Equipment state data management method and system applied to automatic radix astragali seu hedysari refined production line

Abstract

The invention relates to the technical field of data analysis, provides an equipment state data management method and system applied to an automatic radix astragali seu hedysari refined production line, and is used for realizing equipment health precise prediction and active maintenance through whole-process data monitoring management. The method comprises the steps that a production line equipment state data set is acquired, a production line equipment state feature set is extracted from the production line equipment state data set, and the production line equipment state feature set comprises multi-dimensional time sequence features reflecting the equipment operation trend and topological features of the incidence relation between the equipment; training an astragalus membranaceus essence production line state prediction model based on the production line equipment state feature set, wherein the astragalus membranaceus essence production line state prediction model is used for predicting the equipment abnormal probability and the performance degradation trend in a selected time period according to the current equipment state feature; and generating an equipment management strategy set according to an output result of the radix astragali seu hedysari refined production line state prediction model, wherein the equipment management strategy set comprises production parameter adjustment instructions and maintenance node planning schemes for different equipment maintenance priorities.

Description

Technical Field

[0001] The present application belongs to the field of data analysis technology, and specifically relates to a method and system for managing equipment status data applied to an automated astragalus extract production line. Background Art

[0002] In the automated production of traditional Chinese medicine preparations, equipment status monitoring and maintenance decision-making technologies play a critical role in ensuring product quality and production line stability. The equipment health management systems currently in widespread use in the industry primarily rely on single-dimensional real-time data collection, such as obtaining equipment operating parameters through vibration sensors or temperature detection devices, and generating abnormality alarms based on preset thresholds. However, existing solutions often employ independent monitoring of individual equipment, failing to effectively integrate the synergistic parameters between production line equipment. This is particularly true in complex preparation processes, where the dynamic interactions between equipment are often simplified as linear superposition relationships. Furthermore, existing maintenance strategy generation systems often utilize a fixed rule-based decision-making mechanism, failing to dynamically correlate predictions with production process parameters. When faced with raw material batch variations or fluctuating environmental parameters, existing systems struggle to generate adaptive maintenance plans that balance equipment health and production quality requirements. Especially in temperature-sensitive, narrow process parameter windows, such as in the production of astragalus extract, existing maintenance decisions often create a conflict between equipment downtime and product quality control.

[0003] Therefore, the above-mentioned technical defects lead to the common problems of existing technologies in the continuous production scenarios of Astragalus extract preparations, such as insufficient timeliness of equipment abnormality warning, low accuracy of performance degradation prediction, poor adaptability of maintenance strategies and production processes, making it difficult to achieve accurate prediction of equipment health and proactive maintenance. Summary of the Invention

[0004] The present application provides an equipment status data management method and system for an automated Astragalus extract production line, which is used to achieve accurate prediction and proactive maintenance of equipment health through full-process data monitoring and management.

[0005] In the first aspect, an embodiment of the present application provides an equipment status data management method applied to an automated astragalus extract production line, which is applied to an equipment status data management system, the method comprising: obtaining a production line equipment status data set, the production line equipment status data set comprising production parameters, operating status parameters and environmental monitoring parameters of each equipment in multiple continuous production cycles in the automated astragalus extract production line; extracting a production line equipment status feature set from the production line equipment status data set, the production line equipment status feature set comprising multi-dimensional time series features reflecting equipment operation trends and topological features of association relationships between equipment; training an astragalus extract production line status prediction model based on the production line equipment status feature set, the astragalus extract production line status prediction model being used to predict the probability of equipment abnormality and performance degradation trend within a selected time period based on current equipment status characteristics; generating an equipment management policy set based on the output results of the astragalus extract production line status prediction model, the equipment management policy set comprising production parameter adjustment instructions and maintenance node planning schemes for different equipment maintenance priorities.

[0006] In a second aspect, an embodiment of the present application provides a device status data management system, which includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the above method.

[0007] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, which includes a computer program. When the computer program is run on a device status data management system, the computer program is used to enable the device status data management system to execute the steps of the above method.

