Powder warehouse loading metering and scheduling platform fusing machine vision and digital twinning

Abstract

The present application belongs to the technical field of powder warehouse loading metering and scheduling, and discloses a powder warehouse loading metering and scheduling platform fusing machine vision and digital twinning; comprising a manifold feature mapping unit, which continuously collects images during the whole process of vehicle approach and loading, acquires images of the tank opening and the loading opening, and maps the images to a high-dimensional manifold space, calculates the difference between the center of the tank opening and the center of the loading opening in the high-dimensional manifold space, and forms a corresponding manifold deviation vector; a curvature deviation analysis unit calculates the curvature of the manifold deviation vector, and when the curvature change rate is greater than a preset curvature change rate threshold, identifies an unstable alignment state and outputs the curvature change curve of the manifold deviation vector during the loading process; the present application has the adaptability to complex working conditions and different vehicle models, and improves the intelligent level of loading accuracy and scheduling efficiency.

Description

Technical Field

[0001] This invention relates to the field of powder warehouse loading, metering, and scheduling technology, and more specifically, to a powder warehouse loading, metering, and scheduling platform that integrates machine vision and digital twins. Background Technology

[0002] Currently, the alignment of tank openings and loading ports during powder loading relies heavily on manual observation or simple limiting devices, resulting in low efficiency, insufficient alignment accuracy, and significant human error. Some machine vision methods directly calculate pixel differences on the image plane, but these are prone to large errors due to variations in lighting, camera installation angles, and different vehicle models, making stability difficult to guarantee. Furthermore, existing methods are mostly static detection, unable to depict the dynamic alignment trajectory during loading, thus making it difficult to promptly identify risks of misalignment or overflow.

[0003] Existing powder loading technologies typically rely on manual experience or single sensor data, lacking indicators to quantify alignment trajectory curvature and fluctuation amplitude, making it difficult to achieve dynamic monitoring and scientific evaluation of alignment stability. When loading multiple vehicles simultaneously, a first-come, first-served or fixed-order scheduling method is often used, failing to optimize based on the real-time alignment status of the vehicles, resulting in a decrease in overall loading efficiency and accuracy. The deviation characteristics and dynamic alignment stability of different vehicles are not fully utilized, further limiting the effectiveness of scheduling optimization.

[0004] Meanwhile, traditional scheduling methods based on fixed thresholds or rules struggle to cope with complex operating conditions such as vehicle model differences, varying lighting conditions, and material flow disturbances. Digital twin models often employ preset rules and lack self-learning and dynamic correction capabilities based on historical data, making it impossible to continuously optimize loading accuracy and scheduling efficiency over long-term operation. Existing technologies struggle to achieve robust monitoring, risk prediction, and adaptive scheduling optimization of the alignment process under complex operating conditions.

[0005] Based on the above problems, there is an urgent need to propose a powder warehouse loading, metering, and scheduling platform that integrates machine vision and digital twins. Summary of the Invention

[0006] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: a powder warehouse loading, metering, and scheduling platform integrating machine vision and digital twins, comprising:

[0007] The manifold feature mapping unit continuously acquires images of the entire process of vehicle entry and loading, obtains images of the tank opening and loading port, and maps the images to a high-dimensional manifold space. In the high-dimensional manifold space, the difference between the center of the tank opening and the center of the loading port is calculated to form the corresponding manifold deviation vector.

[0008] The curvature deviation analysis unit calculates the curvature of the manifold deviation vector. When the rate of change of curvature is greater than the preset rate of change of curvature threshold, it is identified as an alignment instability state and outputs the curvature change curve of the manifold deviation vector during the loading process.

[0009] The loading evolution unit constructs different sets of invariant operators based on image features to address differences in lighting, angle, and vehicle type, obtains dynamic alignment indicators of manifold deviation vectors, and inputs them into a preset digital twin. When the curvature evolution trend indicates off-loading or overflow risk, it automatically triggers the adjustment of the loading strategy.

[0010] The scheduling optimization unit optimizes material flow alignment in real time based on curvature change curves and dynamic alignment characteristics. Combining the curvature patterns and invariant operator characteristics of different vehicles, it establishes vehicle hierarchical scheduling rules to prioritize the allocation of loading tasks to vehicles with the best alignment indicators.

[0011] The feedback adaptive unit stores the manifold deviation vector and curvature change curve to form a manifold deviation database; based on the manifold deviation database, it continuously corrects the invariant operator parameters, reconstructs the twin alignment rules, and updates the scheduling and loading strategy according to changes in working conditions.

[0012] Specifically, the method for acquiring the images of the tank opening and the filling port includes:

[0013] Industrial CMOS cameras are deployed at key locations in the powder loading area to cover the loading positions, tank openings, and filling ports. Key locations in the powder loading area include the vehicle entrance, above the loading positions, to the side of the loading positions, above the tank openings, to the side of the tank openings, above the filling ports, and to the side of the filling ports.

