Laboratory detection equipment intelligent management system based on edge calculation and reinforcement learning

Abstract

The invention discloses a laboratory detection equipment intelligent management system based on edge calculation and reinforcement learning, and relates to the technical field of equipment intelligent management, the system comprises a multi-dimensional data acquisition module, a model training and feature library construction module, a deviation degree research and judgment module, a model optimization module and an interaction and execution module; according to the invention, through integration of the multi-dimensional data acquisition module and the model training and feature library construction module, all-directional precision monitoring of the operation state of laboratory detection equipment is realized, and various feature data in the operation process of the equipment can be captured in real time through the multi-dimensional sensor array; the model training and feature library construction module uses a deep learning model of a CNN-LSTM mixed structure, combines historical operation data and fault cases of equipment, calculates each feature weight, and dynamically constructs and updates a health state feature vector library of the equipment, thereby improving the accuracy and timeliness of equipment state evaluation, and improving the reliability of equipment state evaluation. Therefore, managers can find potential faults in advance and take preventive maintenance measures.

Description

Technical Field

[0001] This invention relates to the field of intelligent equipment management technology, specifically to an intelligent management system for laboratory testing equipment based on edge computing and reinforcement learning. Background Technology

[0002] With the rapid development of laboratory testing equipment towards high precision, automation, and intelligence, the complexity of equipment operation and the requirements for continuity have significantly increased. In modern laboratories, testing equipment typically needs to operate stably for extended periods to support the continuity of experimental procedures. Its operating status directly affects the reliability of experimental data and research efficiency. However, during operation, equipment may experience performance degradation due to mechanical wear, electrical faults, or environmental factors, or even lead to unplanned shutdowns, resulting in experimental interruptions and data loss. Therefore, developing an intelligent management system for laboratory testing equipment has become a key requirement for improving the reliability of laboratory equipment operation.

[0003] However, traditional laboratory equipment management methods, which mainly rely on regular manual inspections and fixed-cycle maintenance, have significant limitations: their monitoring dimensions are singular, typically focusing only on a few parameters such as temperature and current, making it difficult to comprehensively capture latent fault characteristics such as abnormal vibrations and noise changes during equipment operation, leading to inaccurate fault location; data utilization efficiency is low, with historical equipment operating data and fault cases stored in a scattered manner and not systematically analyzed, failing to form a knowledge base that can be used for predictive maintenance, resulting in the recurrence of similar faults; at the same time, maintenance strategies lack dynamic adaptability, and fixed-cycle maintenance plans cannot respond to changes in the actual state of the equipment, potentially increasing costs due to over-maintenance or accelerating equipment performance degradation due to under-maintenance; in addition, the real-time nature of manual inspections is insufficient, making it difficult to detect transient anomalies during equipment operation, increasing the risk of potential faults developing into serious faults, ultimately affecting the continuous operation capability of laboratory testing equipment and the reliability of experimental data. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an intelligent management system for laboratory testing equipment based on edge computing and reinforcement learning. This invention integrates a multi-dimensional data acquisition module and a model training and feature library construction module to achieve comprehensive and accurate monitoring of the operating status of laboratory testing equipment. Through a multi-dimensional sensor array, various characteristic data during equipment operation can be captured in real time. The model training and feature library construction module utilizes a CNN-LSTM hybrid deep learning model, combined with historical equipment operating data and fault cases, to calculate the weights of each feature, dynamically constructing and updating the equipment's health status feature vector library. This improves the accuracy and timeliness of equipment status assessment, enabling managers to detect potential faults in advance, take preventative maintenance measures, effectively reduce unplanned downtime, and improve the overall operational efficiency of the laboratory.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: an intelligent management system for laboratory testing equipment based on edge computing and reinforcement learning, comprising:

[0006] Multi-dimensional data acquisition module: Based on the structure and operating principle of laboratory testing equipment, a multi-dimensional sensor array is arranged to collect multi-dimensional feature data and integrate it into a standardized multi-dimensional dataset for transmission;

[0007] Model training and feature library construction module: Receives standardized multi-dimensional feature dataset, calculates the weight of each feature based on the equipment's historical operating data and fault case data, and then extracts features using a deep learning model with a CNN-LSTM hybrid structure, taking the equipment's factory standard parameters as the initial baseline, and iteratively trains the model on a periodic basis to build a health status feature vector library for the equipment.

