An AI-based laboratory consumables management system and method

Abstract

This invention provides an AI-based laboratory consumables management system and method, belonging to the field of AI management. The system includes: a model building layer for constructing item models of different consumables; an AI recognition layer for intelligent AI recognition of each consumable item received by students based on the item models; a statistics layer for recording the quantity of consumables issued and the remaining quantity of consumables based on their item attributes and the results of intelligent AI recognition, and simultaneously recording the quantity of corresponding consumables received based on their item attributes; and an early warning layer for formulating a first early warning strategy and issuing an early warning when the item attribute is a consumable attribute, based on the issued and remaining quantities; and for formulating a second early warning strategy and issuing an early warning when the item attribute is a non-consumable attribute, based on the issued, remaining, and received quantities. This system enables the management of medical consumables.

Description

Technical Field

[0001] This invention relates to the field of AI management technology, and in particular to an AI-based laboratory consumables management system and method. Background Technology

[0002] In the medical field, the management of consumables is crucial. Currently, the management of consumables such as medical cotton swabs and scalpels typically relies on manual record-keeping, which is inefficient and prone to errors. For example, medical cotton swabs are for single use only after being issued and cannot be restocked; therefore, the issued and remaining quantities must be accurately recorded for timely replenishment. Scalpels, after being issued, also need to be restocked, requiring the simultaneous recording of issued, remaining, and restocked quantities, making the management process quite complex. Furthermore, traditional management methods cannot develop timely early warning strategies based on the properties of consumables, easily leading to problems such as insufficient inventory or overstocking.

[0003] Therefore, this invention proposes an AI-based experimental consumables management system and method. Summary of the Invention

[0004] This invention provides an AI-based laboratory consumables management system and method to solve the aforementioned technical problems.

[0005] This invention provides an AI-based laboratory consumables management system, comprising:

[0006] The model building layer is used to build item models for different consumable items;

[0007] The AI ​​recognition layer is used to intelligently recognize each consumable item received by a student based on the item model when the student receives the consumable item.

[0008] The statistics layer is used to record the outbound quantity of consumables based on their item attributes and combined with the results of intelligent AI recognition, and to count the remaining quantity of consumables. At the same time, it counts the inbound quantity of corresponding consumables based on their item attributes.

[0009] The early warning layer is used to formulate a first early warning strategy and issue an early warning reminder based on the quantity issued and the remaining quantity when the item attribute is a loss attribute;

[0010] When the item attribute is a non-consumable attribute, a second early warning strategy is formulated and an early warning is issued based on the outbound quantity, remaining quantity, and inbound quantity.

[0011] Preferably, the model building layer includes:

[0012] Architecture building unit, used to build a two-layer item model architecture that includes basic attributes and state attributes;

[0013] A coupling unit for real-time synchronization of inventory change data and quality inspection data through a bidirectional data interface, realizing dynamic coupling of basic attributes and status attributes.

[0014] Preferably, the model construction layer further includes:

[0015] A function construction unit for setting a loss prediction module based on an LSTM neural network and training a loss trend function based on a historical usage data sequence X=(x1,x2,...,xt) σ(Whht-1+Wxxt+b), where ht is the hidden layer state at time t, σ is the activation function; b is the parameter coefficient; x1, x2, xt are the first, second, and t-th sequence values respectively; Wh, Wx are the weighting coefficients;

[0016] A weight update unit for introducing a fuzzy logic rule base to process unstructured data. When a consumable item of a non-loss item type shows a visible abnormality, the weight of εg is dynamically adjusted through the membership function μ(d)=1 / (1+e-k(d-d0)), where d is the degree of the visible abnormality, d0 is the critical threshold of the degree of abnormality, and k is the gradient adjustment coefficient; e is a constant with a value of 2.7;

[0017] A model update unit for realizing uncertainty reasoning of attribute parameters through a Bayesian network and establishing an attribute change probability matrix Updating the corresponding item model, where the dynamic attribute parameters of the item model maintain a mapping with the actual usage status, Respectively representing the change probabilities in different states.

