline radius, ensuring coverage of the air gap area;

[0041] S44, Implement a phased suction strategy according to the air gap evolution characteristics.

[0042] Optionally, the phased suction strategy includes:

[0043] When t < 30s, it is set as the initial stage: Apply the maximum negative pressure P max = 12 kPa in the air gap core area (select the largest connected domain as the core area based on the high-confidence area spatial distribution in the air gap probability distribution map P(x, y, z));

[0044] When 30s ≤ t < 90s, it is set as the middle stage: Distribute the negative pressure (8 - 12 kPa) along the gradient negative pressure curve,

[0045] and gradually adjust the suction intensity;

[0046] When t ≥ 90s, it is set as the later stage: Start pulsed suction, use a 5 Hz square wave modulation, and the negative pressure amplitude is 10 ± 2 kPa to improve the removal efficiency of micro air gaps.

[0047] The unattended crucible loading and unloading operation control system based on visual processing is used to implement the above-mentioned crucible loading and unloading operation control method, and includes the following modules:

[0048] The electrostatic monitoring and neutralization module is used to monitor the electrostatic accumulation value in the fluidized state of graphite powder in real time during the transportation in the negative pressure pipeline, and trigger pulsed ionic wind neutralization treatment when the electrostatic accumulation exceeds the limit to form an electrostatically balanced material flow;

[0049] The polarization vision positioning and alignment deviation calculation module captures the real-time pose of the crucible based on the polarization vision positioning method, and calculates the dynamic alignment deviation between the silo and the crucible opening through a multi-frame motion blur restoration algorithm;

[0050] The spiral auger stepping control and vibration field regulation module is used to adjust the stepping angle and rotation speed of the spiral auger based on the dynamic alignment deviation, and synchronously apply an axial vibration field during the discharging process. The vibration frequency is dynamically adjusted according to the real-time change of the angle of repose of the graphite powder to optimize the feeding uniformity;

[0051] The intelligent degassing control module is used to monitor the temperature gradient distribution of the graphite powder accumulation body through infrared thermal imaging during the degassing and feeding stage, predict the position of the air gap inside the material based on a convolutional neural network, dynamically adjust the vacuum negative pressure value, and form a gradient suction mode to improve the degassing efficiency and the consistency of the filling density.

[0052] The beneficial effects of the present invention:

[0053] In the present invention, by arranging multiple groups of electrostatic induction rings in the negative pressure pipeline, the electrostatic accumulation value of the graphite powder during the fluidized transportation process is monitored in real time, and the sliding window algorithm is used to calculate the electrostatic accumulation amount to ensure the accurate assessment of the electrostatic risk. When the electrostatic accumulation value exceeds the safety threshold, pulsed ionic wind neutralization is triggered, and through frequency-flow rate linkage adjustment and voltage segmented control, adaptive neutralization of different transportation states is achieved, effectively preventing powder agglomeration, transportation pipeline blockage or electrostatic discharge risks caused by electrostatic accumulation, ensuring the safety of the transportation process, and at the same time optimizing the flow stability of the graphite powder.

[0054] In the present invention, through multi-angle polarization filtering and Stokes vector calculation, the influence of dust scattering is effectively eliminated, the detection accuracy of the real-time pose of the crucible is improved, combined with the multi-frame motion blur restoration algorithm, it is ensured that clear positioning data can still be obtained during the high-speed discharging process, and the dynamic alignment deviation between the silo and the crucible opening is accurately calculated. Based on the alignment deviation, the stepping angle and rotation speed of the spiral auger are adaptively adjusted, and an axial vibration field is synchronously applied. The vibration frequency is dynamically adjusted according to the real-time change of the angle of repose of the graphite powder to ensure uniform material flow and prevent powder accumulation or segregation. Improve the alignment accuracy during the filling process, optimize the feeding stability, and improve the overall uniformity of the filling density.