[0008] Deviation assessment module: calls the health status feature vector library, receives a standardized multi-dimensional dataset, calculates the deviation, analyzes the deviation data from historical health operation status to obtain the mean and standard deviation, sets three threshold levels, generates response instructions based on the threshold range of the deviation, and transmits them.

[0009] Model optimization module: Based on standardized multi-dimensional datasets, health status feature vector library and threshold judgment results, a structured database is established for classification and storage, and health baseline parameters are optimized. After optimization, the data is sent back to the model training and feature library construction module.

[0010] Interaction and Execution Module: Receives response commands, displays relevant device information on the terminal through visualization processing, and executes corresponding operations based on the type of response command.

[0011] Furthermore, in the multi-dimensional data acquisition module, the multi-dimensional sensor array includes a current sensor, a vibration sensor, a temperature sensor, and a noise sensor;

[0012] The current sensor collects current characteristic data during equipment operation, including instantaneous current peak value, average current value and current fluctuation amplitude data, with a collection frequency of 1-100Hz.

[0013] The vibration sensor collects vibration characteristic data during equipment operation, including vibration acceleration, vibration displacement, and vibration frequency spectrum data, with a collection frequency of 1-100Hz.

[0014] The temperature sensor collects temperature characteristic data at each monitoring point during equipment operation, including real-time temperature, temperature difference between adjacent points, and temperature change rate data, with a collection frequency of 1-100Hz.

[0015] The noise sensor collects noise characteristic data during equipment operation, including instantaneous noise value, average noise value, and noise frequency distribution data, with a collection frequency of 1-100Hz.

[0016] After each sensor collects multi-dimensional feature data, the analog signals of the multi-dimensional feature data are converted into standardized digital signals and integrated to form a standardized multi-dimensional feature dataset.

[0017] Furthermore, the intelligent management system for laboratory testing equipment based on edge computing and reinforcement learning according to claim 1 is characterized in that, in the model training and feature library construction module, a standardized multi-dimensional feature dataset is received, and based on the equipment's historical operating data and fault case data, the weights of each feature are calculated through the core feature weight allocation formula to quantify the differences in the impact of four types of features—current, vibration, temperature, and noise—on the equipment's health status. Then, feature extraction and baseline learning are completed through a deep learning model with a CNN-LSTM hybrid structure, and the equipment's factory standard parameters are used as the initial baseline for model training.

[0018] The CNN module of the deep learning model outputs spatial sequence features. The CNN module includes an input layer, a convolutional layer, an activation layer, and a pooling layer. The input layer receives standardized multi-dimensional feature data after fusion weights. The convolutional layer extracts local spatial correlation features of different sensor parameters in the feature data. The activation layer performs nonlinear transformation on the feature data. The pooling layer performs downsampling to retain key features and reduce data dimensionality and computational cost.

[0019] The LSTM module of the deep learning model outputs time-series features. The LSTM module includes an input gate, a forget gate, an output gate, and a fully connected layer: the input gate receives the spatial sequence features output by the CNN module and filters the effective feature information at the current time; the forget gate controls the retention and discard of historical feature information based on the importance of historical features; the output gate extracts the key time-series features at the current time; and the fully connected layer converts the time-series features output by the LSTM module into a fixed-dimensional feature vector.

[0020] Each training iteration combines newly added real-time data with updated fault cases to optimize model parameters; during training, extracted spatial and temporal features are integrated to generate a set of feature vectors for the equipment operation phase, and a health status feature vector library for the equipment is constructed and dynamically updated.

[0021] Furthermore, in the model training and feature library construction module, the weights of each feature are calculated using the core feature weight allocation formula, which is as follows: ,in, For the first Weights of class features For the first Influence coefficient of class features For equipment historical faults and the first Number of failures related to class characteristics The target feature index is obtained based on the equipment's historical operating data and fault case data. For equipment historical faults and the first Number of failures related to class characteristics For feature category index, For the first Influence coefficient of class features.

