Engineering material a and b multi-level warehouse supply material ai management system

Abstract

The present application relates to the technical field of engineering material management, and discloses an AI management system for a multi-level warehouse of engineering materials provided by materials A and B, comprising: a multi-modal perception grid module, used for dividing the multi-level warehouse into multiple virtual grids and perceiving the change events of the materials; a dual-state data storage module, used for maintaining physical state data records and ownership state data records; an ownership switching module, used for generating a new ownership state data record when the cumulative consumption exceeds the upper limit of the quota; an AI data flow hard lock module, used for unlocking the controlled hardware when the physical state change event is matched with the business invoice, and keeping the hardware locked and triggering an alarm when the matching fails; and a multi-agent game optimization module, used for limiting the material requisition of the subject determined to have a false reporting behavior. The present application can automatically distinguish the ownership of the materials provided by materials A and B, realize the unattended inventory of the multi-level warehouse, automatic settlement of over-consumption, automatic comparison and hard locking of the physical materials and invoices, and material optimization scheduling in the game environment.

Description

Technical Field

[0001] This invention relates to the field of engineering material management technology, and in particular to an AI management system for a multi-level warehouse supplying engineering materials to suppliers A and B. Background Technology

[0002] In large-scale infrastructure projects, the management of engineering materials faces the following technical challenges.

[0003] First, there's the issue of automatically distinguishing ownership between materials supplied by Party A and those supplied by Party B. Materials supplied by Party A are procured by Party A, while materials supplied by Party B are procured by the construction company. Physically, both are standard products such as steel bars and cement, but the parties responsible for the costs differ. The existing system relies on manual labeling or RFID tags to distinguish ownership. However, these tags are easily damaged under harsh working conditions at construction sites, and the system cannot automatically distinguish ownership when similar materials are mixed together, leading to settlement disputes.

[0004] Secondly, there is the issue of unattended automated inventory management in multi-level warehouses. Construction materials need to be transferred through a primary central warehouse, a secondary work area warehouse, and a tertiary on-site warehouse. The tertiary warehouse is located on the work surface, and materials are randomly stacked and frequently moved. Existing technology relies on personnel actively scanning codes or labels attached to the surface of materials, which cannot be used for automated inventory management in situations where there is no cooperation, no labels, obstructions, or harsh working conditions. This results in long-term discrepancies between the records and actual inventory in the tertiary warehouses.

[0005] Third, there is the issue of automatic settlement for excessive consumption of materials supplied by Party A. Excessive consumption of materials supplied by Party A needs to be deducted from the contractor's payment. Traditional methods rely on manual statistics of consumption, comparison with quotas, and negotiation of deductions. This results in data lag and makes it impossible to trace the specific time and responsible section where the excessive consumption occurred.

[0006] Therefore, this invention proposes an AI management system for multi-level warehouses supplying engineering materials to suppliers A and B. Summary of the Invention

[0007] The purpose of this invention is to solve the problems of difficulty in automatically distinguishing ownership, failure of unattended inventory in multi-level warehouses, and reliance on manual settlement of over-consumption of materials supplied by Party A in the existing technology, and proposes an AI management system for multi-level warehouses for engineering materials supplied by Party A and Party B.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: a multi-level warehouse AI management system for engineering materials supplied by suppliers A and B, comprising:

[0009] The multimodal sensing grid module is used to divide a multi-level warehouse into multiple virtual grids, and to sense changes in materials within each grid in a completely passive and non-contact manner.

[0010] The dual-state data storage module is used to maintain independently stored physical state data records and ownership state data records for each material. The physical state data records are updated by the multimodal sensing grid module and cannot be modified by the business process. The ownership state data records include cost-bearing information and settlement price information.

[0011] The ownership switching module is used to maintain the physical status data record of a material unchanged and generate a new ownership status data record when the physical status data record shows that the cumulative consumption of a material exceeds the quota limit. In this new record, the cost-bearing information is rewritten to the contractor's identifier, the settlement price information is rewritten to the market settlement unit price, and the original ownership status data record is marked as invalid.

[0012] The AI ​​data flow hardware lock module is used to unlock the controlled hardware to allow the process to proceed when the physical state change event of materials matches the business document; when the match fails, the hardware remains locked and an alarm is triggered.

[0013] The multi-agent game optimization module is used to identify inventory overstatement behavior based on the actual consumption data of each construction entity through an adversarial learning mechanism, and to restrict the material requisition of entities that have been found to have engaged in overstatement behavior.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0015] This invention enables fully automatic ownership differentiation between materials supplied by Party A and materials supplied by Party B. The multimodal sensing grid module passively extracts the spectral characteristics of the materials when they first enter the site, and the dual-state data storage module stores them in the physical state data record. When materials of the same type are mixed and stacked, the system automatically compares the spectral characteristics to determine ownership, without the need for manual labeling or warehousing.

[0016] This invention enables fully passive, unattended inventory management in multi-level warehouses. The multimodal sensing grid module senses the volume changes and behavior types of materials within each virtual grid in a fully passive, non-contact manner, without relying on material surface markings or active personnel cooperation. Even under harsh conditions in a three-level warehouse, it can still automatically generate material change ledgers.

[0017] This invention enables automatic settlement of excess consumption of materials supplied by Party A. When the cumulative consumption exceeds the quota limit, the ownership switching module automatically switches the cost-bearing party to the construction party and the settlement price to the market price, and marks the original ownership record as invalid. The excess consumption tracing unit writes the time, location, responsible section and process code of the excess consumption event into the ownership record and automatically matches the settlement item, without the need for manual reconciliation and negotiation.

[0018] This invention enables automatic comparison and physical locking of the loaded goods with the business documents. The AI ​​data flow locking module identifies loading events from physical state change events, compares the type and quantity of the loaded goods with the outbound order, unlocks the controlled hardware to release the goods if they match, and locks and alarms if they do not match, thus preventing the goods from leaving the factory if they do not match the order.

[0019] This invention enables optimized scheduling of materials in a non-cooperative game environment. The multi-agent game optimization module trains an adversarial network based on the deviation between the historically reported inventory and the actual consumption data of each construction entity. It outputs the false reporting probability of each entity and restricts the material requisition of entities that make false reports, thus forcing each entity to honestly report its inventory. Attached Figure Description

[0020] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a schematic diagram of the system configuration provided in an embodiment of the present invention;

[0022] Figure 2 This is a schematic diagram of the system flow provided for an embodiment of the present invention. Detailed Implementation

[0023] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the multi-level warehouse AI management system for engineering materials supplying both parties according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0024] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0025] The following examples are for illustrative purposes and are not intended to limit the scope of the invention.

[0026] The specific solution of the multi-level warehouse AI management system for engineering materials supplied by Party A and Party B, provided by the present invention, is described in detail below with reference to the accompanying drawings.

[0027] Please see Figure 1 and Figure 2 It illustrates a system configuration diagram and a system flow diagram of a multi-level warehouse AI management system for engineering materials supplied by suppliers A and B according to an embodiment of the present invention, including:

[0028] The multimodal sensing grid module is used to divide a multi-level warehouse into multiple virtual grids, and to sense changes in materials within each grid in a completely passive and non-contact manner.

[0029] The dual-state data storage module is used to maintain independently stored physical state data records and ownership state data records for each material. The physical state data records are updated by the multimodal sensing grid module and cannot be modified by the business process. The ownership state data records include cost-bearing information and settlement price information.