[0014] The industrial CMOS camera uses a robotic arm to perform variable angle tracking and continuously acquires images of the entire process of vehicle entry and loading at a preset frame rate. It obtains image sequences of the tank opening and loading port, and adds a timestamp to each frame in the image sequence of the tank opening and loading port to ensure that the vehicle entry and loading actions correspond to the time sequence of the image sequence of the tank opening and loading port.

[0015] Specifically, the method for forming the corresponding manifold deviation vector includes:

[0016] The image sequence of tank opening and filling port is preprocessed, including denoising, image noise reduction, and image enhancement. In the preprocessed image sequence of tank opening and filling port, key feature points of the tank opening and filling port are identified, and the spatial position of the tank opening and filling port is described by the coordinates of the key feature points.

[0017] The processed image of each frame in the image sequence of the tank opening and the loading port is mapped to a high-dimensional manifold space to represent the high-dimensional spatial relationship between the vehicle and the loading port. The local geometric and global structural information of each frame in the image sequence is preserved through the high-dimensional manifold space to represent the relative positional relationship between the tank opening and the loading port. In the high-dimensional manifold space, the spatial deviation between the center of the tank opening and the center of the loading port is calculated and a manifold deviation vector is formed.

[0018] Specifically, the method for identifying the state as an inverted state includes:

[0019] The manifold deviation vector in the high-dimensional manifold space is divided into local windows according to the time sequence. The manifold deviation vector is regarded as a spatial trajectory curve that evolves with time. The curvature of the manifold deviation vector is defined to describe the degree of curvature of the deviation trajectory.

[0020] Based on the local curvature values ​​of consecutive frames, the rate of change of curvature over time is calculated. A preset curvature change rate threshold is set. When the curvature change rate is greater than the preset curvature change rate threshold, the corresponding manifold deviation vector is identified as an in-position unstable state.

[0021] Specifically, the method for outputting the curvature change curve of the manifold deviation vector during the loading process includes:

[0022] During the loading process, the curvature of the manifold deviation vector at each time point is arranged in chronological order to form a curvature sequence. Based on the curvature sequence, a curve of curvature changing with time is plotted along the time axis. This curve is the curvature change curve of the manifold deviation vector.

[0023] Specifically, the method for obtaining the dynamic alignment index of the manifold deviation vector includes:

[0024] Image features are mapped to different feature subspaces according to lighting conditions, shooting angle, and vehicle model differences. A corresponding set of invariant operators is constructed in each feature subspace to eliminate the influence of environmental factors and vehicle model differences on the manifold deviation vector.

[0025] The manifold deviation vector is applied sequentially to the corresponding invariant operator group to obtain the corrected manifold deviation vector, which reflects the actual relative positional relationship between the vehicle tank opening and the loading port in the high-dimensional manifold space. Based on the corrected manifold deviation vector, the dynamic alignment index is calculated. The dynamic alignment index is updated over time to form a sequence of alignment state changes throughout the loading process.

[0026] Specifically, the method for automatically triggering the adjustment of the loading strategy includes:

[0027] The preset digital twin is a virtual model of the loading process that is built in advance, including the location of the vehicle tank opening, the location of the loading port, the material flow characteristics, and the loading strategy rules. The preset digital twin performs real-time evolution analysis on the input dynamic alignment indicators and curvature change curves to determine whether there is a risk of misloading or overflow during the loading process.

[0028] The judgment criteria include curvature trend, curvature change rate, and deviation of alignment indicators from the preset alignment indicator safety threshold. When the digital twin determines that there is a risk of misloading or overflow, it automatically generates a loading strategy adjustment instruction. The loading strategy adjustment instruction is fed back to the powder warehouse loading metering and scheduling platform in real time, driving the preset loading actuator to make automatic adjustments. The adjustment instructions include modifying the material flow delivery speed or direction, adjusting the centering position of the loading port and vehicle tank opening, and controlling the loading sequence of specific areas.

[0029] Specifically, the method of prioritizing the allocation of loading tasks to vehicles with the best positioning indicators includes:

[0030] During the loading process, for each vehicle, the real-time scheduling control quantity is defined using the dynamic alignment index of the manifold deviation vector and the curvature change curve; by optimizing the objective function, the material flow is aligned with the tank opening during the loading process, minimizing the corrected modulus length of the deviation vector of all vehicles, so that the material flow is aligned with the tank opening during the loading process.

[0031] The mean curvature of the curvature sequence of each vehicle is calculated to reflect the overall curvature of the vehicle's deviation trajectory; the dynamic alignment characterization is calculated to quantify the fluctuation range of the dynamic alignment index, reflecting the alignment stability of the vehicle under different working conditions; the mean curvature and the dynamic alignment characterization are combined to form a vehicle scheduling ranking index, and the loading task is preferentially assigned to the vehicle with the smallest vehicle scheduling ranking index.

[0032] Specifically, the method for obtaining the manifold deviation database includes:

[0033] Record the manifold deviation vector sequence of each vehicle during the loading process in chronological order, and store and classify the corresponding curvature curves and characteristic parameters in a hierarchical manner using vehicle number, vehicle type, loading batch and working conditions as index labels.

