Inventory control method and device for medical apparatus and instruments

Abstract

The invention discloses an inventory control method and device for medical instruments, and relates to the field of medical instrument management, and the method comprises the steps: obtaining the information of each medical instrument in an inventory; constructing a knowledge graph, wherein the knowledge graph comprises a medical instrument node, a state node and a hierarchy node; a rule engine is adopted to classify each medical device based on the information of the medical device, nodes of the medical devices are connected to corresponding state nodes, the state nodes are connected to corresponding hierarchical nodes, and the hierarchical nodes comprise a freezing layer, an available layer and a scrap layer; and performing warehouse-out according to the hierarchy priority during inventory scheduling, performing sorting according to the remaining available duration in the available layer, performing sorting according to the remaining freezing duration in the freezing layer, and switching to the available layer after the freezing duration ends. According to the invention, the key factor of sterilization mode difference is systematically introduced for the first time, refined inventory control based on sterilization characteristics is realized, and the warehouse management efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of medical device management, and more specifically, to a method and apparatus for inventory control of medical devices. Background Technology

[0002] Medical device inventory management is a crucial link in medical supply chain management, and its accuracy and timeliness directly impact the availability and safety of medical supplies. Unlike ordinary warehoused products, medical devices undergo rigorous sterilization before entering the warehousing system. Different sterilization methods correspond to different chemical, physical, or radiation process characteristics, which directly determine the device's performance stability period, failure mode, and safe usage cycle. Common sterilization methods include ethylene oxide sterilization, gamma-ray or electron beam irradiation sterilization, autoclaving, and plasma sterilization. Among these, ethylene oxide-sterilized devices require a dissipation period to reduce residual gas concentration; gamma-ray-sterilized devices experience gradual material aging due to irradiation; autoclaving devices have an upper limit to the number of times they can be reused; and plasma-sterilized devices gradually degrade with changes in surface energy. The failure processes caused by these different sterilization methods are non-uniform, making traditional inventory management models based on a uniform expiration date unable to accurately reflect the true usability of the devices.

[0003] Current medical device inventory management primarily relies on static expiration dates, neglecting the impact of sterilization methods and material differences on performance degradation. This can easily lead to two problems: first, some devices still within their safe dissipation period are used prematurely, resulting in excessive residue risks; second, devices nearing their performance threshold are continued to be used because their labels still indicate an expiration date, increasing potential quality risks. Furthermore, traditional warehouse layouts generally use chronological order as the sole criterion for sorting, failing to reflect the dynamic relationship between different levels of devices in terms of outbound priority and storage location. Therefore, how to construct an intelligent inventory control mechanism that can dynamically identify device status and rationally arrange outbound sequence and storage location, considering the diversity of medical device sterilization methods and their evolutionary characteristics, has become a key issue that needs to be addressed in this field. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an inventory control method and apparatus for medical devices, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: An inventory control method for medical devices includes the following steps: Obtain information on each medical device in the inventory, including sterilization method, sterilization time, and device material information; A knowledge graph is constructed, which includes medical device nodes, status nodes, and hierarchical nodes, with attributes corresponding to the information attached to the medical device nodes. A rule engine is used to classify each medical device based on its information, thereby connecting the medical device node to its corresponding status node, and the status node to the corresponding hierarchical node, which includes a frozen layer, an available layer, and a scrapped layer. During inventory scheduling, outbound shipments are executed based on hierarchical priority, with the available layer having a higher priority than the frozen layer, and the frozen layer having a higher priority than the scrapped layer. Outbound shipments only originate from the available layer. Within the available layer, medical devices are prioritized from low to high based on their remaining available time. Within the frozen layer, medical devices are prioritized from low to high based on their remaining frozen time, and are transferred to the available layer when their remaining frozen time ends. Among these, storage cells that are closer to the exit door are assigned to higher priority levels.