[0018] Preferably, the warning layer includes:

[0019] A loss attribute analysis unit for constructing a dynamic consumption prediction model when the item attribute is a loss attribute, extracting the export quantity sequence of the last M cycles through a time sliding window, calculating the average consumption rate, and quantifying the consumption fluctuation through the standard deviation;

[0020] Based on the normal distribution confidence interval, setting a basic safety inventory threshold and introducing a fluctuation compensation term;

[0021] At the same time, combining the department usage priority coefficient to obtain a dynamic safety inventory threshold Sy1;

[0022] When the remaining quantity R satisfies R<1.2Sy1, a yellow warning is triggered;

[0023] When R<Sy1, a red warning is triggered;

[0024] Implementing hierarchical warning through the warning level L1 and associating different response strategies according to the level;

[0025]

[0026] in, This is the floor symbol.

[0027] Preferably, the early warning layer includes:

[0028] The non-loss attribute analysis unit is used to construct a full-cycle inventory status model when the item attribute is a non-loss attribute. It defines the quantity in stock R1, the quantity requested for warehousing Sn, and the quantity in transit Zn, and calculates the effective inventory In = R1 + Sn - Zn.

[0029] An LSTM network is trained based on historical usage sequences to predict the remaining usable usage Nr, combined with a single-use loss coefficient. Given the rated total number of cycles En, calculate the available inventory Iy after loss reduction. ;

[0030] Calculate the average inbound delay based on the outbound-inbound time series D. And introduce a delay compensation factor Dynamically adjust the early warning threshold, among which, Standard delay;

[0031]

[0032]

[0033] in, This is the lower limit of the threshold. This represents the average outbound rate. This is the replenishment cycle; x represents the maximum inventory capacity; This represents the average inbound rate.

[0034] when At that time, a supply warning is triggered;

[0035] when At that time, a backlog warning was triggered.

[0036] Preferably, the basic attribute layer includes at least item type, item identifier, physical specifications, sterilization status, and storage conditions, and the status attribute layer sets the usage frequency association parameter group for consumable items of the consumable item type.

[0037] Set up a multi-dimensional loss assessment parameter group for consumable items that are not consumable items.

[0038] Preferably, the usage frequency-related parameter set includes a dynamic consumption coefficient kt based on a time sliding window;

[0039]

[0040] in, , As a seasonal fluctuation moderating factor; The concentration coefficient for departmental requisition; This represents the cumulative outbound volume over the past n days. This is the standard daily consumption benchmark value.

[0041] Preferably, the multidimensional loss assessment parameter set includes: the geometric loss degree εg, the heat treatment performance degradation degree εh, and the surface coating failure probability εc of the corresponding consumable item, constructing a loss vector E=(εg,εh,εc), and determining the weight ωi of each component through the analytic hierarchy process to form a comprehensive loss index ε=Σ(ωi). εi).

[0042] This invention provides an AI-based method for managing laboratory consumables, comprising:

[0043] Step 1: Construct item models for different consumable items;

[0044] Step 2: When students collect consumables, the system performs intelligent AI recognition on each consumable based on the item model.

[0045] Step 3: Record the quantity of consumables issued based on their attributes and the results of intelligent AI recognition, and calculate the remaining quantity of consumables. At the same time, calculate the quantity of consumables received based on their attributes.

[0046] Step 4: When the item attribute is a loss attribute, formulate a first early warning strategy and issue an early warning reminder based on the quantity issued and the remaining quantity;

[0047] When the item attribute is a non-consumable attribute, a second early warning strategy is formulated and an early warning is issued based on the outbound quantity, remaining quantity, and inbound quantity.

[0048] Preferably, step 1 includes:

[0049] Establish a two-layer item model architecture that includes basic attributes and status attributes;

[0050] By synchronizing inventory change data and quality inspection data in real time through a two-way data interface, dynamic coupling of basic attributes and status attributes is achieved.

[0051] Compared with the prior art, the beneficial effects of this application are as follows:

[0052] Through a closed-loop process encompassing model building, intelligent identification, quantity control, and tiered early warning, the system enables digital mapping of static features and dynamic states in the item model, accurately depicting the entire lifecycle of consumables. Multimodal AI identification breaks through the limitations of single identifiers, quantity statistics adapt to differences in loss / non-loss attributes, and blockchain-based evidence storage ensures data credibility. Early warning strategies embed dynamic formulas (safety stock, balance indicators) and loss prediction, achieving an intelligent upgrade in medical consumables management.