[0030] The ownership switching module is used to maintain the physical status data record of a material unchanged and generate a new ownership status data record when the physical status data record shows that the cumulative consumption of a material exceeds the quota limit. In this new record, the cost-bearing information is rewritten to the contractor's identifier, the settlement price information is rewritten to the market settlement unit price, and the original ownership status data record is marked as invalid.

[0031] The AI ​​data flow hardware lock module is used to unlock the controlled hardware to allow the process to proceed when the physical state change event of materials matches the business document; when the match fails, the hardware remains locked and an alarm is triggered.

[0032] The multi-agent game optimization module is used to identify inventory overstatement behavior based on the actual consumption data of each construction entity through an adversarial learning mechanism, and to restrict the material requisition of entities that have been found to have engaged in overstatement behavior.

[0033] It should be noted that engineering materials refer to all kinds of building materials used in construction projects, including but not limited to steel bars, cement, sand and gravel, concrete, cables, etc., and are divided into materials supplied by Party A and materials supplied by Party B, depending on the purchaser and the party bearing the cost.

[0034] Materials supplied by Party A refer to engineering materials that are purchased and owned by the construction unit (Party A). The construction party uses them according to the quota, and any excess is deducted from the construction party's payment based on the market price.

[0035] Materials supplied by Party B refer to engineering materials that are procured and funded by the construction party itself. The construction party is responsible for the procurement plan, on-site inspection, and management of these materials.

[0036] Multi-level warehouses refer to a warehousing system divided according to the level of material flow, including the primary central warehouse managed by the client's materials department, the secondary work area warehouses managed by each construction section, and the tertiary site warehouses located at the construction work site.

[0037] A virtual grid refers to an independent, non-physically separated data acquisition and computing area within a multi-level warehouse physical space, defined by multimodal sensing grid modules. Each virtual grid is bound to a physical space region and assigned a unique grid identifier, used to subdivide the warehouse space into independently perceptible and manageable units.

[0038] Fully passive means that the multimodal sensing grid module does not rely on the markings attached to the material surface, the active scanning operation by personnel, or the active cooperation of the sensed object when sensing material change events.

[0039] Non-contact refers to the multimodal sensing grid module acquiring material information in a non-contact manner through sensors such as optical and electromagnetic waves, without requiring physical contact between the sensors and the materials.

[0040] Physical state data records refer to data reflecting the real-time physical state of materials, updated by the multimodal sensing grid module. This includes the grid position coordinates, occupied volume, surface material characteristics, and integrity indicators of the materials in the warehouse. Write operations are only performed in response to change events output by the multimodal sensing grid module and do not respond to write requests from any other module.

[0041] Ownership status data records refer to data that reflects the party responsible for material costs and the settlement price. They include cost-bearing information and settlement price information, and are stored independently from physical status data records but are linked through the material's unified identification code.

[0042] The upper limit of the quota refers to the product of the theoretical consumption of a certain material in the project calculated based on the engineering design model or bill of quantities and the preset loss coefficient, which serves as the threshold for judging the overconsumption of materials supplied by the client.

[0043] Ownership switching refers to the process where, when the physical data record shows that the cumulative consumption of a certain material exceeds the quota limit, the system keeps the physical data record unchanged, generates a new ownership data record, changes the cost bearer from the original owner to the contractor, changes the settlement price from the contract unit price to the market settlement unit price, and marks the original ownership data record as invalid.

[0044] Physical state change events refer to events in which materials within a grid are added, removed, displaced, or undergo morphological changes, as sensed by the multimodal sensing grid module. Each event includes a timestamp, the grid identifier where it occurred, the material category, the amount of change, and a behavior type label.

[0045] Controlled hardware refers to physical barriers or locking devices deployed at warehouse entrances and exits that can receive unlocking or locking signals sent by AI data stream hardware lock modules to control vehicle passage.

[0046] Adversarial learning refers to a machine learning paradigm that uses a generator network and a discriminator network for adversarial training. The generator network learns to simulate the false reporting strategies of various construction entities, while the discriminator network learns to distinguish between real consumption data and false reporting data. The two networks are alternately optimized until the discriminator can effectively identify false reporting behavior.

[0047] In one specific implementation, a large hydropower station project has three construction sections: A, B, and C. Materials supplied by the client are procured and stored in a primary central warehouse. Each section applies for and allocates materials to its respective secondary work area warehouse as needed, and finally, these materials are distributed to tertiary on-site warehouses at locations such as the tunnel face and the bottom of the foundation pit. The system divides the primary central warehouse, each secondary work area warehouse, and each tertiary on-site warehouse into multiple virtual grids. Each virtual grid is equipped with a fusion sensing element consisting of millimeter-wave radar and event cameras.

[0048] A batch of rebar supplied by Party A was transferred from the primary central warehouse to the secondary field warehouse shared by sections A and B. Upon the initial arrival of the rebar, the multimodal sensing grid module passively extracted the surface spectral characteristics of the batch and stored them in the physical state data record of the dual-state data storage module. When a subsequent batch of rebar supplied by Party B also entered the same grid, the system automatically compared the spectral characteristics of the rebar to be identified with the stored spectral characteristics of Party A-supplied materials, determined that the batch was supplied by Party B, and wrote the ownership determination result into its ownership state data record.

[0049] During construction, sections A and B successively consumed the batch of rebar supplied by the client. When the cumulative consumption exceeded the upper limit of the quota calculated from the design quantity extracted from the BIM model and the loss coefficient, the over-consumption allocation unit of the ownership switching module divided the over-consumption amount according to the time sequence of the consumption events in the two sections and the proportion of actual consumption. The section that first triggered the over-consumption was identified as the primary responsible party, and a responsibility determination data package bound to the ownership switching signal was generated and written to the blockchain for evidence storage. The over-consumption tracing unit wrote the working face location, time, responsible section, and corresponding WBS process code of the over-consumption event into the ownership record, automatically matched the settlement item of the process from the bill of quantities, and wrote the over-consumption deduction amount into the deduction account.

[0050] After construction was completed, a truck from Section B loaded with remaining materials and prepared to leave the site. The loading event capture unit detected the loading activity and identified the loaded materials as rebar and cables. The single-vehicle material locking unit compared the actual loaded materials with the Section B outbound manifest. The comparison revealed that the outbound manifest listed rebar as the material, but the loaded materials also included cables. The system determined that the categories did not match, and the exit blocking execution unit sent a lock signal to the exit gate, prohibiting the vehicle from leaving. At the same time, it pushed an alarm to the person in charge of Section B and the client's materials department. The alarm information displayed vehicle information, the quantity of rebar and cables loaded, and the discrepancy that the cables did not belong to the category recorded on the outbound manifest. Upon verification, the cables were found to be client-supplied materials, and the construction personnel attempted to smuggle them out, but were intercepted by the system in real time.

[0051] At the material allocation level, both Section A and Section C simultaneously apply for the next batch of cement supplied by the client. The multi-agent game optimization module trains a generative adversarial network based on the discrepancy between the historically reported inventory of the two sections and the actual consumption recorded by the perception grid module. The discriminator network outputs that Section A has a lower probability of false reporting and Section C has a higher probability of false reporting. The system automatically lowers the material application priority of Section C, prioritizing the allocation of client-supplied cement to Section A.