[0006] For medical devices sterilized with ethylene oxide, the node connected to the evaporation state corresponds to the frozen layer, the node connected to the stable state corresponds to the usable layer, and the node connected to the expired state corresponds to the scrapped layer. For medical devices sterilized by gamma rays or electron beams, the stable state node to which it is connected corresponds to the usable layer, and the decay state node to which it is connected corresponds to the scrap layer. For medical devices sterilized by high-pressure steam, the nodes connected to the cyclic available state correspond to the available layer, and the nodes connected to the cyclic over-limit state correspond to the scrap layer. For medical devices sterilized by plasma, the surface stable state nodes to which they are connected correspond to the usable layer, and the surface degraded state nodes to which they are connected correspond to the scrapped layer.

[0007] The rule engine determines state nodes based on a neural network model, specifically including the following steps: The neural network model receives input parameters from the medical device, including sterilization method information, sterilization time, and device material information; The input parameters are standardized and then input into a neural network model. The output of the neural network model is a state node category, which includes dissipation, stability, expiration, decay, cyclic availability, cyclic overrun, surface stability, and surface degradation.

[0008] When the rule engine determines that a certain medical device node is connected to the state node corresponding to the availability layer, the rule engine further uses a neural network model to predict its remaining availability time.

[0009] The neural network model uses differentiated input features for nodes based on different sterilization methods, specifically including: For medical devices sterilized with ethylene oxide, the input parameters include sterilization completion time, material adsorption coefficient, current residual detection value, and ambient temperature and humidity. The output is the remaining usable time before the residual concentration reaches the safety lower limit. For medical devices sterilized by gamma rays or electron beams, the input parameters include sterilization time, material aging coefficient, historical irradiation dose and ambient temperature, and the output is the remaining usable time before the material performance decays to the safety threshold. For medical devices sterilized by high-pressure steam, the input parameters include the cumulative number of cycles, the heat resistance rating of the material, and the frequency of use. The output is the remaining number of usable cycles or the equivalent time before reaching the upper limit of the cycle. For medical devices sterilized by plasma, the input parameters include sterilization time, surface energy decay rate, material type and storage humidity, and the output is the remaining usable time before the surface properties degrade to a safe threshold.

[0010] When the rule engine determines that a certain medical device node is connected to the corresponding state node in the frozen layer, it further predicts its freezing duration through a neural network model. The freezing duration result is used to control the unlocking plan of medical devices in the frozen layer. When the expected freezing duration is reached, the medical device is moved from the frozen layer to the available layer to participate in the subsequent outbound sorting.

[0011] For medical devices sterilized with ethylene oxide, when predicting their freezing time, the input parameters include sterilization completion time, initial residue concentration, material adsorption coefficient, current temperature and humidity, and ventilation rate. The output is the estimated dissipation time required for the residue concentration to drop to a safe threshold.

[0012] The neural network model is trained using supervised learning with historical sample data and has the following three functional modules: Classification module: The model parameters are trained by minimizing the classification error, with the state node category as the output and labeled historical state data as the training samples. Availability prediction module: It is trained by minimizing the difference between the predicted time and the actual usage time, with the remaining available time as the output. Freeze duration prediction module: The module is trained by minimizing the error between the predicted duration and the actual unlock duration, using the actual time required to go from the frozen state to the available state as the output.

[0013] After training, the three modules of the neural network model share the input feature extraction layer and output results independently, which can be called by the rule engine in inventory classification, availability prediction and freeze unlocking plan respectively.

[0014] The present invention also includes an inventory control device for medical devices, comprising: The information acquisition module is used to acquire information about each medical device in the inventory, including sterilization method information, sterilization time, and device material information; The knowledge graph construction module is used to construct a knowledge graph, which includes medical device nodes, status nodes, and hierarchical nodes, and attaches attributes with corresponding information to the medical device nodes. The classification module is used to classify each medical device based on its information using a rule engine, connect the medical device node to the corresponding status node, and connect the status node to the corresponding hierarchical node, wherein the hierarchical node includes a frozen layer, an available layer, and a scrapped layer. The scheduling module is used to execute outbound shipments based on hierarchical priority during inventory scheduling. The priority of the available layer is higher than that of the frozen layer, the priority of the frozen layer is higher than that of the scrapped layer, and outbound shipments only come from the available layer. The sorting and conversion module is used to prioritize medical devices from low to high based on their remaining available time within the available layer, and to prioritize medical devices from low to high based on their remaining frozen time within the frozen layer. It also automatically converts medical devices to the available layer when their remaining frozen time ends. The warehouse layout module is used to determine the spatial layout of storage cells according to hierarchical priority, assigning storage cells closer to the exit door to higher priority hierarchical positions.

