A laboratory internal risk prediction trolley and a risk prediction method thereof

Abstract

The application discloses a laboratory internal risk prediction trolley and a risk prediction method thereof, and the trolley comprises a moving chassis, a sensing module, a computing module, a communication module, an execution module and a power module. By adopting a multi-sensor array and an autonomous moving chassis, the application realizes full-time, full-area and dead-angle-free data acquisition in a laboratory. By a weighting fusion algorithm and an LSTM-CNN fusion risk prediction model built in the computing module, multi-dimensional data such as environment, equipment, personnel and dangerous goods are comprehensively analyzed and judged, and a leap from post-alarm to pre-prediction is realized. By an execution module integrated with an audible-light alarm, a small mechanical arm and a micro-spraying unit, and by linkage with an existing safety system of the laboratory, the risk response time is significantly shortened. By a data storage and model iteration mechanism, the system can adapt to the individualized needs of different laboratories, continuously improve the prediction accuracy and provide reliable and intelligent decision support for laboratory safety management.

Description

Technical Field

[0001] This invention relates to the field of laboratory safety monitoring and intelligent robot technology, specifically to a laboratory internal risk prediction vehicle and its risk prediction method. Background Technology

[0002] With the deep integration of science and technology with industrialization, laboratories, as core locations for scientific research, innovation, and teaching practice, are filled with diverse risk sources, including toxic and hazardous chemicals, high-pressure and high-temperature equipment, precision instruments, and biological samples. Laboratory safety management is directly related to the health and safety of researchers and the orderly conduct of experimental activities. Currently, the mainstream laboratory safety management model still heavily relies on manual inspections and fixed-point monitoring systems; however, this traditional model has revealed many inherent limitations in practical application.

[0003] First, manual inspections suffer from efficiency and timeliness bottlenecks. Limited by the physiological limits, sense of responsibility, and working hours of the inspectors, manual inspections struggle to achieve 24 / 7 comprehensive and continuous monitoring of the laboratory. This easily leads to missed or false detections, and the inspection results are subjective. More importantly, manual inspections are essentially a "post-event" response mechanism; risks are only discovered and addressed after they occur, lacking the ability to anticipate potential risks, resulting in a severely delayed response.

[0004] Secondly, fixed monitoring systems suffer from blind spots and poor adaptability to different scenarios. Fixed cameras or sensors typically only cover specific locations and cannot effectively adapt to dynamic scenarios in laboratories, such as frequent personnel movement, equipment repositioning, and temporary transfer of hazardous materials. For confined or obscured spaces such as inside fume hoods, behind instruments, and in the corners of reagent cabinets, the monitoring capabilities of fixed systems are severely inadequate, creating safety blind spots. Furthermore, this fixed deployment method has poor system scalability and cannot be flexibly relocated or reconfigured according to changes in laboratory functional zoning or layout.

[0005] Secondly, existing monitoring methods have limited analytical capabilities and lack multi-dimensional data fusion. Most monitoring equipment currently available on the market or used in laboratories primarily monitors single parameters, such as ambient temperature, humidity, or a specific gas. This "single-point" monitoring cannot correlate and comprehensively analyze multi-dimensional information such as environmental parameters (temperature, humidity, toxic gas concentration), equipment operating status (vibration, temperature, current), personnel behavior (wearing protective equipment, violations), and hazardous material information (type, shelf life, storage location). Therefore, the system struggles to identify complex and coupled risk sources, easily leading to false alarms (e.g., false alarms triggered by fluctuations in ambient temperature and humidity) or missed alarms (e.g., failure to detect the correlation between early abnormal equipment vibration and gas leaks), resulting in low safety management efficiency.

[0006] Finally, existing inspection robot technology lacks specific adaptation to laboratory scenarios. Although inspection robots have emerged in the industrial sector, their monitoring parameter configurations, risk identification models, path planning algorithms, and physical dimensions are all designed for open environments such as large factories and substations. They cannot be directly adapted to the special environments of laboratories, which are characterized by limited space, dense obstacles, complex risk sources, and frequent personnel movement. These robots generally lack lightweight risk prediction algorithms (such as vibration-temperature joint prediction models for instrument malfunctions) and preliminary linkage response capabilities (such as shutting down valves of small equipment or activating local sprinklers) specifically for laboratory scenarios, making it difficult to meet the full-chain management needs of laboratory safety, including "pre-event prediction, in-event intervention, and post-event traceability."

[0007] In summary, there is an urgent need in the existing technology for an intelligent and adaptive solution that can achieve all-time, all-area monitoring of the laboratory without blind spots, possess multi-dimensional data fusion analysis and risk prediction capabilities, and execute graded early warning and preliminary handling. Summary of the Invention

[0008] In view of the above-mentioned technical problems in related technologies, the present invention proposes a laboratory internal risk prediction vehicle and its risk prediction method, which can overcome the above-mentioned shortcomings of the prior art.

