Intelligent management method and system for intelligent cabinet material taking and returning

Abstract

The present application relates to the technical field of intelligent cabinet management, and particularly relates to an intelligent management method and system for material taking and returning of an intelligent cabinet. A dynamic updating material storage management library is formed by acquiring real-time tracking trajectories of material taking and returning in a storage cabinet, constructing a multi-layer space-time memory layer and a space-time index layer. Based on the material attribute characteristics driven by a work order, fuzzy membership retrieval and feature splitting identification are performed on the storage cabinet to accurately determine the target storage cabinet. In combination with the weight difference obtained by a weight sensor, a dynamic radial basis function and a local aggregation window are introduced to robustly accumulate the material quantity. In the material taking and returning scenarios, an invariant accumulation data node and a time-varying accumulation data node are respectively constructed to realize adaptive calculation and real-time display of the material quantity. The present application realizes cabinet opening positioning guidance and weight sensing counting prompt for material taking and returning of an intelligent cabinet, and improves the material positioning accuracy, measurement reliability and dynamic management capability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent cabinet management, and particularly relates to an intelligent management method and system for material taking and returning of an intelligent cabinet. BACKGROUND

[0002] With the development of manufacturing, maintenance and warehouse management towards fine and informatization, intelligent cabinets are widely used in the storage and management of tools, spare parts and processing materials. The existing intelligent cabinet system usually adopts the mode of fixed grid binding materials, manual code scanning or simple weight threshold judgment to realize the management of material taking and returning, which reduces the cost of manual registration to a certain extent, but there are still obvious deficiencies in actual application. The existing intelligent cabinet management technology generally lacks continuous tracking and space-time modeling of the whole process of material taking and returning, and the material state update depends on the single operation result, which is easy to cause inventory information lag or distortion due to operation omission, abnormal taking and placing or multiple concurrent operations. Secondly, it is difficult to accurately locate the target material storage cabinet according to the material demand of the work order, and it needs to rely on manual experience memory to find, which is easy to cause confusion of the location of material taking or returning. At the same time, in the face of multi-category, multi-specification, similar shape or small volume materials, it is difficult to accurately distinguish the material category and quantity by relying on manual counting confirmation or static weight comparison, which is easy to cause the problems of wrong taking, missing recording or wrong returning, especially in the scene of small material and high frequency taking and returning. In addition, the existing system usually adopts the same metering logic for material taking and returning, and cannot distinguish the weight fluctuation characteristics, lacks effective inhibition mechanism for weight fluctuation, sensor noise and operation disturbance, resulting in insufficient accuracy of material taking and returning. SUMMARY

[0003] The present application overcomes the deficiencies of the prior art and provides an intelligent management method and system for material taking and returning of an intelligent cabinet.

[0004] To achieve the above purpose, the technical solution adopted by the present application is as follows:

[0005] The present application provides an intelligent management method for material taking and returning of an intelligent cabinet, comprising the following steps:

[0006] S1: obtaining the real-time tracking trajectory of material taking and returning in the storage cabinet, constructing a multi-layer space-time memory layer and a space-time index layer of the intelligent cabinet storage partition, inserting the real-time tracking trajectory into the space-time memory layer and dynamically updating and maintaining the space-time index layer, and obtaining a dynamic management library of the intelligent cabinet material storage;

[0007] S2: input the work order of the materials and equipment in the intelligent cabinet, perform fuzzy membership retrieval on the space storage of each storage cabinet based on the attribute characteristics of the required processing materials in the work order, import the dynamic management library, and determine the candidate storage cabinet, and split and identify the current material characteristics in the candidate storage cabinet according to the constraint feature vector of the processing material to determine the target storage cabinet for taking or returning the processing material;

[0008] S3: obtain the weight difference of the material in the target storage cabinet through the weight sensor, preset the dynamic radial basis function of the observation weight scale based on the weight difference, and plan the local aggregation window of the observation real-time weight difference change in the constant cumulative volume field of the total weight based on the dynamic radial basis function;

[0009] S4: if the target storage cabinet performs the taking operation, hash the independent volume characteristics of the to-be-taken processing material to determine the invariant cumulative data node, perform weighted accumulation of the quantity scale based on the local aggregation window of the standard weight value mapping value of the to-be-taken processing material, and determine the taking quantity uploaded to the grid screen terminal display;

[0010] S5: if the target storage cabinet performs the returning operation, predict the current volume characteristics of the to-be-returned processing material through the dynamic factor, determine the time-varying cumulative data node based on the current volume characteristics and map it to the value local aggregation window for quantity accumulation, and determine the return quantity uploaded to the grid screen terminal display.

[0011] More specifically, the S1 specifically includes the following steps:

[0012] Obtain the storage management log of the intelligent cabinet, and extract the real-time tracking trajectory of each storage cabinet in the intelligent cabinet for material storage and return at consecutive time nodes through the storage management log;

[0013] Obtain the structural design drawing of the intelligent cabinet and the cabinet opening number system of the storage cabinet, and set the pattern and three-dimensional distribution of the storage cabinet of the intelligent cabinet in the PROE model software according to the structural design drawing to form a plurality of discrete partitions;

[0014] Based on the plurality of discrete partitions, construct a multi-layer space-time memory layer and a corresponding space-time index layer of the LSM log architecture storage partition of the intelligent cabinet, and insert the real-time tracking trajectory into each space-time memory layer according to the space partition code recorded by the cabinet opening number system;

[0015] In the insertion process, a dynamic log update mechanism is introduced to maintain the local space index of the corresponding space-time index layer in each space-time memory layer, and output the real-time index memory value of the material storage partition tracking; wherein the local space index includes real-time range query, adjacent partition query and space filtering of material storage;

[0016] If the real-time index memory value is greater than the preset memory threshold, the spatial index in the space-time memory layer is serialized at this time, so that the space-time memory is refreshed as an ordered and immutable space-time index, and a space-time index file of the space-time index layer for dynamic storage tracking of the material is generated;

[0017] The above steps are repeated to obtain the space-time index file of each space-time index layer. The spatial index files of adjacent space-time index layers are read according to the order matrix of the spatial partition coding and reorganized, and the old storage index is replaced, so that a dynamic management library of the intelligent cabinet material storage tracking index is obtained.

