Intelligent laboratory full-process management and control method based on multi-modal perception and strict control logic

Abstract

The invention relates to an intelligent laboratory full-process management and control method based on multi-modal perception and strict control logic, and the method comprises the steps: constructing a unified business logic middle platform and a multi-modal perception and strict control logic closed loop; deep fusion and data linkage of instrument reservation, reagent full-life-cycle management, room intelligent management and control, environment and behavior real-time monitoring and abnormal causal reasoning are realized, and the defects of system splitting, information island and passive management in the prior art are effectively eliminated; the system can actively identify intention deviation and causal anomaly in the whole experiment process, and triggers hierarchical intervention measures, thereby effectively improving the resource collaborative scheduling capability, the real-time performance of safety access control and the overall operation and maintenance efficiency of a laboratory, remarkably reducing the management risk and manual intervention cost, and improving the working efficiency of the laboratory. And the requirements of modern laboratories on intelligent and closed-loop safety management and control are met.

Description

Technical Field

[0001] This invention belongs to the field of smart laboratory management technology, and in particular relates to a smart laboratory full-process management and control method based on multimodal perception and strict control logic. Background Technology

[0002] With the continuous expansion of scientific research scale in higher education and the deepening of experimental teaching reform, the management needs of laboratories, as core locations for scientific research and teaching, have been rapidly evolving from traditional paper records and manual inspections towards digitalization and intelligence. In recent years, the advancement of smart campus construction has prompted some laboratories to introduce information management systems, such as LIMS (Laboratory Information Management System), access control and attendance systems, and environmental monitoring platforms. These systems have, to a certain extent, realized functions such as instrument reservation, data recording, and basic safety monitoring.

[0003] However, existing laboratory management technologies still have significant shortcomings: First, functional modules are highly fragmented, with each subsystem (such as instrument reservation, reagent control, room scheduling, and environmental monitoring) often deployed independently, resulting in severe data silos and a lack of a unified business logic platform and real-time communication mechanism, leading to frequent resource scheduling conflicts and delayed information updates. Second, management methods are mainly based on passive recording, lacking the ability to deeply integrate and intelligently analyze multi-source heterogeneous data, and are unable to proactively warn and intervene when high-risk reagent operations, abnormal personnel behavior, or environmental risks occur. Third, the system has poor scalability and compatibility, with inconsistent hardware interfaces and protocol standards, making it difficult to connect devices from multiple vendors, and resulting in high costs for later function upgrades and cross-system integration, making it difficult to adapt to the complexity of multi-dimensional cross-management of laboratory personnel, finances, materials, and affairs.

[0004] These shortcomings collectively lead to low overall laboratory operation and maintenance efficiency and difficulty in real-time closed-loop control of safety hazards, making it impossible to meet the requirements of modern laboratories for high security, efficient collaboration, and low-risk operation and maintenance. Summary of the Invention

[0005] The purpose of this invention is to provide a smart laboratory full-process management and control method based on multimodal perception and strict control logic, so as to solve the shortcomings of existing laboratory management technologies, such as the independence of functional subsystems, the inability to share data in real time, the reliance on manual intervention in management methods, resulting in frequent resource scheduling conflicts, lagging security access control, and low overall operation and maintenance efficiency.

[0006] To achieve one of the aforementioned objectives, one embodiment of the present invention provides a method for full-process management and control of a smart laboratory based on multimodal perception and strict control logic, the method comprising:

[0007] In response to a user's request to reserve laboratory resources, an experimental intent summary token, which includes the expected scope and constraints of the experiment, is generated after the request is approved and sent to the edge computing node and the corresponding control terminal.

[0008] In response to a user approaching the target resource, perform multimodal authentication and location presence verification. Only when all verification results pass, update the resource state machine and unlock the resource.

[0009] During the experiment, a causal relationship model was constructed and dynamically updated in real time based on the collected multi-source heterogeneous sensing data. The experiment intent summary token was used to verify the intent similarity and causal consistency of the current operation state.

[0010] When a deviation from intent or causal anomaly is detected, tiered proactive intervention is triggered based on the severity of the anomaly until the experiment ends and data asset compliance archiving and full-process log generation are completed.

[0011] As a further improvement to one embodiment of the present invention, the method further includes, in that the generation of the experimental intent summary token, which includes the expected experimental range and constraint rules, comprises,

[0012] Extract the experimental objective description, reagent requisition list, and corresponding rules from the preset standard experimental template library from the reservation application;

[0013] The extracted content is encapsulated into an experimental intent summary token, which includes the range of permitted devices, reagent combination constraints, environmental safety constraints, and expected operation sequence characteristics.

[0014] The experimental intent summary token is sent to the edge computing node and the corresponding instrument or reagent cabinet control terminal as a benchmark for subsequent real-time verification.

[0015] As a further improvement to one embodiment of the present invention, the method further includes, in the step of performing multimodal authentication and location presence verification, updating the resource state machine and unlocking the resource only when all verification results are successful, the method includes,

[0016] When a user is detected approaching a target resource, biometric recognition is performed and infrared and RGB images are acquired simultaneously. Texture depth and reflectivity are calculated using the fused image data to complete liveness detection.

[0017] After the liveness detection is passed, dynamic interactive commands are randomly generated and issued, and the degree of matching between the user's facial key point displacement and the command is collected and verified.

[0018] For high-risk reagent cabinets, a dual-person, dual-lock logic is further implemented. Within a preset time window, biometric information of two authorized personnel is collected, causing the state machine to enter a dual-person presence state.

[0019] The real-time location coordinates of the smart badge are compared with the preset electronic fence area. Only when the coordinates are within the valid operating area will the verification pass signal be transmitted to the resource state machine to perform the unlocking operation.

[0020] As a further improvement to one embodiment of the present invention, the method further includes: constructing and dynamically updating a causal relationship model in real time based on the collected multi-source heterogeneous sensing data, and verifying the intent similarity and causal consistency of the current operation state by combining the experimental intent summary token.

[0021] Environmental sensor data is collected via the MQTT protocol, video data is collected via the video stream analysis gateway, and wearable device coordinate data is collected via the positioning gateway. Timestamp alignment and spatiotemporal fusion are performed on all data streams to generate a real-time digital twin snapshot.

[0022] The video stream is analyzed frame by frame using an object detection model to identify the categories of personnel and protective equipment, and the spatial relationship between each detection box is calculated to determine the compliance of the wearing. The compliance determination result is used as the input for updating the causal graph nodes.