[0015] The advantage of this invention over existing technologies lies in its first systematic introduction of the key factor of differences in sterilization methods, enabling refined inventory control based on sterilization characteristics. By constructing a knowledge graph containing medical device nodes, status nodes, and hierarchical nodes, and combining intelligent analysis with a rule engine and neural network model, this invention can automatically determine the stable, dissipation, or degradation phase of medical devices based on the failure mechanisms of different sterilization methods such as ethylene oxide, gamma rays, high-pressure steam, and plasma, and then dynamically allocate them to the usable, frozen, or obsolete layers. The neural network used in this invention can not only predict the remaining usable and frozen time of devices, but also automatically arrange high-priority layers near the exit point in the warehouse layout, thus making the inventory structure more in line with medical safety standards. This invention effectively avoids the problems of premature use or delayed obsolescence caused by neglecting sterilization differences in traditional warehousing, realizing a leap from static recording to intelligent evolution in medical device inventory management, and significantly improving warehouse utilization efficiency and device safety. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the method of the present invention; Figure 2 This is a schematic diagram illustrating the classification of this invention; Figure 3 This is a schematic diagram of the neural network model of the present invention; Figure 4 This is a schematic diagram of the device of the present invention. Detailed Implementation

[0017] The specific embodiments of the present invention will now be described with reference to the accompanying drawings.

[0018] like Figure 1 As shown, the method of the present invention includes: Obtain information on each medical device in the inventory, including sterilization method, sterilization time, and device material information; A knowledge graph is constructed, which includes medical device nodes, status nodes, and hierarchical nodes, with attributes corresponding to the information attached to the medical device nodes. A rule engine is used to classify each medical device based on its information, thereby connecting the medical device node to its corresponding status node, and the status node to the corresponding hierarchical node, which includes a frozen layer, an available layer, and a scrapped layer. During inventory scheduling, outbound shipments are executed based on hierarchical priority, with the available layer having a higher priority than the frozen layer, and the frozen layer having a higher priority than the scrapped layer. Outbound shipments only originate from the available layer. Within the available layer, medical devices are prioritized from low to high based on their remaining available time. Within the frozen layer, medical devices are prioritized from low to high based on their remaining frozen time, and are transferred to the available layer when their remaining frozen time ends. Among these, storage cells that are closer to the exit door are assigned to higher priority levels.

[0019] In a specific embodiment, the present invention collects detailed information on each medical device in inventory through an information acquisition module. This information includes sterilization method, sterilization time, and device material information. Sterilization methods encompass various types, including ethylene oxide sterilization, gamma-ray or electron beam sterilization, autoclaving, and plasma sterilization. Each sterilization method has different impacts on the availability and storage requirements of the device, therefore requiring precise recording. Sterilization time refers to the specific point in time when the device completes sterilization, typically stored as a timestamp, accurate to the minute, to ensure the accuracy of subsequent duration calculations. Device material information includes material type, such as stainless steel, polypropylene, silicone, etc. In further embodiments, it may also include the physical and chemical properties of the material, such as adsorption coefficient, heat resistance rating, or surface energy decay rate. This information is obtained through barcode or RFID tag scanning and combined with device registration information in the database to ensure data integrity and consistency. Furthermore, in further embodiments, environmental parameters, such as the temperature and humidity of the storage area and ventilation rate, may also be recorded to support subsequent status analysis and duration prediction.

[0020] After collecting medical device information, this invention constructs a knowledge graph for organizing and managing inventory data. The knowledge graph consists of three types of nodes: medical device nodes, status nodes, and hierarchical nodes. Each medical device node corresponds to a specific medical device in the inventory. Attributes attached to these nodes include sterilization method, sterilization time, and material information. Further embodiments may include other relevant parameters such as residue concentration and cumulative cycle count. Status nodes represent the current state of the medical device. Depending on the sterilization method, status nodes are categorized as: dissipated state, stable state, expired state, decayed state, cycle-available state, cycle-exceeded state, surface-stable state, and surface-degraded state. Hierarchical nodes are divided into a frozen layer, a usable layer, and a scrapped layer, representing the inventory management hierarchy of medical devices.