[0053] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0054] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0055] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0056] Figure 1 This is a structural diagram of an AI-based experimental consumables management system according to an embodiment of the present invention;

[0057] Figure 2 This is a flowchart of an AI-based experimental consumables management method in an embodiment of the present invention. Detailed Implementation

[0058] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0059] This invention provides an AI-based laboratory consumables management system, such as... Figure 1 As shown, it includes:

[0060] The model building layer is used to build item models for different consumable items;

[0061] The AI ​​recognition layer is used to intelligently recognize each consumable item received by a student based on the item model when the student receives the consumable item.

[0062] The statistics layer is used to record the outbound quantity of consumables based on their item attributes and combined with the results of intelligent AI recognition, and to count the remaining quantity of consumables. At the same time, it counts the inbound quantity of corresponding consumables based on their item attributes.

[0063] The early warning layer is used to formulate a first early warning strategy and issue an early warning reminder based on the quantity issued and the remaining quantity when the item attribute is a loss attribute;

[0064] When the item attribute is a non-consumable attribute, a second early warning strategy is formulated and an early warning is issued based on the outbound quantity, remaining quantity, and inbound quantity.

[0065] In this embodiment, the item model is a digital twin system that integrates basic attributes (static inherent characteristics) and dynamic attributes (characteristics that change with usage / time), supporting the full lifecycle management of consumables:

[0066] Basic attributes: Record inherent information of consumables, such as wear and tear attribute markings for medical cotton swabs, QR code + RFID dual identification (built-in batch and sterilization date), 10cm sterile specification, and initial inventory of 10,000 swabs; non-wear attribute markings for surgical scalpels, laser-etched serial number + RFID (associated with production batch and blade parameters), 22-blade specification, and initial inventory of 50 scalpels. Specifically, basic information is entered through the hospital information system (HIS), connected to supplier data, and blockchain is used to ensure that the data is tamper-proof (such as sterilization date and production batch are recorded on the blockchain to prevent counterfeiting).

[0067] Dynamic attributes: depict changes in the state of consumables.

[0068] Medical cotton swabs: By capturing historical data through the inventory system, the usage frequency coefficient is calculated as: total outbound volume in the past 30 days ÷ 30 (e.g., 6000 swabs ÷ 30 days = 200 swabs / day), reflecting the average daily consumption intensity.

[0069] Surgical scalpels integrate laser edge scanning (to measure wear), RFID usage records (to track usage frequency), and timestamps (to record usage duration). The wear index is dynamically updated using the formula: Wear Index = 0.4 × Number of Uses + 0.3 × Usage Duration + 0.3 × Edge Wear (e.g., a scalpel used 10 times for 5 hours with an edge wear of 0.3, has a wear index of 5.59). Real-time status tracking is achieved through real-time calculations using IoT sensors (scalpel vibration sensors, cotton swab scanning devices).

[0070] In this embodiment, the intelligent AI recognition adopts multimodal fusion technology to simultaneously collect image features (appearance and morphology) and RFID data (electronic identification), and compare them with the item model to achieve accurate judgment. For example, when receiving medical cotton swabs: the binocular camera recognizes the text "sterile cotton swabs" and the blue and white packaging, the ultra-high frequency RFID reader reads the batch "20250701", and the system compares the "specification + sterilization date" in the model. If the confidence level is ≥95%, it is determined to be valid.

[0071] Scalpel Receiving: A camera identifies the outline of the No. 22 blade edge, a laser scans the serial number "SK2025001," and the RFID retrieves the last sterilization date "2025-06-30." A pressure sensor detects the grip feel of the handle (verifying if it's a registered instrument), confirming compliance through multiple dimensions. Hardware: Deploys a binocular camera (to identify subtle features, such as the notch on the scalpel blade) and an UHF RFID reader (for long-range, rapid identification, supporting parallel reading of multiple items). Algorithm: Based on a YOLOv8-trained consumable appearance recognition model, it fuses image and RFID data using DS evidence theory, outputting recognition confidence (e.g., confidence <90% triggers manual verification), supporting simultaneous identification and judgment during batch receivables.

[0072] In this embodiment, the outbound quantity is recorded automatically after a student successfully receives the supplies (e.g., receiving 100 cotton swabs → outbound +100; receiving 5 scalpels → outbound +5). Relying on the batch recognition function of the smart terminal, the system supports one-click deduction when multiple items are received at the same time, triggering real-time inventory updates.