[0034] The characteristic parameters include mean curvature, rate of change of curvature, fluctuation amplitude, and threshold exceedance marker. After each loading task is completed, the newly collected images of the tank opening and loading port are automatically stored in the database and compared with historical images of the tank opening and loading port to dynamically update the database and obtain the manifold deviation database.

[0035] Specifically, the method for updating the scheduling and loading strategy according to changes in operating conditions includes:

[0036] For the cumulative image data of the manifold deviation database, the invariant operator parameters are iteratively corrected by updating the rules; combined with the corrected invariant operator parameters, the alignment rules are reconstructed in the digital twin, and the scheduling and loading strategy is updated and adjusted according to the working conditions to optimize the real-time scheduling control of the vehicle.

[0037] The technical effects and advantages of this invention, a powder silo loading, metering, and scheduling platform integrating machine vision and digital twins, are as follows:

[0038] By mapping image sequences of the tank opening and the filling port to a high-dimensional manifold space, local geometric and global structural information is extracted. The calculation of the manifold deviation vector, compared to two-dimensional centroid difference, more accurately reflects the true spatial relationship between the tank opening and the filling port, avoiding inaccuracies caused by differences in lighting, angle, and vehicle type. High-dimensional manifold mapping ensures the preservation of local and global geometric structures in the image, and combined with a compensation mechanism, it maintains stable alignment even under complex working conditions. It can not only detect deviations in single-frame images but also dynamically monitor the alignment state through curvature evolution curves, identifying unstable trends in advance and reducing the risk of powder misloading and overflow. By injecting the manifold deviation vector and curvature evolution curve as input into the digital twin, it supports real-time adjustment of the filling strategy and optimization of vehicle scheduling, breaking through the static limitations of traditional visual inspection.

[0039] By utilizing manifold deviation vectors and curvature change curves, dynamic monitoring of the deviation between the vehicle's tank opening and loading port is achieved. The objective function is optimized to minimize the modulus of the corrected deviation vector, ensuring precise material flow alignment and reducing the risk of misalignment and spillage. By calculating the mean curvature and dynamic alignment characteristics, the alignment stability and trajectory variation of vehicles under different operating conditions are quantified. Vehicles with excessively high curvature change rates are identified promptly to prevent loading accidents caused by sudden deviations. The mean curvature and dynamic alignment characteristics are combined to form a vehicle scheduling ranking index, and its contribution to ranking is controlled by adjustable weights. Loading tasks are preferentially assigned to vehicles with the best alignment indicators, achieving dynamic optimization of the multi-vehicle loading process and improving overall loading efficiency. By integrating machine vision and digital twin technologies, and considering differences in lighting, angle, and vehicle type, adaptive control of the loading process under different vehicle and environmental conditions is achieved.

[0040] By continuously refining the parameters of the invariant operators, the digital twin can dynamically adapt to environmental disturbances caused by differences in lighting, angle, and vehicle type, ensuring that the alignment rules always match the actual working conditions and significantly improving alignment robustness. Combined with the refined twin alignment rules, the real-time scheduling control of vehicles is optimized and adjusted, making material flow alignment more precise and avoiding the risks of misalignment or overflow caused by vehicle differences or material flow fluctuations, thereby improving loading safety and accuracy. Through the continuous accumulation of the manifold deviation database, the platform can continuously iterate and update operator parameters and scheduling strategies, achieving a closed-loop evolution of data-rules-optimization, possessing long-term self-learning and adaptive capabilities. By comprehensively measuring the alignment performance of different vehicles through dynamic scheduling indicators, tasks are prioritized for allocation to vehicles with the best alignment indicators, achieving efficient resource utilization, shortening loading time, and improving overall scheduling efficiency and fairness. The platform can operate stably for a long time under different working conditions and vehicle types, solving the problems of strong environmental dependence and poor adaptability of existing systems, and improving reliability and promotional value in industrial application scenarios. Attached Figure Description

[0041] Figure 1 This is a schematic diagram of the structure of the powder warehouse loading, metering, and scheduling platform that integrates machine vision and digital twins in this invention.

[0042] Figure 2 This is a schematic diagram of the powder warehouse loading, metering, and scheduling method that integrates machine vision and digital twins in this invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Example 1

[0045] Please see Figure 1 This embodiment provides a powder warehouse loading, metering, and scheduling platform that integrates machine vision and digital twins, specifically including the following steps:

[0046] The manifold feature mapping unit continuously acquires images of the entire process of vehicle entry and loading, obtains images of the tank opening and loading port, and maps the images to a high-dimensional manifold space. In the high-dimensional manifold space, the difference between the center of the tank opening and the center of the loading port is calculated to form the corresponding manifold deviation vector.

[0047] The curvature deviation analysis unit calculates the curvature of the manifold deviation vector. When the rate of change of curvature is greater than the preset rate of change of curvature threshold, it is identified as an alignment instability state and outputs the curvature change curve of the manifold deviation vector during the loading process.