[0021] In a specific embodiment, the knowledge graph is implemented using graph database technology, such as Neo4j, to store nodes and relationships. Medical device nodes are connected to their corresponding state nodes via edges, and state nodes are then connected to their corresponding hierarchical nodes via edges. This structured data organization facilitates rapid querying and updating of device status, while also supporting complex logical reasoning.

[0022] like Figure 2 As shown, in this embodiment of the invention, a rule engine is used to classify medical devices. Based on their information, medical device nodes are connected to corresponding state nodes, and further connected to corresponding hierarchical nodes. The core of the rule engine is a neural network model specifically used for determining state nodes. The neural network model receives input parameters, including sterilization method information, sterilization time, and device material information. These parameters are first standardized, for example, by converting timestamps to relative times and encoding material information into numerical vectors to ensure the consistency of the input data. The output of the neural network model is the state node category, including dissipated, stable, expired, decayed, cycle-available, cycle-overdue, surface stable, and surface degraded.

[0023] Different sterilization methods determine the lifecycle status of medical devices after they leave the factory and their tier in the inventory system. Taking ethylene oxide sterilization as an example, this method achieves sterilization by permeating materials with ethylene oxide gas, but the gas may remain on the device surface or in the micropores of the material after sterilization. To avoid toxic residues threatening safety, medical devices typically enter a residual gas dissipation phase after sterilization. Devices in this phase are not yet usable and are therefore classified as a frozen layer in the system, corresponding to the dissipation state. The rule engine predicts the time required to reach a safety threshold based on the residual concentration estimate output by the neural network model, combined with parameters such as temperature, humidity, and ventilation rate. When the predicted dissipation time ends and the detection data shows that the residual concentration is below the safety lower limit, the device is automatically adjusted to a stable state, corresponding to the usable layer, and can enter the normal outbound scheduling process. If the storage time exceeds the set expiration date, or if the detection shows that the ethylene oxide concentration has increased again, the system adjusts the device status to an expired state and automatically classifies it into the scrap layer to avoid potential risks from reuse.

[0024] In gamma-ray or electron beam sterilization scenarios, the sterilization process relies on high-energy irradiation to destroy the DNA structure of microorganisms. Therefore, after sterilization, the initial state of the medical device can be considered a stable state, corresponding to the usable layer. However, long-term storage or use can cause the molecular chain structure of the material to age and become brittle, especially under unfavorable temperature, humidity, or irradiation conditions, resulting in a slow decline in material performance. When the rule engine detects that the material aging coefficient exceeds the threshold or the predicted lifespan is insufficient based on the neural network model, the device state node will change from a stable state to a decayed state and be automatically classified into the scrap layer to ensure that only medical devices that meet performance requirements are used in the outbound process.

[0025] Medical devices sterilized by autoclaving use high-temperature, high-pressure saturated steam to kill bacteria, and are characterized by their ability to be repeatedly used. After sterilization, these devices are in a cycle-usable state, corresponding to a usable layer, indicating that their performance is still within a safe range. The system records the number of autoclaving cycles for each procedure and, combined with parameters such as the material's heat resistance rating and usage frequency, calculates the current cumulative number of cycles. When the neural network model predicts that the device is about to reach or has exceeded the material's tolerance limit, the device's state node changes from cycle-usable to cycle-over-limit, and it is automatically moved to the scrap layer. This mechanism allows the system to dynamically determine the lifespan of cyclically used devices, preventing excessive reuse from causing structural deformation or sterilization failure.

[0026] Plasma sterilization of medical devices relies on the destruction of microbial cell walls by active ions and free radicals in low-temperature plasma. This method offers good material heat sensitivity, but its sterilization effect depends on the stability of the surface active layer. After sterilization, the device initially has a stable surface, corresponding to a usable layer, indicating that its surface properties meet safety requirements. However, during prolonged storage or in high-humidity environments, surface energy may decay, and the surface active layer gradually degrades, leading to a decrease in post-sterilization protective performance. The system uses a neural network model to analyze characteristics such as storage humidity, material type, and surface energy decay rate. When the model output indicates that performance has degraded below a safe threshold, the state node automatically adjusts to a surface degradation state, corresponding to a discarded layer, thereby achieving dynamic tracking and automated management of surface performance changes.