[0022] Furthermore, the deviation assessment module calls the health status feature vector library to extract the health baseline value, receives a standardized multi-dimensional dataset, converts it into standardized values, and calculates the deviation using the feature deviation calculation formula to quantify the degree of deviation between the device's operating characteristics and the health baseline. The feature deviation calculation formula is as follows: ,in, For deviation, For the first Weights of class features for Time of the first Standardized values ​​of class features for Time of the first Health baseline value of class characteristics.

[0023] Furthermore, the deviation assessment module analyzes and obtains the mean value using deviation data from historical healthy operating states. and standard deviation The three-level thresholds are set using a dynamic three-level threshold setting formula, which is as follows: ,in, The threshold for minor fluctuations and abnormal warnings. This is the critical threshold for abnormal warnings and fault alarms. This is the upper limit threshold for fault alarms.

[0024] Furthermore, in the deviation judgment module, based on the threshold range in which the deviation is determined, a corresponding response command is triggered: when When a slight fluctuation is detected, a trend recording instruction is generated; when When an abnormality is detected, a preventative inspection reminder is generated; when If the alarm is detected, a fault alarm command is generated.

[0025] Furthermore, in the model optimization module, a structured database is established for classified storage based on a standardized multi-dimensional dataset, a health status feature vector library, and threshold judgment results. The data is then verified for integrity and integrated to form a full lifecycle operation archive for the equipment. The health baseline parameters are then optimized through model iterative optimization and update formulas. After optimization, the data is sent back to the model training and feature library construction module to update the health status feature vector library.

[0026] Furthermore, in the model optimization module, the healthy baseline parameters are optimized through a model iterative optimization update formula, which is: ,in, For the updated Time of the first Health baseline values ​​for class characteristics for Time of the first Health baseline values ​​for class characteristics For the number of devices, Index for the number of devices. For the first The contribution weight of each device is obtained through historical data. For the first Taiwanese equipment Time of the first Health baseline value of class characteristics.

[0027] Furthermore, the interaction and execution module receives response commands and visualizes the real-time operating status and various characteristic data of the device on the terminal, generating a visualization interface including real-time change curves and health status radar charts. It then executes corresponding operations based on the type of the received response command: when a trend recording command for slight fluctuations is received, the fluctuation data is stored in chronological order, and a trend analysis chart is generated; when a preventative inspection reminder command for an anomaly warning is received, a preventative inspection reminder is pushed, clearly indicating the type of anomaly, the location of the anomaly, the current deviation value, and the suggested inspection process; when a fault alarm command for a fault alarm is received, a fault diagnosis report is displayed on the terminal interface, and a start command is sent to the standby device, triggering the standby device to enter a standby state.

[0028] Compared with existing technologies, this intelligent management system for laboratory testing equipment based on edge computing and reinforcement learning has the following advantages:

[0029] I. This invention achieves comprehensive and accurate monitoring of the operating status of laboratory testing equipment by integrating a multi-dimensional data acquisition module and a model training and feature library construction module. The multi-dimensional sensor array can capture various feature data during equipment operation in real time. The model training and feature library construction module uses a deep learning model with a CNN-LSTM hybrid structure, combined with historical equipment operating data and fault cases, to calculate the weights of each feature and dynamically build and update the equipment's health status feature vector library. This improves the accuracy and timeliness of equipment status assessment, enabling managers to detect potential faults in advance, take preventive maintenance measures, effectively reduce unplanned downtime, and improve the overall operating efficiency of the laboratory.

[0030] Second, this invention constructs an intelligent decision support system for equipment management through a deviation judgment module and a model optimization module. The deviation judgment module calculates the degree of deviation between the real-time operating data of the equipment and the health baseline, and automatically generates response instructions of different levels by combining three-level threshold settings, including trend recording, preventive inspection reminders, and fault alarms. This makes equipment maintenance more accurate and efficient, avoiding over-maintenance and missed inspections. At the same time, the model optimization module iteratively optimizes the health baseline parameters to ensure that the management strategy can adapt to changes in equipment status, improving the scientificity and effectiveness of equipment management. This reduces the overall maintenance cost of laboratory testing equipment, extends the service life of the equipment, and provides a strong guarantee for the long-term stable operation of the laboratory.