[0048] The loading evolution unit constructs different sets of invariant operators based on image features to address differences in lighting, angle, and vehicle type, obtains dynamic alignment indicators of manifold deviation vectors, and inputs them into a preset digital twin. When the curvature evolution trend indicates off-loading or overflow risk, it automatically triggers the adjustment of the loading strategy.

[0049] The scheduling optimization unit optimizes material flow alignment in real time based on curvature change curves and dynamic alignment characteristics. Combining the curvature patterns and invariant operator characteristics of different vehicles, it establishes vehicle hierarchical scheduling rules to prioritize the allocation of loading tasks to vehicles with the best alignment indicators.

[0050] The feedback adaptive unit stores the manifold deviation vector and curvature change curve to form a manifold deviation database; based on the manifold deviation database, it continuously corrects the invariant operator parameters, reconstructs the twin alignment rules, and updates the scheduling and loading strategy according to changes in working conditions.

[0051] Methods for obtaining images of the tank opening and filling port include:

[0052] Industrial CMOS cameras are deployed at key locations in the powder loading area to cover the loading positions, tank openings, and filling ports. Key locations in the powder loading area include the vehicle entrance, above the loading positions, to the side of the loading positions, above the tank openings, to the side of the tank openings, above the filling ports, and to the side of the filling ports.

[0053] The industrial CMOS camera uses a robotic arm to perform variable angle tracking and continuously acquires images of the entire process of vehicle entry and loading at a preset frame rate. It obtains image sequences of the tank opening and loading port, and adds a timestamp to each frame in the image sequence of the tank opening and loading port to ensure that the vehicle entry and loading actions correspond to the time sequence of the image sequence of the tank opening and loading port.

[0054] Methods for generating the corresponding manifold deviation vector include:

[0055] The image sequence of tank opening and filling port is preprocessed, including denoising, image noise reduction, and image enhancement. In the preprocessed image sequence of tank opening and filling port, key feature points of the tank opening and filling port are identified, and the spatial position of the tank opening and filling port is described by the coordinates of the key feature points.

[0056] The processed image of each frame in the image sequence of the tank opening and the loading port is mapped to a high-dimensional manifold space to represent the high-dimensional spatial relationship between the vehicle and the loading port. The local geometric and global structural information of each frame in the image sequence is preserved through the high-dimensional manifold space to represent the relative positional relationship between the tank opening and the loading port. In the high-dimensional manifold space, the spatial deviation between the center of the tank opening and the center of the loading port is calculated and a manifold deviation vector is formed.

[0057] The manifold deviation vector is: ;in, This represents the spatial deviation vector between the center of the tank opening and the center of the filling port during the loading process, i.e., the relative positional error between the tank opening and the filling port; this deviation is used as a core indicator to determine whether the alignment is accurate. This represents the compensation amount for non-coincidence of installation centers, used to correct fixed deviations caused by target pasting deviations, camera installation deviations, or mechanical assembly errors; it also represents the position coordinate vector of the tank opening center. The coordinate vector representing the position of the center of the vehicle's tank opening is the coordinate of the image center on the CMOS. Here, the image center refers to the geometric center of the tank opening area obtained by target recognition, which is the centroid calculated from the set of pixels of the tank opening target. The position coordinate vector representing the center of the loading port, i.e. the distance to the CMOS center, is essentially the projection coordinate of the loading port center on the CMOS imaging plane, mapped through the intrinsic and extrinsic parameters of the industrial CMOS camera.

[0058] Methods for identifying para-unstable states include:

[0059] The manifold deviation vector in the high-dimensional manifold space is divided into local windows according to the time sequence. The manifold deviation vector is regarded as a spatial trajectory curve that evolves with time. The curvature of the manifold deviation vector is defined to describe the degree of curvature of the deviation trajectory.

[0060] The curvature of the manifold deviation vector is: ;in, This represents a function of the manifold deviation vector over time, used to express the manifold deviation vector as a function of time points. The evolution, at a certain point in time It is the instantaneous manifold deviation vector at that point in time, that is, the relative deviation between the vehicle tank opening and the loading port in the high-dimensional manifold space; Represents the manifold deviation vector At time point The first derivative of the deviation vector, i.e. the rate at which the deviation vector changes with time, can be understood as the deviation change velocity vector, reflecting the direction and speed of the deviation change; Represents the manifold deviation vector At time point The second derivative, i.e. the change in the rate of change of the deviation, can be understood as the deviation acceleration vector, reflecting the curvature trend or the change in the rate of change of the deviation trajectory. It represents the magnitude of the first derivative vector, i.e., the magnitude of the rate of change of the deviation; This represents the normalized curvature, ensuring that the curvature reflects the degree of trajectory bending rather than the magnitude of velocity. Index representing a point in time;

[0061] Based on the local curvature values ​​of consecutive frames, the rate of change of curvature over time is calculated. A preset curvature change rate threshold is set. When the curvature change rate is greater than the preset curvature change rate threshold, the corresponding manifold deviation vector is identified as an in-position unstable state.