[0052] I. Multimodal Sensing Mesh Module

[0053] The multimodal sensing mesh module includes a mesh generation unit, an irregular stack sensing unit, a consumption behavior identification unit, and a residual material status determination unit;

[0054] Grid partitioning unit is used to divide a multi-level warehouse into multiple virtual grids. Each virtual grid is bound to a physical space area and assigned a unique grid identifier.

[0055] The irregular stack sensing unit is used to passively detect engineering materials that have no surface markings, irregular shapes, and irregular stacking within a virtual grid, and to obtain the volume change of the materials before and after the change event.

[0056] The consumption behavior identification unit is used to identify the behavior type corresponding to material change events, and to distinguish between temporary displacement behavior and irreversible consumption behavior. Irreversible consumption behavior includes cutting behavior, mixing behavior and pouring behavior.

[0057] The residual material status determination unit is used to calculate the ratio of the volume of the remaining material after consumption to the volume of the original material when the material change event is identified as an irreversible consumption behavior. When the ratio is greater than the preset material consolidation threshold, the remaining material is marked as usable residual material. When the ratio is less than or equal to the preset material consolidation threshold, the remaining material is marked as waste material.

[0058] It should be noted that the grid identifier is a unique code assigned to each virtual grid, used to distinguish different virtual grids in the system and to bind material change events to the spatial location where the event occurred.

[0059] "No surface marking" refers to the absence of artificial markings such as RFID tags, QR codes, and barcodes on the surface of engineering materials that can be read by sensors, making it impossible for the system to identify the material's identity by reading the markings.

[0060] Irregular shape refers to engineering materials that do not have a standard geometric shape, such as bundles of steel bars, piles of sand and gravel, and stacks of bagged cement, whose shapes are irregular and vary depending on the stacking method.

[0061] Irregular stacking refers to the stacking of engineering materials in a non-standardized manner in a warehouse. For example, steel bars are leaning against the wall, sand and gravel are naturally piled into cones, and waste materials are scattered and piled up randomly. The materials block each other and have no fixed placement posture.

[0062] Passive detection refers to the sensing unit acquiring information about a material by receiving signals reflected or radiated by the material itself, without applying any active stimulation or requiring the material to cooperate.

[0063] The volume change refers to the increase or decrease in the volume of space occupied by materials within the same virtual grid between two adjacent sensing operations. It is calculated by comparing the three-dimensional point cloud data within the grid using the sensing unit.

[0064] A change event refers to the addition, removal, displacement, or morphological change of materials within a virtual grid. Each change event includes a timestamp, grid identifier, event type, material category, and amount of change.

[0065] Temporary displacement behavior refers to the behavior of moving materials from one location to another within a virtual grid without changing their own shape, such as a crane moving steel bars from a storage area to a processing area.

[0066] Irreversible consumption refers to the physical changes in materials that cannot be reversed during construction, including cutting, mixing, and pouring.

[0067] Cutting refers to the act of cutting or severing materials such as steel bars and structural steel by a cutting machine. After cutting, the volume of the original material is reduced and waste material is generated.

[0068] Mixing refers to the act of putting powdery materials such as cement and fly ash into a mixer and mixing them with other materials. After mixing, the original form of the materials disappears.

[0069] The pouring process refers to the act of pouring fluid materials such as concrete into a formwork to form a structure. After pouring, the fluid material solidifies and its original form cannot be restored.

[0070] The material consolidation threshold is a preset volume ratio used to distinguish between usable surplus material and waste material. When the ratio of the volume of the remaining material after consumption to the volume of the original material is greater than the threshold, it is determined to be usable surplus material; when it is less than or equal to the threshold, it is determined to be waste material. The material consolidation threshold ranges from 0 to 1. The lower the value, the more surplus material is determined to be usable surplus material; the higher the value, the more surplus material is determined to be waste material. The material consolidation threshold is determined as follows: based on the specifications and construction process requirements of the material, the minimum length or minimum volume of the material that can be used in the subsequent construction section is obtained, and the ratio of the minimum length or minimum volume to the original length or original volume of the material is used as the material consolidation threshold. For steel reinforcement materials, the material consolidation threshold is the ratio of the shortest cutting length required for the steel reinforcement of that specification in the construction drawings to the original length of the steel reinforcement. For example, for a rebar with an original length of 12 meters, if the shortest cutting length for that specification in the construction drawings is 1.5 meters, then the material consolidation threshold is 1.5 divided by 12, which equals 0.125. For structural steel materials, the material loading threshold is the ratio of the minimum weldable length to the original length of the structural steel; for granular or powdery materials that cannot be determined by length, the material loading threshold is 0, meaning that as long as there is any remaining material, it is marked as usable surplus material.

[0071] Usable surplus materials refer to materials that, after being consumed, still have engineering value and can be used in other construction areas or transferred to other sections.

[0072] Waste materials refer to materials that are unusable due to their small size or irregular shape after consumption, and are disposed of according to the waste disposal process.

[0073] In one specific implementation, the material warehousing system for a large hydropower station's key project includes a primary central warehouse, one secondary work area warehouse each for sections A and B, and six tertiary field warehouses. The system divides the primary central warehouse into four virtual grids, corresponding to the steel reinforcement storage area, cement storage area, sand and gravel storage area, and cable storage area, respectively. The secondary work area warehouse for section A is divided into two virtual grids, and the secondary work area warehouse for section B is divided into three virtual grids. The six tertiary field warehouses are located at the working faces of the water diversion tunnel, the dam foundation pit, and the power plant, with each tertiary field warehouse divided into one virtual grid.

[0074] Within the virtual grid of the three-level on-site storage area at the working face of the water diversion tunnel in Section A, there are threaded steel bars supplied by Party A and I-beams supplied by Party B. Neither material has surface markings, and they are piled together and covered by mud. The irregular stockpile sensing unit uses millimeter-wave radar to penetrate the mud obstruction to acquire three-dimensional point clouds of the two materials. Event cameras capture the actions of workers handling and cutting them, and automatically calculate the volume change when a change event occurs.

[0075] When a worker uses a cutting machine to cut a section of rebar supplied by Party A, the consumption behavior identification unit captures the movement characteristics of the cutting machine's sparks and the rebar's fracture through an event camera, determining that this event is a cutting action, which is an irreversible consumption behavior. The remaining material status determination unit then calculates the ratio of the remaining rebar volume to the original volume. The original volume is the complete volume of a single rebar, and the length of the remaining part after cutting is still 80% of the original length. The system presets a whole material threshold of 0.3, and the actual ratio is 0.8, which is greater than the threshold. Therefore, the remaining rebar is determined to be usable surplus material, marked, and retained in the inventory ledger, and subsequently allocated to the drainage corridor construction site for continued use.

[0076] When the workers used up the last section of rebar that day, the remaining part was only 15% of the original length. The ratio of 0.15 was less than the whole material threshold of 0.3. The remaining material status judgment unit marked it as a waste material head, generated a waste disposal record in the change ledger, and pushed it to the person responsible for waste recycling.

[0077] II. Dual-state data storage module

[0078] The dual-state data storage module includes an independent storage unit, a write control unit, and a traceability record unit;

[0079] Independent storage units are used to store physical state data records and ownership state data records in separate storage areas. The two independent storage areas are linked by a unified material identification code. Physical state data records include the grid position coordinates, occupied volume, surface material characteristics, and integrity identifier of the material.