[0055] In the present invention, an infrared thermal imaging sensor is used to collect temperature gradient data, and the material density distribution is obtained by combining with the X-ray backscattering signal. Through a dual-channel convolutional neural network, multi-physical quantity fusion calculation is carried out to generate a three-dimensional air gap probability distribution map, which can accurately identify air gaps of different scales and positions, improve the detection sensitivity to non-uniformly distributed air gaps. The DC-CNN adopts an attention mechanism to enhance the weight of air gap features, improving the recognition accuracy of effective air gaps, ensuring that the gradient negative pressure control can accurately act on the air gap area, thereby improving the overall degassing effect, reducing the air gap residue rate, and improving the consistency of the loading density.

[0056] In the present invention, according to the three-dimensional air gap probability distribution, a gradient negative pressure curve is dynamically generated to ensure that the negative pressure application range can match the air gap distribution pattern, making the suction effect more accurate. A phased suction strategy is introduced during the negative pressure control process. At the initial stage, a strong negative pressure is applied to large-scale air gaps. At the middle stage, it is dynamically adjusted according to the gradient negative pressure. At the later stage, pulsed suction is used to eliminate micro air gaps. This strategy can match different stages of the unloading process, ensuring that the material maintains a stable flow during degassing, preventing powder collapse or excessive disturbance from causing secondary inflation, thereby improving the unloading consistency and optimizing the uniformity of the loading density. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only those of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0059] Figure 1 It is a schematic flow chart of the operation control method for the embodiment of the present invention;

[0060] Figure 2 It is a schematic diagram of air gap prediction and suction mode adjustment for the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0061] The present invention will be described in detail below with reference to the drawings and specific embodiments. At the same time, it should be noted here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments. For some well-known technologies, those skilled in the art can also adopt other alternative methods for implementation; and the drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0062] It should be noted that in the specification, the mention of "an embodiment", "embodiments", "exemplary embodiments", "some embodiments", etc. indicates that the described embodiments may include specific features, structures or characteristics, but not necessarily every embodiment includes such specific features, structures or characteristics. Additionally, when combining embodiments to describe a specific feature, structure or characteristic, achieving such feature, structure or characteristic in combination with other embodiments (whether explicitly described or not) should be within the knowledge scope of those skilled in the relevant art.

[0063] Generally, terms can be understood at least in part from their use in context. For example, at least in part depending on the context, the term "one or more" used herein can be used to describe any feature, structure or characteristic in a singular sense, or can be used to describe a combination of features, structures or characteristics in a plural sense. Additionally, the term "based on" can be understood as not necessarily aiming to convey a set of exclusive factors, but instead, at least in part depending on the context, allowing for the existence of other factors that may not be explicitly described.

[0064] As Figure 1 - Figure 2 shown, the unattended crucible loading and unloading operation control method based on visual processing includes the following steps:

[0065] S1. When feeding materials to the silo through the negative pressure pipeline, the electrostatic accumulation value during the fluidized transportation process of graphite powder is monitored in real time by a distributed electrostatic sensor. When the electrostatic accumulation value exceeds the safety threshold, pulsed ionic wind neutralization treatment is triggered to generate an electrostatically balanced material flow;

[0066] S2. The real-time pose of the crucible is captured by a polarization vision positioning method resistant to dust interference, positioning data is generated through a multi-frame motion blur restoration algorithm, and the dynamic alignment deviation between the silo and the crucible mouth is calculated;

[0067] S3. Based on the dynamic alignment deviation, the stepping angle and rotational speed of the spiral auger in the silo are controlled, and an axial vibration field is applied synchronously during the unloading process. A dynamic mapping relationship is established between the vibration frequency of the axial vibration field and the repose angle of the graphite powder;

[0068] S4. In the degassing and discharging stage, the temperature gradient distribution of the graphite powder accumulation is monitored by infrared thermal imaging, the position of the air gap inside the material is predicted based on a convolutional neural network, and the vacuum negative pressure value is dynamically adjusted to form a gradient suction mode.

[0069] S1 includes arranging multiple groups (8 - 12 groups) of electrostatic induction rings in an annular array on the inner side of the pipe wall in the fluidized transportation section of the negative pressure pipeline, with the distance between adjacent electrostatic induction rings being 1 times the pipe diameter, and collecting the electrostatic potential distribution data of the graphite powder flow cross-section in real time.