[0062] The rate of change of curvature over time is: ;in, Indicates curvature The increment, that is, the amount of change in curvature within a preset (very short) time interval; This represents the increment of time, i.e., the time interval corresponding to the calculation of the rate of change of curvature;

[0063] Methods for outputting the curvature change curve of the manifold deviation vector during loading include:

[0064] During the loading process, the curvature of the manifold deviation vector at each time point is arranged in chronological order to form a curvature sequence. Based on the curvature sequence, a curve of curvature changing with time is plotted along the time axis. This curve is the curvature change curve of the manifold deviation vector.

[0065] The curvature variation curve visually displays the degree of curvature curvature and dynamic trend of the deviation trajectory during loading, and can be used to analyze the stability of the alignment state and identify potential off-center loading or overflow risks. During curve generation, time periods where the rate of curvature change exceeds a preset threshold can be marked to highlight areas of alignment instability during loading. The generated curvature variation curve and related time information can be stored in a manifold deviation database for loading strategy optimization and scheduling adjustments.

[0066] Methods for obtaining dynamic alignment indices of manifold deviation vectors include:

[0067] Image features are mapped to different feature subspaces according to lighting conditions, shooting angle, and vehicle model differences. A corresponding set of invariant operators is constructed in each feature subspace to eliminate the influence of environmental factors and vehicle model differences on the manifold deviation vector.

[0068] It should be noted that invariance operators are functions or mappings that process the original image features or manifold deviation vectors, ensuring that the output features remain unchanged or change minimally under certain conditions. In vehicle assembly scenarios, they are primarily used to eliminate the influence of lighting differences, camera angle deviations, and differences in the geometry of different vehicle models on the deviation vector. Matrix transformations or function mappings can be used to map the original deviation vector to a normalized space, ensuring the consistency of the deviation vector output under different conditions.

[0069] The manifold deviation vector is applied sequentially to the corresponding invariant operator group to obtain the corrected manifold deviation vector, which reflects the actual relative positional relationship between the vehicle tank opening and the loading port in the high-dimensional manifold space. Based on the corrected manifold deviation vector, the dynamic alignment index is calculated. The dynamic alignment index is updated over time to form a sequence of alignment state changes throughout the loading process.

[0070] Methods for automatically triggering loading strategy adjustments include:

[0071] The preset digital twin is a virtual model of the loading process that is built in advance, including the location of the vehicle tank opening, the location of the loading port, the material flow characteristics, and the loading strategy rules. The preset digital twin performs real-time evolution analysis on the input dynamic alignment indicators and curvature change curves to determine whether there is a risk of misloading or overflow during the loading process.

[0072] The judgment criteria include curvature trend, curvature change rate, and deviation of alignment indicators from the preset alignment indicator safety threshold. When the digital twin determines that there is a risk of misloading or overflow, it automatically generates a loading strategy adjustment instruction. The loading strategy adjustment instruction is fed back to the powder warehouse loading metering and scheduling platform in real time, driving the preset loading actuator to make automatic adjustments. The adjustment instructions include modifying the material flow delivery speed or direction, adjusting the centering position of the loading port and vehicle tank opening, and controlling the loading sequence of specific areas.

[0073] Methods for prioritizing the allocation of loading tasks to vehicles with the best performance indicators include:

[0074] During the loading process, for each vehicle, the real-time scheduling control quantity is defined using the dynamic alignment index of the manifold deviation vector and the curvature change curve; by optimizing the objective function, the material flow is aligned with the tank opening during the loading process, minimizing the corrected modulus length of the deviation vector of all vehicles, so that the material flow is aligned with the tank opening during the loading process.

[0075] The objective function to be optimized is: ;in, Indicates the vehicle at a certain time point. The total correction deviation; Indicates the vehicle at a certain time point. Real-time scheduling and control quantities; Indicates vehicle At the point of time The manifold deviation vector; Index representing the vehicle; Indicates the total number of vehicles involved in loading;

[0076] The mean curvature of the curvature sequence of each vehicle is calculated to reflect the overall curvature of the vehicle's deviation trajectory; the dynamic alignment characterization is calculated to quantify the fluctuation range of the dynamic alignment index, reflecting the alignment stability of the vehicle under different working conditions; the mean curvature and the dynamic alignment characterization are combined to form a vehicle scheduling ranking index, and the loading task is preferentially assigned to the vehicle with the smallest vehicle scheduling ranking index.

[0077] The vehicle dispatching ranking criteria are: ;in, Indicates vehicle The scheduling and ranking indicators are used to measure the priority of the vehicle in the loading task; Indicates vehicle The mean curvature; This represents the curvature weight, used to control the contribution of the mean curvature in the vehicle scheduling ranking index. This represents the dynamic alignment representation weight, used to control the contribution of dynamic alignment representation to the vehicle scheduling ranking index; Indicates vehicle The dynamic alignment representation, i.e., the fluctuation amplitude of the manifold deviation vector within the time window;

[0078] Methods for obtaining manifold deviation databases include:

[0079] Record the manifold deviation vector sequence of each vehicle during the loading process in chronological order, and store and classify the corresponding curvature curves and characteristic parameters in a hierarchical manner using vehicle number, vehicle type, loading batch and working conditions as index labels.