[0027] The neural network model employs an architecture combining convolutional neural networks and fully connected layers. The input layer receives standardized parameters, the convolutional layers extract the latent relationships between parameters, and the fully connected layers output the state category. The model is trained through supervised learning using historical sample data containing labeled state categories. The training objective is to minimize the classification error, employing the cross-entropy loss function and the Adam optimizer. The training data is derived from historical inventory records, covering instruments with various sterilization methods and material types to ensure the model's generalization ability.

[0028] During inventory scheduling, the system executes outbound operations based on hierarchical priority. Available inventory has a higher priority than frozen inventory, which in turn has a higher priority than obsolete inventory, and outbound operations only begin from the available inventory. This design ensures that equipment in a safe and usable state is used first, avoiding waste or the use of unsafe equipment.

[0029] In a further embodiment, within the available layer, the system prioritizes medical devices from low to high based on their remaining available time; that is, the shorter the remaining available time, the higher the priority. The remaining available time is predicted by a neural network model, the specific method of which will be described later. During sorting, devices with shorter remaining available time are released first to reduce inventory backlog and expiration risks. Within the frozen layer, the system sorts devices from low to high based on their remaining frozen time, which is also predicted by a neural network model. When the frozen time expires, the system automatically transfers the devices from the frozen layer to the available layer for subsequent release sorting.

[0030] For medical devices connected to the availability layer state node, the rule engine of this invention further employs a neural network model to predict their remaining availability time. In an improved embodiment, the prediction model uses differentiated input features based on different sterilization methods to ensure prediction accuracy.

[0031] For medical devices sterilized with ethylene oxide, the input parameters include sterilization completion time, material adsorption coefficient, current residue detection value, and ambient temperature and humidity. The output is the remaining usable time before the residue concentration reaches the safety lower limit. The material adsorption coefficient is experimentally determined, reflecting the adsorption capacity of the medical device's materials for ethylene oxide. The residue detection value is acquired in real time by a chemical sensor, while the ambient temperature and humidity are provided by sensors in the storage area. The model analyzes these parameters and combines them with historical data to predict the time required for residues to dissipate to the safety threshold.

[0032] For medical devices sterilized with gamma rays or electron beams, input parameters include sterilization time, material aging factor, historical irradiation dose, and ambient temperature. The material aging factor reflects the rate of performance degradation of the material under irradiation, and the historical irradiation dose records the total radiation received by the device. The model outputs the remaining usable time before the material performance degrades to a safe threshold.

[0033] For medical devices sterilized by autoclaving, input parameters include cumulative cycle count, material heat resistance rating, and usage frequency. The cumulative cycle count records the number of sterilization cycles the device has undergone, and the material heat resistance rating reflects the material's ability to withstand high temperatures. The model outputs the remaining usable cycles or equivalent time before reaching the cycle limit.

[0034] For medical devices sterilized by plasma, input parameters include sterilization time, surface energy decay rate, material type, and storage humidity. The surface energy decay rate is determined experimentally and reflects changes in surface properties. The model outputs the remaining usable time before surface properties degrade to a safe threshold.

[0035] For medical devices connected to the frozen layer state node, the rule engine predicts their freezing duration, primarily targeting ethylene oxide sterilized devices. Input parameters include sterilization completion time, initial residue concentration, material adsorption coefficient, current temperature and humidity, and ventilation rate. The initial residue concentration is obtained through post-sterilization detection, and the ventilation rate is provided by the storage area's ventilation system. The model outputs the estimated dissipation time required for the residue concentration to drop to a safe threshold. This duration is used to develop an unlocking plan; when the freezing time is reached, the device automatically moves from the frozen layer to the available layer and participates in the outbound sequencing.