[0023] On the edge computing node, a small causal graph is constructed and dynamically updated in real time based on the digital twin snapshot. The causal graph includes operation event nodes, environmental variable nodes, reagent status nodes and instrument reading nodes, and the graph structure is dynamically updated according to the preset causal dependency relationship.

[0024] The similarity between the current operation sequence and the experimental intent summary token is calculated, as well as the degree of deviation between the actual readings and the expected causal relationships in the causal graph. Using historical intervention event data, a lightweight online learning model is used to adaptively optimize the verification threshold and the causal graph structure weights.

[0025] As a further improvement to one embodiment of the present invention, the method further includes, when an intention deviation or causal anomaly is detected, triggering graded proactive intervention based on the severity of the anomaly, including,

[0026] When the intent similarity is lower than the expected threshold or the causal deviation is within a slight range, voice warning signals and light guidance signals are generated first, and multi-source feedback data such as user response time, operation correction trajectory and intent recovery status are collected.

[0027] The feedback data is statistically analyzed based on a preset sliding time window to calculate the intention correction rate and abnormal recurrence frequency, and to generate a fitness index that characterizes the effectiveness and interference level of the current warning strategy. The subsequent verification threshold and intervention priority are dynamically adjusted according to the fitness index. When the correction rate is higher than the benchmark and the recurrence frequency is lower than the preset value, the threshold is relaxed to reduce false alarms.

[0028] When the anomaly persists or the causal deviation worsens, based on the updated cause-effect graph and the adjusted intervention strategy, a power supply suspension command for the equipment or a reagent cabinet lock command is generated, and the intervention event is recorded in the process log.

[0029] When the cause-effect graph shows a high-risk cause-effect chain, a reversible forced rollback instruction is generated, including resetting the valve state, restoring the initial flow setting, or disconnecting the power supply, and the rollback execution result is sent back to the resource state machine to update the resource state.

[0030] As a further improvement to one embodiment of the present invention, the method further includes, for the reservation application containing the reagent usage list, performing list constraint verification and causal risk deduction at the time of reagent requisition, specifically including,

[0031] The actual reagent information used is read by RFID and compared with the pre-approved requisition list for consistency of the product categories.

[0032] Based on the comparison results, verify whether the temperature, humidity, concentration and other conditions collected by the current environmental sensors in the cabinet meet the storage requirements of the corresponding reagents in the list;

[0033] If environmental conditions meet the requirements, further testing will be conducted to determine if any mutually exclusive chemicals on the list were removed at the same time period, and a conflict signal will be generated.

[0034] Using the conflict signals, construct or update a small causal graph related to reagents, deduce potential risk paths, and immediately generate and execute intervention commands to prohibit further outbound shipments, lock cabinets, or trigger audible and visual alarms based on the risk level of the deduced paths.

[0035] The results of the intervention are recorded in the overall process log to maintain consistency of data and status throughout the process.

[0036] As a further improvement to one embodiment of the present invention, the method further includes, in the step of completing the compliant archiving of data assets and generating a full-process log, the following steps are taken:

[0037] When a user uploads experimental data, the validity of the data access token and the login status bound to this reservation are verified in real time.

[0038] Once the verification is successful, the experimental data will be written to the storage space strongly bound to the reservation record, and the write event will be recorded in the process log.

[0039] After the experiment, the resource state machine is automatically restored to the idle state, all tokens and locks generated during the reservation process are released, and a complete process log is generated, including multimodal perception event records, causal anomaly detection records, intervention operation records, and rollback operation records, to support subsequent auditing and traceability.

[0040] To achieve one of the above-mentioned objectives, an embodiment of the present invention also provides a smart laboratory full-process management method based on multimodal perception and strict control logic. The system includes an intent summary generation module, a multimodal access verification module, a causal real-time monitoring module, and a hierarchical intervention execution module.

[0041] The intent summary generation module is used to respond to a user's reservation request for laboratory resources. After approval, it generates an experimental intent summary token that includes the expected scope of the experiment and the constraint rules, and sends it to the edge computing node and the corresponding control terminal.

[0042] The multimodal access verification module is used to perform multimodal identity verification and location presence verification in response to a user approaching the target resource. Only when all verification results are passed, the resource state machine is updated and the resource is unlocked.

[0043] The real-time causal monitoring module is used to construct and dynamically update the causal relationship model in real time based on the collected multi-source heterogeneous sensing data during the experiment, and to verify the intent similarity and causal consistency of the current operation status in combination with the experiment intent summary token.

[0044] The tiered intervention execution module is used to trigger tiered proactive intervention based on the severity of the abnormality when a deviation from intent or causal anomaly is detected, until the experiment ends and data asset compliance archiving and full-process log generation are completed.

[0045] To achieve one of the above-mentioned objectives, an embodiment of the present invention also provides an electronic device, including a memory and a processor, characterized in that the memory stores a computer program that can run on the processor, and when the program is executed on the processor, it implements the steps in the smart laboratory full-process management method based on multimodal perception and strict control logic as described above.

[0046] To achieve one of the above-mentioned objectives, an embodiment of the present invention also provides a storage medium storing a computer program, characterized in that, when the computer program is executed by a processor, it implements the steps in the smart laboratory full-process management method based on multimodal perception and strict control logic as described above.

[0047] Compared with existing technologies, this invention provides a smart laboratory end-to-end management method based on multimodal perception and strict control logic. By constructing a unified business logic platform and a closed loop of multimodal perception and strict control logic, it achieves deep integration and data linkage of instrument reservation, reagent lifecycle management, intelligent room control, real-time environmental and behavioral monitoring, and abnormal causal reasoning. This effectively eliminates the defects of system fragmentation, information silos, and passive management in existing technologies. The system can proactively identify intention deviations and causal anomalies throughout the entire experimental process and trigger graded intervention measures, effectively improving resource collaborative scheduling capabilities, real-time security access control, and overall laboratory operation and maintenance efficiency. It significantly reduces management risks and manual intervention costs, meeting the needs of modern laboratories for intelligent and closed-loop safety management. Attached Figure Description

[0048] Figure 1 This is the overall flowchart of the smart laboratory full-process management method based on multimodal perception and strict control logic described in this invention.

[0049] Figure 2 This is a schematic diagram of the architecture of the intelligent laboratory full-process management and control system based on multimodal perception and strict control logic described in this invention. Detailed Implementation

[0050] The present invention will now be described in detail with reference to the specific embodiments shown in the accompanying drawings. However, these embodiments do not limit the present invention, and any structural, methodological, or functional modifications made by those skilled in the art based on these embodiments are included within the scope of protection of the present invention.