[0080] The write control unit is used to control the write operation of physical state data recording to be executed only in response to the change event output by the multimodal sensing grid module, and not in response to write requests from other modules;

[0081] The traceability record unit is used to maintain a valid status flag in the ownership status data record. When ownership switching occurs, the valid status flag of the original ownership status data record is first set to invalid, then a new ownership status data record is generated. The traceability identifier pointing to the original record is written into the new record, and the valid status flag of the new record is set to valid.

[0082] Furthermore, in the independent storage unit, the surface material features are the material spectral features extracted and stored by the multimodal sensing grid module when the material first enters the site. When both material supplied by supplier A and material supplied by supplier B exist in the same grid and are of the same type, the material spectral features of the material to be identified are compared with the stored material spectral features of material supplied by supplier A. Based on the comparison results, the ownership attribute of the material is determined to be supplied by supplier A or supplier B, and the determination result is written into the ownership status data record of the material.

[0083] Furthermore, the traceability recording unit is also used to package the following data into a digitally signed settlement evidence package when the ownership transfer is completed:

[0084] The physical state consumption record sequence corresponding to the current ownership switch, the ownership state data records before and after the ownership switch, and the complete traceability chain tracing back from the ownership state data records involved in this ownership switch to the first ownership state data record when the material initially entered the site;

[0085] The settlement evidence package is used as an immutable data basis for determining the contractor's cost responsibility when settling accounts for excessive consumption of materials supplied by Party A.

[0086] It should be noted that independent storage areas refer to the separate storage spaces occupied by physical data records and ownership data records. The two can be deployed on different storage media. Physical data records are stored on edge computing nodes in the warehouse to ensure real-time write performance, while ownership data records are stored on cloud servers to ensure transaction consistency.

[0087] The material unified identification code is a unique code assigned to each material. The physical state data record and the ownership state data record establish a relationship through this code, realizing the correspondence between the two sets of independently stored data.

[0088] Grid location coordinates refer to the spatial location information of materials in the virtual grid. They are obtained by the multimodal sensing grid module when sensing change events and are used to locate the specific storage location of materials in multi-level warehouses.

[0089] Occupied volume refers to the spatial volume occupied by the material in the virtual grid. It is calculated by the multimodal sensing grid module through three-dimensional point cloud data and is dynamically updated as the material is consumed or moved.

[0090] Surface material characteristics refer to the physical material properties data of the material surface. Specifically, the material spectral characteristics are extracted and stored by the multimodal sensing grid module through the spectral sensor when the material first enters the site, which are used to distinguish materials of the same type but different batches.

[0091] The integrity indicator is a marker that indicates the physical integrity of materials. The value can be either intact, partially consumed, or completely consumed. It is automatically set by the multimodal sensing grid module based on the ratio of the volume of the remaining material after consumption to the volume of the original material.

[0092] A write operation refers to a data operation that creates or updates physical state data records. This operation is only triggered by change events output by the multimodal sensing grid module, and write requests from any other module are rejected.

[0093] The valid status flag is a status field in the ownership status data record. Its value is valid or invalid, indicating whether the ownership record is currently a valid ownership basis for the material. There is only one valid ownership record for the same material at any given time.

[0094] The traceability identifier is an identifier written into the newly generated ownership status data record that points to the original ownership status data record. Through this identifier, it is possible to trace back from the currently valid record to the first ownership status data record generated when the material initially entered the site.

[0095] Digital signature refers to the operation of cryptographically signing a settlement evidence package. After signing, the integrity of the settlement evidence package can be verified, and any modification to the contents of the evidence package can be detected.

[0096] The settlement evidence package refers to a digitally signed data packet generated by the traceability record unit when the ownership switch is completed. It includes the physical state consumption record sequence, ownership state data records before and after the ownership switch, and a complete traceability chain from the current record back to the first record.

[0097] The physical state consumption record sequence refers to the collection of all physical state consumption records related to this ownership switch. They are arranged in chronological order of the consumption events, and each record contains the timestamp of the consumption event, the consumption amount, and the consumption behavior type.

[0098] A complete traceability chain refers to a seamless link that starts from the current ownership status data record, traces back along the traceability markers one by one, until it reaches the first ownership status data record when the material initially entered the site.

[0099] In one specific implementation, the primary central warehouse of a large hydropower station receives a batch of 500 pieces of rebar supplied by Party A. Upon initial arrival, each piece of rebar in the batch is assigned a unique material identification code by an independent storage unit. A multimodal sensing grid module extracts the surface material spectral characteristics of the rebar using a spectral sensor, storing this data in the corresponding physical state data record. Simultaneously, it retrieves ownership information from the purchase order, identifying Party A as the party bearing the cost and the settlement price as the contract unit price, and writes this information into the ownership status data record. The valid status flag of the first ownership status data record is then set to valid.

[0100] Physical state data is recorded and stored on edge computing nodes at the primary central warehouse, while ownership state data is recorded and stored on a cloud server. The two are linked by a unified material identification code. Data is transmitted between the edge computing nodes and the cloud server via an encrypted channel.

[0101] A few days later, 200 pieces of the rebar in this batch were transferred to the Level 3 field warehouse of Section B. During the transfer process, when workers loaded the rebar from the Level 1 central warehouse, the write control unit received a loading change event output by the multimodal sensing grid module, allowing the update of the grid position coordinates in the physical state data record of these 200 pieces of rebar. After loading was completed, the grid position coordinates of these 200 pieces of rebar changed from the Level 1 central warehouse storage area grid to the Section B transport vehicle grid. When the material allocation management personnel attempted to directly modify the grid position coordinates of 50 pieces of rebar to Section A through the business system, the write control unit detected that the write request did not originate from the change event output by the multimodal sensing grid module, rejected the request, and recorded the rejection in the log.

[0102] During the construction of the tunnel face in Section B, workers continuously consumed this batch of rebar. When the consumption exceeded the upper limit of the quota calculated from the design quantity extracted from the BIM model and the loss coefficient, a change of ownership was triggered. The traceability record unit first set the valid status mark of the currently valid ownership data record to invalid, then generated a new ownership data record, rewrote the cost bearer to the identification of the construction party in Section B, rewrote the settlement price to the market settlement unit price, wrote a traceability identifier pointing to the original record in the new record, and set the valid status mark of the new record to valid.

[0103] After the ownership transfer is completed, the traceability recording unit automatically collects the physical state consumption record sequence corresponding to this ownership transfer, including the timestamp, consumption amount, and occurrence grid identifier for each cut. Simultaneously, it extracts the original ownership state data record before the transfer and the new ownership state data record after the transfer, tracing back along the traceability identifiers to the first ownership state data record generated when the material first entered the site, forming a complete traceability chain. The traceability recording unit packages these four parts of data and performs digital signature to generate a settlement evidence package. During the final settlement of the project, the client's auditors retrieve this settlement evidence package, verify the digital signature, confirm that the data has not been tampered with, and, based on the excess consumption amount and market settlement unit price recorded in the evidence package, include the excess consumption deduction amount in the settlement statement for Section B of the project.

[0104] III. Ownership Switching Module

[0105] The ownership switching module includes an over-consumption judgment unit, an over-consumption allocation unit, a switching execution unit, and an over-consumption tracing unit;

[0106] The over-consumption judgment unit is used to determine whether the main body consuming the material is a single construction section or multiple construction sections when the physical state data records show that the cumulative consumption of a certain material exceeds the quota limit, and to generate an ownership switching trigger signal.