[0009] To achieve the above-mentioned technical objectives, the technical solution of the present invention is implemented as follows: A laboratory risk prediction vehicle includes: Mobile chassis, used to achieve autonomous movement, fixed-point docking, and dynamic obstacle avoidance; The sensing module, mounted on the mobile chassis, includes a multi-sensor array for real-time collection of environmental parameters, equipment operating parameters, personnel behavior data, and obstacle information within the laboratory. The computing module is communicatively connected to both the mobile chassis and the sensing module. The computing module has a built-in risk prediction model and uses a weighted fusion algorithm to perform fusion analysis on the multi-dimensional data collected by the sensing module in order to identify risk types and assess risk levels. A communication module, which is communicatively connected to the computing module, is used to enable data interaction with external devices; An execution module, communicatively connected to the computing module, is used to perform risk warning and / or preliminary handling operations according to instructions from the computing module; and A power module is used to provide electrical energy to the vehicle.

[0010] Furthermore, the multi-sensor array of the sensing module includes one or more of the following: lidar, industrial camera, infrared thermal imager, temperature and humidity sensor, gas sensor array, vibration sensor, and RFID reader.

[0011] Furthermore, the calculation module uses a precision-reliability joint assignment method to determine the fusion weight of each sensor in the weighted fusion algorithm, specifically including: calculating the basic weight using the reciprocal of the sensor precision error; correcting the basic weight using a reliability attenuation coefficient to obtain the corrected weight; and normalizing the corrected weight to obtain the final fusion weight.

[0012] Furthermore, the risk prediction model built into the computing module includes an LSTM-CNN fusion neural network model, which adopts a five-layer architecture, namely: feature input layer, LSTM temporal feature extraction layer, CNN spatial feature extraction layer, fusion layer and output layer; The feature input layer is used to receive multi-source monitoring data of equipment vibration and temperature; the LSTM temporal feature extraction layer is used to capture the temporal dependencies of the data; the CNN spatial feature extraction layer is used to extract local anomaly features of the data; the fusion layer is used to fuse temporal features and spatial features; and the output layer is used to output the equipment fault type, occurrence probability and warning time.

[0013] Furthermore, the computing module also integrates a personnel behavior recognition model, which is used to identify dangerous behaviors of laboratory personnel based on image or video data collected by the perception module. The dangerous behaviors include at least not wearing protective equipment and using open flames in violation of regulations.

[0014] Furthermore, the execution module includes one or more of an audible and visual alarm unit, a small robotic arm, and a micro-spray unit; the audible and visual alarm unit is used to issue audible and visual alarms at different frequencies according to different risk levels; the small robotic arm is used to perform operations such as closing valves, removing hazardous materials, or turning off equipment power; and the micro-spray unit is used to perform preliminary cooling measures for fires or equipment overheating.

[0015] Furthermore, the power module includes a lithium battery pack and a charging management unit. The power module is used to monitor the power level in real time and send a low power signal to the calculation module when the power level is lower than a first preset threshold. The calculation module is used to plan a return charging path based on the low power signal, control the mobile chassis to move autonomously to the charging base station for charging, and control the vehicle to return to the inspection interruption position after the power level is charged to a second preset threshold.

[0016] A risk prediction method based on any one of the laboratory internal risk prediction carts includes the following steps: S1 map building and path planning: control the movement of the vehicle and collect laboratory environmental data, build a laboratory map and plan inspection routes; S2 multi-dimensional data acquisition controls the trolley to move along the inspection route and collects environmental parameters, equipment operating parameters, personnel behavior data and obstacle information in real time through the sensing module; S3 data fusion and risk prediction uses a weighted fusion algorithm to fuse multi-dimensional data and calls a pre-trained risk prediction model to identify risk types and assess risk levels. S4 Risk Response and Handling: Based on the assessed risk level, the control execution module performs corresponding early warning and / or preliminary handling operations. S5 data storage and model iteration transmits the collected data, risk prediction results, and handling records to the cloud platform for subsequent iterative optimization of the risk prediction model.

[0017] Furthermore, the risk levels assessed in step S3 include low risk, medium risk, and high risk; If a person at low risk is not wearing protective equipment or if environmental parameters deviate slightly from the standard, the procedure is to record the data and send a reminder message. The medium-risk level corresponds to a toxic gas concentration approaching the threshold or a slow increase in equipment temperature. The handling method is to activate an audible and visual alarm and push a warning message. The high-risk situations correspond to toxic gas leaks, equipment overheating, or small fires. The response measures include continuous audible and visual alarms, emergency warning push notifications, and control of the execution module for initial handling, while simultaneously linking with the laboratory fire protection or emergency system.

