Intelligent scheduling-based vertical optical element tray-based collaborative access method and system

Abstract

The application discloses a vertical optical element tray-based collaborative access method and system based on intelligent scheduling, relates to the field of automatic storage of precise optical elements, and comprises the following steps: S1, constructing an intelligent collaborative control module comprising a three-level cache management module, a hierarchical mixed intelligent scheduling module and a collaborative timing control module; S2, generating an initial distribution point cloud graph matched with the optical element by a double-vision positioning module of a gantry manipulator; S3, generating a corresponding scheduling decision by a real-time response layer to obtain a cache position corresponding to a first-level cache loading / unloading port; S4, performing grabbing by the manipulator, placing the grabbed optical element in a transfer cache position of a second-level cache tray and updating RFID information and element count of the tray; and S5, judging whether the current tray is full, returning to S3 if the tray is not full, and triggering a tray-level transfer process if the tray is full. The application can make the system throughput no longer limited by the slowest operation cycle of a single device, and greatly improve the overall efficiency of the system.

Description

Technical Field

[0001] This invention relates to the field of automated warehousing technology for precision optical components. More specifically, this invention relates to a method and system for collaborative storage and retrieval of vertical optical components using a tray-based system based on intelligent scheduling. Background Technology

[0002] Vertical optical components (such as lenses and prisms) are characterized by a high center of gravity, easy tipping, and stringent surface precision requirements, making their efficient and secure storage and retrieval in automated warehousing a persistent technical challenge. Existing technologies mainly offer the following solutions, but all have significant drawbacks:

[0003] Option 1, independent storage location: (such as the intelligent storage system for photomasks disclosed in patent CN223238669U). This option equips each storage location with an independent indicator light and sensor, using the indicator light for location guidance. While achieving individualized indication, this results in a complex storage rack structure, high manufacturing costs, and low storage density. More importantly, this option requires independent detection and indication components for each storage location, leading to reduced system reliability and frequent maintenance.

[0004] Option 2, Simple Stacking Storage: This option involves directly stacking multiple vertical optical components on ordinary shelves or pallets. While this increases storage density to some extent, the lack of effective isolation and fixation between components makes them highly susceptible to collisions and surface scratches during handling and storage due to shaking and tipping, thus failing to guarantee the surface quality of the optical components.

[0005] Option 3, sorting indication based on polarized light: (such as the polarized light indication system and polarizer disclosed in patent CN113593445A). This option indicates the position of goods by setting polarizers with different polarization angles and a light source controller, aiming to reduce hardware costs and subsequent maintenance costs. However, this option mainly solves the problem of low-cost, extensive picking guidance. Its positioning accuracy is low and cannot meet the precise positioning and compliant gripping requirements required in the storage and retrieval of vertical optical components. At the same time, this option does not involve the design of a dedicated protective structure for the components themselves, and cannot solve the core problems of preventing tipping and collision of vertical optical components.

[0006] Option 4, direct gripping and placing with a robotic arm: This option uses a robotic arm to perform individual and continuous access operations on each component. While this option ensures operational accuracy, the robotic arm is used extremely frequently, leading to rapid mechanical wear, short maintenance cycles, and system throughput limited by the robotic arm's single operation cycle, resulting in low overall efficiency and making it difficult to meet the efficiency requirements of large-scale production scenarios.

[0007] Option 5, Traditional Cache Scheduling Algorithm: Existing automated warehousing systems mostly use fixed rules or simple heuristic algorithms for cache scheduling, which lacks the ability to adaptively learn access patterns and cannot achieve multi-objective optimization of storage efficiency, energy consumption and security under dynamically changing production needs.

[0008] In summary, existing technologies have failed to effectively resolve the contradictions between storage density and security, access efficiency and equipment complexity, and single-processing and batch handling in vertical optical component warehousing. Therefore, there is an urgent need in this field for an innovative technical solution that can achieve high-density, high-efficiency, low-cost, and easily maintained automated storage and handling while ensuring the surface quality of optical components and operational safety, and possess intelligent learning and adaptive optimization capabilities. Summary of the Invention

[0009] One object of the present invention is to solve at least the above-mentioned problems and / or defects, and to provide at least the advantages described below.

[0010] To achieve these objectives and other advantages of the present invention, a method for collaborative access to tray-based vertical optical components based on intelligent scheduling is provided, comprising:

[0011] S1. Construct an intelligent collaborative control module that includes a three-level cache management module, a hierarchical hybrid intelligent scheduling module, and a collaborative timing control module;

[0012] S2. The dual-vision positioning module of the gantry manipulator quickly scans the optical components on the intelligent handling robot AGV to generate an initial distribution point cloud map that matches the optical components by recognizing their position and quantity.

[0013] S3. Activate the real-time response layer of the hierarchical hybrid intelligent scheduling module based on the initial distribution point cloud map to generate the corresponding scheduling decision and obtain the cache position corresponding to the first-level cache inlet and outlet.