[0107] The over-consumption allocation unit is used to, when it is determined that the main body of consumption is multiple construction sections, divide the over-quota consumption of the material into each section according to the consumption event sequence of each section recorded by the multimodal perception grid module, and generate the corresponding ownership switching signal for each section.

[0108] The switching execution unit is used to respond to ownership switching trigger signals or ownership switching signals, keep the physical state data record of the material unchanged, and generate a new ownership state data record;

[0109] The over-consumption traceability unit is used to write the timestamp of the consumption event that triggered the current ownership switch, the virtual grid identifier of the consumption event, the construction section identifier that performed the consumption, and the WBS process code corresponding to the consumption event into the ownership data record when generating a new ownership status data record. Based on the construction section identifier and WBS process code, it matches the corresponding settlement item from the bill of quantities and writes the over-consumption deduction amount of the materials supplied by the client into the deduction amount of the settlement item.

[0110] Furthermore, when the over-consumption allocation unit performs the over-quota consumption breakdown, it simultaneously performs the following operations:

[0111] The sequence of consumption events for the material in each contract section, the timestamp of each consumption event and the corresponding construction contract section identifier, the excess consumption amount allocated to each contract section after splitting and the corresponding market settlement unit price are associated and encapsulated to generate an unalterable responsibility determination data package bound to the material ownership switching signal.

[0112] The liability determination data packet is generated synchronously with the ownership transfer signal. When the ownership transfer signal is triggered, the liability determination data packet is written into the blockchain for evidence storage as the basis for settlement, along with the ownership transfer signal.

[0113] It should be noted that the cumulative consumption refers to the sum of the volume changes corresponding to each irreversible consumption behavior recorded by the multimodal sensing grid module since the first consumption of a certain material. It is obtained by the system from the consumption event sequence recorded in the physical state data.

[0114] The upper limit of the quota refers to the product of the theoretical usage of a certain material in an engineering project calculated based on the engineering design model or bill of quantities and the preset loss coefficient, which serves as the threshold for judging whether the material supplied by Party A is over-consumed.

[0115] The consuming entity refers to the construction section that performs the material consumption behavior. When the same material is consumed by multiple sections, the consuming entity is multiple construction sections.

[0116] The ownership switching trigger signal is a trigger signal generated by the overconsumption judgment unit when the consumption subject is a single construction section. It is used to directly drive the switching execution unit to perform the ownership switching operation.

[0117] The ownership switching signal refers to the trigger signal generated by the over-consumption allocation unit for each construction section after the over-consumption amount is divided into multiple construction sections when the main body of consumption is multiple construction sections. Each signal corresponds to the ownership switching operation of one construction section.

[0118] A consumption event sequence refers to the collection of events recorded by the multimodal sensing grid module in which a certain material is consumed. The events are arranged in chronological order of occurrence. Each event includes a timestamp, a virtual grid identifier, the identifier of the construction section where the consumption was performed, and the consumption amount.

[0119] The actual consumption ratio refers to the ratio of the actual consumption of a material by each construction section to the total consumption of that material when multiple construction sections share the same material.

[0120] Worksheet itemized structure (WBS) codes are unique identifiers for construction procedures within the bill of quantities (BOQ). They are used to associate material consumption with specific construction procedures, enabling accurate collection of over-consumption deductions. The WBS code is determined as follows: during system initialization, the BIM model and BOQ are imported, and a mapping relationship is established between the spatial coordinates of each construction location in the BIM model and the WBS code in the BOQ. The multimodal sensing grid module binds each virtual grid to the spatial coordinates of the construction location in the BIM model. When a material consumption event occurs, the system identifies the construction location based on the virtual grid identifier of the consumption event, through the binding relationship between the virtual grid and the construction location, and then automatically obtains the corresponding WBS code through the mapping relationship between the construction location and the WBS code. If multiple parallel procedures exist for the same construction location, the system further matches the type of material consumed and the type of consumption behavior in the consumption event. For example, the cutting of rebar matches a rebar installation procedure, and the mixing of cement matches a concrete pouring procedure, thus accurately determining the WBS code corresponding to the consumption event. If the system cannot automatically determine the status, it will mark the event as pending confirmation and push it to the construction management personnel's terminal, whereby the construction management personnel will manually select and confirm it from the list of candidate procedures associated with that construction site.

[0121] Settlement items refer to the pricing items in the bill of quantities that correspond to the WBS process codes, including the planned quantity, contract unit price, and deduction fields for that process.

[0122] The responsibility determination data packet refers to an immutable data packet generated synchronously by the over-consumption allocation unit when performing the over-quota consumption splitting, which is bound to the ownership switching signal. It contains the consumption event sequence of each section, the allocated over-quota consumption, and the market settlement unit price, and is used as the basis for settlement after being associated and encapsulated.

[0123] In one specific implementation, sections A and B of a large hydropower station's key project share the same batch of rebar supplied by the client. This batch of rebar is transferred from a primary central warehouse to a shared tertiary on-site warehouse located at the working face of the water diversion tunnel. A multimodal sensing grid module continuously records each consumption event, and the cumulative consumption recorded in the physical state data gradually increases as construction progresses.

[0124] On a certain morning, the cumulative consumption exceeded the upper limit of the quota calculated based on the design quantity of the BIM model and the quota loss coefficient for the first time. The over-consumption judgment unit read the consumption event sequence of this batch of rebar, identified that the consumption events came from two construction entities, A section and B section, and determined that the consumption entities were multiple construction sections. The data was then handed over to the over-consumption allocation unit for processing.

[0125] Based on the consumption event sequence recorded by the sensing grid module, the over-consumption allocation unit calculates that Section A actually consumed 12.5 tons, Section B actually consumed 7.5 tons, and the total consumption is 20 tons. The quota limit is 16 tons, and the over-quota consumption is 4 tons. The over-consumption allocation unit calculates the over-consumption allocation proportionally to the actual consumption, allocating 2.5 tons to Section A and 1.5 tons to Section B. Simultaneously with the splitting, the over-consumption allocation unit associates and encapsulates the consumption event sequence for Sections A and B, the timestamp of each consumption cut, the corresponding section identifier, the allocated over-quota consumption, and the market settlement unit price, generating a responsibility determination data package bound to the ownership transfer signal for this batch of materials. The responsibility determination data package is generated synchronously with the ownership transfer signal and written to the storage record through the system's built-in blockchain interface.

[0126] The switching execution unit responds to the ownership switching signals for sections A and B, maintaining the physical state data records of this batch of rebar unchanged, and generates new ownership state data records for sections A and B respectively. In the new record for section A, the cost-sharing information is rewritten with the contractor's identifier for section A, and the settlement price information is rewritten with the market settlement unit price. In the new record for section B, the cost-sharing information is rewritten with the contractor's identifier for section B, and the settlement price information is rewritten with the market settlement unit price. The original ownership state data records for both sections are marked as invalid.