[0073] Remaining quantity: Loss attribute (e.g., cotton swabs): The formula is remaining quantity = initial inventory - cumulative outbound quantity (e.g., initial 10,000 swabs, cumulative outbound 3,000 swabs → remaining 7,000 swabs).

[0074] Non-destructible attributes (such as scalpels): The formula is: Remaining quantity = Initial inventory - Cumulative outbound quantity + Cumulative inbound quantity (e.g., initial 50 scalpels, cumulative outbound 20 scalpels, inbound 10 scalpels → remaining 40 scalpels). Specifically, inventory data is synchronized to the management system in real time, and each outbound and inbound operation is recorded on the blockchain (recording the operator, time, and quantity) to ensure traceability and tamper-proof.

[0075] Quantity of items received: Only for non-consumable items (such as scalpels). Upon return, items must be identified by AI (confirmation of serial number and appearance) + sterilization test (UV scanning to verify sterility). Once qualified, the inventory will be increased (e.g., 3 returned → +3 received) to prevent the return of unsterilized instruments.

[0076] (1) Loss attribute warning (taking medical cotton swabs as an example)

[0077] Based on the dynamic safety stock formula: the safety stock threshold uses a frequency coefficient, a warning cycle number of days, and a safety factor (e.g., 200 units / day × 7 days × 1.5 = 2100 units). When the remaining quantity is less than the safety stock threshold, a tiered warning is issued.

[0078] Yellow alert (1800 units remaining, 1500 < 1800 < 2100): Initiate regular procurement (e.g., send a purchase requisition to the logistics department).

[0079] Red alert (1500 units remaining, 1500 = 2100 × 0.714): Triggers emergency reserve allocation (automatically generates cross-departmental allocation work order, prioritizing emergency care).

[0080] (2) Non-destructive attribute warning (taking a scalpel as an example)

[0081] Based on the inventory balance model: Inventory balance indicators: remaining quantity, inbound quantity, outbound quantity, initial inventory. Set a threshold range [0.8, 1.2]. If the value exceeds the range, an alert will be issued: replenishment alert (e.g., initial 50 scalpels, remaining 30 scalpels, 5 scalpels inbound, 20 scalpels outbound → balance = 0.3 < 0.8): prompt to replenish inventory, and trigger the scalpel maintenance plan (such as blade sharpening, sterilization).

[0082] Backlog warning (e.g., 60 units remaining, 15 units received, 5 units shipped out → balance = 1.4 > 1.2): This prompts the discontinuation of inefficient inventory (e.g., sealing off idle equipment, detecting performance degradation).

[0083] Advanced logic: Combining the LSTM loss prediction model, the threshold is dynamically adjusted. If the scalpel wear is predicted to accelerate in the next 15 days (e.g., increased blade wear rate), the lower threshold is lowered to 0.7; if the wear is slow (e.g., high idle rate), the upper threshold is raised to 1.3, thus achieving "loss-inventory" linkage control.

[0084] Specifically, the system calculates balance indicators in real time, dynamically adjusts thresholds by integrating loss prediction results, and provides differentiated responses based on early warnings (replenishment → procurement + maintenance; backlog → storage + detection).

[0085] The beneficial effects of the above technical solution are as follows: through the closed-loop process of model building → intelligent identification → quantity control → hierarchical early warning, the item model can achieve digital mapping between static features and dynamic states, accurately depict the entire life cycle of consumables, multimodal AI identification breaks through the limitations of single identifiers, quantity statistics adapt to the differences between loss and non-loss attributes, and blockchain notarization ensures data credibility; the early warning strategy embeds dynamic formulas (safety stock, balance index) and loss prediction, achieving an intelligent upgrade of medical consumable management.

[0086] This invention provides an AI-based laboratory consumables management system, wherein the model building layer includes:

[0087] Architecture building unit, used to build a two-layer item model architecture that includes basic attributes and state attributes;

[0088] The coupling unit is used to synchronize inventory change data and quality inspection data in real time through a bidirectional data interface, thereby achieving dynamic coupling between basic attributes and status attributes.

[0089] Preferably, the basic attribute layer includes at least item type, item identifier, physical specifications, sterilization status, and storage conditions, and the status attribute layer sets the usage frequency association parameter group for consumable items of the consumable item type.