[0014] S4. The robotic arm performs corresponding grasping actions based on scheduling decisions, and places the optical components grasped from the inlet / outlet buffer position into the transfer buffer position corresponding to the secondary buffer tray, and updates the RFID information and component count of the tray.

[0015] S5. The hierarchical hybrid intelligent scheduling module determines whether the current pallet is full. If it is not full, it returns to S3. Otherwise, it triggers the pallet-level transfer process and allocates an optimal storage location for the full pallet on the storage rack of the three-level cache based on the coordination and optimization layer of the hierarchical hybrid intelligent scheduling module. It also performs empty pallet callback and system reset to complete one inbound operation.

[0016] Among them, the strategy learning layer of the hierarchical hybrid intelligent scheduling module runs continuously in the background and periodically collects data from the three-level cache in the current ingestion and historical jobs. It uses the gradient boosting decision tree model to train and update the component access frequency prediction model, and based on system performance monitoring data, it adaptively adjusts the weights in the real-time response layer and coordination optimization layer algorithms through the response surface methodology to achieve continuous optimization of the scheduling strategy.

[0017] Preferably, in S1, the collaborative timing control module decouples the operations of the robotic arm and the extraction vehicle in terms of timing to achieve parallel operation scheduling.

[0018] The hierarchical hybrid intelligent scheduling module includes:

[0019] A real-time response layer that processes burst requests in milliseconds using dynamic priority-based rapid decision-making rules.

[0020] Running on a second-level cycle, the cache scheduling problem is modeled as a coordination optimization layer of a mixed integer programming model;

[0021] Running on a minute-level timescale, a gradient boosting decision tree model is used to predict the probability of element access in the policy learning layer.

[0022] Preferably, the real-time response layer is triggered when a new access request arrives or when the system state undergoes a momentary change;

[0023] The decision logic of the real-time response layer is derived from an immediate decision-making model based on multi-factor weighted evaluation, and the urgency priority in the decision logic is... P ( r The calculation formula is:

[0024]

[0025] In the above formula, W 1. W 2. W 3 is the weighting coefficient, and , V ( r () represents the value coefficient. T d ( r () is the deadline. F ( r () represents the recent access frequency;

[0026] The execution action of the real-time response layer is based on priority, and it immediately judges the access order of optical components and their storage location in the first and second level caches to obtain the corresponding scheduling decision.

[0027] Preferably, the coordination optimization layer is triggered periodically at fixed time intervals, with the goal of minimizing the total system cost, and constrained by buffer capacity, robotic arm load, and operation continuity. It employs a greedy random adaptive search process to obtain the corresponding optimization instructions.

[0028] Preferably, the gradient boosting decision tree model constructs multiple weak decision trees iteratively, with each weak decision tree... h m Based on input features x i Make a prediction, the final predicted probability. It is characterized by the following formula:

[0029]

[0030] In the above formula, As weight, M For the number of decision trees, m It is the index variable of the decision tree.

[0031] Preferably, the system performance monitoring data includes: average response time, hit rate, and energy consumption;

[0032] The response surface methodology establishes the algorithm's internal parameters using the following formula. θ A second-order polynomial prediction model is used to compare the system performance monitoring data with the predicted system performance values. :

[0033]

[0034] In the above formula, β 0、 β i , β ij These are the coefficients of a second-order polynomial prediction model derived from sample fitting. d This represents the total number of internal parameters of the algorithm. θ i and θ j These represent the internal parameters. θ The first in i The and the first j One parameter, β ij θ i θ j Used to describe the i The parameter and the first j The nonlinear effect of the interaction between parameters on system performance;

[0035] The policy learning layer synchronizes the updated second-order polynomial prediction model and optimized parameters to the real-time response layer and the policy learning layer.

[0036] A vertical optical component tray-based collaborative access system includes:

[0037] A tray with RFID tags used to hold and fix optical components;

[0038] A robotic arm used to perform precise grasping and placement operations on individual components;

[0039] A pickup vehicle used to perform bulk storage and horizontal transfer of the entire pallet;

[0040] An intelligent collaborative control module for implementing three-level cache management and intelligent scheduling;

[0041] The tray includes:

[0042] The U-shaped support plate has multiple slots arranged in a U-shape in space.

[0043] Multiple limiting plates that mate with the card slot, each limiting plate having a guiding V-shaped groove on its insertion surface that mates with the optical element;

[0044] The bearing plate has multiple bearing areas separated by limiting plates, and each bearing area has multiple levels of support steps of different heights on both sides.