[0036] like Figure 3As shown, the neural network model architecture of this invention includes a shared input feature extraction layer and three independent functional modules: a classification module, an availability prediction module, and a freeze duration prediction module. The input feature extraction layer employs a multilayer perceptron structure to extract deep features from the input parameters. The classification module outputs the state node category, the availability prediction module outputs the remaining available time, and the freeze duration prediction module outputs the freeze duration. The three modules share the feature extraction layer to improve computational efficiency, but each has an independent output layer to adapt to different tasks. The model is trained using historical sample data; the classification module uses labeled state data, the availability prediction module uses actual usage time data, and the freeze duration prediction module uses actual unlocking time data. The training objective is to minimize the prediction error, employing cross-entropy loss (classification) and mean squared error loss (duration prediction), respectively.

[0037] To improve outbound efficiency, this invention optimizes the spatial layout of storage cells based on hierarchical priority. Storage cells near the exit are assigned to higher-priority levels, i.e., usable layers take precedence over frozen layers, and frozen layers take precedence over scrapped layers. In practice, the warehouse is divided into multiple areas: areas near the exit are allocated to usable layer equipment, middle areas to frozen layer equipment, and areas far from the exit to scrapped layer equipment. Each storage cell is equipped with an RFID reader to track the location and status of the equipment in real time. The system dynamically adjusts cell allocation through a warehouse layout module. When equipment is transferred from a frozen layer to a usable layer, it is prioritized for movement to cells near the exit, ensuring efficient outbound processing.

[0038] like Figure 4 As shown, this invention is implemented through an inventory control device, comprising the following modules: an information acquisition module, responsible for collecting medical device information; a knowledge graph construction module, responsible for constructing and managing the knowledge graph; a classification module, using a neural network model to achieve state classification; a scheduling module, executing outbound operations based on hierarchical priority; a sorting and transformation module, responsible for internal hierarchical sorting and state transformation; and a warehouse layout module, optimizing the layout of storage cells. These modules are integrated through a software system, running on a server or embedded device, and combined with sensor and RFID technology to achieve real-time monitoring and scheduling.

[0039] Through the above-described embodiments, the present invention can efficiently manage medical device inventory, ensure the safety and availability of devices, optimize outbound efficiency, and meet the actual needs of medical scenarios.

[0040] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A medical instrument-oriented inventory control method characterized by, The method comprises the following steps: acquiring information of each medical instrument in the inventory, including sterilization method information, sterilization time and instrument material information; constructing a knowledge graph, the knowledge graph comprising medical instrument nodes, state nodes and hierarchical nodes, the medical instrument nodes being attached with attributes of corresponding information thereof; classifying each medical instrument based on its information by using a rule engine, so as to connect the medical instrument nodes to corresponding state nodes, and connect the state nodes to corresponding hierarchical nodes, wherein the hierarchical nodes comprise a frozen layer, a usable layer and a scrap layer; when scheduling the inventory, performing the outbound operation according to the hierarchical priority, wherein the priority of the usable layer is higher than that of the frozen layer, and the priority of the frozen layer is higher than that of the scrap layer, and the outbound operation is only from the usable layer; the priority of the usable layer is sorted from low to high according to the remaining usable time length of the medical instrument; the priority of the frozen layer is sorted from low to high according to the remaining frozen time length of the medical instrument, and the medical instrument is converted to the usable layer when the remaining frozen time length ends; wherein the storage grid closer to the exit door position is divided into the hierarchical position with higher priority.

2. The medical instrument-oriented inventory control method according to claim 1, characterized by, wherein: for the medical instrument sterilized by ethylene oxide, the state node connected thereto corresponds to the frozen layer, the state node connected thereto corresponds to the usable layer, and the state node connected thereto corresponds to the scrap layer; for the medical instrument sterilized by γ-ray or electron beam, the state node connected thereto corresponds to the usable layer, and the state node connected thereto corresponds to the scrap layer; for the medical instrument sterilized by high-pressure steam, the state node connected thereto corresponds to the usable layer, and the state node connected thereto corresponds to the scrap layer; for the medical instrument sterilized by plasma, the state node connected thereto corresponds to the usable layer, and the state node connected thereto corresponds to the scrap layer.

3. The medical instrument-oriented inventory control method according to claim 2, wherein The rule engine determines the state node based on a neural network model, and the determination comprises the following steps: the neural network model receives input parameters of the medical instrument, the input parameters comprising sterilization method information, sterilization time and instrument material information; after standardizing the input parameters, the neural network model inputs the standardized input parameters, and the output of the neural network model is a state node category, the state node category comprising dissipation, stability, expiration, attenuation, cycle availability, cycle overrun, surface stability and surface degradation.