[0090] Set up a multi-dimensional loss assessment parameter group for consumable items that are not consumable items.

[0091] Preferably, the usage frequency-related parameter set includes a dynamic consumption coefficient kt based on a time sliding window;

[0092]

[0093] in, , As a seasonal fluctuation moderating factor; The concentration coefficient for departmental requisition; This represents the cumulative outbound volume over the past n days. This is the standard daily consumption benchmark value.

[0094] Preferably, the multidimensional loss assessment parameter set includes: the geometric loss degree εg, the heat treatment performance degradation degree εh, and the surface coating failure probability εc of the corresponding consumable item, constructing a loss vector E=(εg,εh,εc), and determining the weight ωi of each component through the analytic hierarchy process to form a comprehensive loss index ε=Σ(ωi). εi).

[0095] Preferably, the model building layer further includes:

[0096] The function building unit is used to set up a loss prediction module based on an LSTM neural network and train it to obtain a loss trend function based on the historical usage data sequence X=(x1,x2,...,xt). σ(Whht-1+Wxxt+b), where ht is the hidden layer state at time t, σ is the activation function, b is the parameter coefficient, x1, x2, and xt are the 1st, 2nd, and tth sequence values, respectively, and Wh and Wx are weighting coefficients;

[0097] The weight update unit is used to introduce fuzzy logic rule base to process unstructured data. When a visible anomaly is detected in a non-consumable item type consumable item, the weight of εg is dynamically adjusted through the membership function μ(d)=1 / (1+ek(d-d0)), where d is the degree of the visible anomaly, d0 is the critical threshold of the anomaly degree, k is the gradient adjustment coefficient, and e is a constant with a value of 2.7.

[0098] The model update unit is used to perform uncertain reasoning of attribute parameters through a Bayesian network and to establish an attribute change probability matrix. The corresponding item models are updated, ensuring that the dynamic attribute parameters of the item models maintain their mapping to their actual usage state. These represent the probabilities of change under different states.

[0099] In this embodiment, the basic attribute layer defines the static inherent characteristics of the consumables, including:

[0100] Item type: Distinguish between consumable items (such as medical cotton swabs, which are consumed as soon as they are issued) and non-consumable items (such as scalpels, which are reusable).

[0101] Item identification: Consumables are uniquely marked using RFID tags, laser-etched serial numbers, etc. (e.g., surgical scalpel serial number "SK2025001").

[0102] Physical specifications: Describe morphological parameters (such as cotton swab length of 10cm, curvature of scalpel blade);

[0103] Sterilization status: Record the sterilization time and expiration date (e.g., cotton swab sterilization date 2025-07-01, expiration date 7 days);

[0104] Storage conditions: Specify the storage environment requirements (e.g., cotton swabs need to be dry and protected from light, and surgical scalpels need to be stored in a rust-proof cabinet).

[0105] State attribute layer: Depicting the dynamic changes of consumables:

[0106] Consumable items (cotton swabs): Set usage frequency-related parameter groups (e.g., dynamic consumption coefficient k). t (This reflects fluctuations in average daily consumption).

[0107] Non-destructive items (surgical scalpel): Set up a multi-dimensional loss assessment parameter group (such as geometric loss degree, performance degradation degree, quantification of wear and aging).

[0108] In this embodiment, two types of data are synchronized in real time through a bidirectional data interface:

[0109] Inventory change data: Outbound and inbound operations trigger status attribute updates (e.g., 100 cotton swabs are issued, and the frequency parameter group is calculated in real time).

[0110] Quality inspection data: By connecting to blade scanners and sterilization monitoring equipment (such as scanning a scalpel to detect blade wear of 0.3mm, and updating the geometric wear degree with multi-dimensional loss parameter groups), the basic attributes (static) and state attributes (dynamic) are continuously linked, and the model always fits the actual state of the consumables.

[0111] In this embodiment, emergency requisition is centralized. =1.5; General Departments =1.

[0112] In this embodiment, after long-term use of the scalpel, the edge wears out and the coating peels off, and ε continuously increases, and the model dynamically captures the loss process.

[0113] In this embodiment, the historical data sequence: X = (x1, x2,..., xt), records the characteristics of each use (such as the first use time of the scalpel x1 = 5 minutes, the second use time x2 = 6 minutes, etc.).