[0045] The present invention has at least the following beneficial effects:

[0046] Firstly, this invention breaks through the bottleneck of serial operation of equipment in traditional automated warehousing, and for the first time establishes a time-series decoupling model between "precise grasping by robotic arms (single-piece level)" and "batch transfer by retrieval carts (pallet level)" in the field of vertical optical component storage and retrieval. By setting up a physical three-level buffer (inbound / outbound port, pallet transfer area, and shelf storage area), the two different granularity operation processes are parallelized, so that the system throughput is no longer limited by the slowest operation cycle of a single device, and a significant improvement in the overall system efficiency is achieved.

[0047] Secondly, to address the challenge of multi-objective optimization scheduling in dynamic environments, this invention innovatively constructs a hierarchical hybrid intelligent scheduling algorithm framework. This framework comprises: a real-time decision-making layer with millisecond-level response, handling bursty requests based on a dynamic priority formula; a coordination optimization layer with second-level cycles, employing a Greedy Randomized Adaptive Search Process (GRASP) to solve a mixed-integer programming model for local layout optimization; and a policy learning layer with minutes or more, applying a Gradient Boosting Decision Tree Model (GBDT) to predict access frequency and adaptively adjusting algorithm parameters using the Response Surface Method (RSM). These three layers of algorithms communicate and coordinate their decisions, achieving an intelligent scheduling scheme that evolves from instantaneous response to long-term evolution.

[0048] Thirdly, this invention upgrades the traditional rule-based static storage location allocation to a dynamic self-optimization strategy based on data prediction. The system utilizes the prediction results of the future access probability of components from the strategy learning layer ( The system guides and coordinates the optimization layer to continuously organize storage locations in the background, dynamically migrating high-frequency components to lower-level areas with better access paths. This allows the system's storage layout to proactively adapt to changes in production rhythm, maintaining optimal access performance over the long term.

[0049] Fourth, the tray of this invention abandons the traditional passive protection or general load-bearing approach, and innovatively proposes a dedicated tray structure that integrates stepped load-bearing, V-shaped guide limiting, spring-loaded clamping, and real-time pressure sensing. This design actively adapts to and constrains vertical optical components of different sizes from a physical principle perspective, realizing active protection against tipping, collision, and overpressure damage during storage, and elevating component safety to a deterministic level.

[0050] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0051] Figure 1 This is a schematic diagram illustrating the working principle of the three-level caching mechanism used in this invention.

[0052] Figure 2 This is a complete flowchart of the optical component storage method of the present invention;

[0053] Figure 3 This is a schematic diagram of the overall structure of the vertical optical element tray-based collaborative access system in this invention;

[0054] Figure 4 This is a schematic diagram of the tray subsystem in this invention;

[0055] Figure 5 for Figure 4 A diagram from another perspective;

[0056] Figure 6 This is a schematic diagram of the robotic arm in this invention;

[0057] Figure 7 This is a schematic diagram of the structure from which the vehicle is extracted in this invention;

[0058] Among them, 1-cabinet, 2-retrieval vehicle, 20-dual ground rails, 21-forklift device, 22-mechanical positioning ruler, 3-inlet / outlet, 4-AGV, 5-pallet, 50-steps, 51-slot, 52-limiting plate, 6-robotic arm, 60-truss mechanism, 61-global camera, 62-macro camera, 63-adaptive gripper mechanism, 7-storage rack location, 8-optical components. Detailed Implementation

[0059] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0060] A collaborative access method for tray-based vertical optical components based on intelligent scheduling includes the following core processes:

[0061] 1. Palletized Batch Storage Process

[0062] a) Component feature digitization and classification: Based on the component size, weight, value and historical access frequency, the ABC classification method is used for initial classification, and the classification results are converted into digital feature vectors as input to the scheduling algorithm.

[0063] b) Pallet Configuration and Information Binding: Based on the physical specifications of the components, the system automatically selects and assembles appropriate limit modules. RFID tags embedded in the pallet are used to write or update information such as the component list and access timestamps in real time, achieving synchronization between information flow and logistics.

[0064] c) Pre-allocation and dynamic adjustment of storage locations: The intelligent scheduling algorithm comprehensively considers the real-time access queue, the status of each buffer bit, and the prediction of the future access frequency of components to dynamically allocate appropriate storage locations for each pallet or component. High-frequency access components are allocated to lower-level storage locations with shorter access paths, while low-frequency components are stored in higher-level locations.

[0065] 2. Robotic arm gripping and placement process

[0066] a) Visual hierarchical positioning and pose calculation: First, the approximate position of the component is quickly identified by the visual camera, and then the macro camera is guided to accurately position the target component and calculate its precise three-dimensional coordinates and pose.

[0067] b) Adaptive compliant gripping: The grippers automatically adjust the gripping force within a preset safety force range based on the size characteristics of the target component. During gripping, pressure sensors provide real-time feedback, forming a closed-loop force control to prevent pinching or slippage.