4. The medical instrument-oriented inventory control method according to claim 3, wherein When the rule engine determines that a certain medical instrument node is connected to the state node corresponding to the usable layer, the rule engine further predicts the remaining usable time length of the medical instrument node by using the neural network model.

5. The medical instrument-oriented inventory control method according to claim 4, wherein The neural network model adopts different input features according to different sterilization methods, and the input features comprise: for the medical instrument sterilized by ethylene oxide, the input parameters comprise sterilization completion time, material adsorption coefficient, current residual detection value and environmental temperature and humidity, and the output is the remaining usable time length before the residual concentration reaches the lower limit of safety; for the medical instrument sterilized by γ-ray or electron beam, the input parameters comprise sterilization time, material aging coefficient, historical irradiation dose and environmental temperature, and the output is the remaining usable time length before the material performance attenuation reaches the safety threshold; For high-pressure steam sterilized medical devices, input parameters include cumulative cycle number, material heat resistance level, and usage frequency, and output is the remaining available number or equivalent time before reaching the cycle upper limit. For plasma sterilized medical devices, input parameters include sterilization time, surface energy decay rate, material type, and storage humidity, and output is the remaining available time before the surface performance degrades to the safety threshold.

6. The medical instrument-oriented inventory control method according to claim 4, wherein When the rule engine determines that a certain medical device node is connected to the state node corresponding to the frozen layer, it further predicts the freezing duration of the medical device node through the neural network model. The freezing duration result is used to control the unlocking plan of the medical device in the frozen layer. When the predicted freezing duration is reached, the medical device is moved from the frozen layer to the available layer to participate in subsequent warehouse sorting.

7. The medical instrument-oriented inventory control method according to claim 6, wherein For ethylene oxide sterilized medical devices, when predicting the freezing duration, input parameters include sterilization completion time, initial concentration of residual, material adsorption coefficient, current temperature and humidity, and ventilation rate. The output is the predicted dissipation duration required for the residual concentration to drop to the safety threshold.

8. The medical instrument-oriented stock control method according to any one of claims 3 to 7, characterized by, The neural network model is trained through historical sample data through supervised learning, and simultaneously has the following three functional modules: Classification module: taking the state node category as the output, using the labeled historical state data as the training sample, and training the model parameters by minimizing the classification error; Available period prediction module: taking the remaining available time as the output, and training by minimizing the difference between the predicted time and the actual usage time; Freezing duration prediction module: taking the actual time required from the frozen state to the available state as the output, and training by minimizing the error between the predicted duration and the actual unlocking duration.

9. The medical device-oriented inventory control method according to claim 8, wherein, After training, the three modules of the neural network model share the input feature extraction layer and independently output results, which are called by the rule engine in inventory classification, available period prediction, and frozen unlocking plan, respectively.

10. An inventory control device for medical instruments, characterized by, It includes: An information acquisition module for acquiring information of each medical device in the inventory, including sterilization method information, sterilization time, and device material information; A knowledge graph construction module for constructing a knowledge graph, the knowledge graph including medical device nodes, state nodes, and hierarchical nodes, and attaching attributes of corresponding information to the medical device nodes; A classification module for classifying each medical device based on its information using a rule engine, connecting the medical device node to the corresponding state node, and connecting the state node to the corresponding hierarchical node, wherein the hierarchical node includes a frozen layer, an available layer, and a scrap layer; A scheduling module for performing warehouse-out based on hierarchical priority, wherein the available layer has higher priority than the frozen layer, the frozen layer has higher priority than the scrap layer, and the warehouse-out only comes from the available layer; An ordering and conversion module for prioritizing medical devices in the available layer from low to high based on their remaining available time, prioritizing medical devices in the frozen layer from low to high based on their remaining freezing time, and automatically converting the medical devices to the available layer when the remaining freezing time ends; A warehouse layout module for determining the spatial layout of storage cells based on hierarchical priority, and dividing storage cells closer to the exit door to higher priority hierarchical positions.

Images