[0114] In this embodiment, represents the known state at time t, represents the predicted state at time t + 1. For example, the current state (such as the scalpel loss index 0.3, the sterilization state "effective");

[0115] is the future state (such as the loss index 0.35, the sterilization state "invalid"). For example, the sterilization validity period of the scalpel is 7 days, and on the 6th day is "sterilization effective". Through Bayesian inference, calculate the probability that on the 7th day is "sterilization invalid" (such as 0.8), and dynamically update the model.

[0116] The beneficial effects of the above technical solution are: through the closed-loop design of static-dynamic double-layer modeling → real-time data coupling → AI prediction + fuzzy correction + Bayesian inference, it empowers a more accurate and intelligent medical consumable management system.

[0117] The present invention provides an AI-based experimental consumable management system. The warning layer includes:

[0118] A loss attribute analysis unit, which is used to construct a dynamic consumption prediction model when the item attribute is a loss attribute, extract the export quantity sequence of the last M cycles through a time sliding window, calculate the average consumption rate, and quantify the consumption fluctuation through the standard deviation;

[0119] Based on the confidence interval of the normal distribution, set the basic safety stock threshold and introduce a fluctuation compensation term;

[0120] At the same time, combine the department requisition priority coefficient to obtain the dynamic safety stock threshold Sy1;

[0121] When the remaining quantity R satisfies R < 1.2Sy1, a yellow warning is triggered;

[0122] When R < Sy1, a red warning is triggered;

[0123] Achieve hierarchical warning through the warning level L1, and associate different response strategies according to the level;

[0124]

[0125] Among them, This is the floor symbol.

[0126] Preferably, the early warning layer includes:

[0127] The non-loss attribute analysis unit is used to construct a full-cycle inventory status model when the item attribute is a non-loss attribute. It defines the quantity in stock R1, the quantity requested for warehousing Sn, and the quantity in transit Zn, and calculates the effective inventory In = R1 + Sn - Zn.

[0128] An LSTM network is trained based on historical usage sequences to predict the remaining usable usage Nr, combined with a single-use loss coefficient. Given the rated total number of cycles En, calculate the available inventory Iy after loss reduction. ;

[0129] Calculate the average inbound delay based on the outbound-inbound time series D. And introduce a delay compensation factor Dynamically adjust the early warning threshold, among which, Standard delay;

[0130]

[0131]

[0132] in, This is the lower limit of the threshold. This represents the average outbound rate. This is the replenishment cycle; x represents the maximum inventory capacity; This represents the average inbound rate.

[0133] when At that time, a supply warning is triggered;

[0134] when At that time, a backlog warning was triggered.

[0135] In this embodiment, the time sliding window selects the outbound data of the most recent M cycles (e.g., 30 days) to form a sequence (e.g., the outbound volume of medical cotton swabs in the past 30 days [100, 120, 95, ..., 110]) to capture short-term consumption patterns.

[0136] Average consumption rate v: The average daily consumption within the calculation period, the formula is v=Σqᵢ / M (e.g., 3000 units out of the warehouse in 30 days, v=100 units / day), reflecting the normal consumption intensity.

[0137] Standard deviation σ: quantifies consumption fluctuations (e.g., σ=15 indicates that daily average consumption fluctuates between 85 and 115 items), revealing demand uncertainty.

[0138] In this embodiment, the basic safety stock is based on the replenishment cycle T (such as 7 days) to ensure basic supply.

[0139] The fluctuation compensation term introduces the normal distribution confidence coefficient k (1.96 for 95% confidence), and the formula is Sᵥ = k×σ×√T (for example, Sᵥ ≈ 80 pieces) to compensate for the fluctuation risk.

[0140] In this embodiment, the trigger conditions are as follows:

[0141] Yellow warning: The remaining quantity R < 1.2Sy1 (for example, R = 1000 < 1123, indicating regular replenishment);

[0142] Red warning: R < Sy1 (for example, R = 900 < 936, triggering emergency allocation).

[0143] Specifically: Extract data through sliding query in the database, calculate statistics using Pandas, mark the department priority in the system configuration interface, and trigger email / sms warnings through the workflow engine.