[0051] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0052] In Embodiment 1 of the present invention, the present invention provides a smart laboratory full-process management and control method based on multimodal perception and strict control logic, such as... Figure 1 As shown, the method includes,

[0053] S1: In response to a user's request to reserve laboratory resources, after approval, an experimental intent summary token including the expected scope and constraint rules is generated and sent to the edge computing node and the corresponding control terminal.

[0054] S2: In response to a user approaching the target resource, perform multimodal authentication and location presence verification. Only when all verification results are successful, update the resource state machine and unlock the resource.

[0055] S3: During the experiment, a causal relationship model is built and dynamically updated in real time based on the collected multi-source heterogeneous sensing data. The experiment intent summary token is used to verify the intent similarity and causal consistency of the current operation state.

[0056] S4: When intention deviation or causal anomaly is detected, trigger tiered proactive intervention based on the severity of the anomaly until the experiment ends and data asset compliance archiving and full-process log generation are completed.

[0057] In one specific embodiment of the present invention, an experimental intent summary token including the expected experimental range and constraint rules is generated, specifically as follows:

[0058] Extract the experimental objective description, reagent requisition list, and corresponding rules from the preset standard experimental template library from the reservation application;

[0059] The extracted content is encapsulated into an experimental intent summary token, which includes the range of permitted devices, reagent combination constraints, environmental safety constraints, and expected operation sequence characteristics.

[0060] The experimental intent summary token is sent to the edge computing node and the corresponding instrument or reagent cabinet control terminal as a benchmark for subsequent real-time verification.

[0061] It should be noted that the purpose of generating an experiment intent summary token is to transform the approved reservation application from a static "approval result" into a "dynamic experiment intent benchmark" that can be understood and executed in real time by edge computing nodes. This provides a unified and quantifiable reference standard for subsequent multimodal perception, causal consistency verification, and proactive intervention. Essentially, this token is a structured, lightweight digital carrier that explicitly and standardizes the business intent and security constraints in the reservation application. It effectively compensates for the shortcomings of traditional laboratory management systems where approved intent information exists only in a human-readable form and is difficult for the system to perceive and verify in real time. This enables the system to shift from passive response to proactive, strict control based on intent.

[0062] Furthermore, extracting the experimental objective description, reagent requisition list, and corresponding rules from the pre-set standard experimental template library from the reservation application is crucial for comprehensively collecting and structuring the business intent and compliance constraints of this experiment. The experimental objective description (e.g., "catalyst performance testing") provides the semantic context of the experiment; the reagent requisition list clarifies the permitted types, quantities, and expected combinations of chemicals; and the standard experimental template library supplements the general safety specifications for this type of experiment (e.g., upper temperature limits, prohibited mixtures, protective equipment requirements, etc.). This information is then extracted and standardized using natural language processing or rule parsing engines to form a set of computable intent features.

[0063] Furthermore, encapsulating the extracted content into an experimental intent summary token transforms scattered and heterogeneous intent information into a unified, transmissible, and verifiable digital structured token. This token uses JSON or a custom binary format and contains the following key fields: Permitted equipment range: listing approved instruments and their usage permissions (e.g., "X-ray diffractometer only, high-temperature furnace not allowed"); Reagent combination constraint rules: defining allowed / prohibited reagent combinations (e.g., "Oxidizing and reducing agents cannot be used simultaneously"); Environmental safety constraints: setting threshold ranges for temperature, humidity, VOC concentration, etc.; Expected operation sequence features: extracting key feature vectors of typical operation paths (e.g., the sequence pattern of heating → stirring → settling) for subsequent intent similarity calculations. Through encapsulation, this originally scattered information is integrated into an "intent fingerprint" that can be directly loaded by edge-side rule engines or lightweight models, achieving a transformation from "human-readable" to "machine-executable," providing an accurate and comparable benchmark for real-time verification.

[0064] Furthermore, distributing the experiment intent summary token to edge computing nodes and control terminals ensures that the intent benchmark is deployed in real-time and reliably to the forefront of experiment execution (the edge side). This guarantees that verification occurs near the data generation location, achieving low latency and high reliability. Figure 1 Consistency assessment. The distribution process typically employs a secure channel (such as MQTT over TLS or a dedicated API) and incorporates a digital signature mechanism to prevent tampering. Upon reaching the edge node, the token is loaded into the causal real-time monitoring module, serving as a core reference benchmark for subsequent causal graph construction, operation sequence comparison, and anomaly detection; simultaneously, it is distributed to the instrument / reagent cabinet control terminal for hardware-level permission filtering (e.g., operation commands exceeding the device's scope are directly rejected).

[0065] In one specific embodiment of the present invention, multimodal authentication and location presence verification are performed. Only when all verification results are successful is the resource state machine updated and the resource unlocked. Specifically,

[0066] When a user is detected approaching a target resource, biometric recognition is performed and infrared and RGB images are acquired simultaneously. Texture depth and reflectivity are calculated using the fused image data to complete liveness detection.

[0067] After the liveness detection is passed, dynamic interactive commands are randomly generated and issued, and the degree of matching between the user's facial key point displacement and the command is collected and verified.

[0068] For high-risk reagent cabinets, a dual-person, dual-lock logic is further implemented. Within a preset time window, biometric information of two authorized personnel is collected, causing the state machine to enter a dual-person presence state.

[0069] The real-time location coordinates of the smart badge are compared with the preset electronic fence area. Only when the coordinates are within the valid operating area will the verification pass signal be transmitted to the resource state machine to perform the unlocking operation.

[0070] It should be noted that the purpose of performing multimodal identity verification and location presence verification is to achieve multi-dimensional and strongly correlated verification of the operator's true identity, presence status and operation legality through multi-source and heterogeneous biological and physical sensing methods, thereby effectively preventing safety risks such as identity theft, remote operation, and single-person unauthorized operation of high-risk reagents that exist in traditional laboratory management.

[0071] Furthermore, by calculating texture depth and reflectivity using the fused image data to complete liveness detection, the system reliably verifies the operator's biometric identity and immediately excludes common non-liveness attack methods such as photos, video playback, and masks. Through the simultaneous acquisition and fusion of infrared and RGB images, the system calculates the texture depth (based on thermal imaging differences) and reflectivity (based on spectral characteristics) of the facial region, thereby achieving passive, non-contact detection of liveness features.

[0072] Furthermore, collecting and verifying the matching degree between the user's facial key point displacement and the commands is aimed at further enhancing anti-spoofing capabilities and countering advanced forgery attacks (such as deepfake videos or 3D masks). The system randomly generates dynamic commands such as "blink twice" or "turn your head 30 degrees to the left," and captures the displacement trajectory of facial key points (eyes, nose tip, corners of mouth, etc.) in real time through video streams, calculating the matching degree with the commands (usually using Euclidean distance or vector similarity of key points). Only when the matching degree reaches a preset threshold is it determined to be a genuine person actively cooperating.