[0008] During the implementation of this application, accurate prediction of equipment health and proactive maintenance are achieved through closed-loop management of full-process data. In detail, a comprehensive status portrait is constructed based on multi-source equipment parameters and environmental monitoring data to support multi-dimensional feature extraction. In addition, by integrating the temporal characteristics and associated topological characteristics of equipment operation trends, the dynamic evolution law and synergistic mechanism of the equipment are revealed, and the model learning efficiency is enhanced. Furthermore, the trained astragalus extract production line status prediction model can simultaneously identify equipment abnormality risks and performance degradation trends, and achieve early warning of faults and prediction of degradation paths. Finally, based on the generated differentiated equipment management strategy set, process parameters and intelligent planning maintenance nodes can be dynamically adjusted to reduce the risk of unplanned downtime and optimize maintenance resource allocation. The embodiment of the present application forms a closed-loop control from data perception to decision execution, improves the stability of production line operation and the full life cycle management capabilities of equipment, and ensures the continuity and quality consistency of astragalus extract production. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1A flow chart of a method for managing equipment status data applied to an automated astragalus extract production line provided in an embodiment of the present application.

[0010] Figure 2 A schematic diagram of the structure of a device status data management system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0011] In order to make the purpose, technical solutions and advantages of the embodiments of this application clearer, the technical solutions of this application will be described clearly and completely below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of the technical solutions of this application, but not all of them. Based on the embodiments recorded in this application document, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the technical solutions of this application. See Figure 1 , which is a device status data management method applied to an automated astragalus extract production line provided in an embodiment of the present application. This method can be applied to a device status data management system, and the specific process is as shown in steps 110 to 140.

[0012] Step 110: Acquire a production line equipment status data set, wherein the production line equipment status data set includes production parameters, operating status parameters, and environmental monitoring parameters of each device in multiple continuous production cycles in the automated astragalus extract production line.

[0013] In an embodiment of the present application, the equipment status data management system collects and stores multi-source heterogeneous data in real time through a sensor network deployed at each key node of the automated astragalus extract production line. For example, in the astragalus raw material extraction section, the temperature sensor records the temperature change curve of the liquid medium in the extraction tank at a frequency of once per minute, the pressure transmitter continuously monitors the steam pressure value of the tank jacket, and the flow meter accumulates and counts the injection volume of ethanol solvent per hour. In the centrifugal separation equipment, the vibration sensor captures the three-axial vibration waveform of the drum bearing at a sampling rate of 500Hz, and the current transformer synchronously collects the three-phase working current of the drive motor.

[0014] In addition, the deployed environmental monitoring module continuously records and extracts the air temperature and humidity distribution data in different areas of the workshop through distributed temperature and humidity probes, and measures the concentration of suspended particulates in the operating area through a laser particle counter. All data can be structured and stored according to the production batch number, forming a complete data set containing the equipment serial number, timestamp, parameter type and original value, covering the operation archives of all equipment in multiple consecutive production cycles. For example, in the third production cycle, the equipment status data management system fully recorded the dynamic process of the outlet pressure of the vacuum pump in the concentration section gradually increasing from the initial value of -0.095MPa to -0.088MPa during the continuous operation of 72 hours, and at the same time associated and stored the inlet and outlet water temperature difference data of the cooling water circulation system during this period.

[0015] Step 120: extracting a production line equipment status feature set from the production line equipment status data set, wherein the production line equipment status feature set includes multi-dimensional time series features reflecting equipment operation trends and topological features of association relationships between equipment.

[0016] In the embodiments of this application, the equipment status data management system uses feature engineering algorithms to deeply analyze raw data. For example, for the temperature time series data of the extraction tank, the equipment status data management system uses a sliding window Fourier transform to extract the frequency domain features of each 15-minute period, and simultaneously calculates the standard deviation between adjacent windows as an indicator of the intensity of temperature fluctuations. For the vibration signal of the centrifuge, wavelet packet decomposition is used to extract the energy proportion characteristics of specific frequency bands and generate a histogram of the vibration energy distribution along the X / Y / Z axes.

[0017] It can be understood that at the level of equipment association analysis, the equipment status data management system establishes an equipment topology diagram based on the material transmission path, calculates the cross-correlation function between the opening time of the extraction tank outlet valve and the speed of the concentration tank feed pump, and quantifies the linkage strength between the two. For example, during the feature extraction process, when the agitator speed of the alcohol precipitation process is increased to 120rpm, the pressure difference growth rate of the downstream tubular filter shows a significant nonlinear relationship with the stirring duration. This feature is marked as a key association parameter and included in the feature set. Furthermore, the equipment status data management system can also align the key event timestamps of multiple devices through a dynamic time warping algorithm and generate a state transition matrix across devices.