[0018] Furthermore, it also includes step S6: automatic recharging, which monitors the power of the power module in real time. When the power is lower than a preset threshold, the vehicle is controlled to move autonomously to the charging base station to complete automatic charging, and returns to the inspection interruption position after charging is completed.

[0019] The beneficial effects of this invention are as follows: By employing a multi-sensor array and an autonomous mobile chassis, this invention achieves all-weather, all-area data acquisition in the laboratory without blind spots, fundamentally overcoming the problems of low efficiency in manual inspections and blind spots in fixed monitoring. Through the weighted fusion algorithm built into the computing module and the LSTM-CNN fusion risk prediction model, it comprehensively analyzes multi-dimensional data on the environment, equipment, personnel, and hazardous materials, thus solving the problems of false alarms and missed alarms caused by single-parameter monitoring, achieving a leap from "post-event alarm" to "pre-event prediction." By integrating the execution modules of audible and visual alarms, small robotic arms, and micro-spray units, and linking them with the existing laboratory safety system, it enables tiered early warning and initial rapid response, significantly shortening risk response time. Simultaneously, through data storage and model iteration mechanisms, the system can adapt to the personalized needs of different laboratories, continuously improving prediction accuracy and providing reliable and intelligent decision support for laboratory safety management. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

[0021] A laboratory risk prediction vehicle according to an embodiment of the present invention includes: Mobile chassis, used to achieve autonomous movement, fixed-point docking, and dynamic obstacle avoidance; The sensing module, mounted on the mobile chassis, includes a multi-sensor array for real-time collection of environmental parameters, equipment operating parameters, personnel behavior data, and obstacle information within the laboratory. The computing module is communicatively connected to both the mobile chassis and the sensing module. The computing module has a built-in risk prediction model and uses a weighted fusion algorithm to perform fusion analysis on the multi-dimensional data collected by the sensing module in order to identify risk types and assess risk levels. A communication module, which is communicatively connected to the computing module, is used to enable data interaction with external devices; An execution module, communicatively connected to the computing module, is used to perform risk warning and / or preliminary handling operations according to instructions from the computing module; and A power module is used to provide electrical energy to the vehicle.

[0022] Preferably, the multi-sensor array of the sensing module includes one or more of the following: lidar, industrial camera, infrared thermal imager, temperature and humidity sensor, gas sensor array, vibration sensor, and RFID reader.

[0023] Preferably, the calculation module uses a precision-reliability joint assignment method to determine the fusion weight of each sensor in the weighted fusion algorithm, specifically including: calculating the basic weight using the reciprocal of the sensor precision error; correcting the basic weight using a reliability attenuation coefficient to obtain the corrected weight; and normalizing the corrected weight to obtain the final fusion weight.

[0024] Preferably, the risk prediction model built into the computing module includes an LSTM-CNN fusion neural network model, which adopts a five-layer architecture, namely: feature input layer, LSTM temporal feature extraction layer, CNN spatial feature extraction layer, fusion layer and output layer; The feature input layer is used to receive multi-source monitoring data of equipment vibration and temperature; the LSTM temporal feature extraction layer is used to capture the temporal dependencies of the data; the CNN spatial feature extraction layer is used to extract local anomaly features of the data; the fusion layer is used to fuse temporal features and spatial features; and the output layer is used to output the equipment fault type, occurrence probability and warning time.

[0025] Preferably, the computing module also integrates a personnel behavior recognition model, which is used to identify dangerous behaviors of laboratory personnel based on image or video data collected by the sensing module. The dangerous behaviors include at least not wearing protective equipment and using open flames in violation of regulations.

[0026] Preferably, the execution module includes one or more of an audible and visual alarm unit, a small robotic arm, and a micro-spray unit; the audible and visual alarm unit is used to issue audible and visual alarms at different frequencies according to different risk levels; the small robotic arm is used to perform operations such as closing valves, removing hazardous materials, or turning off equipment power; and the micro-spray unit is used to perform preliminary cooling measures for fires or equipment overheating.

[0027] Preferably, the power module includes a lithium battery pack and a charging management unit. The power module is used to monitor the power level in real time and send a low power signal to the calculation module when the power level is lower than a first preset threshold. The calculation module is used to plan a return charging path based on the low power signal, control the mobile chassis to move autonomously to the charging base station for charging, and control the vehicle to return to the inspection interruption position after the power level is charged to a second preset threshold.