[0073] Furthermore, for high-risk reagent cabinets, a dual-person, dual-lock logic is implemented. This is to enforce the "dual-person supervision" principle in operation scenarios involving highly dangerous reagents such as those for making toxic or explosive substances, preventing safety accidents caused by single-person violations. The system continuously collects and compares the biometric features (face + liveness detection results) of two different individuals within a preset time window (e.g., 60 seconds), while simultaneously verifying whether both individuals possess the appropriate permissions. Only when the state machine successfully transitions to the "dual-person presence" state is subsequent location verification allowed.

[0074] Furthermore, the verification pass signal is only transmitted to the resource state machine to execute the unlocking operation when the coordinates are within the valid operating area. This is achieved through UWB or Bluetooth positioning technology to accurately verify the physical location of personnel, effectively solving the risk of remote fraud or proxy operation when "biometric features pass but the person is not present". The system presets an electronic fence area (usually with a radius of 1-2 meters) in front of the target resource (such as a reagent cabinet) and compares the coordinates of the smart badge in real time. If the coordinates are within the valid area and all the aforementioned verifications pass, a final "verification pass signal" is generated, triggering the resource state machine to transition from "reserved / pre-locked" to "in use" and issuing an unlocking command.

[0075] In one specific embodiment of the present invention, a causal relationship model is constructed and dynamically updated in real time based on collected multi-source heterogeneous sensing data. The current operation state is then verified for intent similarity and causal consistency using an experimental intent summary token. Specifically,

[0076] Environmental sensor data is collected via the MQTT protocol, video data is collected via the video stream analysis gateway, and wearable device coordinate data is collected via the positioning gateway. Timestamp alignment and spatiotemporal fusion are performed on all data streams to generate a real-time digital twin snapshot.

[0077] The video stream is analyzed frame by frame using an object detection model to identify the categories of personnel and protective equipment, and the spatial relationship between each detection box is calculated to determine the compliance of the wearing. The compliance determination result is used as the input for updating the causal graph nodes.

[0078] On the edge computing node, a small causal graph is constructed and dynamically updated in real time based on the digital twin snapshot. The causal graph includes operation event nodes, environmental variable nodes, reagent status nodes and instrument reading nodes, and the graph structure is dynamically updated according to the preset causal dependency relationship.

[0079] The similarity between the current operation sequence and the experimental intent summary token is calculated, as well as the degree of deviation between the actual readings and the expected causal relationships in the causal graph. Using historical intervention event data, a lightweight online learning model is used to adaptively optimize the verification threshold and the causal graph structure weights.

[0080] It should be noted that the purpose of constructing a causal relationship model in real time based on multi-source heterogeneous sensing data and performing intent-causal consistency verification is to transform multi-source heterogeneous real-time sensing data into a computable causal knowledge representation. Based on this, the current operational behavior and the approved experimental intent are subjected to dual quantitative verification of intent similarity and causal consistency, thereby realizing the transformation from passive threshold alarm to proactive causal reasoning and intent-driven intervention.

[0081] Furthermore, performing timestamp alignment and spatiotemporal fusion on all data streams to generate a real-time digital twin snapshot aims to construct a unified and comparable real-time scene representation, serving as the common data foundation for all subsequent inference. Environmental sensor data (temperature, humidity, VOC concentration, toxic gases, etc.) provides physical state information; video data provides behavioral and visual information; and location data provides spatial location information. These three types of data streams are independently collected via the MQTT protocol, video stream analysis gateway, and location gateway. Precise timestamp alignment (typically using NTP synchronization or software timestamp interpolation) and spatiotemporal fusion (coordinate system unification, spatial coordinate mapping to the same reference system) are then performed at the edge computing node, ultimately generating a real-time digital twin snapshot containing multimodal information.

[0082] Furthermore, object detection models are used to analyze the video stream frame by frame to identify personnel and protective equipment categories, and the spatial relationships between detection boxes are calculated to determine compliance with protective equipment requirements. This aims to extract highly semantic behavioral and compliance features from the video stream, providing key inputs for updating "operational events" and "personnel status" nodes in the causal graph. Using YOLOv8 or other object detection models with transfer learning, multi-object detection of personnel and protective equipment (coats, gloves, masks, goggles, etc.) is performed on each frame of the video. Spatial constraints (e.g., "hands not wearing gloves and close to reagent bottles") are calculated based on the IoU, overlap, and relative positions of the detection boxes to determine compliance with protective equipment requirements. The determination result is output in the form of structured events (e.g., "violation event: operation without gloves"), directly serving as the update input for the "operational event node" in the causal graph.

[0083] Furthermore, the real-time construction and dynamic updating of small causal graphs based on the digital twin snapshots at edge computing nodes transforms multi-source snapshot data into a knowledge representation capable of causal reasoning, enabling real-time modeling of potential risk chains in the experimental process. The small causal graph employs a lightweight structure (typically with 10–30 nodes). Nodes include operational events, environmental variables (temperature, humidity, concentration, etc.), reagent status (reagent location and type read by RFID), and instrument readings (power, temperature, rotation speed, etc.). Edges represent preset causal dependencies (e.g., "heating power increases → expected temperature increase" or "oxidant removed → prohibited from coexisting with reducing agent"). The system dynamically adds / updates node values, adjusts edge weights, or discovers new causal paths based on the latest snapshot data. The construction of the small causal graph can employ constraint-based PC algorithms or lightweight variants of NOTEARS for structure learning, combined with preset chemical / physical domain knowledge rules for initialization and edge constraint correction, to achieve real-time updates in environments with limited edge computing resources.

[0084] Furthermore, calculating the similarity between the current operation sequence and the experimental intent summary token, as well as the degree of deviation between the actual readings and the expected causal relationship in the causal graph, involves performing a two-dimensional quantitative verification based on the aforementioned causal graph and intent summary token, outputting anomaly signals that can directly drive intervention decisions. Specifically, this includes: intent similarity calculation, which involves vectorizing the features of the current operation sequence (the sequence of operation events extracted from the causal graph) and the expected operation sequence in the intent summary token (e.g., embedding vectors or sequence encoding), and comparing them using cosine similarity or other sequence matching algorithms; and causal consistency deviation calculation, which involves traversing the causal paths in the causal graph and comparing the deviation between the actual readings and the expected causal relationship (e.g., the percentage deviation between the actual temperature change rate and the expected change rate, or the causal link probability being lower than the expected threshold). When any dimension exceeds a preset threshold, it is determined to be an intent deviation or causal anomaly, and the quantitative results (similarity value, degree of deviation, anomaly type) are output to the hierarchical intervention execution module.