[0018] More specifically, the S2 specifically includes the following steps:

[0019] By using face recognition or verification code on the interactive screen terminal of the intelligent cabinet, the material work order or equipment work order is inputted after authentication login for material processing or material returning;

[0020] The structure design drawing of the intelligent cabinet is analyzed and drawn by PROE model software to construct a space storage model of the intelligent cabinet, and the sub-space storage units in which the storage cabinet grids are located are divided on the space storage model;

[0021] The space feature extraction algorithm is introduced to analyze the material work order and the equipment work order, extract the attribute feature factor items of the processed materials required by the work order and the attribute weight values corresponding to each attribute feature factor item, establish a fuzzy search cluster center based on the attribute feature factor items, plan the uncertain search direction of the fuzzy search cluster center based on the attribute weight values, and generate the fuzzy attribute coverage domain of different processed materials for the material work order or the equipment work order;

[0022] Based on the fuzzy attribute coverage domain, the material work order or the equipment work order is imported into the dynamic management library for material attribute retrieval, and one or more sub-space storage units that can provide the processed materials required by the work order are determined and defined as fuzzy space storage units;

[0023] In the retrieval process, the Gaussian membership function is used to estimate the storage membership between the processed materials required by the work order and the corresponding storage cabinet grids of each fuzzy space storage unit to obtain the work order retrieval membership; and the attribute membership between the processed materials and the current materials stored in the corresponding storage cabinet grids of each fuzzy space storage unit to obtain the material retrieval membership;

[0024] The work order retrieval membership and the material retrieval membership are combined and fused to generate a fuzzy matching score of the work order storage retrieval, and only the storage cabinet grid corresponding to the fuzzy space storage unit with the maximum fuzzy matching score is extracted as a candidate storage cabinet grid;

[0025] The experience characteristic template of the current material in the candidate storage cabinet is obtained based on the model specification parameters and the function information, and a monotone constraint characteristic vector of the work order is constructed based on the characteristic information of the processed material, the gradient characteristic vector of the experience characteristic template is split and identified by the monotone constraint characteristic vector, and the target storage cabinet that meets the taking or lending of the processed material is determined according to the leaf node gain.

[0026] More specifically, the experience characteristic template of the current material in the candidate storage cabinet is obtained based on the model specification parameters and the function information, and a monotone constraint characteristic vector of the work order is constructed based on the characteristic information of the processed material, the gradient characteristic vector of the experience characteristic template is split and identified by the monotone constraint characteristic vector, and the target storage cabinet that meets the taking or lending of the processed material is determined according to the leaf node gain, and specifically includes the following steps:

[0027] The model specification parameters and the function information of the current material stored in the candidate storage cabinet are obtained, and the experience characteristic templates of different current materials are obtained based on the model specification parameters and the function information in the big data retrieval;

[0028] The characteristic information of the processed material required by the material work order or the equipment work order is extracted to obtain the target sample characteristic and the sample characteristic value, and a monotone constraint characteristic vector of the material work order or the equipment work order is constructed based on the target sample characteristic and the sample characteristic value;

[0029] The first-order gradient of recursive iteration is calculated according to the partition index architecture of the candidate storage cabinet in the spatial storage model of the intelligent cabinet, and the root node of the monotone gradient identification is anchored based on the first-order gradient;

[0030] The characteristic gradient of the experience characteristic template of the corresponding current material stored in each candidate storage cabinet is internally aggregated and counted starting from the root node as the starting source point, and the gradient characteristic vector of each experience characteristic template is described;

[0031] When the approximation degree of the monotone constraint characteristic vector and the gradient characteristic vector is less than the preset approximation degree, the normal gradient splitting is performed on the experience characteristic template to generate a type of sub-constraint identification node; when the approximation degree of the monotone constraint characteristic vector and the gradient characteristic vector is greater than the preset approximation degree, the monotone constraint splitting is performed on the experience characteristic template to generate a type of sub-constraint identification node; and the processed material identification tree is generated by combining the type of sub-constraint identification node and the type of sub-constraint identification node;

[0032] The leaf node distribution pattern of the processed material identification tree and the gain coefficient of each leaf node are stripped, and the candidate storage cabinet where the current material of the processed material is located is determined according to the leaf node division of the maximum gain coefficient, and the target storage cabinet is determined.

[0033] More specifically, the S3 specifically includes the following steps:

[0034] The total weight value and the terminal weight value of the material in the target storage cabinet are detected by the weight sensor, and the constant cumulative volume field of the material scale before the uncollected management or the un-borrowed and returned management is created based on the total weight value;

[0035] The weight difference is calculated by subtracting the terminal weight value from the total weight value, and the real-time weight difference value is obtained. A dynamic radial basis function for observing the weight scale is preset according to the real-time weight difference value, which changes with the material taking management operation or the borrowed and returned management operation;

[0036] The window scale limit of the non-fixed connected component is preset based on the dynamic radial basis function, and the volume range of the real-time weight difference value is dynamically planned in the constant cumulative volume field until the window scale limit is reached, thereby generating a local aggregation window for observing the change of the real-time weight difference value;

[0037] The cumulative boundary of the local aggregation window is peeled off, and a cumulative counter is virtually embedded in the local aggregation window in the constant cumulative volume field.

[0038] More specifically, the S4 specifically includes the following steps:

[0039] If the target storage cabinet performs the taking management operation, the material in the target storage cabinet is defined as the to-be-taken processed material, and the standard weight value and the independent volume characteristic of different to-be-taken processed materials are obtained through the material procurement specification;

[0040] The hash volume encoding character of the independent volume characteristic is introduced by using a hash algorithm, and the hash volume encoding character of a single to-be-taken processed material is used as an unchanged cumulative data node to continuously map into the local aggregation window;

[0041] During the cumulative mapping process, the node contribution weight value is set according to the standard weight value, and the unchanged cumulative data node located in the local aggregation window is maximized and accumulated through the traversal weighted summation based on the node contribution weight value;

[0042] When the mapping of the node reaches the cumulative boundary, the continuous mapping operation is immediately stopped, and the final local cumulative response value reflecting the unit scale volume of the to-be-taken processed material in the local aggregation window is output;

[0043] The taking quantity of the to-be-taken processed material is determined according to the final local cumulative response value when the target storage cabinet detects the real-time weight difference, and the taking quantity of the material is uploaded to the grid screen terminal for display.

[0044] More specifically, the S5 specifically includes the following steps:

[0045] If the target storage cabinet is performing a material return management operation, the material in the target storage cabinet is defined as a material to be returned for processing. The storage management log is used to extract several past borrowing and returning image data, past borrowing and returning purpose information, and corresponding past borrowing and returning duration of the material to be returned for processing within a continuous preset historical period.

[0046] Extract the K historical fixed volume features of the materials to be returned for processing from the past application and return time sequence of the image data of each borrowing and returning, and construct a volume state variable model of the materials to be returned for processing through state equations.

[0047] The information on past borrowing and returning uses is defined as a potential disturbance factor, and the duration of past borrowing and returning is set as a factor disturbance intensity index. The potential disturbance factor and the factor disturbance intensity index are combined to form a factor disturbance load matrix that drives the change in the borrowing and returning volume of the processing materials to be returned.

[0048] Based on the factor perturbation load matrix, the Kalman filter algorithm is used in the volume state variable model to estimate the factors of each historical fixed volume feature and perform regression training to generate a dynamic factor prediction model for the dynamic volume change of materials.

[0049] Obtain the current borrowing and returning purpose information and current borrowing and returning duration of the processing materials to be returned, and import the current borrowing and returning purpose information and current borrowing and returning duration into the dynamic factor prediction model for prediction to obtain the current volume characteristics of the processing materials to be returned.

[0050] By querying the material procurement specifications, the degree of matching between the current volume characteristics and each independent volume characteristic is calculated one by one. Only the standard weight value corresponding to the independent volume characteristic with the highest degree of matching is extracted and marked as the approximate weight value of the material to be returned for processing.