[0018] Step 130: training an astragalus extract production line state prediction model based on the production line equipment state feature set, wherein the astragalus extract production line state prediction model is used to predict the equipment abnormality probability and performance degradation trend within a selected time period according to the current equipment state features.

[0019] In an embodiment of the present application, the device status data management system can construct a dual-channel prediction model comprising a long-short-term memory network and a graph convolutional network as a status prediction model for the Astragalus Extract production line. The temporal feature channel uses a three-layer LSTM structure to process the temperature fluctuation sequence of the extraction tank, with each layer containing 128 memory cells, to learn the time domain propagation patterns of temperature anomaly patterns. The topological feature channel uses a graph convolutional network to embed the device association matrix and capture potential patterns of coordinated failure of multiple devices.

[0020] For example, during the training process of the Astragalus Extract production line status prediction model, the equipment status data management system used the data from the 1st to 30th production cycles as the training set and the data from the 31st to 36th cycles as the validation set, using the early stopping method to prevent overfitting. After 150 rounds of iterative training, the Astragalus Extract production line status prediction model can accurately predict the remaining life of centrifuge bearings. For example, when the high-frequency energy ratio of the vibration signal exceeds the threshold baseline for three consecutive periods, the Astragalus Extract production line status prediction model determines that the probability of the bearing becoming stuck in the subsequent 48 hours of operation has increased to a high-risk level. For the vacuum degree indicator of the concentration tank, the Astragalus Extract production line status prediction model can predict the extent of performance degradation in the next 72 hours based on the current downward slope, providing a decision-making basis for preventive maintenance.

[0021] Step 140: Generate an equipment management strategy set according to the output result of the Astragalus Extract production line status prediction model, wherein the equipment management strategy set includes production parameter adjustment instructions and maintenance node planning solutions for different equipment maintenance priorities.

[0022] In an embodiment of the present application, the equipment status data management system generates a hierarchical response strategy based on the output results of the astragalus extract production line status prediction model. For example, when it is predicted that there is a risk of abnormal fluctuation in the jacket pressure of the extraction tank, the equipment status data management system automatically generates a temperature control parameter optimization plan: it is recommended to adjust the heating rate of the third stage from 2°C per minute to 1.5°C, and extend the insulation time by 15 minutes. For highly correlated equipment groups, such as the centrifuge and concentration tank linkage system, the equipment status data management system comprehensively formulates maintenance windows, plans to give priority to replacing centrifuge seals during regular batch intervals, and arranges vacuum detection operations for the concentration tank at the same time.

[0023] For example, if the Astragalus Extract production line's status prediction model identifies critical risk equipment, such as a vibrating screen spring support, with structural fatigue warnings, the equipment status data management system immediately generates a red alert, triggering the production line to slow down and dynamically rescheduling production tasks from the relevant sections to backup equipment. All strategies are virtually verified using the digital twin system before being issued to the on-site control system for execution, ensuring that policy adjustments do not affect the overall production rhythm.

[0024] It can be seen that the device status data management method for the automated Astragalus Extract production line provided in the embodiment of the present application realizes intelligent monitoring and predictive maintenance of the entire process. To further understand the technical solution described in the above embodiment of the present application, the following is an introduction and explanation through a complete application scenario example.

[0025] Taking the typical application of the Huangqijing oral liquid production line of a pharmaceutical company as an example, during the 36th production cycle of the Huangqijing oral liquid production line of a pharmaceutical company, the equipment status data management system collected multi-source heterogeneous data in real time through the sensor network deployed in the three major sections of extraction, separation, and concentration. In the raw material extraction stage, the temperature sensor records the temperature curve of the liquid medium in the extraction tank with minute-level accuracy. When it was detected that the fifth batch of materials had an abnormal fluctuation of ±1.2°C in the constant temperature stage of 75°C, the equipment status data management system synchronously retrieved the jacket steam pressure data and found that the pressure value dropped sharply from 0.35MPa to 0.28MPa in the corresponding period. Combined with the hourly injection volume recorded by the ethanol solvent flow meter, the deviation was 8.7%, which constituted the characteristics of multi-dimensional data abnormal events. At the same time, the vibration sensor in the centrifugal section captured the Z-axis vibration acceleration of the drum bearing at a sampling rate of 500Hz, which dropped from 4.3m / s in three consecutive batches. 2 Increased to 6.8m / s 2 , the energy proportion of its high frequency band (8-12kHz) exceeded the historical threshold baseline of 15%, triggering in-depth analysis of the feature extraction mechanism.