[0085] Furthermore, by leveraging historical intervention event data, a lightweight online learning model is used to adaptively optimize the verification threshold and the weights of the causal graph structure, thereby constructing a closed-loop verification mechanism with self-evolutionary capabilities. The system uses each intervention event (including voice warning broadcasts, automatic device locking, and permission freezing) as a direct feedback signal, inputting it into a lightweight online learning model employing an incremental decision tree or online support vector machine (Online SVM) architecture. This model evaluates the effectiveness of the intervention event in real time with low latency and dynamically adjusts the similarity threshold in subsequent verification processes based on the evaluation results and feedback data, while adaptively correcting the weight parameters of the connections between nodes in the causal graph. Through this feedback-based dynamic adjustment, the system can continuously adapt to complex anomaly patterns in the laboratory, ensuring accurate identification of potential risks in different experimental scenarios, thereby continuously improving the accuracy of causal consistency verification and dynamically reducing the false alarm rate. The aforementioned threshold and weight adjustment parameters are configurable and can be adaptively optimized according to different laboratory scenarios or through statistical analysis of historical intervention data to adapt to different risk preferences and experimental types.

[0086] In one specific embodiment of the present invention, when a deviation from intent or causal anomaly is detected, a tiered proactive intervention is triggered based on the severity of the anomaly. Specifically,

[0087] When the intent similarity is lower than the expected threshold or the causal deviation is within a slight range, voice warning signals and light guidance signals are generated first, and multi-source feedback data such as user response time, operation correction trajectory and intent recovery status are collected.

[0088] The feedback data is statistically analyzed based on a preset sliding time window to calculate the intention correction rate and abnormal recurrence frequency, and to generate a fitness index that characterizes the effectiveness and interference level of the current warning strategy. The subsequent verification threshold and intervention priority are dynamically adjusted according to the fitness index. When the correction rate is higher than the benchmark and the recurrence frequency is lower than the preset value, the threshold is relaxed to reduce false alarms.

[0089] When the anomaly persists or the causal deviation worsens, based on the updated cause-effect graph and the adjusted intervention strategy, a power supply suspension command for the equipment or a reagent cabinet lock command is generated, and the intervention event is recorded in the process log.

[0090] When the cause-effect graph shows a high-risk cause-effect chain, a reversible forced rollback instruction is generated, including resetting the valve state, restoring the initial flow setting, or disconnecting the power supply, and the rollback execution result is sent back to the resource state machine to update the resource state.

[0091] It should be noted that when the system detects intent deviation or causal anomalies through intent similarity and causal consistency checks, it triggers tiered proactive intervention. This aims to proactively block or mitigate the chain of danger before the risk escalates into an accident, employing a gradual, controllable, and reversible intervention strategy based on the severity of the anomaly and its real-time evolution. Simultaneously, the intervention results are fed back to the causal model in real time, forming a closed-loop protection mechanism with self-evolving capabilities. This mechanism effectively addresses the shortcomings of traditional laboratory safety systems, which only issue alarms or require manual intervention after an accident occurs and lack tiered response and feedback optimization capabilities. It endows the system with complete proactive protection capabilities, including prediction, early warning, blocking, recovery, and continuous optimization.

[0092] Furthermore, when the anomaly is at a minor deviation stage, a non-intrusive, low-interference alert method is used to guide operators to proactively correct their behavior and prevent further escalation of risks. Based on the verification results, the system determines the anomaly to be minor (e.g., intent similarity slightly below the threshold, causal deviation within a preset minor range), immediately generating a voice warning signal (such as "Please confirm that protective equipment is complete" or "Current operation deviates from the expected path") and issuing guidance signals (such as flashing red warning lights) through on-site lighting equipment. After intervention, the system collects multi-source feedback data in real time, including user response time, operation correction trajectory, and intent recovery status, and constructs this feedback data into the current interaction feature vector as input for subsequent statistical analysis.

[0093] Furthermore, based on a preset sliding time window, cumulative statistical analysis is performed on the interaction feature vectors to calculate the intent correction rate and abnormal recurrence frequency within a specific time window, and an fitness index characterizing the effectiveness and interference level of the current warning strategy is generated. This aims to achieve dynamic adaptation of the intervention strategy to user behavior habits. The system dynamically adjusts subsequent verification thresholds and intervention priorities according to the fitness index: when the intent correction rate is higher than the baseline value and the abnormal recurrence frequency is lower than the preset threshold, the intent similarity threshold is appropriately relaxed to reduce the false alarm rate; conversely, the threshold is tightened and the response priority of the warning signal is increased, thereby continuously optimizing the protection strategy according to user behavior habits. This adjustment process is implemented through a lightweight evaluation algorithm to ensure that the intervention measures minimize interference with normal experiments while ensuring safety.

[0094] Furthermore, when mild intervention fails to effectively correct the anomaly and the risk has entered a moderate stage, physical restrictive measures are taken to promptly cut off the further development of the hazard source. Based on the updated causal graph and adjusted intervention strategy, the system reassesses the risk path. If the deviation worsens (e.g., causal deviation exceeds the preset moderate range, or the duration of the anomaly exceeds the preset window), targeted blocking instructions are generated: a power supply suspension instruction is issued to precision instruments or heating equipment (achieved through a soft power-off via the instrument control interface); a locking instruction is issued to reagent cabinets (achieved through an electronic lock hardware interface). Simultaneously, the intervention event (triggering conditions, intervention type, execution time, execution result, and fitness indicators) is fully recorded in the process log, providing verifiable data for subsequent auditing, tracing, and system optimization.

[0095] Furthermore, when an anomaly evolves into a high-risk causal chain (e.g., a high-probability explosion risk path in the causal graph such as "oxidant removal + simultaneous heating + lack of protective gear"), a reversible forced recovery operation is immediately executed to pull the system state back to a safe baseline as much as possible, preventing irreversible accidents. Based on the probability of a risk link in the causal graph exceeding a preset high threshold, the system generates targeted rollback instructions: reset valve status (close gas / liquid valves); restore initial flow settings (pump or gas flow controller reset); disconnect power (power off the main control relay). All rollback operations are designed to be reversible and standardized to ensure rapid resumption of the experimental process after the anomaly is resolved. The rollback execution results (success / failure, comparison of states before and after execution) are transmitted back to the resource state machine in real time, updating the resource from "in use" to "abnormally paused" or "recovered" state, and simultaneously updating the risk path weights in the causal graph.