[0028] A risk prediction method based on the aforementioned laboratory internal risk prediction vehicle includes the following steps: S1 map building and path planning: control the movement of the vehicle and collect laboratory environmental data, build a laboratory map and plan inspection routes; S2 multi-dimensional data acquisition controls the trolley to move along the inspection route and collects environmental parameters, equipment operating parameters, personnel behavior data and obstacle information in real time through the sensing module; S3 data fusion and risk prediction uses a weighted fusion algorithm to fuse multi-dimensional data and calls a pre-trained risk prediction model to identify risk types and assess risk levels. S4 Risk Response and Handling: Based on the assessed risk level, the control execution module performs corresponding early warning and / or preliminary handling operations. S5 data storage and model iteration transmits the collected data, risk prediction results, and handling records to the cloud platform for subsequent iterative optimization of the risk prediction model.

[0029] Preferably, the risk levels assessed in step S3 include low risk, medium risk, and high risk; If a person at low risk is not wearing protective equipment or if environmental parameters deviate slightly from the standard, the procedure is to record the data and send a reminder message. The medium-risk level corresponds to a toxic gas concentration approaching the threshold or a slow increase in equipment temperature. The handling method is to activate an audible and visual alarm and push a warning message. The high-risk situations correspond to toxic gas leaks, equipment overheating, or small fires. The response measures include continuous audible and visual alarms, emergency warning push notifications, and control of the execution module for initial handling, while simultaneously linking with the laboratory fire protection or emergency system.

[0030] Preferably, the method further includes step S6: automatic recharging, which involves real-time monitoring of the power module's charge level. When the charge level is lower than a preset threshold, the method controls the vehicle to move autonomously to the charging base station to complete automatic charging, and then returns to the inspection interruption position after charging is completed.

[0031] To facilitate understanding of the above technical solutions of the present invention, the following detailed description of the above technical solutions of the present invention will be provided through specific usage methods.

[0032] A laboratory risk prediction vehicle includes a mobile chassis, a sensing module, a computing module, a communication module, an execution module, and a power supply module.

[0033] The mobile chassis adopts a Mecanum wheel structure and is equipped with a 24V DC motor drive unit and an odometer. The motor supports stepless speed regulation to adapt to the inspection speed requirements. The mobile chassis receives path planning instructions from the computing module, enabling autonomous movement, fixed-point docking, and dynamic obstacle avoidance. The movement speed can be adjusted according to the scenario to meet the inspection rhythm requirements of different laboratories.

[0034] The sensing module employs a multi-sensor array design, with core components including a 16-line LiDAR, a 1080P industrial camera, an infrared thermal imager, temperature and humidity sensors, a gas sensor array, a vibration sensor, and an RFID reader. The LiDAR has a ranging range of 0.5-50m and an accuracy of ±1cm, used to capture information about surrounding obstacles to assist in map building and dynamic obstacle avoidance. The industrial camera has a 1080P resolution and a frame rate of 30fps, used to collect personnel behavior data and equipment operating status images to support the identification of personnel violations. The infrared thermal imager has a temperature measurement range of -20℃ to 150℃ and can capture temperature changes exceeding 0.1℃, used to detect potential hazards such as equipment overheating and abnormal ambient temperatures. The temperature and humidity sensors have a measurement range of 0-50℃ and 20%-90%RH, with an accuracy of ±0.5℃ and ±5%RH. H. Real-time monitoring of laboratory environmental temperature and humidity parameters; The gas sensor array adopts a four-in-one design, which can identify a variety of toxic and harmful gases such as CO, CH4, NH3, VOC, etc., with a detection accuracy of ≤0.1mg / m3, and is used to monitor potential gas leaks; The vibration sensor has a measurement range of 0-10g and a frequency of 10-1000Hz, and is used to collect equipment vibration data to help determine equipment malfunctions; The RFID reader supports reading RFID tags on hazardous reagent bottles to obtain information such as hazardous product type, shelf life, and storage requirements, realizing full life cycle monitoring of hazardous products.

[0035] Data processing: The data collected by the sensors first undergoes noise reduction and calibration preprocessing to filter out abnormal data before being transmitted to the computing module to ensure data accuracy. The computing module uses a weighted fusion algorithm to assign fusion weights based on the detection accuracy and operational reliability of different sensors. The weights are determined using a joint accuracy-reliability assignment method, with the reciprocal of the sensor accuracy error as the base weight, corrected by a reliability attenuation coefficient, and the final weights are obtained after normalization.

[0036] The specific weight calculation example is as follows: Assume that the system contains two core sensors: Sensor 1 and Sensor 2, which are used to collect the same physical quantity. The key parameters of each sensor are as follows: 1. Determine the basic parameters: The detection accuracy error of sensor 1 is 0.5%, and the operational reliability attenuation coefficient is 0.95. The attenuation coefficient ranges from 0 to 1. The closer the value is to 1, the higher the reliability. The detection accuracy error of sensor 2 is 1.0%, and the operational reliability attenuation coefficient is 0.90.

[0037] 2. Calculate the basic weight: The basic weight is calculated based on the reciprocal of the sensor accuracy error. The calculation formula is: Basic weight W0 = 1 / Detection accuracy error (%).