[0080] The characteristic parameters include mean curvature, rate of change of curvature, fluctuation amplitude, and threshold exceedance marker. After each loading task is completed, the newly collected images of the tank opening and loading port are automatically stored in the database and compared with historical images of the tank opening and loading port to dynamically update the database and obtain the manifold deviation database.

[0081] Methods for updating and adjusting the scheduling and loading strategy according to changes in operating conditions include:

[0082] For the cumulative image data in the manifold deviation database, the parameters of the invariant operator are iteratively corrected using update rules;

[0083] The update rules are as follows: ;in, Indicates the updated invariant operator parameters; Represents the parameters of the original invariant operator; This represents the learning rate coefficient; Manifold Deviation Database The gradient of the loss function in the dimension of invariant operator parameters enables the digital twin alignment rules to adapt to dynamic working condition changes by continuously updating the invariant operator parameters.

[0084] By combining the modified invariant operator, the alignment rules are reconstructed in the digital twin, and the scheduling and loading strategy is updated and adjusted according to the changes in working conditions, thereby optimizing the real-time scheduling control of vehicles.

[0085] The optimization and adjustment rules are as follows: ;in, This represents the real-time dispatch and control quantity of vehicles after optimization and adjustment. Indicates the scheduling learning rate;

[0086] Ultimately, through long-term operation, the platform can continuously accumulate data to achieve adaptive correction of invariant operators, dynamic reconstruction of twin rules, and iterative updates of scheduling strategies. This enables the platform to maintain long-term robustness to complex working conditions and vehicle differences, while also possessing closed-loop self-learning capabilities to continuously improve loading accuracy and scheduling efficiency.

[0087] The preset curvature change rate threshold is set by staff based on historical data analysis results. This historical analysis process includes the system collecting multiple curvature change rates and calculating their average value as a reference to obtain the preset curvature change rate threshold. Similarly, the preset alignment index safety threshold is also set by staff based on the system's historical operating data and the specific application scenario requirements.

[0088] In this embodiment, by mapping the image sequences of the tank opening and the filling port to a high-dimensional manifold space, local geometric and global structural information is extracted. Through the calculation of the manifold deviation vector, compared to two-dimensional centroid difference, the true spatial relationship between the tank opening and the filling port can be reflected more accurately, avoiding inaccuracies caused by differences in lighting, angle, and vehicle type. High-dimensional manifold mapping ensures the preservation of the local and global geometric structure of the image, and combined with a compensation mechanism, it can maintain stable alignment even under complex working conditions. It can not only detect deviations in a single frame image but also dynamically monitor the alignment state through curvature evolution curves, identifying unstable trends in advance and reducing the risk of powder misloading and overflow. By injecting the manifold deviation vector and curvature evolution curve as input into the digital twin, it supports real-time adjustment of the filling strategy and optimization of vehicle scheduling, breaking through the static limitations of traditional visual inspection.

[0089] By utilizing manifold deviation vectors and curvature change curves, dynamic monitoring of the deviation between the vehicle's tank opening and loading port is achieved. The objective function is optimized to minimize the modulus of the corrected deviation vector, ensuring precise material flow alignment and reducing the risk of misalignment and spillage. By calculating the mean curvature and dynamic alignment characteristics, the alignment stability and trajectory variation of vehicles under different operating conditions are quantified. Vehicles with excessively high curvature change rates are identified promptly to prevent loading accidents caused by sudden deviations. The mean curvature and dynamic alignment characteristics are combined to form a vehicle scheduling ranking index, and its contribution to ranking is controlled by adjustable weights. Loading tasks are preferentially assigned to vehicles with the best alignment indicators, achieving dynamic optimization of the multi-vehicle loading process and improving overall loading efficiency. By integrating machine vision and digital twin technologies, and considering differences in lighting, angle, and vehicle type, adaptive control of the loading process under different vehicle and environmental conditions is achieved.

[0090] By continuously refining the parameters of the invariant operators, the digital twin can dynamically adapt to environmental disturbances caused by differences in lighting, angle, and vehicle type, ensuring that the alignment rules always match the actual working conditions and significantly improving alignment robustness. Combined with the refined twin alignment rules, the real-time scheduling control of vehicles is optimized and adjusted, making material flow alignment more precise and avoiding the risks of misalignment or overflow caused by vehicle differences or material flow fluctuations, thereby improving loading safety and accuracy. Through the continuous accumulation of the manifold deviation database, the platform can continuously iterate and update operator parameters and scheduling strategies, achieving a closed-loop evolution of data-rules-optimization, possessing long-term self-learning and adaptive capabilities. By comprehensively measuring the alignment performance of different vehicles through dynamic scheduling indicators, tasks are prioritized for allocation to vehicles with the best alignment indicators, achieving efficient resource utilization, shortening loading time, and improving overall scheduling efficiency and fairness. The platform can operate stably for a long time under different working conditions and vehicle types, solving the problems of strong environmental dependence and poor adaptability of existing systems, and improving reliability and promotional value in industrial application scenarios.