[0068] c) Motion trajectory optimization and collision avoidance planning: Based on the current position, target position and environmental obstacle information, the robot controller plans a smooth, efficient and collision-free motion trajectory in three-dimensional space, and smooths the acceleration to ensure stable motion.

[0069] 3. Multi-level caching collaborative operation process

[0070] a) Primary buffer (inbound / outbound) operation: Serving as the interface between the system and external AGVs, this buffer buffer is used to buffer mismatches in their operating cycles. This buffer has a small capacity and extremely fast turnover; its core function is to "smooth out peaks and fill valleys," ensuring continuous operation of the robotic arm.

[0071] b) Secondary Buffer (Panel Transfer Area) Operation: This acts as a "decoupling buffer" between the robotic arm and the retrieval vehicle. The robotic arm places components one by one into the tray in this area. When the tray is full, the retrieval vehicle transfers the entire tray at once. This enables parallel operations of "fine-grained grasping" and "coarse-grained handling," greatly improving system throughput.

[0072] c) Level 3 Cache (Shelf Storage Area) Operation: As a long-term storage area, it adopts "dynamic and static partitioning" management. Based on the component usage frequency predicted by intelligent algorithms, its distribution on the upper (low-frequency zone) and lower (high-frequency zone) shelves is dynamically adjusted, so that high-frequency components are always in easily accessible positions, thereby reducing the average access time.

[0073] 4. Workflow of the intelligent scheduling algorithm

[0074] The core innovation of this workflow lies in the embedded hierarchical hybrid intelligent scheduling algorithm. This algorithm consists of three layers from top to bottom, working collaboratively, such as... Figure 1 As shown, and described in detail below:

[0075] First layer: Real-time response layer

[0076] (1) Trigger: Triggered when a new access request arrives or when the system state changes instantaneously.

[0077] (2) Decision-making logic: An immediate decision-making model based on multi-factor weighted evaluation is adopted. For request r, its urgency priority is determined. P ( r From the formula The calculation shows that, V ( r () represents the value coefficient. Td ( r () is the deadline. F ( r () indicates the recent access frequency.

[0078] W 1. W 2. W 3 is the weighting coefficient, and satisfies .

[0079] (3) Action: Based on priority, immediately determine the access order of components and their storage location in the L1 and L2 caches. This decision is completed in milliseconds, with a computational complexity of O(n log n). ,in, This is the complexity calculation function, where N is the number of components to be processed, ensuring agile system response.

[0080] Second layer: Coordination and optimization layer

[0081] (1) Trigger: Triggered periodically at fixed time intervals.

[0082] (2) Decision logic: The scheduling problem in a short future period is modeled as a mixed integer programming model. The model aims to minimize the total system cost (such as comprehensive access time, energy consumption, and equipment wear and tear), and is constrained by factors such as cache capacity, robotic arm load, and operation continuity.

[0083] (3) Solution and Actions: A Greedy Randomized Adaptive Search (GRASP) process is used for the solution. In each iteration, the algorithm first constructs a candidate list containing high-quality operations based on the current state, then randomly selects a candidate operation to execute, followed by local search improvements. Through multiple iterations, the algorithm gradually approaches a better scheduling scheme. The solution output is a series of optimized instructions, such as adjusting the position allocation of components in the buffer and optimizing the movement path of the robotic arm. This optimization is completed in seconds, with a computational complexity of O(n log n). ,in, I max Number of iterations K Where N is the number of storage locations and N is the number of components to be processed. L search It is the local search depth, which aims to eliminate the local optima that may be caused by real-time decision-making and improve the system's medium-term efficiency.

[0084] Third layer: Strategy learning layer

[0085] (1) Trigger: Trigger at longer time intervals, or after enough new data has been accumulated.

[0086] (2) Decision-making logic:

[0087] a. Demand Forecasting: The probability of element access is predicted using a Gradient Boosting Decision Tree (GBDT) model. This model iteratively builds multiple weak decision trees and weights-sums their predictions to form a powerful forecasting model. Specifically, each weak decision tree... h m Based on input features x i (Including component attributes, historical access characteristics, time characteristics, etc.) are used for prediction, and the final prediction probability is ,in, As weight, M For the number of decision trees, m This is the index variable of the decision tree. The computational complexity of model training is... ,in, T The number of training samples. h This represents the maximum depth of the decision tree.

[0088] b. Strategy Evaluation and Parameter Optimization: Monitor system performance metrics (such as average response time, hit rate, and energy consumption). Utilize Response Surface Methodology (RSM) to analyze algorithm internal parameters (such as priority weights). W i The relationship between cost function coefficients and performance indicators.