[0144] For the in - stock quantity R1, the applied - in - stock quantity Sn, and the in - transit quantity Zn, calculate the effective inventory In = R1 + Sn - Zn;

[0145] In this embodiment, the in - stock quantity R1: Currently available consumables (for example, 40 scalpels in stock);

[0146] The applied - in - stock quantity Sn: Consumables that have been returned and are awaiting inspection (for example, 5 pieces returned after surgery);

[0147] The in - transit quantity Zn: Consumables that have been issued but not yet returned (for example, 10 pieces used during surgery and not yet returned);

[0148] LSTM prediction : Based on the historical usage times sequence (such as [5, 6, 4,...]), train the LSTM to predict the remaining available times (for example, = 20 times) to predict future losses.

[0149] In this embodiment, the WMS system synchronizes the inventory status, the AI server deploys the LSTM model, the edge - scanner measures the loss in real - time, and the inventory cost model is associated with the warning response.

[0150] The beneficial effects of the above technical solutions are: Through the accurate modeling of the dynamic consumption - hierarchical response of consumables and the full - cycle status - loss quantification - delay compensation of non - consumables, the upgrade of medical consumable management from quantity control to value control is achieved.

[0151] The present invention provides an AI - based management method for experimental consumables, as Figure 2 shown, including:

[0152] Step 1: Construct item models for different consumable items;

[0153] Step 2: When students collect consumables, the system performs intelligent AI recognition on each consumable based on the item model.

[0154] Step 3: Record the quantity of consumables issued based on their attributes and the results of intelligent AI recognition, and calculate the remaining quantity of consumables. At the same time, calculate the quantity of consumables received based on their attributes.

[0155] Step 4: When the item attribute is a loss attribute, formulate a first early warning strategy and issue an early warning reminder based on the outbound quantity and the remaining quantity; when the item attribute is a non-loss attribute, formulate a second early warning strategy and issue an early warning reminder based on the outbound quantity, the remaining quantity, and the inbound quantity.

[0156] Preferably, step 1 includes:

[0157] Establish a two-layer item model architecture that includes basic attributes and status attributes;

[0158] By synchronizing inventory change data and quality inspection data in real time through a two-way data interface, dynamic coupling of basic attributes and status attributes is achieved.

[0159] The beneficial effects of the above technical solution are as follows: through the closed-loop process of model building → intelligent identification → quantity control → hierarchical early warning, the item model can achieve digital mapping between static features and dynamic states, accurately depict the entire life cycle of consumables, multimodal AI identification breaks through the limitations of single identifiers, quantity statistics adapt to the differences between loss and non-loss attributes, and blockchain notarization ensures data credibility; the early warning strategy embeds dynamic formulas (safety stock, balance index) and loss prediction, achieving an intelligent upgrade of medical consumable management.

[0160] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An AI-based laboratory consumables management system, characterized in that, Comprising: A model construction layer for constructing item models of different consumable items; An AI recognition layer for, when a student receives a consumable item, performing intelligent AI recognition on each received consumable item based on the item model; A statistics layer for recording the outbound quantity of consumables according to the item attributes of the consumable items and combining the intelligent AI recognition results, and for counting the remaining quantity of consumable items. At the same time, the inbound quantity of the corresponding consumable items is counted according to the item attributes; An early warning layer for, when the item attribute is a loss attribute, formulating a first early warning strategy and giving an early warning reminder according to the outbound quantity and the remaining quantity; When the item attribute is a non-loss attribute, formulating a second early warning strategy and giving an early warning reminder according to the outbound quantity, the remaining quantity, and the inbound quantity; The model construction layer includes: The function building unit is used to set up a loss prediction module based on an LSTM neural network and train it to obtain a loss trend function based on the historical usage data sequence X=(x1,x2,...,xt). σ(Whht-1+Wxxt+b), where ht is the hidden layer state at time t, σ is the activation function, b is the parameter coefficient, x1, x2, and xt are the 1st, 2nd, and tth sequence values, respectively, and Wh and Wx are weighting coefficients; A weight update unit for introducing a fuzzy logic rule base to process unstructured data. When a visible abnormality occurs in a consumable item of a non-loss item type, the weight of εg is dynamically adjusted through the membership function μ(d)=1 / (1+e-k(d-d0)), where d is the degree of visible abnormality, d0 is the critical threshold of the degree of abnormality, k is the gradient adjustment coefficient; e is a constant with a value of 2.7; εg corresponds to the geometric loss degree of the consumable item; The model update unit is used to perform uncertain reasoning of attribute parameters through a Bayesian network and to establish an attribute change probability matrix. The corresponding item models are updated, ensuring that the dynamic attribute parameters of the item models maintain their mapping to their actual usage state. These represent the probabilities of change under different states; The early warning layer includes a non-loss attribute analysis unit and a loss attribute analysis unit; The non-loss attribute analysis unit is used for, when the item attribute is a non-loss attribute, constructing a full-cycle inventory status model, defining the in-stock quantity R1, the applied inbound quantity Sn, and the in-transit quantity Zn, and calculating the effective inventory In = R1 + Sn - Zn; An LSTM network is trained based on historical usage sequences to predict the remaining usable usage Nr, combined with a single-use loss coefficient. Given the rated total number of cycles En, calculate the available inventory Iy after loss reduction. ; Calculate the average inbound delay based on the outbound-inbound time series D. And introduce a delay compensation factor Dynamically adjust the early warning threshold, among which, Standard delay; in, This is the lower limit of the threshold. This represents the average outbound rate. This is the replenishment cycle; x represents the maximum inventory capacity; This represents the average inbound rate. when At that time, a supply warning is triggered; when At that time, a backlog warning was triggered.