[0091] Example 2

[0092] Please see Figure 2 As shown, the parts not described in detail in this embodiment are described in Embodiment 1. A method for loading, metering, and scheduling powder materials into a warehouse, integrating machine vision and digital twins, is provided, including:

[0093] S1. Continuously acquire images of the entire process of vehicle entry and loading, obtain images of the tank opening and the loading port, and map the images to a high-dimensional manifold space. Calculate the difference between the center of the tank opening and the center of the loading port in the high-dimensional manifold space to form the corresponding manifold deviation vector.

[0094] S2. Calculate the curvature of the manifold deviation vector. When the rate of change of curvature is greater than the preset rate of change of curvature threshold, it is identified as an alignment instability state, and the curvature change curve of the manifold deviation vector during the loading process is output.

[0095] S3. Based on the differences in lighting, angle and vehicle type, construct different groups of invariant operators based on image features, obtain the dynamic alignment index of manifold deviation vector, and input it into the preset digital twin. When the curvature evolution trend indicates the risk of misloading or overflow, the loading strategy adjustment is automatically triggered.

[0096] S4. Based on the curvature change curve and dynamic alignment characterization, optimize the material flow alignment in real time. Combine the curvature patterns and invariant operator characteristics of different vehicles to establish vehicle hierarchical scheduling rules and prioritize the allocation of loading tasks to vehicles with the best alignment indicators.

[0097] S5. Store the manifold deviation vector and curvature change curve to form a manifold deviation database; continuously correct the invariant operator parameters based on the manifold deviation database, reconstruct the twin alignment rules, and update the scheduling and loading strategy according to the working conditions.

[0098] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0099] The above description is merely a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for users of ordinary technical skills, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A powder storage loading, metering, and scheduling platform integrating machine vision and digital twins, characterized in that: include: The manifold feature mapping unit continuously acquires images of the entire process of vehicle entry and loading, obtains images of the tank opening and loading port, and maps the images to a high-dimensional manifold space. In the high-dimensional manifold space, the difference between the center of the tank opening and the center of the loading port is calculated to form the corresponding manifold deviation vector. The curvature deviation analysis unit calculates the curvature of the manifold deviation vector. When the rate of change of curvature is greater than the preset rate of change of curvature threshold, it is identified as an alignment instability state and outputs the curvature change curve of the manifold deviation vector during the loading process. The method for identifying the state as an inverted state includes: The manifold deviation vector in the high-dimensional manifold space is divided into local windows according to the time sequence. The manifold deviation vector is regarded as a spatial trajectory curve that evolves with time. The curvature of the manifold deviation vector is defined to describe the degree of curvature of the deviation trajectory. Based on the local curvature values ​​of consecutive frames, the rate of change of curvature over time is calculated. A preset curvature change rate threshold is set. When the curvature change rate is greater than the preset curvature change rate threshold, the corresponding manifold deviation vector is identified as an instable state. The loading evolution unit constructs different sets of invariant operators based on image features to address differences in lighting, angle, and vehicle type, obtains dynamic alignment indicators of manifold deviation vectors, and inputs them into a preset digital twin. When the curvature evolution trend indicates off-loading or overflow risk, it automatically triggers the adjustment of the loading strategy. The scheduling optimization unit optimizes material flow alignment in real time based on curvature change curves and dynamic alignment characteristics. Combining the curvature patterns and invariant operator characteristics of different vehicles, it establishes vehicle hierarchical scheduling rules to prioritize the allocation of loading tasks to vehicles with the best alignment indicators. The feedback adaptive unit stores the manifold deviation vector and curvature change curve to form a manifold deviation database; based on the manifold deviation database, it continuously corrects the invariant operator parameters, reconstructs the twin alignment rules, and updates the scheduling and loading strategy according to changes in working conditions. The method for updating and adjusting the loading scheduling strategy according to changes in operating conditions includes: For the cumulative image data of the manifold deviation database, the invariant operator parameters are iteratively corrected by updating the rules; combined with the corrected invariant operator parameters, the alignment rules are reconstructed in the digital twin, and the scheduling and loading strategy is updated and adjusted according to the working conditions to optimize the real-time scheduling control of the vehicle.

2. The powder warehouse loading, metering, and scheduling platform integrating machine vision and digital twins as described in claim 1, characterized in that, The method for acquiring the images of the tank opening and the filling port includes: Industrial CMOS cameras are deployed at key locations in the powder loading area to cover the loading positions, tank openings, and filling ports. Key locations in the powder loading area include the vehicle entrance, above the loading positions, to the side of the loading positions, above the tank openings, to the side of the tank openings, above the filling ports, and to the side of the filling ports. The industrial CMOS camera uses a robotic arm to perform variable angle tracking and continuously acquires images of the entire process of vehicle entry and loading at a preset frame rate. It obtains image sequences of the tank opening and loading port, and adds a timestamp to each frame in the image sequence of the tank opening and loading port to ensure that the vehicle entry and loading actions correspond to the time sequence of the image sequence of the tank opening and loading port.