[0127] When generating new ownership data records, the over-consumption tracing unit writes the timestamp of the cutting consumption event that triggered the over-consumption at 9:32 AM, the virtual grid identifier of the water diversion tunnel face grid where the event occurred, the section identifier of the execution of the consumption, and the WBS process code of the water diversion tunnel lining steel reinforcement installation process corresponding to the event into the ownership data records of sections A and B, respectively. Subsequently, the over-consumption tracing unit matches the settlement item for the water diversion tunnel lining steel reinforcement installation process from the bill of quantities based on the section identifier and WBS process code, and writes the over-consumption deduction amounts of 2.5 tons for section A and 1.5 tons for section B into the deduction amounts of their respective settlement items. During monthly project settlement, the financial system automatically reads the deduction amount and deducts it directly from the project payment for both sections.

[0128] IV. AI Data Flow Hardware Lock Module

[0129] The AI ​​data stream hardware lock module includes a loading event capture unit, a single vehicle material locking unit, and an exit blocking execution unit;

[0130] The loading event capture unit is used to identify events in which materials are loaded onto transport vehicles from the physical state change events output by the multimodal perception grid module, and to obtain the type, quantity, and time of loading of the materials.

[0131] The single-vehicle material locking unit is used to compare the material category and quantity corresponding to this loading event with the material category and quantity recorded in the outbound order to be executed. The comparison includes whether the material category loaded is consistent with the material category recorded in the outbound order and whether the material quantity loaded is within the preset deviation range of the material quantity recorded in the outbound order.

[0132] The exit blocking execution unit is used to send an unlock signal to the controlled hardware to allow the vehicle carrying the batch of materials to leave the warehouse when the single vehicle material locking unit matches. When the matching does not match, it sends a lock signal to the controlled hardware to prevent the vehicle from leaving and pushes an alarm containing vehicle information, loaded material information and matching difference information to the designated terminal.

[0133] Furthermore, the loading event capture unit is also used to send a pause loading signal to the exit blocking execution unit and push instruction information to the designated terminal to adjust the placement of materials when it detects that materials are obscured or stacked, causing some materials to be unidentifiable during the process of identifying loading events.

[0134] After the materials are adjusted, the single-vehicle material locking unit will re-perform the comparison until all the identified materials match the materials recorded in the outbound order to be executed, and then send an unlock signal to the outbound blocking execution unit.

[0135] It should be noted that the loading event refers to the behavioral event of materials being loaded onto a transport vehicle. It is identified by the multimodal perception grid module from physical state change events and includes the type, quantity, and time of loading of the materials.

[0136] The category of loaded materials refers to the type of materials actually loaded onto the vehicle during the loading event. This is determined by the multimodal perception grid module through material spectral feature comparison or visual recognition, such as rebar, cement, and cables.

[0137] The loading time refers to the specific point in time when the loading event is detected by the multimodal sensing grid module, which is automatically recorded by the system.

[0138] An outbound slip is a document that has been approved in the system and is awaiting execution, recording the type and quantity of materials to be shipped out.

[0139] The preset deviation range refers to the allowable error range between the actual quantity of goods loaded and the quantity recorded on the delivery note, as preset by the system. If the error exceeds this range, it is judged as a discrepancy.

[0140] Controlled hardware refers to physical barrier gate devices deployed at warehouse exits, which can receive unlocking or locking signals sent by the system, corresponding to two states: open for passage and closed for prohibition of passage, respectively.

[0141] The unlock signal is the opening command sent by the system to the controlled hardware. After receiving the unlock signal, the controlled hardware raises the barrier arm, allowing the vehicle to leave the warehouse.

[0142] A lock-up signal is a command sent by the system to the controlled hardware to keep the gate closed. Upon receiving the lock-up signal, the controlled hardware maintains the gate arm in a lowered state, preventing vehicles from leaving.

[0143] The comparison discrepancy information refers to the discrepancy record generated when the material locking unit of a single vehicle does not match, which includes specific differences in category or quantity between the materials recorded on the outbound slip and the actual loaded materials.

[0144] The pause loading signal refers to the pause command sent by the loading event capture unit to the exit blocking execution unit when it detects that the materials are obstructed or stacked, making some materials unidentifiable.

[0145] The instruction message for adjusting the placement of materials refers to the prompt message pushed to the designated terminal by the loading event capture unit when it detects that materials are obstructed or stacked, guiding the loading personnel to rearrange the materials to eliminate obstruction.

[0146] In one specific implementation, after the completion of Section B of a large hydropower station's key project, a batch of remaining materials needed to be loaded and dispatched from the secondary work area warehouse of Section B. The dispatch slip recorded that the materials to be dispatched were 8 tons of rebar. At 2 PM, a truck entered the loading area of ​​the secondary work area warehouse of Section B. The multimodal perception grid module monitored the change events within the virtual grid of the loading area in real time. When workers operated the crane to lift the materials into the truck bed, the loading event capture unit identified the loading behavior from the change event stream, obtaining that the types of materials loaded this time were rebar and cables, with a rebar loading volume of 8.2 tons and a cable loading volume of 0.5 tons, and the loading occurred between 2:15 PM and 2:30 PM.

[0147] After receiving the loading event record, the single-vehicle material locking unit queries the system for the pending outbound order for Section B. The outbound order records 8 tons of rebar. The single-vehicle material locking unit compares the actual loaded goods with the outbound order. Category comparison reveals that the outbound order only lists rebar, but the actual loaded goods include both rebar and cable; cable is not a category recorded on the outbound order. Quantity comparison shows that the rebar loading of 8.2 tons is within the preset deviation range of 8 tons recorded on the outbound order. Due to the inconsistency in category comparison, the single-vehicle material locking unit generates comparison discrepancy information—0.5 tons of cable, not recorded on the outbound order, are present in the loaded goods.

[0148] Upon receiving the discrepancy result, the exit blocking unit sent a lock signal to the controlled hardware at the exit of the secondary work area warehouse in Section B. The barrier gate remained in the lowered state, preventing the truck from leaving. Simultaneously, an alarm message was pushed to the designated terminals of the construction supervisor of Section B and the client's materials department. The alarm message included the license plate number, physical information about the load of 8.2 tons of rebar and 0.5 tons of cable, and the discrepancy that the cable did not belong to the category recorded on the outbound manifest. Upon receiving the alarm, the client's materials department immediately went to the site to verify and confirmed that the cable was supplied by the client. Construction personnel attempted to mix the client-supplied cable with the supplier-supplied rebar to leave the site, but this was intercepted by the system in real time.

[0149] In another loading and unloading operation, a truck loaded with rebar was preparing to leave the secondary work area warehouse of Section A. During the loading process, workers tightly stacked multiple bundles of rebar in the truck bed, with some of the lower-layer rebar completely obscured by the upper-layer rebar. When identifying the loading event, the loading event capture unit detected that the material obstruction in the bottom area of ​​the truck bed was causing some rebar to be unidentifiable in terms of type and quantity. The loading event capture unit immediately sent a stop loading signal to the unloading blocking execution unit, and simultaneously pushed an instruction to the designated terminal at the loading area to adjust the material placement, prompting the loading personnel to temporarily lift the upper-layer rebar, exposing the lower-layer rebar to the detection range of the sensing grid.

[0150] Following instructions, the loading personnel adjusted the placement of the rebar, spreading it evenly layer by layer. After the adjustment, the single-vehicle material locking unit re-performed the comparison. The multimodal sensing grid module identified each rebar individually, confirming that all loaded materials were rebar, the total quantity was within the preset deviation range recorded on the outbound slip, and all materials matched the outbound slip. The single-vehicle material locking unit sent the matching result to the exit blocking execution unit, which then sent an unlock signal to the controlled hardware. The barrier gate lifted, and the vehicle departed normally.