[0038] Based on this calculation, the fundamental weight W of sensor 1 can be obtained. 01 =1 / 0.5=2.0; the basic weight W of sensor 202 =1 / 1.0=1.0.

[0039] 3. Weight Correction: The basic weight is corrected using a reliability decay coefficient. The corrected weight W1 = basic weight W0 × reliability decay coefficient k.

[0040] Based on this calculation, the corrected weight W of sensor 1 can be obtained. 11 =2.0 × 0.95 = 1.9; Corrected weight W of sensor 2 12 =1.0 × 0.90 = 0.9.

[0041] 4. Normalization: Normalize the corrected weights to obtain the final fusion weight W. The normalization calculation formula is: W = corrected weight W1 / sum of corrected weights of all sensors.

[0042] First, calculate the sum of the corrected weights: 1.9 + 0.9 = 2.8; then obtain the final weights: the final weight of sensor 1 is W1 = 1.9 / 2.8 ≈ 0.679, and the final weight of sensor 2 is W2 = 0.9 / 2.8 ≈ 0.321.

[0043] The calculation module performs weighted fusion calculation on the data collected by the two sensors based on the final weights mentioned above. The fusion result is calculated as: Sensor 1 data × 0.679 + Sensor 2 data × 0.321. This achieves accurate data fusion based on precision and reliability, improves the accuracy and stability of data output, and performs fusion analysis on multi-dimensional data to avoid misjudgments caused by biases in data from a single sensor.

[0044] The computing module utilizes the NVIDIA Jetson AGX Xavier edge computing platform, running Ubuntu 20.04 and the ROS 2 Humble framework. It incorporates the CartographerSLAM algorithm and an LSTM-CNN fusion risk prediction model. The autonomous navigation algorithm employs a SLAM algorithm that combines LiDAR and vision, achieving centimeter-level positioning and laboratory map construction. It supports semantic navigation and dynamic obstacle avoidance. Semantic understanding involves collecting environmental data and user commands through multiple sensors, deconstructing the command intent and segmenting the environmental semantics using a deep learning model, and then aligning and matching it with the semantic map. Semantic navigation, based on this semantic understanding, constructs a 3D semantic map containing geometric information and semantic labels through semantic SLAM. This is combined with semantic feature matching, multi-sensor fusion positioning, and path planning algorithms to achieve "meaning-based navigation." Dynamic obstacle avoidance uses multi-sensor fusion to perceive dynamic obstacle parameters, combines historical motion data and semantic types to predict trajectories, generates local obstacle avoidance paths based on global path constraints, and adjusts the device state without deviating from the semantic navigation target. Inspection routes can be optimized according to the functional zoning of the laboratory. The risk prediction model integrates a machine learning model and an LSTM-CNN fusion neural network algorithm. The machine learning model is trained based on historical laboratory risk data and is used to predict the probability of fires and gas leaks. The LSTM-CNN fusion fault prediction model adopts a five-layer architecture: "feature input layer - temporal feature extraction layer - spatial feature extraction layer - fusion layer - output layer". The core functions of each layer are as follows, realizing equipment vibration and temperature trend analysis and fault prediction: 1. Model Architecture Details (1) Feature input layer: Receive multi-source monitoring data of equipment vibration (three-dimensional acceleration, displacement) and temperature (core components, environment), and after standardization preprocessing, organize them into a multi-dimensional time series dataset as input samples.

[0045] (2) LSTM temporal feature extraction layer: a bidirectional LSTM structure is adopted to capture the temporal dependency of vibration and temperature data, mine the time change trend features, and transform them into high-dimensional feature vectors to solve the problem that a single CNN cannot capture long-term temporal correlations.

[0046] (3) CNN spatial feature extraction layer: adopts a multi-layer convolution-pooling structure to extract local abnormal features in vibration and temperature data, supplement temporal features, and enhance the model’s ability to identify anomalies.

[0047] (4) Fusion layer: The feature splicing + attention mechanism is adopted to fuse temporal and spatial features, highlight the key features of fault prediction, and output a comprehensive feature vector.

[0048] (5) Output layer: Through the fully connected layer and the corresponding activation function, output the equipment fault type, occurrence probability, early warning time and vibration and temperature trend prediction curves to provide a reference for equipment maintenance.

[0049] 2. Explicitly define input and output parameters Input parameters: three-dimensional vibration signal of the equipment, core component and ambient temperature data. The input format is a time series matrix, and the time step can be set as needed.

[0050] Output parameters: Fault prediction results (category, probability, warning time), vibration / temperature trend prediction values ​​and fault level. The output is structured data and can be connected to a monitoring platform to achieve early warning push.

[0051] It integrates a lightweight CNN+YOLO human behavior recognition model, which shares hardware resources with the fault prediction model to achieve collaborative monitoring without additional costs.