[0089] RSM establishes a second-order polynomial prediction model between parameters and performance:

[0090]

[0091] It should be noted that the second-order polynomial prediction model is a vector used to approximate the internal parameter vector of the algorithm. θ A surrogate model representing the functional relationship between the system and overall performance metrics. System performance monitoring data (such as average response time, cache hit rate, energy consumption, etc.) serves as the input to the second-order polynomial prediction model. β 0、 β i , β ij These are the coefficients of a second-order polynomial prediction model fitted from the samples, used to transform fragmented performance monitoring data into an analytical function form. d This represents the total number of internal parameters of the algorithm. θ i and θ j These represent the internal parameters. θ The first in i The and the first j One parameter, β ij θ i θ j Used to describe the i The parameter and the first j The interaction effects between parameters have a nonlinear impact on system performance; the gradient ascent method is used to find the optimal parameter combination direction and automatically adjust the parameters to approach optimal performance. For example, if a warehousing system has two adjustable parameters: θ 1 = Component value weight, θ 2 = Deadline urgency weight, in order to know θ 1. θ 2. When considering the combined effects on system throughput, if only the individual effects (linear terms) are taken into account: Increase θ 1 will increase throughput and increase θ 2 will also increase throughput, and the two effects can be simply added together. However, in practice, increasing both simultaneously may actually lead to a decrease in throughput. This is because the system doesn't know which to prioritize; therefore, this "1+1≠2" effect is called interaction. θ i θ j This product term is designed to capture this "additional effect on the result when both parameters change together." That is, when the coefficients... β ij If the number is positive, it indicates that the two parameters have a synergistic effect; if the number is negative, it indicates that they have a mutually canceling or competing effect.

[0092] (3) Action: Synchronize the updated prediction model and optimized parameters to the first and second layers to make their decisions forward-looking. For example, move components predicted to "become more frequent" from the upper layer to the lower layer in advance.

[0093] Example 1

[0094] This embodiment mainly describes the overall structure and collaborative operation mode of the vertical optical element tray-based collaborative storage system.

[0095] like Figure 3 As shown, the system mainly consists of three hardware components: a pallet, a robotic arm, and a retrieval vehicle. These three components work around the cabinet 1, which has storage rack positions 7. Specifically, the retrieval vehicle 2 operates within the cabinet 1, cooperating with the robotic arm 6 to retrieve the pallet 5 from the storage rack positions 7. The interaction between the AGV 4 and the inlet / outlet is achieved from the location of the inlet / outlet 3. The actions of each component are coordinated under the unified scheduling of the intelligent collaborative control module. The following is a separate description of each component of the system:

[0096] 1. The tray is the core carrier for supporting and fixing optical components.

[0097] like Figures 4-5 As shown, the dedicated tray 5 is made of high-strength lightweight material. The bottom of the tray is designed with multiple steps 50 to accommodate optical components of different heights and ensure their center of gravity is stable. The tray is set with a U-shaped structure and is provided with a standardized slot 51 that is adapted to the U-shaped structure for installing or quickly replacing the limiting plate 52. The limiting plate is made of wear-resistant material, and the V-shaped groove on the inner side of the limiting plate 52 can guide the component to automatically center.

[0098] An RFID tag is embedded on the side of the tray, serving as its unique digital identity in the system. The RFID tag integrates the storage tray ID, capacity specifications, and usage history.

[0099] 2. The robotic arm is mainly responsible for performing precise grasping and placement operations on individual optical components.

[0100] like Figures 5-6 As shown, the robotic arm mainly includes:

[0101] A gantry mechanism 60 with high repeatability positioning accuracy, employing multi-axis servo control, is used to provide a wide range of high-precision motion capabilities;

[0102] The dual-vision positioning module includes a global camera 61 and a macro camera 62. The dual-vision positioning module is used to achieve hierarchical positioning and provide operation process guidance for the gantry robot. First, the global camera quickly locates the approximate position of the component group on the AGV within a large field of view. Then, the macro camera performs local high-precision positioning of the target optical component to obtain its precise three-dimensional coordinates and attitude information, guiding the robot's end effector (adaptive gripper) to the correct position.

[0103] The gripping force is adjustable, and the gripping surface is covered with a low-friction coefficient material. The adaptive gripping mechanism 63 automatically adjusts the gripping force according to the component size and forms a closed-loop control with real-time pressure feedback to ensure a smooth and reliable gripping process. It should be noted that the structure and internal working method of the robot are existing technologies, so its structural equipment connection relationship and working principle will not be described here.

[0104] 3. The retrieval vehicle is mainly responsible for performing batch storage, retrieval, and horizontal transfer of the entire pallet.

[0105] like Figure 7 As shown, the extraction vehicle adopts a robust double-column and double-rail structure, which has high load capacity and good operational stability.

[0106] The forklift device 21 of the pickup truck is equipped with a high-performance servo drive and anti-shake control algorithm for closed-loop control to achieve smooth acceleration, deceleration and precise stopping.