[0051] Based on the current volume characteristics, the hash volume encoded characters are proposed as time-varying cumulative data nodes. Based on the approximate weight value, the node contribution weight is mapped to the time-varying cumulative data nodes to the local aggregation window for weighted accumulation of volume scale, and the cumulative result is obtained.

[0052] Based on the cumulative results, determine the return count of the materials to be processed when the target storage cabinet detects a real-time weight difference, and upload the return count to the cabinet screen terminal for display.

[0053] A second aspect of the present invention provides an intelligent management system for intelligent cabinet material requisition and return. The system includes: a memory, a processor, and a communication interface. The memory includes an intelligent management method program for intelligent cabinet material requisition and return. The communication interface is used for data connection communication between the memory and the processor. When the intelligent management method program is executed by the processor, it implements any of the steps of the intelligent management method described in the present invention.

[0054] The present application solves the technical defects in the background art, and has the beneficial technical effects of:

[0055] By modeling the multi-layer space-time memory and space-time index of the real-time tracking trajectory of the entire process of material taking and returning, a dynamically evolving material management library is constructed, continuous perception and self-update of the storage partition state are realized, and the problems of misplacement and misjudgment caused by information lag in traditional intelligent cabinets are effectively avoided; by combining the work order driven material attribute constraint with fuzzy membership retrieval, the candidate storage cabinet is split, identified and finely matched, the target grid can be quickly and accurately determined in the scene where multiple categories and multiple specifications coexist, and the positioning efficiency and reliability of material taking and returning are significantly improved; the dynamic radial basis function and local aggregation window based on weight difference are introduced, the real-time weight change is robustly aggregated under the cumulative volume constraint of constant total weight, and the influence of sensing noise and human operation disturbance on the counting result is effectively suppressed; further, the invariant cumulative data node and time-varying cumulative data node are constructed for the two operation modes of material taking and returning, respectively, to realize adaptive quantity accumulation and accurate measurement of standard materials and volume changing materials; finally, the instant feedback of the grid terminal is realized. The present application solves the problems of material taking or returning quantity error, material taking or returning position confusion and other problems caused by manual counting, relying on experience memory to find material position or manual recording information in traditional material management, greatly improves the accuracy of material taking and returning; in addition, by setting the weight sensor, the quantity and weight change of the material can be accurately detected, the quantity of the taken and returned material can be automatically calculated, replacing the traditional manual counting and large estimation error process, and solving the cumbersome counting problem of small items with large quantity. BRIEF DESCRIPTION OF DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can also obtain other drawings of embodiments according to these drawings without creative labor.

[0057] Figure 1 A first method flowchart of an intelligent management method of an intelligent cabinet material taking and returning is shown;

[0058] Figure 2 A second method flowchart of an intelligent management method of an intelligent cabinet material taking and returning is shown;

[0059] Figure 3 A system framework diagram of an intelligent management system of an intelligent cabinet material taking and returning is shown. DETAILED DESCRIPTION

[0060] In order to enable the above-mentioned objects, features and advantages of the present application to be clearer, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0061] In the following description, a large number of specific details are set forth in order to facilitate a thorough understanding of the present application, however, the present application can also be implemented in other manners different from those described herein, and therefore, the protective scope of the present application is not limited to the specific embodiments disclosed below.

[0062] The first aspect of the present application provides an intelligent management method for taking and returning materials of an intelligent cabinet, as shown in the following figure, comprising the following steps: Figure 1

[0063] S1: obtaining a real-time tracking trajectory of a material in a storage cabinet grid, constructing a multi-layer space-time memory layer and a space-time index layer of an intelligent cabinet storage partition, inserting the real-time tracking trajectory into the space-time memory layer and dynamically updating and maintaining the space-time index layer, and obtaining a dynamic management library of intelligent cabinet material storage;

[0064] S2: authenticating a work order for inputting materials and equipment into the intelligent cabinet, performing fuzzy membership retrieval on the space storage of each storage cabinet grid based on the attribute features of the required processed materials in the work order, formulating candidate storage cabinet grids, and splitting and identifying the current material features in the candidate storage cabinet grids based on the constraint feature vector of the processed materials to determine the target storage cabinet grid for taking or returning the processed materials;

[0065] S3: obtaining a weight difference of the materials in the target storage cabinet grid through a weight sensor, planning a local aggregation window for observing the real-time weight difference change in the constant cumulative volume field of the total weight based on a dynamic radial basis function of a preset observation weight scale;

[0066] S4: if the target storage cabinet grid performs a taking operation, formulating an invariant cumulative data node for the independent volume features of the to-be-taken processed materials, performing weighted accumulation of the quantity scale based on the local aggregation window of the mapping value of the standard weight value of the to-be-taken processed materials, and determining the taking quantity to upload to the grid screen terminal display;

[0067] S5: if the target storage cabinet grid performs a returning operation, predicting the current volume features of the to-be-returned processed materials through a dynamic factor, formulating a time-varying cumulative data node based on the current volume features and mapping the quantity accumulation in the value local aggregation window, and determining the returning quantity to upload to the grid screen terminal display.

[0068] More specifically, the S1 specifically comprises the following steps:

[0069] ​Obtain the storage management log of the intelligent cabinet, and extract the real-time tracking trajectory of each storage compartment in the intelligent cabinet for material storage, taking and returning at consecutive time nodes through the storage management log;

[0070] Obtain the structural design drawing of the intelligent cabinet and the cabinet opening number system of the storage compartment, and set the pattern and three-dimensional distribution of the storage compartment of the intelligent cabinet into a number of discrete partitions based on the structural design drawing in the PROE model software.

[0071] Based on the number of discrete partitions, construct a multi-layer space-time memory layer and a corresponding space-time index layer of the LSM log architecture of the intelligent cabinet, and insert the real-time tracking trajectory into each space-time memory layer according to the space partition code recorded by the cabinet opening number system.

[0072] During the insertion process, a dynamic log update mechanism is introduced to maintain the local space index of the corresponding space-time index layer in each space-time memory layer, and output the real-time index memory value of the material storage partition tracking; wherein the local space index includes real-time range query, adjacent partition query and space filtering of material storage.

[0073] If the real-time index memory value is greater than the preset memory threshold, then the space index in the space-time memory layer is serialized at this time, so that the space-time memory is refreshed to an ordered and immutable space-time index, and a space-time index file of the space-time index layer for dynamic storage tracking of the material is generated.

[0074] Repeat the above steps to obtain the space-time index file of each space-time index layer, read the space index file of the adjacent space-time index layer according to the order matrix of the space partition code and reorganize it, replace the old storage index, and obtain the dynamic management library of the intelligent cabinet material storage tracking index.