[0026] In this application scenario, the feature engineering module uses adaptive sliding window technology to perform a joint time-frequency analysis of the extraction tank temperature series. For example, the Fourier transform identifies the 0.05Hz low-frequency interference component implicit in the temperature fluctuations. Combined with the wavelet packet decomposition algorithm, 32 sets of frequency band energy features are extracted from the centrifuge vibration signal. It is found that the energy value of the 16th frequency band (corresponding to the characteristic frequency of bearing retainer failure) is 3.2 times higher than that under normal operating conditions. Furthermore, the equipment association analysis layer, based on the material transfer topology, calculates the Pearson correlation coefficient between the vacuum pump outlet pressure of the concentration tank and the cooling water circulation temperature difference to -0.87, revealing a strong negative correlation between the two. Furthermore, the dynamic time warping algorithm further aligns the time lag characteristics of the agitator speed increase event in the alcohol precipitation process and the pressure difference change of the downstream filter, establishing a nonlinear response model with a 23-second delay between the two. This feature is marked as a critical process control node.

[0027] Next, based on the above feature set, a dual-channel prediction model can be used to perform multi-dimensional state deduction. For example, when the long short-term memory network (LSTM) processes the temperature fluctuation sequence of the extraction tank, it captures the similarity between the abnormal fluctuation pattern and the historical failure case of the third production cycle in the memory unit, and outputs a probability of temperature out of control of 78.6% within the next 24 hours. For another example, the graph convolutional network identifies the synergistic failure mode of abnormal centrifuge vibration and decreased vacuum degree of the concentration system through the device topology embedding representation, and predicts that if it continues to run for more than 48 hours, the linkage system performance will decline by 42%. In the model verification stage, the data from the 31st to 35th production cycles confirmed that the prediction error for the remaining life of the centrifuge bearing was controlled within ±8 hours, and the determination coefficient R for the prediction of the vacuum pump performance degradation trend was 0. 2 Reached 0.93.

[0028] Then, in response to high-risk warnings output by the dual-channel prediction model, the strategy generation module of the equipment status data management system initiates a multi-objective optimization algorithm. For the extraction tank temperature control issue, parameter adjustment instructions are generated: the third-stage heating rate is reduced from 2°C / min to 1.3°C / min, and the 75°C holding phase is extended to 1.5 times the original duration. This solution, verified by digital twin system simulations, reduces the standard deviation of temperature fluctuation by 64%. Regarding maintenance planning, the equipment status data management system coordinates centrifuge bearing replacement and concentrate tank cooler cleaning tasks within batch intervals based on the equipment correlation matrix. This dynamic scheduling algorithm reduces production line downtime to 3.2 hours. When a structural fatigue warning is detected on the vibrating screen spring support, the equipment status data management system immediately triggers a three-level response mechanism: reducing the production line speed to 70% of the rated value, activating a backup filter unit to take over production, and sending a maintenance work order to a mobile terminal, guiding technicians to complete flaw detection inspections of critical parts within 45 minutes. All strategies are seamlessly integrated with the on-site control system via the OPC-UA protocol, ensuring stable production continuity indicators (OEE).

[0029] With this design, the embodiments of the present application significantly improve the reliability and maintenance efficiency of the Astragalus Extract production line. For example, during the target production cycle, the equipment status data management system predicts the risk of seal failure in the alcohol precipitation tank's agitator shaft dozens of hours in advance. By adjusting process parameters and replacing spare parts in advance, batch material losses can be effectively avoided. As another example, the embodiments of the present application can significantly reduce unplanned equipment downtime and improve the first-pass yield of products.

[0030] In an optional embodiment, the step of extracting a production line equipment status feature set from the production line equipment status data set in step 120 includes:

[0031] Step 121: Segment the production line equipment status data set to obtain equipment status data subsets corresponding to each production cycle.

[0032] In an embodiment of the present application, the equipment status data management system performs a segmented processing operation based on the production cycle on the production line equipment status data set, specifically by cutting and grouping the original data according to the start and end timestamps of each production cycle.