3. The powder warehouse loading, metering, and scheduling platform integrating machine vision and digital twins as described in claim 2, characterized in that, The method for forming the corresponding manifold deviation vector includes: The image sequence of tank opening and filling port is preprocessed, including denoising, image noise reduction, and image enhancement. In the preprocessed image sequence of tank opening and filling port, key feature points of the tank opening and filling port are identified, and the spatial position of the tank opening and filling port is described by the coordinates of the key feature points. The processed image of each frame in the image sequence of the tank opening and the loading port is mapped to a high-dimensional manifold space to represent the high-dimensional spatial relationship between the vehicle and the loading port. The local geometric and global structural information of each frame in the image sequence is preserved through the high-dimensional manifold space to represent the relative positional relationship between the tank opening and the loading port. In the high-dimensional manifold space, the spatial deviation between the center of the tank opening and the center of the loading port is calculated and a manifold deviation vector is formed.

4. The powder warehouse loading, metering, and scheduling platform integrating machine vision and digital twins as described in claim 3, characterized in that, The method for outputting the curvature change curve of the manifold deviation vector during the loading process includes: During the loading process, the curvature of the manifold deviation vector at each time point is arranged in chronological order to form a curvature sequence. Based on the curvature sequence, a curve of curvature changing with time is plotted along the time axis. This curve is the curvature change curve of the manifold deviation vector.

5. The powder warehouse loading, metering, and scheduling platform integrating machine vision and digital twins as described in claim 4, characterized in that, The method for obtaining the dynamic alignment index of the manifold deviation vector includes: Image features are mapped to different feature subspaces according to lighting conditions, shooting angle, and vehicle model differences. A corresponding set of invariant operators is constructed in each feature subspace to eliminate the influence of environmental factors and vehicle model differences on the manifold deviation vector. The manifold deviation vector is applied sequentially to the corresponding invariant operator group to obtain the corrected manifold deviation vector, which reflects the actual relative positional relationship between the vehicle tank opening and the loading port in the high-dimensional manifold space. Based on the corrected manifold deviation vector, the dynamic alignment index is calculated. The dynamic alignment index is updated over time to form a sequence of alignment state changes throughout the loading process.

6. The powder warehouse loading, metering, and scheduling platform integrating machine vision and digital twins as described in claim 5, characterized in that, The method for automatically triggering the adjustment of the loading strategy includes: The pre-built digital twin is a virtual model of the loading process, including the vehicle's tank opening location, loading port location, material flow characteristics, and loading strategy rules. Real-time evolution analysis is performed on the input dynamic alignment indicators and curvature change curves within the pre-built digital twin to determine whether there is any risk of misalignment or overflow during the loading process. The judgment criteria include curvature trend, curvature change rate, and deviation of alignment indicators from the preset alignment indicator safety threshold. When the digital twin determines that there is a risk of misloading or overflow, it automatically generates a loading strategy adjustment instruction. The loading strategy adjustment instruction is fed back to the powder warehouse loading metering and scheduling platform in real time, driving the preset loading actuator to make automatic adjustments. The adjustment instructions include modifying the material flow delivery speed or direction, adjusting the centering position of the loading port and vehicle tank opening, and controlling the loading sequence of specific areas.

7. The powder warehouse loading, metering, and scheduling platform integrating machine vision and digital twins as described in claim 6, characterized in that, The method of prioritizing the allocation of loading tasks to vehicles with the best positioning indicators includes: During the loading process, for each vehicle, the real-time scheduling control quantity is defined using the dynamic alignment index of the manifold deviation vector and the curvature change curve; by optimizing the objective function, the material flow is aligned with the tank opening during the loading process, minimizing the corrected modulus length of the deviation vector of all vehicles, so that the material flow is aligned with the tank opening during the loading process. The mean curvature of the curvature sequence of each vehicle is calculated to reflect the overall curvature of the vehicle's deviation trajectory; the dynamic alignment characterization is calculated to quantify the fluctuation range of the dynamic alignment index, reflecting the alignment stability of the vehicle under different working conditions; the mean curvature and the dynamic alignment characterization are combined to form a vehicle scheduling ranking index, and the loading task is preferentially assigned to the vehicle with the smallest vehicle scheduling ranking index.

8. The powder warehouse loading, metering, and scheduling platform integrating machine vision and digital twins as described in claim 7, characterized in that, The method for obtaining the manifold deviation database includes: Record the manifold deviation vector sequence of each vehicle during the loading process in chronological order, and store and classify the corresponding curvature curves and characteristic parameters in a hierarchical manner using vehicle number, vehicle type, loading batch and working conditions as index labels. The characteristic parameters include mean curvature, rate of change of curvature, fluctuation amplitude, and threshold exceedance marker. After each loading task is completed, the newly collected images of the tank opening and loading port are automatically stored in the database and compared with historical images of the tank opening and loading port to dynamically update the database and obtain the manifold deviation database.

Images