[0075] It should be noted that the material storage position in the daily management of the intelligent cabinet is not fixed in a certain storage cabinet grid, but is updated and adjusted according to the use process or classification standard and other requirements, which makes the operator need to rely on past experience to remember to pick up or return the material, which easily leads to the confusion of the positioning of the material position, greatly reducing the accuracy of the material picking or returning. In view of this, the method forms the continuous time sequence behavior record of the material in the intelligent cabinet by calling the time stamp, cabinet opening number and operation state corresponding to the material taking and returning behavior of each storage cabinet grid at the continuous time node, that is, the real-time tracking trajectory as input, which is the basis and premise of the dynamic flow expression of the material. In the PROE model software, the physical cabinet structure is digitally reconstructed, and the storage cabinet grid is divided into several discrete space partitions in the three-dimensional space by scalar gridding, so that each storage cabinet grid has a clear space partition code and three-dimensional position attribute, so as to solve the problem of lack of spatial semantics in the traditional intelligent cabinet which is only managed by number, and optimize the tracking granularity of the material trace change, and provide spatial foundation semantic support for subsequent dynamic query and index of material position based on spatial proximity relationship and range relationship. Since different storage cabinet grids record or write a large amount of index data about the material position, in order to realize the time sequence replacement of the material index data, the method introduces the LSM log architecture idea to construct a multi-layer space-time memory layer and its corresponding space-time index layer, wherein the space-time memory layer is used to write the spatial index data of different materials located in different storage cabinet grids at the current time, and the space-time index layer is used to fix and replace the spatial index frequently, realize the rapid landing of the trajectory data in the high-frequency spatial index writing scene, and avoid the management performance bottleneck caused by frequent random replacement. At this time, according to the space partition code mapped by the cabinet opening number system, the real-time tracking trajectory data is written into each layer of space-time memory, so as to change the static index of the traditional intelligent cabinet into a log-type space-time index structure suitable for high-frequency taking of materials, and significantly improve the dynamic expandability and tracking stability of the intelligent cabinet in the high-concurrency adjustment scene of the material position.

[0076] It needs to be explained that by introducing a dynamic log update mechanism, each layer of spatio-temporal index layer is enabled with the ability of dynamic monitoring and updating of spatial index, and the local spatial index in each spatio-temporal memory layer is synchronously maintained, which is used to support material storage range query, adjacent partition query and spatial filtering operation, and ensure that the management system still has real-time spatial perception ability under high concurrent write state, so as to realize efficient spatial positioning and index updating effect in memory state, and improve the real-time traceability accuracy of material storage location. If the real-time index memory value is greater than the preset memory threshold, it means that the space index loading of the spatio-temporal memory layer is close to a certain upper limit, and there is an inflation accumulation phenomenon of historical invalid index, which is equivalent to setting a regular maintenance mechanism for the intelligent cabinet according to the time sequence step to update the spatial index. On the one hand, it can realize the smooth transfer of the storage cabinet for the material persistent index, guarantee the time sequence and accuracy of the material spatial index structure, and on the other hand, avoid the invalid intervention and real-time noise interference of the old storage index, and provide stable and reliable dynamic positioning maintenance. Through the method, the material taking behavior can be converted into a sustainable maintenance spatio-temporal index structure based on the LSM log architecture, so as to provide a spatial tracking maintenance and dynamic update summary management database for the current material storage location in the intelligent cabinet, so that the material behavior and current state can be uniformly monitored, managed and traced in real time, and the storage cabinet positioning accuracy of material taking and returning is optimized.

[0077] More specifically, the S2 specifically includes the following steps:

[0078] By using face recognition or verification code on the interactive screen terminal of the intelligent cabinet for authentication login, and inputting the material work order or equipment work order after authentication login;

[0079] By using PROE model software to analyze and draw the structural design drawings of the intelligent cabinet, a spatial storage model of the intelligent cabinet is constructed, and the sub-space storage units in which the storage cabinet is located are divided on the spatial storage model;

[0080] A spatial feature extraction algorithm is introduced to analyze the material work order and equipment work order, extract the attribute feature factor items of the processed material required by the work order and the attribute weight values corresponding to each attribute feature factor item, establish a fuzzy retrieval cluster center based on the attribute feature factor items, plan the uncertain retrieval direction of the fuzzy retrieval cluster center based on the attribute weight values, and generate the fuzzy attribute coverage domain of different processed materials for the material work order or equipment work order;

[0081] Based on the fuzzy attribute coverage domain, the material work order or equipment work order is imported into the dynamic management library for material attribute retrieval, one or more sub-space storage units that can provide the processed material required by the work order are determined, and are defined as fuzzy space storage units;

[0082] In the retrieval process, the Gaussian membership function is used to estimate the storage membership between the processing material required by the work order and the storage cabinet corresponding to each fuzzy space storage unit, to obtain the work order retrieval membership; and the attribute membership between the processing material and the current material stored in the storage cabinet corresponding to each fuzzy space storage unit, to obtain the material retrieval membership;

[0083] The work order retrieval membership and the material retrieval membership are combined for weighted fusion processing to generate the fuzzy matching score of the work order storage retrieval, and only the storage cabinet corresponding to the fuzzy space storage unit with the maximum fuzzy matching score is extracted as the candidate storage cabinet.

[0084] Based on the model specification parameters and the function information, the experience characteristic template of the current material in the candidate storage cabinet is obtained, and based on the characteristic information of the processing material, a monotone constraint feature vector of the work order is constructed. The gradient feature vector of the experience characteristic template is split and identified by the monotone constraint feature vector, and the target storage cabinet that meets the processing material taking or returning is determined according to the leaf node gain.

[0085] It should be noted that the material taking and returning service is for the work order demand, and since the material characteristics required by different material work orders or equipment work orders are inconsistent, and the existing intelligent cabinet has limitations and large deviations in understanding and analyzing the material characteristics of different work orders, the retrieval positioning of the storage partition of the processing material required by the work order in the intelligent cabinet is prone to errors and confusion, increasing the error rate of material taking and returning. In view of this, the method first performs semantic analysis on the processing material of the input material work order or equipment work order by a spatial feature extraction algorithm, thereby directionally extracting the attribute characteristics of the processing material required by the work order, i.e. the attribute characteristic factor, and the weight corresponding to the representation of each attribute characteristic factor, i.e. the attribute weight value. Since the attribute characteristics clearly define the fuzzy demand profile of the processing material, the fuzzy attribute coverage domain converts the work order demand from rigid parameter matching to fuzzy attribute space representation with directionality and weight distribution, depicting the uncertainty of the processing material attributes rather than rigid positioning, which enhances the robustness of the management system to incomplete, inaccurate or alternative scenarios of work order demand, enabling the intelligent cabinet to have a human-like fuzzy understanding retrieval capability. Then the material work order or equipment work order is imported into the dynamic management library for retrieval of the material attributes based on the fuzzy attribute coverage domain, reducing the calculation scale of subsequent fine matching from the full cabinet range to a local candidate space set, significantly reducing the calculation scale of subsequent fine matching. The fuzzy space storage unit is actually the storage partition where the suspicious storage cabinet of the processing material of the work order may exist, realizing the preliminary and general retrieval positioning of the fuzzy storage area of the work order.