[0070] The electrostatic accumulation value is calculated using a sliding window algorithm, and the calculation is as follows:

[0071] Among them, E accumulate is the static electricity accumulation value, and ΔE i is the static electricity potential difference between adjacent induction rings. ΔE i = E i+1 - E i , and t i represents the time for the material to pass between the static electricity induction rings i to i + 1. T window is the sliding time window, with a value range of 20 - 40 ms, and l is the number of sampling points within the sliding window;

[0072] When E accumulate > 800 V / m (i.e., the safety threshold), trigger the pulsed ionic wind neutralization treatment, activate the pulsed ionic wind nozzles arranged at the pipe bends, optimize the release strategy of the ionic wind, generate a static electricity balanced material flow with stable potential, and control the spraying parameters as follows:

[0073] 1. Frequency - flow velocity linkage relationship: Among them, f is the pulsed ionic wind spraying frequency, v is the current flow velocity, and v max is the maximum flow velocity during the fluidized transportation process;

[0074] 2. Piece - wise linear relationship between voltage and static electricity accumulation value:

[0075]

[0076] Among them, U is the ionic wind spraying voltage;

[0077] 3. Optimization of the spraying direction: The ionic wind flow direction forms an angle θ with the material movement direction, satisfying:

[0078] 25° ≤ θ ≤ 55°; This design can generate a vortex effect at the bend, improve the neutralization uniformity, and generate a static electricity balanced material flow.

[0079] Specifically, S2 includes:

[0080] S21, Polarization imaging unit arrangement: Arrange three groups of polarization imaging units below the silo. Each group includes linearly polarized filters distributed in a ring, and cooperate with an 850 nm near - infrared polarization light source to irradiate the moving crucible. The angles of the linearly polarized filters distributed in a ring are 0°, 60°, and 120°;

[0081] S22, Stokes vector calculation: Generate a polarization feature map with dust interference removed based on Stokes vector calculation, expressed as:

[0082] Among them, S is the polarization feature vector, and I θFor the image gray values corresponding to the polarization angles (0°, 60°, 120°), extract the intrinsic polarization features of the object material and effectively suppress the interference of dust scattering;

[0083] S23, Motion blur restoration: Input the polarization feature maps of multiple consecutive frames (5 frames can be selected) into the motion blur restoration neural network. The motion blur restoration neural network adopts an encoder-decoder structure and embeds the temporal optical flow constraint to output a high-definition positioning image after displacement compensation;

[0084] The temporal optical flow constraint correlates the motion trajectories of adjacent frames to achieve:

[0085] 5-frame sliding window processing to balance the computational complexity and accuracy;

[0086] Displacement compensation output to improve the positioning accuracy of the crucible opening;

[0087] S24, Based on the above high-definition positioning image (removing dust interference and motion blur), provide high-quality visual input, 3D template matching and dynamic alignment deviation calculation: Solve the 3D pose of the crucible opening through the 3D template matching method and calculate the dynamic alignment deviation, expressed as:

[0088] ΔP = K·∥T actual -T target ∥+(1 - K)·arccos(q actual ·q target ), where

[0089] ΔP is the dynamic alignment deviation, T actual , T target are the translation vectors of the actual position of the current crucible opening and the target crucible opening respectively, q actual , q target are the actual rotation quaternion of the current crucible opening attitude and the rotation quaternion representation of the target crucible opening attitude respectively. K is the weight coefficient, and its value range is 0.7 - 0.9, which is used to balance the contributions of translation error and rotation error to the alignment deviation and is adaptively adjusted according to the material loading amount. arccos(q actual ·q target ) represents the calculation of the quaternion rotation error.

[0090] Let the input be a sequence of 5 consecutive polarization feature maps {S t |t = 1,..., 5}, and the network processing flow is as follows:

[0091] 1. Encoder part, use two convolutional blocks + max pooling for feature extraction:

[0092]

[0093] where ConvBlock(x) = ReLU(BN(Conv 3×3 (x))), Conv 3×3 represents a 3×3 convolutional layer, BN is batch normalization, and ReLU(·) is the ReLU activation function.