[0033] Taking the automated Astragalus Extract production line as an example, the equipment status data management system defines the data subset corresponding to the third production cycle as all equipment parameters collected between 08:00 on May 12, 2023, and 08:00 on May 15, 2023. This includes 2,880 minute-by-minute temperature data points recorded by temperature sensors in the extraction section, 108 million sampling points generated by centrifuge vibration sensors, and 216 hourly pressure monitoring values from the vacuum pumps in the concentration section. Each data subset is linked to the equipment serial number and production batch code to ensure strict timeline alignment of data from each device within the same cycle and to eliminate overlapping data points from inter-cycle handover periods.

[0034] Step 122: extracting equipment operation trend features for each equipment status data subset, wherein the equipment operation trend features include the fluctuation amplitude of the production parameters within a continuous time window, the parameter change rate, and the cumulative duration of the parameter deviation from the standard threshold.

[0035] In this embodiment of the present application, the equipment status data management system extracts multidimensional equipment operation trend features from segmented equipment status data subsets. For the extraction tank temperature parameter, the equipment status data management system uses a sliding time window analysis method to calculate the temperature range within each 15-minute period as a fluctuation amplitude feature. For example, during the 18th hour window of the third production cycle, the temperature rose from 85.3°C to 87.9°C and then fell back to 84.1°C, with a range feature value of 3.8°C.

[0036] The equipment status data management system also calculates the parameter change rate through first-order differences. For example, if the centrifuge drive motor current increases by 0.12A / s for five consecutive minutes, a current rate-of-change-exceed limit flag is generated. For the ethanol solvent flow parameter, the system accumulates statistics on the duration of time that the measured value deviates from the process standard value (150L / h) by more than ±5% per hour. If insufficient flow occurs for three consecutive hours during the concentration phase of the fifth production cycle, this cumulative deviation duration triggers a warning indicator.

[0037] Step 123: Generate a production line topology network based on the physical connection relationship between devices, and extract inter-device association features based on the production line topology network. The inter-device association features include a synchronization index of parameter transmission between adjacent devices and a coupling index of the operating status of the device group.

[0038] In an embodiment of the present application, the equipment status data management system constructs a production line topology network based on the material transmission path and signal interaction relationship between the equipment. The production line topology network uses extraction tanks, centrifuges, and concentration tanks as nodes, and connecting pipelines and control system signal lines as edges to form a directed weighted graph structure.

[0039] For example, the equipment status data management system calculates the synchronization index of the transmission of adjacent equipment parameters through cross-correlation analysis. For example, the time lag coefficient between the opening event of the extraction tank's liquid outlet valve and the speed increase of the concentration tank's feed pump is 2.3 seconds, and its synchronization index reaches 0.92.

[0040] For example, for the coupling index of the operating status of the equipment group, the equipment status data management system uses the principal component analysis method to evaluate the coordinated operation efficiency of equipment such as agitators, temperature controllers, and liquid level gauges in the alcohol precipitation section. When the agitator speed increases from 100 rpm to 120 rpm, the coupling index of the related equipment decreases by 15%, indicating that there is a risk of operating status mismatch.

[0041] Step 124: Fusing the equipment operation trend features with the inter-equipment association features to obtain the production line equipment status feature set.

[0042] In this embodiment of the present application, the device status data management system performs multimodal fusion of device operation trend features and inter-device correlation features. Specifically, the device status data management system normalizes the extraction tank temperature fluctuation amplitude features and the centrifuge vibration energy distribution features, and generates a cross-device joint feature vector through feature concatenation. Furthermore, the device status data management system encodes synchronization indicators and coupling indicators in the topological network into graph embedding vectors, spatially aligning them with time series features.

[0043] For example, when integrating the characteristic data of the third production cycle, the equipment status data management system performs a weighted combination of the temperature change rate characteristic of the extraction tank and the vacuum drop characteristic of the concentration tank to form a composite feature that reflects the linkage effect of the extraction-concentration section. The weight coefficient is dynamically adjusted according to the calculation results of the mutual information.

[0044] In an optional embodiment, the step 130 of training the Astragalus Extract production line state prediction model based on the production line equipment state feature set includes:

[0045] Step 131: Divide the production line equipment status feature set into a training data set and a validation data set, wherein the training data set includes historical equipment status features and corresponding equipment maintenance records.