[0096] The aforementioned tiered proactive intervention process continues until the anomaly is eliminated or the experiment ends naturally. At this point, the system automatically performs compliant archiving of data assets and generates a full-process log, forming a complete closed-loop management system. The threshold ranges, time windows, and priority adjustment parameters mentioned above are all configurable and can be adaptively optimized based on different laboratory scenarios or through statistical analysis of historical intervention data to adapt to different risk preferences and experiment types.

[0097] In one specific embodiment of the present invention, for the reservation application including the reagent usage list, the method further includes performing list constraint verification and causal risk deduction when the reagents are requisitioned. Specifically,

[0098] The actual reagent information used is read by RFID and compared with the pre-approved requisition list for consistency of the product categories.

[0099] Based on the comparison results, verify whether the temperature, humidity, concentration and other conditions collected by the current environmental sensors in the cabinet meet the storage requirements of the corresponding reagents in the list;

[0100] If environmental conditions meet the requirements, further testing will be conducted to determine if any mutually exclusive chemicals on the list were removed at the same time period, and a conflict signal will be generated.

[0101] Using the conflict signals, construct or update a small causal graph related to reagents, deduce potential risk paths, and immediately generate and execute intervention commands to prohibit further outbound shipments, lock cabinets, or trigger audible and visual alarms based on the risk level of the deduced paths.

[0102] The results of the intervention are recorded in the overall process log to maintain consistency of data and status throughout the process.

[0103] It should be noted that the purpose of implementing list constraint verification and causal risk simulation in the actual reagent requisition process is to transform the approved reagent usage plan from a static list into a dynamic and executable safety constraint benchmark, and to conduct multi-dimensional compliance and risk link verification of the actual operation at the moment of requisition, thereby blocking the source of violations and potential risk links in the high-risk reagent circulation process.

[0104] Furthermore, reading the actual reagent usage information and comparing it with the list allows for the physical capture of the reagent's true identity and usage behavior, and performs a basic matching and verification with the approval intent. The system uses the RFID reader / writer on the smart reagent cabinet to read the label information of the retrieved reagents (including name, batch number, specifications, etc.) in real time, and immediately performs a precise comparison with the approved usage list in the reservation application (consistency in category, quantity, and specifications).

[0105] Furthermore, verifying whether the environment inside the reagent cabinet meets the storage requirements of the reagents involves verifying the compliance of the physical environment in which the reagents are located, ensuring that the storage state of the reagents before use meets their chemical property requirements. The system reads real-time data from environmental sensors inside the cabinet (temperature, humidity, VOC concentration, toxic gases, etc.) and compares it with the preset storage condition thresholds in the requisition list (e.g., "nitric acid needs to be stored at a low temperature of 0–5℃").

[0106] Furthermore, after basic compliance is achieved, conflict detection is performed on the inherent incompatibilities between chemicals to prevent high-risk combinations from appearing simultaneously in the operating space. The system queries all reagent records that have been issued or are being issued within the current time window (e.g., the last 5 minutes) and matches them against the mutual exclusion rule base (a preset chemical incompatibility table, such as strong oxidants and flammable materials, acids and bases) in the issuance list. If mutually exclusive combinations are found to be retrieved simultaneously, a structured "conflict signal" (including conflicting categories, timestamps, and risk levels) is generated.

[0107] The small causal graph related to the reagent can be used as a submodule or a dynamic extension of the overall experimental causal graph, sharing some nodes (such as environmental variable nodes) with the global causal graph, thereby enabling unified reasoning about reagent risk and overall experimental risk.

[0108] Furthermore, the detected conflict signals are transformed into a reasonable risk knowledge representation, and downstream accident chains are deduced based on this to achieve proactive blocking. The system uses conflict signals as triggering conditions to quickly construct or update a small causal graph focused on reagent interactions (nodes include reagents already used, current environmental variables, and operational events; edges represent chemical reaction dependencies or hazard amplification paths, such as "oxidant removal + simultaneous presence of reducing agent → exothermic reaction risk"). By traversing the causal graph, the probability and severity of risk chains are calculated, and intervention commands are immediately generated and issued based on preset level thresholds: prohibiting further outbound shipments (locking remaining outbound channels); locking the entire cabinet (electronic lock forcibly closing); triggering audible and visual alarms (on-site warning lights + buzzer).

[0109] Intervention commands are typically issued using standardized industrial protocols or vendor SDK interfaces to ensure compatibility with instruments from different brands.

[0110] Furthermore, the system provides a complete and auditable record of the entire reagent requisition process, and provides a data foundation for subsequent system optimization and anomaly tracing. The system will record the triggering conditions, execution time, execution results (success / failure, comparison of states before and after execution), and related causal graph snapshots of intervention commands in the process log.

[0111] In one specific embodiment of the present invention, the data asset compliance archiving and full-process log generation are completed, specifically as follows:

[0112] When a user uploads experimental data, the validity of the data access token and the login status bound to this reservation are verified in real time.

[0113] Once the verification is successful, the experimental data will be written to the storage space strongly bound to the reservation record, and the write event will be recorded in the process log.

[0114] After the experiment, the resource state machine is automatically restored to the idle state, all tokens and locks generated during the reservation process are released, and a complete process log is generated, including multimodal perception event records, causal anomaly detection records, intervention operation records, and rollback operation records, to support subsequent auditing and traceability.

[0115] It should be noted that the completion of data asset compliance archiving and full-process log generation aims to ensure that all data assets generated by the experiment strictly adhere to the principle of "data follows the event, rights follow the reservation," and to achieve mandatory binding of data ownership, integrity, and traceability. At the same time, it provides a complete and auditable record of perception, verification, intervention, and rollback behaviors throughout the entire life cycle of the experiment, thereby providing a credible and traceable chain of evidence for subsequent scientific research integrity review, intellectual property protection, accident liability tracing, and continuous system optimization.

[0116] Furthermore, when users upload experimental data, the validity of the data access token bound to the current reservation and the login status are verified in real time. This serves as the entry point for data asset archiving, ensuring that uploads are strictly limited to the approved reservation time window and initiated by legitimately authorized users, thereby preventing data tampering, unauthorized uploads, or the infiltration of unreserved experimental data. When the system receives an upload request, it first extracts the data access token strongly bound to the current reservation and verifies its two core attributes in real time: validity of the time: the token's validity period must cover the current upload time and not have expired (usually synchronized with the reservation end time); login status: verifying whether the current session is maintained by the reservation applicant or authorized role. Only when both pass the verification are subsequent write processes allowed; otherwise, the upload is directly rejected and the violation is recorded.