[0086] It should be noted that when the system searches, on the one hand, the degree of matching between the processing materials and the corresponding storage cabinet compartments of each fuzzy space storage unit is measured by the work order search membership, from this macroscopic plane, the potential storage point of the work order materials is roughly collected; on the other hand, the attribute similarity between the processing materials and the current materials stored in the cabinet is measured by the material search membership, from this microscopic plane, the search accuracy of the potential storage point is further refined, forming a double fuzzy search effect of the attributes of the work order materials, quantifying the matching relationship between the work order demand and the storage cabinet compartment state. Through this method, the storage cabinet compartments can be searched based on the attribute characteristics of the processing materials required by the work order, so as to preliminarily lock the candidate storage cabinet compartments that may contain or store all the processing materials matching, thereby improving the global positioning accuracy of the storage partitions formed by the different cabinet compartments of the intelligent cabinet, and avoiding the positioning errors or omissions caused by traditional hard threshold judgment.

[0087] More specifically, the experience characteristic template of the current material in the candidate storage cabinet compartment is obtained based on the model specification parameters and functional information, and a monotone constraint feature vector of the work order is constructed based on the characteristic information of the processing material, so as to split and identify the gradient feature vector of the experience characteristic template based on the monotone constraint feature vector, and determine the target storage cabinet compartment that meets the processing material taking or borrowing according to the leaf node gain, which specifically includes the following steps:

[0088] The model specification parameters and functional information of the current material stored in the candidate storage cabinet compartment are obtained, and the experience characteristic templates of different current materials are obtained based on the model specification parameters and functional information in big data search;

[0089] The characteristic information of the processing material required by the material work order or the equipment work order is extracted to obtain the target sample feature and sample feature value, and a monotone constraint feature vector of the material work order or the equipment work order is constructed based on the target sample feature and the sample feature value;

[0090] The first-order gradient of recursive iteration is calculated according to the partition index architecture of the candidate storage cabinet compartment in the spatial storage model of the intelligent cabinet, and the root node of the monotone gradient identification is anchored based on the first-order gradient;

[0091] The feature gradient of the experience characteristic template of the corresponding current material stored in each candidate storage cabinet compartment is started to be aggregated and counted from the root node as the starting source point, and the gradient feature vector of each experience characteristic template is described;

[0092] When the approximation degree of the monotone constraint feature vector and the gradient feature vector is less than the preset approximation degree, the normal gradient splitting is performed on the experience characteristic template to generate a type of constraint identification node; when the approximation degree of the monotone constraint feature vector and the gradient feature vector is greater than the preset approximation degree, the monotone constraint splitting is performed on the experience characteristic template to generate a type of constraint identification node; and the processing material identification tree is generated by combining the type of constraint identification node and the type of constraint identification node.

[0093] The leaf node distribution pattern of the stripping processing material recognition tree and the gain coefficient of each leaf node are identified, and the candidate storage cabinet grid where the current material of the processing material is located is determined according to the leaf node division of the maximum gain coefficient, and the target storage cabinet grid is determined.

[0094] It should be noted that after the operator uploads the material work order or the equipment work order, he may not take or return all the recorded processing materials, but will choose some specified required materials, and the requirement for material positioning accuracy is more stringent. To this end, the method obtains the experience characteristic model of different current materials in each storage cabinet grid through big data retrieval. The experience characteristic model is a standardized and stereotyped characteristic template summarized from the experience cases of the current material. Since the taking and returning of the processing material required by the work order is specified or directed according to its characteristics, the characteristic vector of the processing material is set as a characteristic keynote of the work order positioning retrieval, that is, the monotonic constraint characteristic vector of the work order or the equipment work order is constructed based on the target sample characteristics and sample characteristic values of the processing material, so as to guide the monotonicity direction of subsequent characteristic retrieval using the vector constraint, so that the retrieval result of the current material can more closely approximate the expected characteristics of the processing material. Then, the spatial partition index of the storage cabinet grid is anchored to the root node to provide a starting reference point for subsequent gradient analysis and characteristic splitting. The internal aggregation statistics of each experience characteristic model are performed from the root node, so as to map the complex and multi-source experience characteristic data to a comparable gradient vector space, thereby revealing and visualizing the gradient characteristic vector of the current material in each candidate storage cabinet grid, forming a retrieval target object. If the approximation degree of the monotonic constraint characteristic vector and the gradient characteristic vector is less than the preset approximation degree, it indicates that the characteristics of the current material and the characteristics of the processing material are not well matched, indicating that the basic characteristics of the current material have deviated from the material requirement of the work order, so the experience characteristic model is split according to the normal gradient, effectively excluding the deviated vector retrieval item. Otherwise, it indicates that the matching degree is high and can meet the material requirement of the work order, so the experience characteristic model is split into an independent monotonicity constraint path, so that the recognition tree model has flexibility and constraint awareness while maintaining the discrimination ability. Through this method, the decision-making process from "preliminary candidate screening" to "unique grid selection" can be made according to the constraint of the processing material characteristics, compared with the traditional manual search for material location based on memory, the intelligent cabinet can accurately determine the storage cabinet grid where the target material is located under the specified work order requirement, and ensure the accuracy of material taking or returning.

[0095] It should be noted that after the target storage cabinet compartment is determined, the corresponding cabinet door is managed to pop open, the LED light of the small screen terminal below the cabinet door starts to flash to prompt, and the small screen terminal displays the material name to be picked up or returned, so that the material location can be quickly prompted to the material picking or returning personnel, the operator can be quickly positioned to the material location, and the situation of picking or returning wrong materials can be avoided.

[0096] More specifically, the S3 specifically includes the following steps:

[0097] The total weight value of the material in the target storage cabinet compartment is detected by the weight sensor, and a constant cumulative volume field of the material scale before the material is not picked up or managed or not borrowed or returned is created based on the total weight value;

[0098] The weight difference is calculated by subtracting the termination weight value from the total weight value, the real-time weight difference value is obtained, and a dynamic radial basis function for observing the weight scale according to the real-time weight difference value is preset to change with the material picking management operation or the borrowing and returning management operation;

[0099] A window scale limit of a non-fixed connected component is preset based on the dynamic radial basis function, the volume range of the real-time weight difference value is dynamically planned in the constant cumulative volume field until the window scale limit is reached, and a local aggregation window for observing the change of the real-time weight difference value is generated.

[0100] The cumulative boundary of the local aggregation window is peeled off, and a cumulative counter is virtually embedded in the local aggregation window in the constant cumulative volume field.