[0094] The encoder outputs a feature map where H and W are the height and width of the input image respectively, is the feature map output by the encoder;

[0095] 2. Temporal optical flow constraint module:

[0096]

[0097] where is the optical flow estimation sub-network (including 3 layers of dilated convolution) to extract long-range motion information, is the true displacement based on the mechanical encoder data, TV(·) is the total variation regularization term used to remove noise, λ = 0.1 is the regularization parameter, is the optical flow loss function;

[0098] 3. Decoder part, LSTM fuses time series features and restores a clear image through the decoder:

[0099] where LSTM is the LSTM cell, and the LSTM cell fuses temporal features: is the decoder, and the decoder contains 4 levels of upsampling + skip connections to gradually restore sharpness and output the displacement compensation image (with the 3rd frame as the reference), h t is the LSTM hidden state, h t-1 is the previous moment's LSTM hidden state used to capture long-term temporal dependency information, and ΔF t→t+1 serves as the LSTM constraint term to enhance time series consistency;

[0100] 4. Perform 3D matching based on the output image to estimate the pose of the crucible opening:

[0101] where P i is the i-th feature point of the preset crucible opening CAD model (a total of N = 32 key points), π(·) is the camera projection model that projects 3D points into the 2D image space, and e i (·) is the coordinate of the corresponding feature point extracted from , ρ is the Hube robust loss function used to suppress the influence of outliers on the optimization, T is the translation vector, and q is the quaternion pose.

[0102] S3 specifically includes:

[0103] S31. Establish a mapping relationship between the dynamic alignment deviation ΔP and the control parameters of the screw auger, and adjust the stepping angle and rotation speed of the screw auger. The stepping angle is non-linearly adjusted based on the reference angle to avoid overshoot or oscillation, and the rotation speed adopts an exponential decay method and is adaptively reduced according to the size of the deviation.

[0104] S32. Adaptive frequency control of the axial vibration field: Embed a piezoelectric actuator array at the axis of the screw auger to generate an axial vibration field, and the vibration frequency of the axial vibration field is adjusted according to the change of the angle of repose of the graphite powder.

[0105] S33. Vibration-flow coupling control strategy: Construct a vibration-flow coupling control model. When the angle of repose of the graphite powder exceeds the set threshold, activate the high-frequency pulse vibration mode based on the vibration-flow coupling control model. The frequency of the high-frequency pulse adopts a stepped activation method and is increased in stages according to the increment of the angle of repose, and compensation is considered according to the change trend of the rotation speed of the screw auger.

[0106] In S31, the mapping relationship between the dynamic alignment deviation ΔP and the control parameters of the screw auger is expressed as:

[0107]

[0108] where θ is the stepping angle of the screw auger, θ0 is the reference angle, k θ is the adjustment gain, with a value range of 5° - 8°, ΔP is the dynamic alignment deviation, ΔP max is the maximum allowable deviation, with a value of 5 mm, n is the real-time rotation speed of the screw auger, n base is the reference rotation speed, with a value range of 30 - 50 rpm, and γ is the rotation speed decay coefficient, with a value range of 0.2 - 0.3.

[0109] The vibration frequency f of the axial vibration field in S32 satisfies: where f is the real-time frequency of the axial vibration field, f0 is the reference frequency, with a value of 80 Hz, α is the angle of repose of the graphite powder measured in real time, α c is the critical angle of repose, with a value of 35°, α max , α min is the range of change of the angle of repose, Δf is the adjustment bandwidth, with a value of 120 Hz, A is the vibration amplitude, with a value range of 0.1 - 0.3 mm, and is inversely proportional to the feeding speed.