[0107] The retrieval vehicle employs a dual verification method combining a barcode reader and a mechanical positioning ruler 22 for multiple positioning safeguards, ensuring millimeter-level accuracy in docking with the shelf location. It should be noted that the structure and internal workings of the retrieval vehicle are existing technologies, therefore its structural equipment connections and working principles will not be elaborated here.

[0108] The three hardware components mentioned above do not operate independently, but are deeply integrated and scheduled through an intelligent collaborative control module. This intelligent collaborative control module is the "brain" of the entire system. On one hand, it receives data from all sensors (such as vision, RFID, pressure, and position data; vision data is used for component positioning, identification, and counting, providing real-time response layer for grasping decisions, while historical vision data accumulation is used for pattern recognition in the strategy learning layer; RFID data is used for pallet identification, component list confirmation, and location tracking, providing coordination and optimization layer for pallet scheduling and location allocation, and also for updating inventory records; pressure sensor data is used for gripper force control closed-loop, providing real-time feedback to the robot controller to ensure compliant grasping; position sensor data is used for precise positioning and stroke control of the retrieval cart and robot, providing path planning to the coordination and optimization layer). On the other hand, it runs a core hierarchical hybrid intelligent scheduling algorithm, sending coordinated and orderly control commands to the robot and retrieval cart based on real-time status and task objectives, achieving optimal utilization of hardware resources and the most efficient connection of work processes.

[0109] Example 2

[0110] Combination Figure 2 The flowchart below, taking component warehousing as an example, details the execution steps of this method:

[0111] S1, Task Initialization

[0112] The AGV transports the vertical optical components to the inlet / outlet, and uses a QR code for secondary precise positioning. The system receives the task instruction and confirms that the component batch information matches the preset task order.

[0113] S2, Vision system pre-scan

[0114] The global positioning camera of the gantry robot quickly scans the components on the AGV, identifies the approximate location and quantity of the components, and generates an initial point cloud map of the component distribution, providing coarse guidance for subsequent fine grasping.

[0115] S3, Intelligent Scheduling Decision Trigger

[0116] The real-time response layer of the intelligent scheduling algorithm is activated. For the first target element identified by the scan, the algorithm determines its value based on its preset value. V ( r ), Task urgency (based on deadline)T d ( r (reciprocal representation of) and historical access frequency F ( r ), through formula Calculate its dynamic operation priority P ( r Based on this priority and the real-time occupancy status of the first-level cache (inbound / outbound port), the algorithm instantly determines which specific pallet and which limit point the component should be placed in.

[0117] S4, Fine Positioning and Adaptive Grasping

[0118] Based on the scheduling decision, the macro-precision positioning camera accurately locates the target component, acquiring its three-dimensional coordinates and orientation. The adaptive gripper calls upon the preset clamping force according to the component's size characteristics and performs compliant gripping at a smooth approach speed. During the gripping process, the pressure sensor provides real-time feedback, forming a force closed-loop control.

[0119] S5, Cache Placement and State Update

[0120] The robotic arm precisely places the successfully grasped components into the inlet / outlet buffer positions designated by the S3 decision. After placement, the system automatically updates the RFID information and component count of the pallet.

[0121] S6, Loop Decisions and Flow Branches

[0122] The system determines whether the current pallet is full (usually 5 components). If it is not full, it returns to step S2 and continues to grab the next component on the AGV; if it is full, it triggers the pallet-level transfer process and proceeds to step S7.

[0123] S7, retrieval vehicle collaborative disk retrieval

[0124] Upon receiving the signal that the pallet is full, the pickup vehicle moves to the inlet / outlet. Through dual verification using its barcode reader and the mechanical scale of the pallet location, precise alignment with the buffer position is achieved. The forks extend smoothly, retrieving the full pallet.

[0125] S8, Intelligent Storage Location Allocation and Inbound

[0126] The coordinated optimization layer algorithm is invoked. This algorithm constructs a mixed-integer programming model based on the current system state (such as the occupancy of each buffer zone, shelf space distribution, and equipment location) with the objective of minimizing global access costs, and uses a greedy stochastic adaptive search process (GRASP) for efficient solution. The solution assigns an optimal storage location to the fully loaded pallet (e.g., the low-level high-frequency zone, column B, level 2). The retrieval vehicle then transports the pallet to the designated storage location according to this instruction and completes the storage.

[0127] S9, Empty Tray Recall and System Reset

[0128] After completing the storage of the current pallet, the retrieval vehicle immediately retrieves an empty pallet from the empty pallet buffer slot in the third-level buffer area according to the scheduling instructions, and transports and places it in an available buffer slot at the inlet / outlet. The system resets, ready to receive the next batch of components delivered by the AGV or to begin executing an outbound task.