[0052] The model takes on-site monitoring images / videos as input, extracts human features and combines them with target detection to accurately identify two types of dangerous behaviors: not wearing protective equipment and illegally using open flames. It outputs the location and duration of the behavior and triggers an early warning.

[0053] In summary, this system improves the accuracy of fault prediction by using a simplified and optimized LSTM-CNN fusion model, integrates personnel behavior recognition function to achieve dual protection, is applicable to multiple industrial scenarios, and has outstanding practicality and innovation.

[0054] The communication module supports Wi-Fi 6 and Bluetooth 5.2 communication, and an optional 5G module is available to adapt to different network environments in the laboratory. The communication module enables two-way communication with the laboratory's host computer, cloud platform, and management personnel's terminal devices to complete data transmission (collected data, risk prediction results, and handling records), instruction reception (remote control, inspection parameter adjustment), and early warning information push.

[0055] The execution module includes an audible and visual alarm unit, an optional small robotic arm (load ≤ 1kg), and a miniature sprinkler unit. The audible and visual alarm unit uses a red audible and visual alarm, adjusting the alarm frequency according to the risk level; low-risk situations only require a light warning, while medium- and high-risk situations trigger simultaneous audible and visual alarms. The small robotic arm is used to handle minor hazards, such as closing small valves, removing light hazardous materials, and shutting off power to small equipment. The miniature sprinkler unit is used for initial cooling of small fires and equipment overheating, and it can be linked to the laboratory fire suppression system to improve response efficiency.

[0056] The power module uses a lithium battery pack and a charging management unit, supporting fast charging. The power module monitors the lithium battery level in real time. When the level falls below a preset threshold (20% recommended), it sends a low-level signal to the computing module. The computing module plans the shortest return path and controls the vehicle to autonomously move to the charging base station to complete automatic charging. When the battery level reaches the preset threshold (90% recommended), charging is complete. The computing module then controls the vehicle to return to the interrupted inspection location to continue the inspection task, ensuring inspection continuity.

[0057] Through the collaborative work of the aforementioned hardware and software, accurate prediction and rapid handling of laboratory risks can be achieved, a multi-dimensional data fusion and analysis system can be constructed, and comprehensive risk prediction of environment, equipment, personnel, and hazardous materials can be realized, upgrading from "post-event alarm" to "pre-event prediction", providing strong support for laboratory safety management decisions.

[0058] A risk prediction method based on a laboratory-based risk prediction vehicle includes the following steps: Step S1: Map Building and Route Planning After the vehicle starts, the LiDAR and visual camera in the perception module simultaneously collect data on the laboratory environment, including aisle width, equipment location, and wall layout. The calculation module calls the SLAM algorithm (Cartographer) to process the collected data and complete the construction of a laboratory map with a positioning accuracy of ±1cm. Staff can set laboratory risk level zones via a host computer: high risk: reagent storage area; medium risk: experimental operation area; low risk: office area. The calculation module plans the optimal inspection route based on the zoning results, sets key inspection points such as reagent cabinets, heating equipment, and high-pressure equipment, and allows for customized inspection cycles (e.g., once every 1-2 hours).

[0059] Step S2: Multi-dimensional data collection The vehicle moves autonomously along a planned route at a speed of 0.1-0.3 m / s, while the sensing module collects various types of data in real time at a preset frequency. The collected data types include: environmental parameters (temperature, humidity, concentration of toxic and harmful gases), equipment operating parameters (vibration, surface temperature, current), personnel behavior data (operational actions, protective equipment wearing status), and obstacle information. The sensing module performs noise reduction and calibration preprocessing on the collected data, eliminating abnormal data such as sensor false triggering data, before transmitting it to the computing module to ensure data reliability.

[0060] Step S3: Data Fusion and Risk Prediction The calculation module employs a weighted fusion algorithm, allocating weights based on the accuracy and reliability of different sensors. LiDAR and visual cameras have the highest weight, followed by gas sensors and vibration sensors. This multi-dimensional data fusion analysis avoids misjudgments caused by biases in data from a single sensor. A pre-trained risk prediction model is used, combined with historical laboratory risk data, to identify and assess various risk types, categorizing risks into low, medium, and high levels. (1) Low risk: personnel not wearing protective equipment, environmental parameters slightly deviating from the standard, equipment slight vibration, etc., with no direct safety hazards; (2) Medium risk: Potential safety hazards exist, such as toxic gas concentrations approaching the threshold, equipment temperature rising slowly, and minor improper storage of hazardous materials. (3) High risk: toxic gas leaks, equipment overheating (sudden temperature rise in a short period of time), violent equipment vibration, small fires, etc., which require immediate handling.