[0110] The vibration-flow coupling control model is expressed as:

[0111] where the high-frequency pulse mode is activated when α > 40°, f pulseIt represents the vibration frequency when the high-frequency pulse mode is activated. The high-frequency vibration is activated in stages. sgn(dn / dt) represents the sign function of the rotational speed change rate (+1 indicates acceleration, -1 indicates deceleration. When accelerating, the frequency is increased to compensate for the inertial effect; when decelerating, the frequency is decreased to prevent excessive disturbance). d is the clearance of the spiral blade, and its value range is 1.2 - 2.0 mm. represents the floor function, which means rounding down to trigger high-frequency pulse modes of different intensities:

[0112] When α = 38°, after rounding down there are no additional pulses at this time (i.e., it is still low-frequency vibration);

[0113] When α = 40°, a high-frequency pulse at the 200 Hz level is triggered;

[0114] When α = 42°, a high-frequency pulse at the 400 Hz level is triggered;

[0115] This staged activation mechanism is smoother than single triggering, can avoid system oscillation, and prevent excessive or insufficient vibration.

[0116] S4 specifically includes:

[0117] S41, Axial temperature gradient calculation: A plurality of groups of infrared thermal imaging sensors are arranged circumferentially at the discharge port of the silo to form an annular monitoring array, and the temperature field data of the graphite powder accumulation body during the discharging process are synchronously collected to calculate the axial temperature gradient distribution: Among them, is the axial temperature gradient, T top , T bottom are the average temperatures at 20 cm and 80 cm away from the material port respectively, H is the monitoring spacing with a value of 60 cm, and t is the discharging duration;

[0118] S42, Three-dimensional air gap probability distribution calculation: A dual-channel convolutional neural network (DC-CNN) is constructed. The input layer simultaneously receives:

[0119] Channel 1: Axial temperature gradient distribution matrix (20×20 grid data);

[0120] Channel 2: X-ray backscattering signal (density distribution map, corresponding to the material density information in the spatial region);

[0121] The network output is the air gap probability distribution map P(x, y, z). When P > 0.7, mark this area as an effective air gap;

[0122] S43, Gradient negative pressure curve generation: Generate a gradient negative pressure curve dynamically according to the air gap distribution:

[0123] where P vac (r) is the vacuum negative pressure at the radial coordinate r, P base is the reference negative pressure, r is the radial coordinate of the vacuum pipeline, r i is the center position of the i-th air gap, w i is the air gap influence weight, with a value range of 0.3 - 0.6, σ is the air gap diffusion influence range, with a value of 0.2R, where R is the pipeline radius, ensuring coverage of the air gap area;

[0124] S44. According to the air gap evolution characteristics, implement a phased pumping strategy.

[0125] The phased pumping strategy includes:

[0126] When t < 30s, it is set as the initial stage: In this stage, the air gap is large and concentrated, and the negative pressure acts on the core area intensively. Apply the maximum negative pressure P max = 12 kPa;

[0127] When 30s ≤ t < 90s, it is set as the middle stage: The air gap distribution tends to be uniform, and the negative pressure is adjusted according to the gradient distribution. Distribute the negative pressure (8 - 12 kPa) along the gradient negative pressure curve and gradually adjust the pumping intensity;

[0128] When t ≥ 90s, it is set as the later stage: The residual air gap decreases, start pulse pumping, use a 5Hz square wave modulation, and the negative pressure amplitude is 10 ± 2 kPa to improve the removal efficiency of micro air gaps.

[0129] DC-CNN consists of an input layer, a feature extraction layer, a feature fusion layer, and an air gap prediction output layer. The detailed structure is as follows:

[0130] 1) Input layer:

[0131] Channel 1 (temperature gradient): Single-channel grayscale image;

[0132] Channel 2 (density distribution): Single-channel density map.

[0133] 2) Feature extraction layer:

[0134] Channel 1 (temperature gradient data):

[0135] Adopt a two-layer 3×3 convolution structure to extract the spatial distribution features of the temperature gradient;

[0136] Enhance the feature expression ability through batch normalization (BN) and ReLU activation function;

[0137] The max pooling layer (MaxPool) gradually reduces the dimension to capture multi-scale heat conduction anomalies;

[0138] Channel 2 (X-ray density data):

[0139] Use a combination of 5×5 and 3×3 convolution kernels to identify density mutation regions;

[0140] Introduce the LeakyReLU activation function to enhance the sensitivity to low-density regions;

[0141] Implement feature map downsampling through convolution with a stride of 2×2 to focus on the air gap edge features.