[0101] It should be noted that the intelligent cabinet is additionally provided with a weight sensor for accurately measuring the weight change when the material is taken or returned, and then calculating the quantity of the taken or returned material according to the weight difference change. However, the traditional weight calculation method is not accurate, and the counting result has a large deviation, especially for the counting accuracy of a large number of small items. For this reason, the method first obtains the weight data of the stored material in the target storage cabinet through the detection of the weight sensor, including the total weight (total weight value) before the operation starts and the weight remaining (termination weight value) after the taking or returning operation, and the weight difference between the two is the weight reflection of the quantity of the taken or returned material. Since the weight difference is the direct loss of the total weight value, the method quantifies the initial inventory state of the material as a stable and time-invariant volume reference space according to the total weight value, forms a constant cumulative volume field for observing the initial weight scale, and provides a loss scale reference for the dynamic change analysis of the weight difference, avoiding the phenomenon of misjudgment or counting fluctuation caused by sensor instantaneous fluctuation or operation process disturbance. With the continuous progress of the material taking or returning operation, the weight difference will change in real time, so that the weight difference is given a dynamic attribute, and the discrete weight change is mapped to a continuous radial function form according to the real-time weight difference value, so that the weight difference real-time change process has a range of adaptive adjustable smooth response expression, and is constructed as a dynamic radial basis function for observing the weight scale of the material taking management operation or the borrowing management operation, avoiding the rigid judgment deviation of the traditional threshold method to the dynamic weight change, and enhancing the weight specification ability under different material quantity levels and different operation rhythm conditions.

[0102] It should be noted that the window scale limit of the preset non-fixed connected component is constrained under the dynamic radial basis function, which reveals the real-time jump range of the dynamic weight difference, and dynamically plans the volume range of the real-time weight difference in the constant cumulative volume field. In short, it can be understood as planning a two-dimensional observation window (local aggregation window) that will adaptively geometrically stretch in a fixed two-dimensional space (constant cumulative volume field), so as to accurately capture the effective weight variation interval, realize the local sensitivity and global stability of the real-time weight difference change, provide an accurate and reliable weight difference limit framework for the subsequent material counting, and ensure the explainable statistical expression of the weight change when the material is taken and returned. Through the method, adaptive perception and stable statistics of the weight change of the material in the storage cabinet can be realized, and the counting accuracy and robustness of the weight monitoring can be effectively improved.

[0103] More specifically, the S4, as shown in Figure 2 specifically includes the following steps:

[0104] If the target storage cabinet grid performs a pick-up management operation, the material in the target storage cabinet grid is defined as a to-be-picked processed material, and the standard weight value and the independent volume characteristic of different to-be-picked processed materials are obtained through the material procurement specification;

[0105] The hash volume encoding character of the hash algorithm is introduced to compile the independent volume characteristic, and the hash volume encoding character of a single to-be-picked processed material is used as an invariant accumulation data node to continue to be mapped into a local aggregation window;

[0106] In the accumulation mapping process, the node contribution weight value is set according to the standard weight value, and the invariant accumulation data node located in the local aggregation window is subjected to a maximum accumulation of a traversal weighted summation based on the node contribution weight value;

[0107] When the mapping of the node reaches the accumulation boundary, the continuous mapping operation is immediately stopped, and a final local accumulation response value reflecting the unit scale volume of the to-be-picked processed material owned in the local aggregation window is outputted;

[0108] According to the final local accumulation response value, the pick-up quantity of the to-be-picked processed material when the real-time weight difference is detected by the target storage cabinet grid is determined, and the pick-up material quantity is uploaded to the grid screen terminal for display.

[0109] It should be noted that for the taking operation, since the target storage cabinet usually stores the same batch of processing materials, the model, specification, size and other parameters of each processing material are the same, so the standard weight and its weight error consideration are almost the same. Therefore, the method obtains the standard weight value of the to-be-taken processing material and the known specification independent volume characteristics; wherein the independent volume characteristics include length, width, height, density and shape. Then the independent volume characteristics are compiled using a hash algorithm, so as to convert them from continuous physical quantities to stable discrete coding nodes, i.e. invariant cumulative data nodes, to realize unique identification and anti-interference expression of the material volume characteristics, and improve the equality and consistency of cumulative calculation. The invariant cumulative data nodes of different to-be-taken processing materials are continuously mapped to the previously constructed local aggregation window to form the weight aggregation effect of the same material in the two-dimensional observation window. The standard weight value defines the weight contribution degree of each to-be-taken processing material, effectively superimposes the contributions of multiple material volume nodes, and makes the cumulative result truly reflect the material volume scale in the local aggregation window, avoiding the problem that the traditional simple counting method ignores the weight difference, so that the taking quantity determination takes into account both the quantity rationality and the weight consistency. When the mapping of the node touches the cumulative boundary, the continuous mapping operation is immediately stopped, so that a stable and unique local volume cumulative result can be obtained under the premise of not exceeding the real-time weight difference constraint, effectively preventing the phenomenon of over-counting or mis-counting, and making the system maintain strict consistency between weight change and material quantity mapping. Finally, the accurate cumulative quantity of the taken material is obtained. Through the method, the standard weight of the to-be-taken processing material can be weighted and maximized in the local aggregation window, so that the real-time weight change of the material taking can be stably and reliably quantified as a high-credibility material quantity, improving the management accuracy of the material taking.

[0110] More specifically, the S5 specifically includes the following steps:

[0111] If the target storage cabinet performs a material returning management operation, the material in the target storage cabinet is defined as a to-be-returned processing material, and the to-be-returned processing material is extracted from the storage management log for a plurality of past borrowing and returning image data, past borrowing and returning purpose information and corresponding past borrowing and returning time length within a continuous preset historical period;

[0112] The K historical fixed volume characteristics of the to-be-returned processing material are extracted from each past borrowing and returning image data after the past application and at any time sequence continuously recursive at the time of returning, and a volume state variable model of the applied to-be-returned processing material is constructed through a state equation;

[0113] The past borrowing and returning purpose information is defined as a potential disturbance factor, the past borrowing and returning time length is set as a factor disturbance intensity index, and the potential disturbance factor and the factor disturbance intensity index are combined to form a factor disturbance load matrix driving the borrowing and returning volume change of the to-be-returned processing material;

[0114] Factor estimation and regression training are performed on each historical fixed volume feature in the volume state variable model based on the factor disturbance load matrix using the Kalman filter algorithm to generate a dynamic factor prediction model of material dynamic volume change;

[0115] The current borrowing and lending purpose information and the current borrowing and lending time length of the material to be returned for processing are obtained, and the current borrowing and lending purpose information and the current borrowing and lending time length are imported into the dynamic factor prediction model for prediction to obtain the current volume feature of the material to be returned for processing;

[0116] The degree of fit between the current volume feature and each independent volume feature is calculated one by one by querying the material procurement specification, and only the standard weight value corresponding to the independent volume feature with the maximum degree of fit is extracted to be calibrated as the approximate weight value of the material to be returned for processing;

[0117] The hash volume coding character based on the current volume feature is drafted as a time-varying cumulative data node, and the node is mapped to a local aggregation window based on the approximate weight value to perform weighted accumulation of the volume scale to obtain a cumulative result;

[0118] The return count of the material to be returned for processing when the real-time weight difference is detected by the target storage cabinet grid is determined according to the cumulative result, and the return count is uploaded to the grid screen terminal for display.