[0031] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0033] Figure 1 A flowchart of an intelligent management system for laboratory testing equipment based on edge computing and reinforcement learning;

[0034] Figure 2 This is a framework diagram of an intelligent management system for laboratory testing equipment based on edge computing and reinforcement learning. Detailed Implementation

[0035] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0036] Example 1:

[0037] Multi-dimensional data acquisition module: For the intelligent management scenario of quantitative PCR instruments in a biosafety level 2 laboratory, a multi-dimensional sensor array is deployed in the nucleic acid testing area of ​​the biosafety level 2 laboratory, targeting the core operating units of three quantitative PCR instruments. Specifically, a temperature sensor is installed near the metal bath of the heating unit; a current sensor and a vibration sensor are deployed next to the motor drive unit controlling the sample tray rotation; and a noise sensor is fixed to the inner wall of the equipment side. The current sensor collects instantaneous peak current, average current value, and current fluctuation amplitude data at a frequency of 50Hz when the heating wire is energized, focusing on capturing current changes during the heating phase. The vibration sensor collects vibration acceleration, vibration displacement, and vibration frequency spectrum data of the rotating sample tray at a frequency of 80Hz to monitor motor operational stability. The temperature sensor collects real-time temperature, temperature difference between adjacent points, and temperature change rate data of the metal bath surface, the inner wall of the heating chamber, and the heat dissipation vents at a frequency of 30Hz to ensure temperature control accuracy. The noise sensor collects instantaneous noise value, average noise value, and noise frequency distribution data throughout the entire operation of the equipment at a frequency of 40Hz to identify abnormal friction sounds. The analog signals from each sensor are then converted into standardized digital signals and integrated into a standardized multi-dimensional dataset containing four types of features for transmission.

[0038] Model training and feature library construction module: Receives a standardized multi-dimensional dataset, combined with historical operational data of this batch of PCR instruments since their commissioning. This data covers the entire process parameters of daily testing, including sample loading, heating, annealing, and extension, as well as past fault case data, such as temperature control drift in the heating unit and sample tray stoppage caused by motor jamming. The module calculates the weights of each feature using a core feature weight allocation formula, quantifying the differences in the impact of four types of features on the equipment's health status. The core feature weight allocation formula is as follows: ,in, For the first Weights of class features For the first Influence coefficient of class features For equipment historical faults and the first Number of failures related to class characteristics The target feature index is obtained based on the equipment's historical operating data and fault case data. For equipment historical faults and the first Number of failures related to class characteristics For feature category index, For the first The influence coefficient of class features; and using the heating temperature benchmark value, standard operating current, vibration amplitude, and noise decibel marked at the time of equipment delivery as the initial baseline, inputting them into a deep learning model with a CNN-LSTM hybrid structure, where the CNN module outputs spatial sequence features, receives the dataset after fusion weights through the input layer, the convolutional layer extracts the local spatial correlation features between the heating unit temperature and the drive motor current, the activation layer performs nonlinear transformation, and the pooling layer downsamples to retain key features to reduce computation; the LSTM module outputs time series features, the input gate filters the spatial sequence features output by the CNN, the forget gate controls the retention and discarding of information based on historical data, the output gate extracts key time series features within the detection period, the fully connected layer converts them into fixed-dimensional feature vectors, and periodically combines newly added real-time operating data and updated fault cases to optimize model parameters, and fuses spatial and temporal features to generate a feature vector set for the equipment operation stage, constructing and dynamically updating the health status feature vector library of the PCR instrument in this special area, such as Figure 1 As shown.