[0142] 3) Feature fusion layer:

[0143] Concatenate the feature vectors of the two channels to form a unified fused feature;

[0144] Attention mechanism: Based on the air gap saliency weights in different regions, enhance the influence of air gap region features and suppress background noise;

[0145] Cross-modal feature fusion: Multiply the temperature features and density features element by element to achieve multi-modal data complementarity, and further integrate the features through 1×1 convolution to generate a high-dimensional fused feature map;

[0146] The fused feature map is input into the decoder, and the three-dimensional space is reconstructed through 5-level upsampling to reconstruct the three-dimensional air gap probability distribution.

[0147] 4) Air gap prediction output layer: Finally, the DC-CNN generates a three-dimensional air gap probability distribution map.

[0148] An unattended crucible loading and unloading operation control system based on visual processing, used to implement the above operation control method, including the following modules:

[0149] Electrostatic monitoring and neutralization module, used to monitor the electrostatic accumulation value in the fluidized state of graphite powder in real time during the negative pressure pipeline transportation process, and trigger pulse-type ion wind neutralization treatment when the electrostatic accumulation exceeds the limit to form an electrostatically balanced material flow;

[0150] Polarized vision positioning and alignment deviation calculation module, based on the polarized vision positioning method to capture the real-time pose of the crucible, and calculate the dynamic alignment deviation between the silo and the crucible mouth through the multi-frame motion blur restoration algorithm;

[0151] Screw auger stepping control and vibration field regulation module, used to adjust the stepping angle and rotation speed of the screw auger based on the dynamic alignment deviation, and synchronously apply an axial vibration field during the unloading process, and the vibration frequency is dynamically adjusted according to the real-time change of the angle of repose of graphite powder to optimize the feeding uniformity;

[0152] The intelligent degassing control module is used to monitor the temperature gradient distribution of the graphite powder accumulation body through infrared thermal imaging during the degassing and blanking stage, predict the position of the air gap inside the material based on the convolutional neural network, dynamically adjust the vacuum negative pressure value, form a gradient suction mode, and improve the consistency of degassing efficiency and loading density.

[0153] The present invention covers any alternatives, modifications, equivalent methods, and solutions made to the essence and scope of the present invention. For the public to have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, and those skilled in the art can fully understand the present invention without the description of these details. Additionally, well-known methods, processes, procedures, components, and circuits, etc., are not described in detail to avoid unnecessary confusion to the essence of the present invention.

[0154] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. An unattended crucible loading and unloading control method based on visual processing, characterized in that: The following steps are involved: S1. When the negative pressure pipeline is loaded to the silo, the static electricity accumulation value in the process of graphite powder fluidization transportation is monitored in real time through the distributed static electricity sensor. When the static electricity accumulation value exceeds the safety threshold, the pulsed ion wind neutralization treatment is triggered to generate an electrostatically balanced material flow; S2, using polarization vision positioning method to capture the real-time position of the crucible, generating positioning data through multi-frame motion blur recovery algorithm, and calculating the dynamic alignment deviation between the silo and the crucible mouth; S3, based on the dynamic alignment deviation, the step angle and speed of the spiral auger in the silo are controlled, and an axial vibration field is synchronously applied during the unloading process, and a dynamic mapping relationship is established between the vibration frequency of the axial vibration field and the repose angle of the graphite powder; S4. During the degassing and feeding stage, the temperature gradient distribution of the graphite powder accumulation body is monitored by infrared thermal imaging, the position of the air gap inside the material is predicted based on the convolutional neural network, and the vacuum negative pressure value is dynamically adjusted to form a gradient suction mode.