[0046] In an embodiment of the present application, the equipment status data management system performs a data partitioning operation on the production line equipment status feature set, and uses 2,400 sets of equipment operation trend features and associated topological features extracted from the 1st to 30th production cycles as a training data set, which includes 1,856 equipment anomaly labels that strictly correspond to historical maintenance records. The 480 sets of feature data generated from the 31st to 36th production cycles are used as a validation data set to evaluate the generalization ability of the model. The equipment status data management system uses stratified sampling to ensure that the distribution ratio of each equipment type in the training set and the validation set is consistent. For example, centrifuge-related features account for 23.7% of the training set and 23.5% of the validation set.

[0047] Step 132: Generate a multi-layer neural network model including a temporal attention mechanism, wherein the temporal attention mechanism is used to capture the weight distribution of device state features at different time steps.

[0048] In an embodiment of the present application, the device status data management system constructs a multi-layer neural network model that includes a temporal attention mechanism. The temporal feature processing channel of the multi-layer neural network model adopts a bidirectional long short-term memory network structure, and the number of hidden layer units is set to 128.

[0049] For example, the temporal attention mechanism dynamically assigns importance weights to features at each time step through a learnable parameter matrix. For example, when processing a 72-hour temperature series of an extraction tank, the multi-layer neural network model assigns an attention weight of 0.68 to the abnormal temperature rise period in the 48th hour, significantly higher than the 0.05-0.12 weights for normal periods. The output vector of the attention mechanism is concatenated with the device-associated feature vector at the fusion layer to form the final state representation.

[0050] Step 133: A dynamic graph convolution module is used to process the inter-device association features, where the dynamic graph convolution module is used to update the node connection weights of the production line topology network according to real-time changes in the equipment operating status.

[0051] In an embodiment of the present application, the device status data management system deploys a dynamic graph convolution module to process inter-device association features. This module dynamically updates the edge weight parameters of the production line topology network based on real-time operating data. For example, when the vibration amplitude of a centrifuge exceeds a threshold, its association weight with the downstream concentrator vacuum pump is automatically increased from 0.75 to 0.91. The dynamic graph convolution operation uses a message passing mechanism. The state update function of each graph node combines the feature vectors and edge weights of its adjacent nodes. For example, the state of the concentrator node is determined by a weighted combination of its own vacuum characteristics and the vibration characteristics of the upstream centrifuge.

[0052] Step 134: Iteratively adjust the model parameters of the multi-layer neural network model by jointly optimizing the prediction loss function and the association constraint function, wherein the prediction loss function is used to measure the prediction error of the device abnormality probability, and the association constraint function is used to constrain the contribution of the association features between devices to the prediction results to meet the preset threshold.

[0053] In an embodiment of the present application, the device status data management system trains a prediction model using a joint optimization strategy. The prediction loss function uses weighted cross entropy to calculate the prediction error for device anomaly probability, with a weight factor of 3 applied to high-risk device categories. The association constraint function uses the L2 norm to restrict the contribution of inter-device association features to the prediction result to no more than a preset threshold of 0.45. During training, the optimizer synchronously updates the temporal attention parameters and graph convolution weights. For example, at the 120th iteration, the association contribution of the centrifuge vibration feature was constrained to 0.43, meeting the preset conditions.

[0054] Step 135: When the prediction accuracy of the verification data set reaches the convergence condition, stop training the multi-layer neural network model and output the astragalus extract production line status prediction model.

[0055] In an embodiment of the present application, the equipment status data management system monitors the prediction accuracy change trend of the validation data set. When the accuracy fluctuation range of 20 consecutive iterations is less than 0.5%, the model is judged to have reached the convergence condition. For example, in the 145th round of training, the prediction accuracy of the validation set for vacuum leakage of the concentration tank stabilized in the range of 93.7%-94.2%. The equipment status data management system immediately terminated the training process and exported the model parameters. The F1 value of the astragalus extract production line status prediction model that completed training for the prediction of centrifuge bearing jamming events reached 0.89, an improvement of 17% over the baseline model.

[0056] In an optional embodiment, the step 140 of generating a set of equipment management strategies based on the output results of the astragalus extract production line state prediction model includes:

[0057] Step 141: Determine the maintenance priority ranking of each device according to the distribution of device abnormality probability output by the astragalus extract production line status prediction model.