[0039] Deviation Assessment Module: After nucleic acid testing is initiated, it retrieves the baseline health value at the corresponding time from the health status feature vector library, receives the real-time standardized multi-dimensional dataset transmitted by the multi-dimensional data acquisition module and converts it into standardized values, and calculates the deviation using the feature deviation calculation formula. The feature deviation calculation formula is as follows: ,in, For deviation, For the first Weights of class features for Time of the first Standardized values ​​of class features for Time of the first The baseline health value of the class characteristics is used; then, based on the deviation data of the PCR instrument under historical healthy operating conditions, the mean and standard deviation are analyzed, and a three-level threshold is set through a three-level threshold dynamic setting formula, which is as follows: ,in, The threshold for minor fluctuations and abnormal warnings. This is the critical threshold for abnormal warnings and fault alarms. The upper limit threshold for fault alarms is set; and the threshold range is determined based on the deviation, triggering the corresponding response command: when When a slight fluctuation is detected, a trend recording instruction is generated; when When an abnormality is detected, a preventative inspection reminder is generated; when If the alarm is detected, a fault alarm command is generated.

[0040] Model optimization module: Based on standardized multi-dimensional datasets, health status feature vector libraries, and threshold assessment results, a structured database is established for classified storage. The integrity of the stored data is verified, and invalid data caused by poor sensor contact is removed. Then, the equipment operating parameters are correlated and integrated with the assessment results to form a full lifecycle operating record for each PCR instrument from its inception to the present. Finally, the health baseline parameters are optimized through model iterative optimization and update formulas. The model iterative optimization and update formulas are as follows: ,in, For the updated Time of the first Health baseline values ​​for class characteristics for Time of the first Health baseline values ​​for class characteristics For the number of devices, Index for the number of devices. For the first The contribution weight of each device is obtained through historical data. For the first Taiwanese equipment Time of the first The baseline health value of the class feature; after optimization, it is sent back to the model training and feature library construction module to update the health status feature vector library.

[0041] Interaction and Execution Module: Upon receiving a response command, the laboratory monitoring terminal immediately performs visualization processing, displaying the real-time heating temperature change curve, motor operating current fluctuation curve, and a health status radar chart including four characteristics: current, vibration, temperature, and noise, on the terminal interface. When a trend recording command is received, the fluctuation data of this test is automatically stored in chronological order, and a deviation trend analysis chart is generated. When a preventive inspection reminder command is received, a reminder is pushed through a pop-up window on the terminal, clearly indicating the type of abnormal feature, the location of the abnormality, the current deviation value, and the suggested inspection procedure of first checking the temperature sensor calibration status and then checking the heating wire wiring. When a fault alarm command is received, the terminal interface immediately displays a detailed fault diagnosis report, and at the same time, a start command is sent to the backup PCR instrument through the equipment control interface, triggering the backup equipment to enter the standby state to ensure the continuous operation of nucleic acid testing.

[0042] In summary, the intelligent management of the real-time PCR instrument in a biosafety level 2 laboratory is achieved by deploying sensors on core units such as heating and motor drives to accurately collect four types of characteristic data, including current and vibration, and transmit them in a standardized manner. Feature weights are calculated by combining historical operating and fault data of the equipment, and spatiotemporal features are extracted using a CNN-LSTM model to dynamically construct a health feature vector library. Based on historical data, three-level thresholds are set to accurately generate different response commands. The data is then integrated to form a full lifecycle archive of the equipment, and baseline parameters are iteratively optimized. The equipment status is visualized, and trend recording, inspection reminders, and emergency fault handling are completed according to commands, achieving intelligent control of the entire PCR instrument process.

[0043] Example 2:

[0044] Multi-dimensional data acquisition module: For intelligent management scenarios of total organic carbon (TOC) analyzers in environmental monitoring laboratories, a multi-dimensional sensor array is deployed for the key operating units of four TOC analyzers in the water quality testing area of ​​an urban environmental monitoring laboratory. Current and vibration sensors are installed at the motor of the sample pump; temperature sensors are deployed both externally and internally within the reactor chamber; and a noise sensor is installed on the top of the equipment. The current sensor collects instantaneous peak current, average current value, and current fluctuation amplitude data at a frequency of 60Hz during sample pump motor operation to monitor load changes. The vibration sensor collects vibration acceleration, vibration displacement, and vibration frequency spectrum data generated by the impeller rotation of the sample pump at a frequency of 70Hz to identify early signs of impeller blockage. The temperature sensor collects real-time temperature, temperature difference between adjacent points, and temperature change rate data of the reactor heating zone, insulation layer, and exhaust port at a frequency of 20Hz to ensure sufficient high-temperature oxidation reaction. The noise sensor collects instantaneous noise value, average noise value, and noise frequency distribution data during equipment operation at a frequency of 35Hz to capture abnormal noises caused by abnormal reactor pressure. The analog signals collected by each sensor are converted into standardized digital signals and integrated to form a standardized multi-dimensional dataset.