[0061] It generates risk prediction results and targeted handling suggestions, and simultaneously generates a risk heat map to visually display high-risk areas in the laboratory, making it easy for managers to quickly locate them. Risk level Determining the object Quantization / State Threshold Nature of risk Response time Low risk Personnel protection, environmental parameters, equipment vibration Without protective equipment, the parameter deviation is ≤ ±5% of the standard value, and the vibration displacement is ≤ 0.2 mm. No direct safety hazards Routine inspections, rectification within 24 hours Medium risk Toxic gas equipment temperature hazardous materials storage Concentration reaches 80%~99% of the threshold; temperature rise ≤5℃ / h (slow rise); minor storage violations (mixed storage / missing label). Potential safety hazards Immediate alert, check within 4 hours High risk Toxic gas equipment status fire situation Concentration ≥ 100% of threshold (leakage); Temperature rise ≥ 10℃ / h (sudden rise) / overheating; Vibration displacement ≥ 1.0mm / severe shaking; Small fire / smoke alarm. Immediate security risks Immediate response and coordinated alarm

[0062] Low-risk handling: Only record data and push reminder information to the management personnel terminal, without activating audible and visual alarms, and continuously track data changes; Medium-risk response: Activate audible and visual alarms, push early warning information to management personnel terminals, indicate the location of the risk, the type of risk, and preliminary response suggestions, and link the laboratory ventilation system (if the gas concentration is abnormal). High-risk response: Continuously activate audible and visual alarms, push emergency warnings, and simultaneously control the execution module to carry out preliminary measures such as micro-spray cooling and robotic arm to shut down equipment power. Link up with laboratory fire protection and emergency systems, and upload real-time images of the scene to facilitate remote analysis and on-site response by management personnel. After the disposal is completed, the disposal process and results are recorded to form a disposal ledger for subsequent model iterations.

[0063] Step S5: Data Storage and Model Iteration The computing module transmits the collected data, risk prediction results, and treatment records to the laboratory cloud platform through the communication module. The storage period can be set, and data query and traceability are supported. The risk prediction model is iteratively optimized periodically (e.g., monthly) using newly added risk data and treatment records, and the algorithm parameters are adjusted to improve the accuracy of risk prediction and reduce the probability of false alarms and missed alarms. The model training data can be updated according to changes in the laboratory scenario (e.g., adding equipment or adjusting the layout) to adapt to personalized needs.

[0064] Step S6: Automatic recharging The power module monitors the lithium battery level in real time. When the level is lower than the preset threshold (20% recommended), it sends a low-level signal to the computing module. The computing module plans the shortest return path and controls the vehicle to move autonomously to the charging base station to complete automatic charging. When the level is charged to the preset threshold (90% recommended), charging is complete. The computing module controls the vehicle to return to the inspection interruption position and continue to perform the inspection task to ensure the continuity of the inspection.

[0065] Through the above steps, the laboratory can achieve full-time, full-area monitoring without blind spots, build a multi-dimensional data fusion and analysis system, realize comprehensive risk prediction of environment, equipment, personnel and hazardous materials, upgrade from "post-event alarm" to "pre-event prediction", have the ability of graded early warning and preliminary handling, shorten risk response time, link with existing safety systems, and reduce accident losses.

[0066] In summary, by employing the above-mentioned technical solutions of this invention, the present invention achieves all-weather, all-area data acquisition in the laboratory without blind spots by using a multi-sensor array and an autonomous mobile chassis, thereby fundamentally overcoming the problems of low efficiency in manual inspections and blind spots in fixed monitoring. Through the weighted fusion algorithm built into the computing module and the LSTM-CNN fusion risk prediction model, comprehensive analysis of multi-dimensional data on the environment, equipment, personnel, and hazardous materials is performed, thus solving the problems of false alarms and missed alarms caused by single-parameter monitoring, achieving a leap from post-event alarm to pre-event prediction. By integrating the execution modules of audible and visual alarms, small robotic arms, and micro-spray units, and linking them with the existing laboratory safety system, graded early warning and initial rapid response become possible, significantly shortening risk response time. Simultaneously, through data storage and model iteration mechanisms, the system can adapt to the personalized needs of different laboratories, continuously improving prediction accuracy and providing reliable and intelligent decision support for laboratory safety management.

[0067] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A laboratory internal risk prediction cart, characterized by, include: Mobile chassis, used to achieve autonomous movement, fixed-point docking, and dynamic obstacle avoidance; The sensing module, mounted on the mobile chassis, includes a multi-sensor array for real-time collection of environmental parameters, equipment operating parameters, personnel behavior data, and obstacle information within the laboratory. The computing module is communicatively connected to both the mobile chassis and the sensing module. The computing module has a built-in risk prediction model and uses a weighted fusion algorithm to perform fusion analysis on the multi-dimensional data collected by the sensing module in order to identify risk types and assess risk levels. A communication module, which is communicatively connected to the computing module, is used to enable data interaction with external devices; An execution module, which is communicatively connected to the computing module, is used to perform risk warning and / or preliminary handling operations according to the instructions of the computing module; as well as A power module is used to provide electrical energy to the vehicle.