[0151] V. Multi-Agent Game Optimization Module

[0152] The multi-agent game optimization module includes an adversarial example generation unit and an anti-game scheduling unit;

[0153] The adversarial example generation unit is used to train a generator network based on the deviation between the inventory data reported by each construction entity in history and the actual consumption data obtained by the multimodal perception grid module, so as to simulate the false reporting strategy of each construction entity in the next cycle and generate the estimated false inventory data of each construction entity.

[0154] The anti-game scheduling unit is used to input the estimated false inventory data and the real consumption data obtained by the multimodal perception grid module into the discriminator network, output the false reporting probability of each construction entity, and reduce the material application priority or restrict the application request of entities whose false reporting probability exceeds the preset false reporting threshold.

[0155] It should be noted that the historically reported inventory data refers to the material inventory data submitted to the system by each construction entity in each historical period, including the current inventory quantity filled in when applying for requisition.

[0156] Actual consumption data refers to the actual consumption of materials by each construction entity in each time period, as perceived and recorded by the multimodal sensing grid module in a fully passive and non-contact manner.

[0157] Deviation refers to the numerical difference between the inventory data reported by the construction entity and the actual consumption data obtained by the multimodal sensing grid module, reflecting the degree to which the entity has overstated its inventory.

[0158] Generator networks refer to neural networks in adversarial learning frameworks that are responsible for generating simulated data. The training objective is to learn the false reporting behavior patterns of various construction entities and output realistically estimated false inventory data.

[0159] The overstatement strategy refers to the behavior pattern adopted by construction entities when reporting inventory, which involves concealing the true inventory and exaggerating the demand. This strategy is learned by the generator network from historical deviation data.

[0160] The estimated false inventory data refers to the generator network's prediction of the amount of inventory that each construction entity may falsely report in the next period, based on the false reporting strategies learned by each entity.

[0161] A discriminator network is a neural network in an adversarial learning framework that is responsible for distinguishing between real data and generated data. The training objective is to accurately identify whether the input data is real consumed data or predicted false data generated by the generator network.

[0162] The false reporting probability refers to the probability value output by the discriminator network after analyzing the input data, which represents the probability that a construction entity is currently making a false reporting behavior. The value is between 0 and 1.

[0163] The preset false alarm threshold refers to the critical value of false alarm probability pre-set by the system. When the false alarm probability output by the discriminator network exceeds this threshold, restrictive measures are triggered. The preset false alarm threshold is determined as follows: the threshold is set to a default value of 0.5 during the initial deployment of the system. During system operation, historical samples marked as false alarms by the discriminator network and confirmed by manual verification are collected from each construction entity as positive samples, and historical samples marked as false alarms but confirmed by manual verification are collected as false alarm samples. Each candidate threshold is traversed in increments of 0.05 between 0 and 1, and the false alarm recognition accuracy and false alarm rate under each candidate threshold are calculated. The candidate threshold with the highest false alarm recognition accuracy and a false alarm rate lower than the preset false alarm tolerance limit is selected as the adjusted preset threshold. The preset threshold ranges from 0 to 1. The lower the value, the stricter the false alarm restriction, and the higher the value, the more lenient the false alarm restriction. Manual adjustment is supported according to the material shortage and management strictness of the project. The higher the material shortage or the stricter the management, the lower the threshold setting.

[0164] Material application priority refers to the ranking weight of each construction entity when allocating materials. Entities with higher credit scores or lower probability of false reporting are ranked higher and given priority in material allocation.

[0165] In one specific implementation, three construction entities—Sections A, B, and C—of a large hydropower station's key project simultaneously apply for the next batch of cement supplied by the client. The primary central warehouse currently has limited cement inventory, necessitating a decision on allocation priorities.

[0166] The adversarial example generation unit retrieved historical reported inventory data from the past three months for the three contract sections and actual consumption data recorded by the multimodal perception grid module for the same period. The data shows that the historical reported inventory and actual consumption deviations for Contract Section A were relatively small, with a monthly average deviation within 3%. For Contract Section B, the historical reported inventory and actual consumption deviations fluctuated significantly, with smaller deviations during the dry season and off-season construction, but a significant increase during the rush to complete work before the main flood season, reaching 18% in the most recent month. For Contract Section C, the historical reported inventory and actual consumption deviations remained consistently high, with a monthly average deviation exceeding 25%.

[0167] The adversarial example generation unit uses this historical bias data to train the generator network. The generator network learns the behavioral pattern of Section B tending to overstate inventory to hoard materials during peak construction periods, as well as Section C's consistent habit of overstating inventory. It also identifies a pattern where Section A's reported data is largely consistent with actual consumption. In this requisition cycle, the generator network outputs estimated overstated inventory data for the three sections: Section B's estimated overstated inventory is 15% of the declared amount, Section C's is 22%, and Section A's is close to zero.

[0168] The anti-game scheduling unit inputs the estimated false inventory data output by the generator network and the actual consumption data obtained by the multimodal perception grid module from the three-level field database of each section into the discriminator network. After analysis and comparison, the discriminator network outputs a false alarm probability of 0.05 for section A, 0.18 for section B, and 0.28 for section C. The preset threshold is 0.2.

[0169] The anti-game scheduling unit makes scheduling decisions based on the false alarm probability output by the discriminator network. For segment A, the false alarm probability is 0.05, below the threshold, maintaining normal material requisition priority. For segment B, the false alarm probability is 0.18, close to but not exceeding the threshold, marking it as an observation target, maintaining normal priority but limiting the maximum amount that can be requisitioned at one time. For segment C, the false alarm probability is 0.28, exceeding the threshold, reducing the material requisition priority to the lowest level, and allocation is only made if there is still remaining inventory after normal requisitions in each segment are fulfilled.

[0170] Based on the aforementioned priority ranking, the primary central warehouse allocated the current cement inventory to Sections A and B. Section A received the full requested quantity, Section B received the allocation within the upper limit of the requested quantity, and Section C, due to its lowest priority, did not receive any allocation this time. The construction manager of Section C adjusted its reporting strategy in subsequent requisition cycles, gradually bringing the reported inventory closer to the accurate value, with the probability of false reporting decreasing to below 0.2 per month, and the material requisition priority returning to normal.

[0171] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. An AI management system for multi-level warehouses supplying engineering materials to suppliers A and B, characterized in that: include: The multimodal sensing grid module is used to divide a multi-level warehouse into multiple virtual grids, and to sense changes in materials within each grid in a completely passive and non-contact manner. The dual-state data storage module is used to maintain independently stored physical state data records and ownership state data records for each material. The physical state data records are updated by the multimodal sensing grid module and cannot be modified by the business process. The ownership state data records include cost-bearing information and settlement price information. The ownership switching module is used to maintain the physical status data record of a material unchanged and generate a new ownership status data record when the physical status data record shows that the cumulative consumption of a material exceeds the quota limit. In this new record, the cost-bearing information is rewritten to the contractor's identifier, the settlement price information is rewritten to the market settlement unit price, and the original ownership status data record is marked as invalid. The AI ​​data flow hardware lock module is used to unlock the controlled hardware to allow the process to proceed when the physical state change event of materials matches the business document; when the match fails, the hardware remains locked and an alarm is triggered. The multi-agent game optimization module is used to identify inventory overstatement behavior based on the actual consumption data of each construction entity through an adversarial learning mechanism, and to restrict the material requisition of entities that have been found to have engaged in overstatement behavior.