2. The unattended crucible loading and unloading operation control method based on visual processing according to claim 1 is characterized in that: The S1 includes arranging a plurality of groups of electrostatic induction rings in a circular array on the inner side of the pipe wall of the fluidized conveying section of the negative pressure pipe, with the spacing between adjacent electrostatic induction rings being 1 times the pipe diameter, to collect the electrostatic potential distribution data of the graphite powder flow cross section in real time.

3. The unattended crucible loading and unloading operation control method based on visual processing according to claim 2 is characterized in that: The static electricity accumulation value is calculated using a sliding window algorithm, which is calculated as follows: Among them, E accumulate is the static electricity accumulation value, ΔE i is the electrostatic potential difference between adjacent induction rings, ΔE i =E i+1 -E i , t i Indicates the time between when the material passes through the electrostatic induction ring i to i+1, T window is the sliding time window, l is the number of sampling points in the sliding window; When E accumulate When the voltage is >800V / m, the pulsed ion wind neutralization treatment is triggered, the pulsed ion wind nozzle arranged at the bend of the pipeline is activated, the ion wind release strategy is optimized, and an electrostatically balanced material flow with stable potential is generated.

4. The unattended crucible loading and unloading operation control method based on visual processing according to claim 1 is characterized in that: The S2 specifically includes: S21, polarization imaging unit arrangement: three groups of polarization imaging units are arranged below the silo, each group includes a circularly distributed linear polarization filter, and cooperates with an 850nm near-infrared polarized light source to irradiate the moving crucible; S22, Stokes vector calculation: Generates the polarization characteristic diagram without dust interference based on Stokes vector calculation, expressed as: Where S is the polarization eigenvector, I θ is the image gray value corresponding to the polarization angle; S23, motion blur recovery: input the motion blur recovery neural network into the continuous multi-frame polarization feature map. The motion blur recovery neural network adopts an encoder-decoder structure and embeds the temporal optical flow constraint to output a high-definition positioning image after displacement compensation. S24, three-dimensional template matching and dynamic alignment deviation calculation: The three-dimensional position of the crucible mouth is solved by the three-dimensional template matching method, and the dynamic alignment deviation is calculated, which is expressed as: ΔP=K·∥T actual -T target ∥+(1-K)·arccos(q actual ·q target ),That In the figure, ΔP is the dynamic alignment deviation, T actual ,T target are the actual position translation vector of the current crucible mouth and the target crucible mouth position translation vector, respectively, actual ,q target are the actual rotation quaternion of the current crucible mouth posture and the rotation quaternion of the target crucible mouth posture, respectively. K is the weight coefficient, which is used to balance the contribution of translation error and rotation error to the alignment deviation. It is adaptively adjusted according to the material loading amount. arccos(q actual ·q target ) represents the quaternion rotation error calculation.

5. The unattended crucible loading and unloading operation control method based on visual processing according to claim 1 is characterized in that: The S3 specifically includes: S31, establish a mapping relationship between the dynamic alignment deviation ΔP and the control parameters of the spiral auger, adjust the step angle and speed of the spiral auger, the step angle is nonlinearly adjusted based on the reference angle to avoid overshoot or oscillation, and the speed adopts an exponential decay method to adaptively reduce according to the deviation size; S32, adaptive frequency control of axial vibration field: a piezoelectric actuator array is embedded in the axis of the spiral auger to generate an axial vibration field, and the vibration frequency of the axial vibration field is adjusted according to the change of the repose angle of the graphite powder; S33, vibration-flow coupling control strategy: construct a vibration-flow coupling control model. When the repose angle of graphite powder exceeds the set threshold, the high-frequency pulse vibration mode is activated based on the vibration-flow coupling control model. The frequency of the high-frequency pulse is activated in a step-by-step manner and increased in stages according to the increment of the repose angle. Compensation is performed considering the speed change trend of the spiral auger.

6. The unattended crucible loading and unloading operation control method based on visual processing according to claim 5 is characterized in that: In S31, the mapping relationship between the dynamic alignment deviation ΔP and the auger control parameter is expressed as: Among them, θ is the step angle of the spiral auger, θ0 is the reference angle, k θ is the gain adjustment, ΔP is the dynamic alignment deviation, ΔP max is the maximum allowable deviation, n is the real-time speed of the spiral auger, n base is the reference speed, and γ is the speed attenuation coefficient.