[0045] Model training and feature library construction module: Receives a standardized multi-dimensional dataset, combined with historical operating data from the batch of total organic carbon analyzers, including records of equipment parameter changes during the testing of different water quality samples such as surface water and groundwater, as well as fault case data, including sample pump blockage, reactor temperature runaway, and detection circuit short circuit. It calculates the weights of each feature using a core feature weight allocation formula to quantify the differences in the impact of the four types of features on the equipment's health status. The core feature weight allocation formula is as follows: The model is constructed using the standard current of the injection pump, the reference temperature of the reactor, the vibration amplitude, and the noise level set at the time of manufacture as initial baselines. A deep learning model with a CNN-LSTM hybrid structure is input. The CNN module outputs spatial sequence features, the input layer receives the dataset after weight fusion, the convolutional layer extracts the local spatial correlation features between the injection pump current and vibration parameters, the activation layer performs nonlinear transformation, and the pooling layer downsamples and retains key features. The LSTM module outputs time-series features, the input gate filters the spatial sequence features output by the CNN, the forget gate controls the retention and discarding of information based on the importance of historical features, the output gate extracts key time-series features within the detection period, and the fully connected layer converts the time-series features into fixed-dimensional feature vectors. The model parameters are periodically optimized by combining newly added real-time operating data and updated fault cases. The spatial and temporal features are fused to generate a feature vector set for the equipment's operating phase, constructing and dynamically updating a health status feature vector library for the total organic carbon analyzer in this region.

[0046] Deviation assessment module: When detecting total organic carbon in surface water samples, it calls the health status feature vector library to extract the health baseline value at the corresponding time, receives the real-time transmitted standardized multi-dimensional dataset and converts it into standardized values, and calculates the deviation using the feature deviation calculation formula, which is as follows: Based on the deviation data from the historical healthy operating conditions of the total organic carbon analyzer, the mean and standard deviation are analyzed. A three-level threshold dynamic setting formula is then used to set the three-level thresholds. The three-level threshold dynamic setting formula is as follows: And based on the deviation, determine the threshold range and trigger the corresponding response command: when When a slight fluctuation is detected, a trend recording instruction is generated; when When an abnormality is detected, a preventative inspection reminder is generated; when If the alarm is detected, a fault alarm command is generated.

[0047] Model optimization module: Based on standardized multi-dimensional datasets, health status feature vector libraries, and threshold assessment results, a structured database is established for classified storage. Data integrity is verified, and invalid data caused by sample contamination of sensors is removed. Then, equipment operating parameters are correlated and integrated with the assessment results to form a full lifecycle operating record for each analyzer. Finally, the health baseline parameters are optimized through model iterative optimization and update formulas. The model iterative optimization and update formulas are as follows: After optimization, the data is sent back to the model training and feature library construction module to update the health status feature vector library, such as... Figure 2 As shown.

[0048] Interaction and Execution Module: After receiving response commands, the laboratory data monitoring terminal performs visualization processing, displaying the analyzer's real-time injection pump current curve, reactor temperature change curve, and health status radar chart. When a trend recording command is received, it stores fluctuation data in chronological order and generates a deviation trend chart. When a preventative inspection reminder command is received, a pop-up reminder is pushed to the terminal, indicating the type of abnormality, the location of the abnormality, the current deviation value, and a suggested inspection procedure for disassembling the injection pump to check for impeller blockage and replacing worn seals. When a fault alarm command is received, the terminal interface displays a fault diagnosis report and simultaneously sends a start command to the backup total organic carbon analyzer, triggering the backup equipment to start the preheating program and enter standby mode to avoid interruption of the detection work.