2. The AI ​​management system for multi-level warehouses supplying engineering materials to suppliers A and B according to claim 1, characterized in that, The multimodal sensing mesh module includes: Grid partitioning unit is used to divide a multi-level warehouse into multiple virtual grids. Each virtual grid is bound to a physical space area and assigned a unique grid identifier. The irregular stack sensing unit is used to passively detect engineering materials that have no surface markings, irregular shapes, and irregular stacking within a virtual grid, and to obtain the volume change of the materials before and after the change event. The consumption behavior identification unit is used to identify the behavior type corresponding to the material change event, and to distinguish between temporary displacement behavior and irreversible consumption behavior. The irreversible consumption behavior includes cutting behavior, mixing behavior and pouring behavior. The residual material status determination unit is used to calculate the ratio of the volume of the remaining material after consumption to the volume of the original material when the material change event is identified as an irreversible consumption behavior. When the ratio is greater than the preset material consolidation threshold, the remaining material is marked as usable residual material. When the ratio is less than or equal to the preset material consolidation threshold, the remaining material is marked as waste material.

3. The multi-level warehouse AI management system for engineering materials supplied by suppliers A and B according to claim 1, characterized in that, The dual-state data storage module includes: An independent storage unit is used to store physical state data records and ownership state data records in separate storage areas. The two independent storage areas are linked by a unified material identification code. The physical state data records include the grid position coordinates, occupied volume, surface material characteristics, and integrity identifier of the material. The write control unit is used to control the write operation of physical state data recording to be executed only in response to the change event output by the multimodal sensing grid module, and not in response to write requests from other modules; The traceability record unit is used to maintain a valid status flag in the ownership status data record. When ownership switching occurs, the valid status flag of the original ownership status data record is first set to invalid, then a new ownership status data record is generated. The traceability identifier pointing to the original record is written into the new record, and the valid status flag of the new record is set to valid.

4. The multi-level warehouse AI management system for engineering materials supplied by suppliers A and B according to claim 3, characterized in that, In the independent storage unit, the surface material feature is the material spectral feature extracted and stored by the multimodal sensing grid module when the material first enters the site. When there are both materials supplied by supplier A and supplier B in the same grid and they are of the same type, the material spectral feature of the material to be identified is compared with the material spectral feature of supplier A that has been stored. Based on the comparison result, the ownership attribute of the material is determined to be supplied by supplier A or supplier B, and the determination result is written into the ownership status data record of the material.

5. The AI ​​management system for multi-level warehouses supplying engineering materials to suppliers A and B according to claim 3, characterized in that, The traceability recording unit is also used to package the following data into a digitally signed settlement evidence package when the ownership transfer is completed: The physical state consumption record sequence corresponding to the current ownership switch, the ownership state data records before and after the ownership switch, and the complete traceability chain tracing back from the ownership state data records involved in this ownership switch to the first ownership state data record when the material initially entered the site; The settlement evidence package is used as an immutable data basis for determining the contractor's cost responsibility when settling accounts for excessive consumption of materials supplied by Party A.

6. The AI ​​management system for multi-level warehouses supplying engineering materials to suppliers A and B according to claim 1, characterized in that, The ownership switching module includes: The over-consumption judgment unit is used to determine whether the main body consuming the material is a single construction section or multiple construction sections when the physical state data records show that the cumulative consumption of a certain material exceeds the quota limit, and to generate an ownership switching trigger signal. The over-consumption allocation unit is used to, when it is determined that the main body of consumption is multiple construction sections, divide the over-quota consumption of the material into each section according to the consumption event sequence of each section recorded by the multimodal perception grid module, and generate the corresponding ownership switching signal for each section. The switching execution unit is used to respond to ownership switching trigger signals or ownership switching signals, keep the physical state data record of the material unchanged, and generate a new ownership state data record; The over-consumption traceability unit is used to write the timestamp of the consumption event that triggered the current ownership switch, the virtual grid identifier of the consumption event, the construction section identifier that performed the consumption, and the WBS process code corresponding to the consumption event into the ownership data record when generating a new ownership status data record. Based on the construction section identifier and WBS process code, it matches the corresponding settlement item from the bill of quantities and writes the over-consumption deduction amount of the materials supplied by the client into the deduction amount of the settlement item.

7. The multi-level warehouse AI management system for engineering materials supplied by suppliers A and B according to claim 6, characterized in that, When the excess consumption allocation unit performs the splitting of excess consumption, it simultaneously performs the following operations: The sequence of consumption events for the material in each contract section, the timestamp of each consumption event and the corresponding construction contract section identifier, the excess consumption amount allocated to each contract section after splitting and the corresponding market settlement unit price are associated and encapsulated to generate an unalterable responsibility determination data package bound to the material ownership switching signal. The responsibility determination data packet is generated synchronously with the ownership switching signal. When the ownership switching signal is triggered, the responsibility determination data packet is written into the blockchain for evidence storage as the basis for settlement, along with the ownership switching signal.

8. The multi-level warehouse AI management system for engineering materials supplied by suppliers A and B according to claim 1, characterized in that, The AI ​​data stream hardware lock module includes: The loading event capture unit is used to identify events in which materials are loaded onto transport vehicles from the physical state change events output by the multimodal perception grid module, and to obtain the type, quantity, and time of loading of the materials. The single-vehicle material locking unit is used to compare the material category and quantity corresponding to the current loading event with the material category and quantity recorded in the outbound order to be executed. The comparison includes whether the material category loaded is consistent with the material category recorded in the outbound order and whether the material quantity loaded is within the preset deviation range of the material quantity recorded in the outbound order. The exit blocking execution unit is used to send an unlock signal to the controlled hardware to allow the vehicle carrying the batch of materials to leave the warehouse when the single vehicle material locking unit matches. When the matching does not match, it sends a lock signal to the controlled hardware to prevent the vehicle from leaving and pushes an alarm containing vehicle information, loaded material information and matching difference information to the designated terminal.

9. The multi-level warehouse AI management system for engineering materials supplied by suppliers A and B according to claim 8, characterized in that, The loading event capture unit is also used to send a pause loading signal to the exit blocking execution unit and push instruction information to the designated terminal to adjust the placement of materials when it detects that materials are blocked or stacked, causing some materials to be unidentifiable during the process of identifying loading events. The single-vehicle material locking unit re-executes the comparison after the material is adjusted, until all the identified materials match the materials recorded in the outbound order to be executed, and then sends an unlock signal to the outbound blocking execution unit.

10. The AI ​​management system for multi-level warehouses supplying engineering materials to suppliers A and B according to claim 1, characterized in that, The multi-agent game optimization module includes: The adversarial example generation unit is used to train a generator network based on the deviation between the inventory data reported by each construction entity in history and the actual consumption data obtained by the multimodal perception grid module, so as to simulate the false reporting strategy of each construction entity in the next cycle and generate the estimated false inventory data of each construction entity. The anti-game scheduling unit is used to input the estimated false inventory data and the real consumption data obtained by the multimodal perception grid module into the discriminator network, output the false reporting probability of each construction entity, and reduce the material application priority or restrict the application request of entities whose false reporting probability exceeds the preset false reporting threshold.

Images