[0117] Furthermore, upon successful verification, the experimental data is written to a storage space strongly bound to the reservation record, and the write event is recorded in the process log. This achieves a dual binding of physical storage and logical ownership of data assets, ensuring that every piece of experimental data is traceable to a specific reservation record, forming an undeniable principle of "data follows reservation, rights are bound to reservation." After successful verification, the system writes the uploaded experimental data (including raw files, metadata, analysis results, etc.) to a pre-allocated cloud drive space or dedicated storage partition. The access path or index key of this storage space is strongly bound to the reservation record ID (e.g., through a database foreign key or encrypted token). Simultaneously, key information of the write event (upload time, file hash value, token verification result, write path, operator ID, etc.) is fully recorded in the process log.

[0118] Furthermore, after the experiment, the resource state machine is automatically restored to an idle state, and a complete process log is generated. This ensures the standardized restoration of resource states and reliable evidence storage of the entire process, guaranteeing timely release of system resources for subsequent use. It also generates an immutable and complete audit trail for the entire experiment lifecycle. Upon detecting an experiment end signal (user manual confirmation, appointment time expiration, or completion of all data uploads), the system automatically performs the following operations: restoring the resource state machine from "in use" or "abnormally paused" to "idle"; releasing all temporary tokens (intent digest tokens, data access tokens), distributed locks (Redis red locks), and hardware lock states generated during the appointment process; and summarizing and generating a complete process log. This log records the following key events in a structured format (e.g., JSON or a dedicated log table): multimodal awareness events (environmental parameter changes, video violation detection, UWB location anomalies, etc.); causal anomaly detection records (intent similarity, causal deviation value, anomaly trigger time point); intervention operation records (voice alerts, power outages, cabinet locking, etc.); and rollback operation records (valve reset, flow restoration, power disconnection, etc.). The log is strongly correlated with the appointment record ID and the experimental data writing event, forming a traceable chain of evidence from appointment to completion.

[0119] The complete process log is stored in a structured format (such as JSON logs or relational database log tables) in a unified log service, supporting encrypted storage, time-series indexing, and access control queries to meet compliance requirements for scientific research integrity review, intellectual property protection, and accident liability tracing.

[0120] In a second embodiment of the present invention, the present invention provides a smart laboratory full-process management method based on multimodal perception and strict control logic, such as... Figure 2 As shown, the system includes an intent summary generation module 1, a multimodal access verification module 2, a causal real-time monitoring module 3, and a hierarchical intervention execution module 4;

[0121] The intent summary generation module 1 is used to respond to a user's reservation request for laboratory resources, and after approval, generate an experimental intent summary token including the expected scope of the experiment and constraint rules, and send it to the edge computing node and the corresponding control terminal.

[0122] The multimodal access verification module 2 is used to perform multimodal identity verification and location presence verification in response to a user approaching the target resource. Only when all verification results are passed, the resource state machine is updated and the resource is unlocked.

[0123] The real-time causal monitoring module 3 is used to construct and dynamically update the causal relationship model in real time based on the collected multi-source heterogeneous sensing data during the experiment, and to verify the intent similarity and causal consistency of the current operation state by combining the experiment intent summary token.

[0124] The graded intervention execution module 4 is used to trigger graded proactive intervention based on the severity of the abnormality when a deviation from intent or causal anomaly is detected, until the experiment ends and the data asset compliance archiving and full-process log generation are completed.

[0125] In a third embodiment of the present invention, the present invention provides an electronic device, including a memory and a processor, characterized in that the memory stores a computer program that can run on the processor, and when the program is executed on the processor, it implements the steps in the smart laboratory full-process management method based on multimodal perception and strict control logic as described above.

[0126] In Embodiment 4 of the present invention, the present invention provides a storage medium storing a computer program, characterized in that, when the computer program is executed by a processor, it implements the steps in the smart laboratory full-process management method based on multimodal perception and strict control logic as described above.

[0127] In summary, this invention provides a smart laboratory end-to-end management method based on multimodal perception and strict control logic. By constructing a unified business logic platform and a closed loop of multimodal perception and strict control logic, it achieves deep integration and data linkage of instrument reservation, reagent lifecycle management, intelligent room control, real-time environmental and behavioral monitoring, and abnormal causal reasoning. This effectively eliminates the defects of system fragmentation, information silos, and passive management in existing technologies. The system can proactively identify intention deviations and causal anomalies throughout the entire experimental process and trigger tiered intervention measures, effectively improving resource collaborative scheduling capabilities, real-time security access control, and overall laboratory operation and maintenance efficiency. It significantly reduces management risks and manual intervention costs, meeting the needs of modern laboratories for intelligent, closed-loop safety management.

[0128] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the modules described above can be referred to the corresponding process in the aforementioned method implementation, and will not be repeated here.

[0129] The modules described as separate components may or may not be physically separate. Similarly, the components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0130] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in a combination of hardware and software functional modules.

[0131] The integrated modules implemented as software functional modules described above can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer system (which may be a personal computer, server, or network system, etc.) or processor to execute some steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0132] Finally, it should be noted that the above 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.

Claims

1. A smart laboratory end-to-end management and control method based on multimodal perception and strict control logic, characterized in that: include, In response to a user's request to reserve laboratory resources, an experimental intent summary token, which includes the expected scope and constraints of the experiment, is generated after the request is approved and sent to the edge computing node and the corresponding control terminal. In response to a user approaching the target resource, perform multimodal authentication and location presence verification. Only when all verification results pass, update the resource state machine and unlock the resource. During the experiment, a causal relationship model was constructed and dynamically updated in real time based on the collected multi-source heterogeneous sensing data. The experiment intent summary token was used to verify the intent similarity and causal consistency of the current operation state. When a deviation from intent or causal anomaly is detected, tiered proactive intervention is triggered based on the severity of the anomaly until the experiment ends and data asset compliance archiving and full-process log generation are completed.

2. The intelligent laboratory full-process management and control method based on multimodal perception and strict control logic according to claim 1, characterized in that: The generated experimental intent summary token, which includes the expected experimental range and constraint rules, includes: Extract the experimental objective description, reagent requisition list, and corresponding rules from the preset standard experimental template library from the reservation application; The extracted content is encapsulated into an experimental intent summary token, which includes the range of permitted devices, reagent combination constraints, environmental safety constraints, and expected operation sequence characteristics. The experimental intent summary token is sent to the edge computing node and the corresponding instrument or reagent cabinet control terminal as a benchmark for subsequent real-time verification.