2. The laboratory internal risk prediction dolly of claim 1, wherein, The multi-sensor array of the sensing module includes one or more of the following: lidar, industrial camera, infrared thermal imager, temperature and humidity sensor, gas sensor array, vibration sensor, and RFID reader.

3. The laboratory internal risk prediction dolly of claim 1, wherein, The calculation module uses a precision-reliability joint assignment method to determine the fusion weight of each sensor in the weighted fusion algorithm. Specifically, this includes: calculating the basic weight using the reciprocal of the sensor precision error; correcting the basic weight using a reliability attenuation coefficient to obtain the corrected weight; and normalizing the corrected weight to obtain the final fusion weight.

4. The laboratory internal risk prediction dolly of claim 1, wherein, The risk prediction model built into the computing module includes an LSTM-CNN fusion neural network model, which adopts a five-layer architecture, namely: feature input layer, LSTM temporal feature extraction layer, CNN spatial feature extraction layer, fusion layer and output layer. The feature input layer is used to receive multi-source monitoring data of equipment vibration and temperature; the LSTM temporal feature extraction layer is used to capture the temporal dependencies of the data; the CNN spatial feature extraction layer is used to extract local anomaly features of the data; the fusion layer is used to fuse temporal features and spatial features; and the output layer is used to output the equipment fault type, occurrence probability and warning time.

5. The laboratory internal risk prediction vehicle according to claim 1, characterized in that, The computing module also integrates a personnel behavior recognition model, which is used to identify dangerous behaviors of laboratory personnel based on image or video data collected by the perception module. The dangerous behaviors include at least not wearing protective equipment and using open flames in violation of regulations.

6. The laboratory internal risk prediction vehicle according to claim 1, characterized in that, The execution module includes one or more of an audible and visual alarm unit, a small robotic arm, and a micro-spray unit; the audible and visual alarm unit is used to issue audible and visual alarms at different frequencies according to different risk levels; the small robotic arm is used to perform operations such as closing valves, removing hazardous materials, or turning off equipment power; the micro-spray unit is used to perform preliminary cooling measures for fires or equipment overheating.

7. The laboratory internal risk prediction vehicle according to claim 1, characterized in that, The power module includes a lithium battery pack and a charging management unit. The power module is used to monitor the power level in real time and send a low power signal to the computing module when the power level is lower than a first preset threshold. The computing module is used to plan a return charging path based on the low power signal, control the mobile chassis to move autonomously to the charging base station for charging, and control the vehicle to return to the inspection interruption position after the power level is charged to a second preset threshold.

8. A risk prediction method based on a laboratory risk prediction vehicle according to any one of claims 1 to 7, characterized in that, Includes the following steps: S1 map building and path planning: control the movement of the vehicle and collect laboratory environmental data, build a laboratory map and plan inspection routes; S2 multi-dimensional data acquisition controls the trolley to move along the inspection route and collects environmental parameters, equipment operating parameters, personnel behavior data and obstacle information in real time through the sensing module; S3 data fusion and risk prediction uses a weighted fusion algorithm to fuse multi-dimensional data and calls a pre-trained risk prediction model to identify risk types and assess risk levels. S4 Risk Response and Handling: Based on the assessed risk level, the control execution module performs corresponding early warning and / or preliminary handling operations. S5 data storage and model iteration transmits the collected data, risk prediction results, and handling records to the cloud platform for subsequent iterative optimization of the risk prediction model.

9. The risk prediction method for the laboratory internal risk prediction vehicle according to claim 8, characterized in that, The risk levels assessed in step S3 include low risk, medium risk, and high risk; If a person at low risk is not wearing protective equipment or if environmental parameters deviate slightly from the standard, the procedure is to record the data and send a reminder message. The medium-risk level corresponds to a toxic gas concentration approaching the threshold or a slow increase in equipment temperature. The handling method is to activate an audible and visual alarm and push a warning message. The high-risk situations correspond to toxic gas leaks, equipment overheating, or small fires. The response measures include continuous audible and visual alarms, emergency warning push notifications, and control of the execution module for initial handling, while simultaneously linking with the laboratory fire protection or emergency system.

10. The risk prediction method for the laboratory internal risk prediction vehicle according to claim 8, characterized in that, It also includes step S6: automatic recharging, which monitors the power of the power module in real time. When the power is lower than a preset threshold, the vehicle is controlled to move autonomously to the charging base station to complete automatic charging, and then returns to the inspection interruption position after charging is completed.