[0129] S10: Background Learning and Parameter Tuning

[0130] While the hardware executes the above process, the policy learning layer continues to run in the background. It periodically collects data from the current and historical tasks, and uses the Gradient Boosting Decision Tree (GBDT) model to train and update the component access frequency prediction model. Meanwhile, based on system performance monitoring data, the key parameters (such as weights) in the real-time response layer and coordination optimization layer algorithms are adaptively adjusted using the response surface methodology (RSM). W 1. W 2. W 3) Achieve continuous optimization of scheduling strategies.

[0131] Example 2 primarily illustrates a hierarchical closed-loop control system to ensure operational reliability. Specifically, addressing the stringent reliability requirements of precision optical components, this invention constructs a multi-level control system encompassing "global positioning, local fine-tuning, and force-position closed-loop." From dual-camera vision-based positioning to real-time feedback adjustment from the gripper force sensor, a complete perception-decision-execution closed loop is formed. The real-time response layer of the intelligent scheduling algorithm makes millisecond-level decisions based on dynamically changing system states, ensuring that each operational command is based on the latest and most accurate environmental information. This minimizes accidental errors in the operation process, guaranteeing extreme reliability and repeatability.

[0132] Example 3

[0133] This embodiment focuses on how the intelligent scheduling algorithm works in deep collaboration with the physical three-level cache structure to achieve dynamic optimization of system performance. For example... Figure 1 , Figure 4 As shown:

[0134] Real-time scheduling of the first-level cache (inbound / outbound): This cache has the smallest capacity and directly faces the uncertainty of external input. The real-time response layer is specifically responsible for managing this area, and its core function is to ensure response speed. It performs immediate priority assessment and decision-making for each arriving component or task, ensuring that high-priority tasks are processed immediately and determining the temporary storage location of the component. Its decision complexity is... It provides a millisecond-level response time.

[0135] Periodic optimization of the second-level cache (tray transit area): This area is the core of job decoupling. The coordination optimization layer plays a crucial role here, periodically reviewing the status of all transit trays. This is achieved by establishing and solving a scheduling optimization model (using the GRASP algorithm, with a complexity of O(n log n)). (with a response time of seconds), it determines the order of pallet access, its placement in the buffer area, and its allocation to the retrieval vehicle. The goal is to maximize the parallel working efficiency of the robotic arm and the retrieval vehicle while minimizing equipment waiting time.

[0136] Long-term prediction and optimization of the Level 3 cache (shelf storage area): This area is the system's long-term storage repository. The policy learning layer provides strategic guidance here. It utilizes the GBDT model (training complexity is O(n log n)). The system predicts the future frequency of each component (with a response time of minutes). Based on the predictions, the coordination and optimization layer dynamically adjusts high-frequency components to lower-level storage locations with convenient access, and archives low-frequency components to higher-level storage locations when performing storage or sorting tasks. This prediction-based dynamic storage location allocation strategy is key to reducing the system's long-term average access time.

[0137] Example 3 primarily addresses the problem in traditional solutions where the precise grasping of robotic arms and the batch transfer of materials by handling equipment are typically performed sequentially, resulting in significant idle waiting time. This example creatively designs a three-level buffer architecture, decoupling and parallelizing the workflow of the robotic arm and the retrieval cart. The robotic arm can focus on filling pallets in the intermediate buffer area, while the retrieval cart independently performs batch storage and retrieval of fully loaded pallets. The intelligent scheduling algorithm driving this process uses a coordination optimization layer that periodically solves a mixed integer programming model (using the greedy stochastic adaptive search process GRASP for efficient solution), continuously optimizing the flow path and timing of pallets between the buffer area and the shelf, thereby improving the overall equipment utilization and material flow efficiency of the system.

[0138] Through detailed explanations of three embodiments, this invention reveals how it combines innovative modular hardware design with advanced hierarchical intelligent scheduling algorithms to construct a secure, efficient, intelligent, and scalable automated access system for vertical optical components. Under the precise scheduling of the algorithm, each hardware component works collaboratively as an organic whole, while the algorithm itself continuously learns and optimizes, ultimately achieving a comprehensive improvement in storage security, access efficiency, and system intelligence.

[0139] The above solution is merely an illustration of a preferred example and is not limited thereto. When implementing this invention, appropriate substitutions and / or modifications can be made according to the user's needs.

[0140] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. It can be applied to various fields suitable for the present invention. Other modifications can be readily made by those skilled in the art. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and examples shown and described herein.