3. The intelligent laboratory full-process management method based on multimodal perception and strict control logic according to claim 2, characterized in that: The process of performing multimodal authentication and location presence verification, including updating the resource state machine and unlocking the resource only when all verification results are successful, includes... When a user is detected approaching a target resource, biometric recognition is performed and infrared and RGB images are acquired simultaneously. Texture depth and reflectivity are calculated using the fused image data to complete liveness detection. After the liveness detection is passed, dynamic interactive commands are randomly generated and issued, and the degree of matching between the user's facial key point displacement and the command is collected and verified. For high-risk reagent cabinets, a dual-person, dual-lock logic is further implemented. Within a preset time window, biometric information of two authorized personnel is collected, causing the state machine to enter a dual-person presence state. The real-time location coordinates of the smart badge are compared with the preset electronic fence area. Only when the coordinates are within the valid operating area will the verification pass signal be transmitted to the resource state machine to perform the unlocking operation.

4. The intelligent laboratory full-process management method based on multimodal perception and strict control logic according to claim 3, characterized in that: The process of constructing and dynamically updating a causal relationship model in real time based on collected multi-source heterogeneous sensing data, and verifying the intent similarity and causal consistency of the current operation state by combining experimental intent summary tokens, includes... Environmental sensor data is collected via the MQTT protocol, video data is collected via the video stream analysis gateway, and wearable device coordinate data is collected via the positioning gateway. Timestamp alignment and spatiotemporal fusion are performed on all data streams to generate a real-time digital twin snapshot. The video stream is analyzed frame by frame using an object detection model to identify the categories of personnel and protective equipment, and the spatial relationship between each detection box is calculated to determine the compliance of the wearing. The compliance determination result is used as the input for updating the causal graph nodes. On the edge computing node, a small causal graph is constructed and dynamically updated in real time based on the digital twin snapshot. The causal graph includes operation event nodes, environmental variable nodes, reagent status nodes and instrument reading nodes, and the graph structure is dynamically updated according to the preset causal dependency relationship. The similarity between the current operation sequence and the experimental intent summary token is calculated, as well as the degree of deviation between the actual readings and the expected causal relationships in the causal graph. Using historical intervention event data, a lightweight online learning model is used to adaptively optimize the verification threshold and the causal graph structure weights.

5. The intelligent laboratory full-process management method based on multimodal perception and strict control logic according to claim 4, characterized in that: The step of triggering tiered proactive intervention based on the severity of the abnormality when a deviation from intent or causal anomaly is detected includes: When the intent similarity is lower than the expected threshold or the causal deviation is within a slight range, voice warning signals and light guidance signals are generated first, and multi-source feedback data such as user response time, operation correction trajectory and intent recovery status are collected. The feedback data is statistically analyzed based on a preset sliding time window to calculate the intention correction rate and abnormal recurrence frequency, and to generate a fitness index that characterizes the effectiveness and interference level of the current warning strategy. The subsequent verification threshold and intervention priority are dynamically adjusted according to the fitness index. When the correction rate is higher than the benchmark and the recurrence frequency is lower than the preset value, the threshold is relaxed to reduce false alarms. When the anomaly persists or the causal deviation worsens, based on the updated cause-effect graph and the adjusted intervention strategy, a power supply suspension command for the equipment or a reagent cabinet lock command is generated, and the intervention event is recorded in the process log. When the cause-effect graph shows a high-risk cause-effect chain, a reversible forced rollback instruction is generated, including resetting the valve state, restoring the initial flow setting, or disconnecting the power supply, and the rollback execution result is sent back to the resource state machine to update the resource state.

6. The intelligent laboratory full-process management method based on multimodal perception and strict control logic according to claim 1 or 4, characterized in that: For the reservation application that includes the reagent usage list, the process also includes checking the list constraints and performing causal risk deduction when the reagents are requisitioned, specifically including: The actual reagent information used is read by RFID and compared with the pre-approved requisition list for consistency of the product categories. Based on the comparison results, verify whether the temperature, humidity, concentration and other conditions collected by the current environmental sensors in the cabinet meet the storage requirements of the corresponding reagents in the list; If environmental conditions meet the requirements, further testing will be conducted to determine if any mutually exclusive chemicals on the list were removed at the same time period, and a conflict signal will be generated. Using the conflict signals, construct or update a small causal graph related to reagents, deduce potential risk paths, and immediately generate and execute intervention commands to prohibit further outbound shipments, lock cabinets, or trigger audible and visual alarms based on the risk level of the deduced paths. The results of the intervention are recorded in the overall process log to maintain consistency of data and status throughout the process.

7. The intelligent laboratory full-process management and control method based on multimodal perception and strict control logic according to claim 1, characterized in that: The completion of data asset compliance archiving and full-process log generation includes... When a user uploads experimental data, the validity of the data access token bound to this reservation and the login status are verified in real time. Once the verification is successful, the experimental data will be written to the storage space strongly bound to the reservation record, and the write event will be recorded in the process log. After the experiment, the resource state machine is automatically restored to the idle state, all tokens and locks generated during the reservation process are released, and a complete process log is generated, including multimodal perception event records, causal anomaly detection records, intervention operation records, and rollback operation records, to support subsequent auditing and traceability.

8. A smart laboratory end-to-end management and control method based on multimodal perception and strict control logic, characterized in that: It includes an intent summarization generation module, a multimodal access verification module, a causal real-time monitoring module, and a tiered intervention execution module; The intent summary generation module is used to respond to a user's reservation request for laboratory resources. After approval, it generates an experimental intent summary token that includes the expected scope of the experiment and the constraint rules, and sends it to the edge computing node and the corresponding control terminal. The multimodal access verification module is used to perform multimodal identity verification and location presence verification in response to a user approaching the target resource. Only when all verification results are passed, the resource state machine is updated and the resource is unlocked. The real-time causal monitoring module is used to construct and dynamically update the causal relationship model in real time based on the collected multi-source heterogeneous sensing data during the experiment, and to verify the intent similarity and causal consistency of the current operation status in combination with the experiment intent summary token. The tiered intervention execution module is used to trigger tiered proactive intervention based on the severity of the abnormality when a deviation from intent or causal anomaly is detected, until the experiment ends and data asset compliance archiving and full-process log generation are completed.

9. An electronic device, comprising a memory and a processor, characterized in that: The memory stores computer programs that can run on the processor, and when the program is executed on the processor, it implements the steps in the smart laboratory full-process management method based on multimodal perception and strict control logic as described in any one of claims 1-7.

10. A storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, it implements the steps in the smart laboratory full-process management method based on multimodal perception and strict control logic as described in any one of claims 1-7.

Images