[0049] In summary, an intelligent management solution was developed for the total organic carbon analyzer in the environmental monitoring laboratory. This solution involves deploying sensors on key components such as the sample pump and reactor to collect four types of characteristic data and transmit them in a standardized manner. Then, feature weights are calculated based on historical equipment data and failure cases. Spatiotemporal features are extracted using a CNN-LSTM model, and the health feature vector library is dynamically updated. Thresholds are set based on historical deviation data to generate corresponding response commands. The data is then integrated to form a full lifecycle archive of the equipment, baseline parameters are optimized and updated, and finally, the equipment status is visualized. The system performs trend analysis, inspection reminders, and backup equipment activation according to commands, ensuring continuous and stable water quality testing.

[0050] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. Laboratory detection equipment intelligent management system based on edge computing and reinforcement learning, characterized in that, The system comprises: A multi-dimensional data acquisition module: according to the structure and operating principle of the laboratory detection equipment, a multi-dimensional sensor array is arranged, multi-dimensional characteristic data is collected, and a standardized multi-dimensional data set is formed for transmission; A model training and feature library construction module: receives the standardized multi-dimensional characteristic data set, calculates the feature weight based on the equipment historical operation data and fault case data, extracts the features and iteratively trains them according to the period, and constructs the health state feature vector library of the equipment; A deviation judgment module: calls the health state feature vector library, receives the standardized multi-dimensional data set, calculates the deviation, analyzes the mean and standard deviation based on the deviation data under the historical health running state, sets three threshold values, generates a response instruction according to the threshold interval where the deviation is located, and transmits it; A model optimization module: based on the standardized multi-dimensional data set, the health state feature vector library and the threshold judgment result, a structured database is established for classified storage, and the health baseline parameters are optimized, after optimization, it is returned to the model training and feature library construction module; An interactive and execution module: receives the response instruction, displays the equipment related information on the terminal through visual processing, and executes the corresponding operation according to the response instruction type. 2.The edge computing and reinforcement learning based intelligent management system for laboratory detection devices according to claim 1, wherein, In the multi-dimensional data acquisition module, the multi-dimensional sensor array includes current sensor, vibration sensor, temperature sensor and noise sensor; The current sensor: collects current characteristic data during equipment operation, including instantaneous current peak value, average current value and current fluctuation amplitude data, with a collection frequency of 1-100Hz; The vibration sensor: collects vibration characteristic data during equipment operation, including vibration acceleration, vibration displacement and vibration frequency spectrum data, with a collection frequency of 1-100Hz; The temperature sensor: collects temperature characteristic data of each monitoring point during equipment operation, including real-time temperature, adjacent point temperature difference and temperature change rate data, with a collection frequency of 1-100Hz; The noise sensor: collects noise characteristic data during equipment operation, including instantaneous noise value, average noise value and noise frequency distribution data, with a collection frequency of 1-100Hz; After each sensor collects multi-dimensional characteristic data, the analog signal of the multi-dimensional characteristic data is converted into standardized digital signal, and the standardized multi-dimensional characteristic data set is formed. 3.The edge computing and reinforcement learning based intelligent management system for laboratory detection devices according to claim 1, wherein, In the model training and feature library construction module, the standardized multi-dimensional characteristic data set is received, and the feature weight is calculated based on the equipment historical operation data and fault case data through the core feature weight distribution formula, the influence difference of current, vibration, temperature and noise four types of features on the equipment health state is quantified, and the feature extraction and baseline learning are completed through the CNN-LSTM hybrid structure deep learning model, and the equipment standard parameters are used as the initial baseline for model training; The CNN module of the deep learning model outputs spatial sequence features, and the CNN module includes an input layer, a convolution layer, an activation layer, and a pooling layer: the input layer receives the standardized multi-dimensional feature data after fusion weight, the convolution layer extracts the local spatial correlation features of different sensor parameters in the feature data, the activation layer performs nonlinear transformation on the feature data, and the pooling layer performs down-sampling to retain key features and reduce data dimensions and computational complexity; The LSTM module of the deep learning model outputs time sequence features, and the LSTM module includes an input gate, a forgetting gate, an output gate, and a fully connected layer: the input gate receives the spatial sequence features output by the CNN module to filter the effective feature information at the current time; the forgetting gate controls the retention and discard of historical feature information according to the importance of historical features; the output gate extracts key time sequence features at the current time; and the fully connected layer converts the time sequence features output by the LSTM module into a fixed-dimensional feature vector. Each training combines new real-time data and updated fault cases to optimize model parameters; the spatial features and time sequence features extracted during training are fused to generate a set of device running phase feature vectors, and a health state feature vector library of the device is constructed and dynamically updated. 4.The edge computing and reinforcement learning based intelligent management system for laboratory detection devices according to claim 3, wherein, In the model training and feature library construction module, the weights of each feature are calculated using the core feature weight allocation formula, which is as follows: ,in, For the first Weights of class features For the first Influence coefficient of class features For equipment historical faults and the first Number of failures related to class characteristics The target feature index is obtained based on the equipment's historical operating data and fault case data. For equipment historical faults and the first Number of failures related to class characteristics For feature category index, For the first Influence coefficient of class features. 5.The edge computing and reinforcement learning based intelligent management system for laboratory detection devices according to claim 1, wherein, In the deviation degree judgment module, the health state feature vector library is called, the health baseline value is extracted, and the standardized multidimensional data set is received, which is converted into a standardized value, the deviation degree is calculated through a feature deviation degree calculation formula, and the deviation degree of the equipment operation feature from the health baseline is quantified, and the feature deviation degree calculation formula is: wherein, is the deviation degree, is the weight of the i-th feature, is the standardized value of the i-th feature at the t-th moment, is the health baseline value of the i-th feature at the t-th moment. ​​​​​ 6.The edge computing and reinforcement learning based intelligent management system for laboratory detection devices according to claim 1, wherein, In the deviation degree judgment module, the mean value and the standard deviation are obtained by analyzing the deviation degree data in the historical health running state , and the three-level threshold is set by a three-level threshold dynamic setting formula , which is: , wherein is a critical threshold for slight fluctuation and abnormal early warning, is a critical threshold for abnormal early warning and fault alarm, is an upper threshold for fault alarm.