[0119] It should be noted that for the return operation, due to the high autonomy of the material returning personnel and the difference in the use of the material, it is difficult to ensure that the weight of each returned material of the same batch is completely consistent, and the weight balance standard is nonlinear. If the previous unified standard weight is used for weighted cumulative processing, it may cause large-scale deviation of the final material returning count result. In view of this, the method builds a dynamic volume change prediction model for the return and use of the processed material to be returned. Specifically, first, the existing past return and use image data of the processed material to be returned is used as dynamic data input. The past return and use image data records the predetermined volume state of the processed material to be returned before and after use. Because different processing purposes and use time will cause different degrees of wear, loss or damage of the material, which is the direct cause of the volume state change. For example, a screw thread part is partially engraved during long-term (use time) repeated processing and tightening (processing purpose), so its weight will decrease. As can be seen, the processing purpose is a direct disturbance factor, and the use time determines the duration of the disturbance factor. Therefore, the method sets the past return and use information as a potential disturbance factor, and sets the disturbance strength index of the corresponding potential disturbance factor according to the past return and use time. The two form a factor disturbance load matrix that drives the return and use volume change of the processed material to be returned, and reveals the dynamic variable contribution of the factor to each historical fixed volume feature. Finally, using the Kalman filtering algorithm, the factor disturbance load matrix is estimated and regressed in the volume state variable model to train each historical fixed volume feature, realize recursive estimation and error correction of the volume state, and obtain a stable and robust dynamic volume change prediction model. Through the method, the weight of the processed material to be returned after different volume changes can be accumulated and counted, so as to calculate the accurate batch quantity of the non-uniform returned material weight, and realize accurate management of material return.

[0120] The second aspect of the present application provides an intelligent management system for material taking and returning of an intelligent cabinet, as shown in the accompanying drawings, the system comprises a memory 301, a processor 302, and a communication interface 303, the memory 301 comprises an intelligent management method program for material taking and returning of an intelligent cabinet, the communication interface 303 is used for data connection communication between the memory 301 and the processor 302, and the intelligent management method program is executed by the processor 302 to realize the steps of the intelligent management method. Figure 3

[0121] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.​

Claims

1. A smart cabinet material requisition and return intelligent management method, characterized in that, Includes the following steps: S1: Obtain the real-time tracking trajectory of material retrieval and return in the storage cabinet, construct a multi-layer spatiotemporal memory layer and spatiotemporal index layer for the intelligent cabinet storage partition, insert the real-time tracking trajectory into the spatiotemporal memory layer and dynamically update and maintain the spatiotemporal index layer to obtain the dynamic management library of intelligent cabinet material storage. S2: Authentication and login to the smart cabinet inputs work orders for materials and equipment. Based on the attribute characteristics of the materials to be processed in the work order, the dynamic management library is imported to perform fuzzy membership retrieval of the space storage of each storage cabinet, candidate storage cabinets are proposed, and the current material characteristics in the candidate storage cabinets are identified by splitting the constraint feature vector of the processed materials, and the target storage cabinet for the processing materials to be picked up or borrowed is determined. S3: Obtain the weight difference of the materials in the target storage cabinet through the weight sensor, preset the dynamic radial basis function of the observed weight scale based on the weight difference, and plan the local aggregation window for observing the real-time weight difference change in the constant cumulative volume domain of the total weight based on the dynamic radial basis function. S4: If a retrieval operation is performed on the target storage cabinet, the independent volume characteristics of the material to be processed are hash-coded to determine an invariant cumulative data node. Based on the standard weight value mapping value of the material to be processed, a local aggregation window is used to perform a weighted accumulation of the quantity scale, and the retrieval count is determined and uploaded to the cabinet screen terminal for display. S5: If the target storage cabinet performs a material return operation, the current volume characteristics of the material to be returned are predicted by dynamic factors. Based on the current volume characteristics, time-varying cumulative data nodes are proposed and mapped to the quantity accumulation in the local value aggregation window. The quantity to be returned is then uploaded to the cabinet screen terminal for display. Specifically, S3 includes the following steps: The total weight and final weight of the materials in the target storage cabinet are obtained by detecting the weight sensor. Based on the total weight, a constant cumulative volume of the material is created in the domain before the material is picked up or borrowed and returned. The weight difference is calculated by subtracting the final weight value from the total weight value to obtain the real-time weight difference. A dynamic radial basis function is preset based on the real-time weight difference to observe the weight scale as it changes with material requisition management operations or borrowing and returning management operations. Based on the window scale limit of the non-fixed connected components preset by the dynamic radial basis function, the volume range of the real-time weight difference is dynamically planned in the constant cumulative volume domain until the window scale limit is reached, and a local aggregation window for observing changes in the real-time weight difference is generated. Extract the cumulative boundary of the local aggregation window and virtually embed an accumulator counter for the local aggregation window within the constant cumulative volume domain.

2. The intelligent management method for material requisition and return in a smart cabinet according to claim 1, characterized in that, S1 specifically includes the following steps: Obtain the storage management logs of the smart cabinet, and extract the real-time tracking trajectory of material storage, retrieval and return for each storage compartment in the smart cabinet at continuous time nodes through the storage management logs; Obtain the structural design drawings of the smart cabinet and the cabinet numbering system of the storage compartments. Based on the structural design drawings, in the PROE model software, set the storage compartment layout and three-dimensional distribution scalar rasterization of the smart cabinet into several discrete partitions. Based on several discrete partitions, a multi-layered spatiotemporal memory layer and a corresponding spatiotemporal index layer of intelligent cabinet LSM log architecture storage partition are constructed. Real-time tracking trajectories are inserted into each spatiotemporal memory layer according to the spatial partition code recorded in the cabinet numbering system. During the insertion process, a dynamic log update mechanism is introduced to maintain the local spatial index of the corresponding spatiotemporal index layer in each spatiotemporal memory layer, and output the real-time index memory value of material storage partition tracking; among which, the local spatial index includes real-time range query, adjacent partition query and spatial filtering of material storage; If the real-time index memory value is greater than the preset memory threshold, then the spatial index in the spatiotemporal memory layer is serialized, so that the spatiotemporal memory is refreshed into an ordered and immutable spatiotemporal index, and a spatiotemporal index file for dynamic storage tracking of materials is generated. Repeat the above steps to obtain the spatiotemporal index file for each spatiotemporal index layer. Read the spatial index files of adjacent spatiotemporal index layers according to the spatial partitioning coding order matrix and reorganize and iterate them to replace the old storage index, thereby obtaining the dynamic management library of the intelligent cabinet material storage tracking index.