Claims

1. A method for collaborative access to tray-based vertical optical components based on intelligent scheduling, characterized in that, include: S1. Construct an intelligent collaborative control module that includes a three-level cache management module, a hierarchical hybrid intelligent scheduling module, and a collaborative timing control module; S2. The dual-vision positioning module of the gantry manipulator quickly scans the optical components on the intelligent handling robot AGV to generate an initial distribution point cloud map that matches the optical components by recognizing their position and quantity. S3. Activate the real-time response layer of the hierarchical hybrid intelligent scheduling module based on the initial distribution point cloud map to generate the corresponding scheduling decision and obtain the cache position corresponding to the first-level cache inlet and outlet. S4. The robotic arm performs corresponding grasping actions based on scheduling decisions, and places the optical components grasped from the inlet / outlet buffer position into the transfer buffer position corresponding to the secondary buffer tray, and updates the RFID information and component count of the tray. S5. The hierarchical hybrid intelligent scheduling module determines whether the current pallet is full. If it is not full, it returns to S3; otherwise, it triggers the pallet-level transfer process and allocates an optimal storage location for the full pallet on the storage rack of the three-level cache based on the coordination and optimization layer of the hierarchical hybrid intelligent scheduling module. It also performs empty pallet callback and system reset to complete one inbound operation. Among them, the strategy learning layer of the hierarchical hybrid intelligent scheduling module runs continuously in the background and periodically collects data from the three-level cache in the current ingestion and historical jobs. It uses the gradient boosting decision tree model to train and update the component access frequency prediction model, and based on system performance monitoring data, it adaptively adjusts the weights in the real-time response layer and coordination optimization layer algorithms through the response surface methodology to achieve continuous optimization of the scheduling strategy.

2. The method for collaborative access to vertical optical components based on intelligent scheduling as described in claim 1, characterized in that, In S1, the collaborative timing control module decouples the operations of the robotic arm and the extraction vehicle in terms of timing to achieve parallel operation scheduling. The hierarchical hybrid intelligent scheduling module includes: A real-time response layer that processes burst requests in milliseconds using dynamic priority-based rapid decision-making rules. Running on a second-level cycle, the cache scheduling problem is modeled as a coordination optimization layer of a mixed integer programming model; Running on a minute-level timescale, a gradient boosting decision tree model is used to predict the probability of element access in the policy learning layer.

3. The method for collaborative access to vertical optical components based on intelligent scheduling as described in claim 2, characterized in that, The real-time response layer is triggered when a new access request arrives or when the system state undergoes a momentary change; The decision logic of the real-time response layer is derived from an immediate decision-making model based on multi-factor weighted evaluation, and the urgency priority in the decision logic is... P ( r The calculation formula is: In the above formula, W 1 、W 2 、W 3 is the weighting coefficient, and , V ( r () represents the value coefficient. T d ( r () is the deadline. F ( r () represents the recent access frequency; The execution action of the real-time response layer is based on priority, and it immediately judges the access order of optical components and their storage location in the first and second level caches to obtain the corresponding scheduling decision.

4. The method for collaborative access to vertical optical components based on intelligent scheduling as described in claim 2, characterized in that, The coordination optimization layer is triggered periodically at fixed time intervals, with the goal of minimizing the total system cost. It is constrained by buffer capacity, robotic arm load, and operation continuity, and uses a greedy random adaptive search process to obtain the corresponding optimization instructions.

5. The method for collaborative access to vertical optical components based on intelligent scheduling as described in claim 2, characterized in that, The gradient boosting decision tree model constructs multiple weak decision trees iteratively, each weak decision tree... h m Based on input features x i Make a prediction, the final predicted probability. It is characterized by the following formula: In the above formula, As weight, M For the number of decision trees, m It is the index variable of the decision tree.

6. The method for collaborative access to tray-based vertical optical components based on intelligent scheduling as described in claim 2, characterized in that, System performance monitoring data includes: average response time, hit rate, and energy consumption; The response surface methodology establishes the algorithm's internal parameters using the following formula. θ A second-order polynomial prediction model is used to compare the system performance monitoring data with the predicted system performance values. : In the above formula, β 0、 β i , β ij These are the coefficients of a second-order polynomial prediction model derived from sample fitting. d This represents the total number of internal parameters of the algorithm. θ i and θ j Representing internal parameters θ The first in i The and the first j One parameter, β ij θ i θ j Used to describe the i The parameter and the first j The nonlinear effect of the interaction between parameters on system performance; The policy learning layer synchronizes the updated second-order polynomial prediction model and optimized parameters to the real-time response layer and the policy learning layer.

7. A tray-based collaborative access system for vertical optical components, applied to the intelligent scheduling-based tray-based collaborative access method for vertical optical components as described in any one of claims 1-6, characterized in that, include: A tray with RFID tags used to hold and fix optical components; A robotic arm used to perform precise grasping and placement operations on individual components; A pickup vehicle used to perform bulk storage and horizontal transfer of the entire pallet; An intelligent collaborative control module for implementing three-level cache management and intelligent scheduling; The tray includes: The U-shaped support plate has multiple slots arranged in a U-shape in space. Multiple limiting plates that mate with the card slot, each limiting plate having a guiding V-shaped groove on its insertion surface that mates with the optical element; The bearing plate has multiple bearing areas separated by limiting plates, and each bearing area has multiple levels of support steps of different heights on both sides.

Images