7. The edge computing and reinforcement learning based intelligent management system for laboratory detection devices according to claim 6, wherein, In the deviation degree judgment module, the threshold interval is determined according to the deviation degree, and the corresponding response instruction is triggered: when the slight fluctuation is determined, the trend record instruction is generated; when the abnormal early warning is determined, the preventive inspection reminding instruction is generated; and when the fault alarm is determined, the fault alarm instruction is generated. 8.The edge computing and reinforcement learning based intelligent management system for laboratory detection devices according to claim 1, wherein, In the model optimization module, based on the standardized multi-dimensional data set, the health state feature vector library, and the threshold judgment result, a structured database is established for classified storage, and the data is subjected to integrity verification and associated integration to form a device full life cycle operation file; then the health baseline parameters are optimized by model iteration optimization update formula, and after optimization, the health state feature vector library is updated by returning to the model training and feature library construction module. 9.The edge computing and reinforcement learning based intelligent management system for laboratory detection devices according to claim 8, wherein, In the model optimization module, the healthy baseline parameters are optimized through a model iterative optimization update formula, which is: ,in, For the updated Time of the first Health baseline values ​​for class characteristics for Time of the first Health baseline values ​​for class characteristics For the number of devices, Index for the number of devices. For the first The contribution weight of each device is obtained through historical data. For the first Taiwanese equipment Time of the first Health baseline value of class characteristics. 10.The edge computing and reinforcement learning based intelligent management system for laboratory detection devices according to claim 1, wherein, In the interaction and execution module, the device real-time running state and each feature data are visualized on the terminal to generate a visual interface including real-time change curve and health state radar chart, and corresponding operations are performed according to the received response instruction type: when a slight fluctuation trend record instruction is received, the fluctuation data is stored in chronological order and a trend analysis chart is generated; When an abnormal early warning preventive inspection reminder instruction is received, a preventive inspection reminder is pushed, and the reminder content clearly marks the abnormal feature type, abnormal occurrence position, current deviation value, and recommended inspection process; when a fault alarm fault alarm instruction is received, a fault diagnosis report is displayed on the terminal interface, and a start instruction is sent to the standby device to trigger the standby device to enter a standby running state.

Images