3. The intelligent management method for material requisition and return in a smart cabinet according to claim 1, characterized in that, S2 specifically includes the following steps: Authentication and login can be performed using facial recognition or verification code on the interactive screen terminal of the smart cabinet. After authentication and login, material work orders or equipment work orders for material requisition or return processing can be entered. The structural design drawings of the smart cabinet are analyzed and drawn using PROE modeling software to construct a spatial storage model of the smart cabinet. The spatial storage model is then divided into sub-space storage units corresponding to each storage compartment. A spatial feature extraction algorithm is introduced to analyze the material work order and equipment work order, extract the attribute feature factors of the materials to be processed in the work order and the attribute weight values ​​corresponding to each attribute feature factor, establish the centroid of the fuzzy retrieval cluster based on the attribute feature factors, and start planning the uncertain retrieval direction of the fuzzy retrieval cluster centroid based on the attribute weight values, and generate the fuzzy attribute coverage domain of the material work order or equipment work order for different processed materials. Based on the fuzzy attribute coverage domain, material work orders or equipment work orders are imported into the dynamic management database for material attribute retrieval, and one or more subspace storage units that can provide the materials required for work order processing are identified and defined as fuzzy space storage units. During the retrieval process, the Gaussian membership function is used to estimate the storage membership degree between the materials to be processed in the work order and the corresponding storage cabinets of each fuzzy space storage unit, thus obtaining the work order retrieval membership degree; and the attribute membership degree between the processed materials and the current materials stored in the corresponding storage cabinets of each fuzzy space storage unit, thus obtaining the material retrieval membership degree. By combining the membership degree of work order retrieval and the membership degree of material retrieval with weighted fusion processing, a fuzzy matching score for work order storage retrieval is generated. Only the storage cabinet corresponding to the fuzzy spatial storage unit with the maximum fuzzy matching score is extracted and designated as a candidate storage cabinet. Based on model specifications and functional information, obtain the empirical feature template of the current material in the candidate storage cabinet, and construct the monotonic constraint feature vector of the work order based on the feature information of the processed material. Use the monotonic constraint feature vector to split and identify the gradient feature vector of the empirical feature template, and determine the target storage cabinet that meets the requirements for processing material retrieval or borrowing / returning based on the leaf node gain.

4. The intelligent management method for material requisition and return in a smart cabinet according to claim 3, characterized in that, The process of obtaining an empirical feature template of the current material in the candidate storage cabinet based on model specifications and functional information, constructing a monotonic constraint feature vector of the work order based on the feature information of the processed material, splitting and identifying the gradient feature vector of the empirical feature template using the monotonic constraint feature vector, and determining the target storage cabinet that meets the requirements for processing material retrieval or borrowing / returning based on the leaf node gain, specifically includes the following steps: Obtain the model, specification parameters, and functional information of the current material stored in the candidate storage cabinet; and retrieve experience feature templates of different current materials based on the model, specification parameters, and functional information through big data retrieval. Extract the characteristic information of the materials to be processed in the material work order or equipment work order, obtain the target sample features and sample feature values, and construct the monotonic constraint feature vector of the material work order or equipment work order based on the target sample features and sample feature values. The first-order gradient of the recursive iteration is calculated based on the partition index architecture of the candidate storage cell in the spatial storage model of the smart cabinet, and the root node of the monotonic gradient identification is anchored based on the first-order gradient. Starting from the root node, the feature gradients of the empirical feature templates corresponding to the current material stored in each candidate storage cabinet are aggregated and statistically analyzed internally, describing the gradient feature vector of each empirical feature template. When the approximation between the monotonic constraint feature vector and the gradient feature vector is less than the preset approximation, normal gradient splitting is performed on the empirical feature template to generate a first-class sub-constraint identification node; when the approximation between the monotonic constraint feature vector and the gradient feature vector is greater than the preset approximation, monotonic constraint splitting is performed on the empirical feature template to generate a second-class sub-constraint identification node; the first-class sub-constraint identification node and the second-class sub-constraint identification node are combined to generate a processed material identification tree; The leaf node distribution pattern and gain coefficient of each leaf node in the material identification tree are extracted. Based on the leaf node with the largest gain coefficient, the candidate storage cabinet where the current material is located is determined and identified as the target storage cabinet.

5. The intelligent management method for material requisition and return in a smart cabinet according to claim 1, characterized in that, S4 specifically includes the following steps: If the target storage cabinet is performing a retrieval operation, the materials in the target storage cabinet are defined as materials to be retrieved and processed. The standard weight values ​​and independent volume characteristics of different materials to be retrieved and processed are obtained through the material procurement specifications. A hash algorithm is introduced to compile hash volume encoding characters for independent volume characteristics. The hash volume encoding character of a single material to be processed is designated as an invariant cumulative data node and continuously mapped to the local aggregation window. During the cumulative mapping process, node contribution weights are set according to standard weight values, and the cumulative data nodes located within the local aggregation window are maximized by performing a traversal weighted summation based on the node contribution weights. When the mapping of a node reaches the cumulative boundary, the continuous mapping operation is stopped immediately, and the final local cumulative response value reflecting the unit size of the materials to be processed within the local aggregation window is output. Based on the final local cumulative response value, determine the quantity of materials to be processed when the target storage cabinet detects a real-time weight difference, and upload the quantity of materials to be processed to the cabinet screen terminal for display.

6. The intelligent management method for material requisition and return in a smart cabinet according to claim 1, characterized in that, S5 specifically includes the following steps: If the target storage cabinet is performing a material return operation, the material in the target storage cabinet is defined as the material to be returned for processing. The storage management log is used to extract several past borrowing and returning image data, past borrowing and returning purpose information and corresponding past borrowing and returning duration of the material to be returned for processing within a continuous preset historical period. Extract the K historical fixed volume features of the materials to be returned for processing from the past application and return time sequence of the image data of each borrowing and returning, and construct a volume state variable model of the materials to be returned for processing through state equations. The information on past borrowing and returning uses is defined as a potential disturbance factor, and the duration of past borrowing and returning is set as a factor disturbance intensity index. The potential disturbance factor and the factor disturbance intensity index are combined to form a factor disturbance load matrix that drives the change in the borrowing and returning volume of the processing materials to be returned. Based on the factor perturbation load matrix, the Kalman filter algorithm is used in the volume state variable model to estimate the factors of each historical fixed volume feature and perform regression training to generate a dynamic factor prediction model for the dynamic volume change of materials. Obtain the current borrowing and returning purpose information and current borrowing and returning duration of the processing materials to be returned, and import the current borrowing and returning purpose information and current borrowing and returning duration into the dynamic factor prediction model for prediction to obtain the current volume characteristics of the processing materials to be returned. By querying the material procurement specifications, the degree of matching between the current volume characteristics and each independent volume characteristic is calculated one by one. Only the standard weight value corresponding to the independent volume characteristic with the highest degree of matching is extracted and marked as the approximate weight value of the material to be returned for processing. Based on the current volume characteristics, the hash volume encoded characters are proposed as time-varying cumulative data nodes. Based on the approximate weight value, the node contribution weight is mapped to the time-varying cumulative data nodes to the local aggregation window for weighted accumulation of volume scale, and the cumulative result is obtained. Based on the cumulative results, determine the return count of the materials to be processed when the target storage cabinet detects a real-time weight difference, and upload the return count to the cabinet screen terminal for display.

7. An intelligent management system for intelligent cabinet material requisition and return, characterized in that, The system includes: a memory, a processor, and a communication interface. The memory includes a program for intelligent management of material requisition and return in an intelligent cabinet. The communication interface is used for data connection and communication between the memory and the processor. When the intelligent management program is executed by the processor, it implements the steps of the intelligent management method as described in any one of claims 1-6.

Images