2. The system according to claim 1, characterized in that, The model construction layer further includes: An architecture establishment unit for establishing a two-layer item model architecture including basic attributes and status attributes; A coupling unit for synchronizing inventory change data and quality inspection data in real time through a bidirectional data interface to achieve dynamic coupling of basic attributes and status attributes.

3. The system according to claim 1, wherein The loss attribute analysis unit of the early warning layer is used for, when the item attribute is a loss attribute, constructing a dynamic consumption prediction model, extracting the outbound quantity sequence of the recent M cycles through a time sliding window, calculating the average consumption rate, and quantifying the consumption fluctuation through the standard deviation; Based on the normal distribution confidence interval, setting a basic safety inventory threshold and introducing a fluctuation compensation term; At the same time, obtaining a dynamic safety inventory threshold Sy1 in combination with the department requisition priority coefficient; When the remaining quantity R satisfies R < 1.2Sy1, triggering a yellow early warning; When R < Sy1, triggering a red early warning.

4. The system according to claim 2, characterized in that, The basic attribute layer at least includes item type, item identifier, physical specifications, sterilization status, and storage conditions. The status attribute layer sets a usage frequency association parameter group for consumable items of the loss item type and sets a multi-dimensional loss evaluation parameter group for consumable items of the non-loss item type.

5. The system according to claim 4, characterized in that, The usage frequency association parameter group includes a dynamic consumption coefficient kt based on a time sliding window; in, , As a seasonal fluctuation moderating factor; The concentration coefficient for departmental requisition; This represents the cumulative outbound volume over the past n days. This is the standard daily consumption benchmark value.

6. The system according to claim 5, characterized in that, The multidimensional loss assessment parameter set includes: the geometric loss degree εg, the heat treatment performance degradation degree εh, and the surface coating failure probability εc of the corresponding consumable item. A loss vector E=(εg,εh,εc) is constructed, and the weight ωi of each component is determined using the analytic hierarchy process (AHP) to form the comprehensive loss index ε=Σ(ωi). εi).

7. An AI-based method for managing laboratory consumables, characterized in that, Implemented through the AI-based experimental consumable management system according to claim 1, including: Step 1: Construct item models of different consumable items; Step 2: When students collect consumables, the system performs intelligent AI recognition on each consumable based on the item model. Step 3: Record the quantity of consumables issued based on their attributes and the results of intelligent AI recognition, and calculate the remaining quantity of consumables. At the same time, calculate the quantity of consumables received based on their attributes. Step 4: When the item attribute is a loss attribute, formulate a first early warning strategy and issue an early warning reminder based on the quantity issued and the remaining quantity; When the item attribute is a non-consumable attribute, a second early warning strategy is formulated and an early warning is issued based on the outbound quantity, remaining quantity, and inbound quantity.

8. The method according to claim 7, characterized in that, Step 1 includes: Establish a two-layer item model architecture that includes basic attributes and status attributes; By synchronizing inventory change data and quality inspection data in real time through a two-way data interface, dynamic coupling of basic attributes and status attributes is achieved.

Images