7. The unattended crucible loading and unloading operation control method based on visual processing according to claim 5 is characterized in that: The vibration frequency f of the axial vibration field in S32 satisfies Among them, f is the real-time frequency of the axial vibration field, f0 is the reference frequency, α is the real-time measured angle of repose of graphite powder, α c is the critical angle of repose, α max ,α min is the range of the repose angle, Δf is the adjustment bandwidth, and A is the vibration amplitude.

8. The unattended crucible loading and unloading operation control method based on visual processing according to claim 1 is characterized in that: The S4 specifically includes: S41, axial temperature gradient calculation: multiple groups of infrared thermal imaging sensors are arranged around the silo discharge port to form a ring-shaped monitoring array, which synchronously collects the temperature field data of the graphite powder accumulation during the discharge process and calculates the axial temperature gradient distribution: in, is the axial temperature gradient, T top ,T bottom are the average temperatures at 20cm and 80cm from the material port, respectively; H is the monitoring interval, and t is the duration of unloading; S42, Calculation of three-dimensional air gap probability distribution: Construct a dual-channel convolutional neural network, and the input layer simultaneously receives: Channel 1: axial temperature gradient distribution matrix; Channel 2: X-ray backscatter signal; The network output is the air gap probability distribution map P(x,y,z). When P>0.7, the area is marked as a valid air gap. S43, gradient negative pressure curve generation: Dynamically generate a gradient negative pressure curve based on air gap distribution: Among them, P vac (r) is the vacuum negative pressure at the radial coordinate r, P base is the reference negative pressure, r is the radial coordinate of the vacuum pipe, r i is the center position of the ith air gap, w i is the air gap influence weight, with a value range of 0.3-0.6, σ is the air gap diffusion influence range, with a value of 0.2R, where R is the pipe radius, to ensure that the air gap area is covered; S44, implement a phased suction strategy according to the air gap evolution characteristics.

9. The unattended crucible loading and unloading operation control method based on visual processing according to claim 8 is characterized in that: The phased aspiration strategy includes: When t<30s, it is set as the initial stage: the maximum negative pressure P is applied to the core area of ​​the air gap max =12kPa; When 30s≤t<90s, it is set as mid-term: negative pressure (8-12kPa) is distributed along the gradient negative pressure curve, and the suction force is gradually adjusted; When t≥90s, it is set to the later stage: start pulse suction, use 5Hz square wave modulation, and the negative pressure amplitude is 10±2kPa to improve the removal efficiency of tiny air gaps.

10. An unattended crucible loading and unloading operation control system based on visual processing, used to implement the unattended crucible loading and unloading operation control method based on visual processing as claimed in any one of claims 1 to 9, characterized in that: Includes the following modules: The static electricity monitoring and neutralization module is used to monitor the static electricity accumulation value of graphite powder in the fluidized state in real time during the negative pressure pipeline transportation process, and trigger the pulsed ion wind neutralization treatment when the static electricity accumulation exceeds the limit to form an electrostatically balanced material flow; Polarization vision positioning and alignment deviation calculation module, which captures the real-time position of the crucible based on the polarization vision positioning method, and calculates the dynamic alignment deviation between the silo and the crucible mouth through a multi-frame motion blur recovery algorithm; The spiral auger stepping control and vibration field control module is used to adjust the stepping angle and speed of the spiral auger based on the dynamic alignment deviation, and synchronously apply the axial vibration field during the unloading process. The vibration frequency is dynamically adjusted according to the real-time change of the graphite powder repose angle; The intelligent degassing control module is used to monitor the temperature gradient distribution of the graphite powder accumulation body through infrared thermal imaging during the degassing and feeding stage, and predict the position of the air gap inside the material based on the convolutional neural network, dynamically adjust the vacuum negative pressure value, and form a gradient suction mode.

Images