[0058] In an embodiment of the present application, the equipment status data management system generates a maintenance priority ranking based on the abnormal probability distribution output by the prediction model. For example, the risk matrix assessment method is used to multiply the abnormal probability value by the equipment criticality coefficient to obtain a priority score. When the abnormal probability of the concentration tank vacuum pump is 0.82 and the criticality coefficient is 1.5, its priority score reaches 1.23, which is significantly higher than the extraction tank agitator with a score of 0.76. Furthermore, the equipment status data management system generates a list of equipment arranged in descending order of score, of which the top three are the centrifuge spindle (1.35), vacuum pump (1.23) and ethanol delivery pump (1.12).

[0059] Step 142: Classify equipment maintenance types based on the analysis results of the performance degradation trend, and the maintenance types include emergency shutdown maintenance, production interval maintenance, and preventive maintenance.

[0060] In an embodiment of the present application, the equipment status data management system divides maintenance types according to performance degradation trends. For centrifuge bearing components with a predicted remaining life of less than 8 hours, the equipment status data management system marks it as an emergency shutdown maintenance type, requiring immediate interruption of the production process. When the performance degradation curve of the extraction tank seal shows that the leakage risk growth rate is lower than the threshold within the next 48 hours, the equipment status data management system classifies it as a production gap maintenance type. For equipment with regular degradation characteristics such as vacuum pumps, the equipment status data management system sets a preventive maintenance plan based on historical maintenance cycles, for example, automatically triggering a lubrication maintenance instruction after every 500 hours of operation.

[0061] Step 143: Match different equipment maintenance types with a preset production parameter adjustment rule library to generate parameter adjustment instructions that are adapted to the current production line operation status.

[0062] In this embodiment of the present application, the equipment status data management system matches maintenance types with a parameter adjustment rule library. The rule library contains 213 adjustment logics defined by process experts. For example, when a centrifuge requires emergency maintenance, the equipment status data management system automatically matches rule CTL-09 to reduce the feed flow rate from 300 L / min to 200 L / min to reduce the equipment load. For preventive maintenance needs on the alcohol precipitation tank, the equipment status data management system calls rule ALC-17 to adjust the stirring speed setpoint from 120 rpm to 110 rpm and extend the mixing time by 8 minutes to compensate for the temporary decrease in stirring efficiency.

[0063] Step 144: combining the maintenance priority ranking with the production plan data, allocating maintenance resources within the preset maintenance time window and generating a maintenance node planning scheme.

[0064] In an embodiment of the present application, the equipment status data management system generates a maintenance node planning scheme in combination with the production plan data. For example, it can access the enterprise resource planning system to obtain production batch information for the next 72 hours, and give priority to inserting maintenance tasks during the scheduled equipment idle periods. For another example, according to the 37th production cycle plan, the system arranges the centrifuge seal replacement operation during the equipment cleaning period of 02:00-04:00 on May 18, and the required human resources and spare parts inventory are locked in advance through the material requirement planning module. For complex maintenance needs across work sections, the equipment status data management system uses the critical path analysis method to determine the optimal execution sequence, and schedules the vacuum pump maintenance and extraction tank overhaul in parallel, shortening the overall downtime by 2.3 hours.

[0065] Step 145: Performing a logic check on the parameter adjustment instruction and the maintenance node planning scheme, excluding strategies with resource conflicts or parameter contradictions, and then outputting the device management strategy set.

[0066] In an embodiment of the present application, the equipment status data management system performs a strategy logic verification function and verifies the compatibility of parameter adjustment instructions with maintenance plans through discrete event simulation. For example, when a batch of production tasks requires the simultaneous execution of extraction tank cooling and centrifuge speed-up operations, the equipment status data management system detects a capacity conflict in the cooling water circulation system and automatically postpones the centrifuge speed-up instruction until the cooling stage is completed. All strategies must be verified through three-dimensional visualization of the digital twin system to ensure that the motion trajectory of the robotic arm does not interfere with the maintenance personnel's work space. The final output equipment management strategy set contains 17 parameter adjustment instructions and 9 maintenance planning schemes, which are issued to the on-site control system for execution after a security review.

[0067] In an optional embodiment, the method further includes:

[0068] Step 210: The latest equipment status data of the automated astragalus extract production line is obtained in real time, and current equipment status features corresponding to the latest equipment status data are extracted and input into the astragalus extract production line status prediction model.

[0069] In step 210, the equipment status data management system collects the latest equipment status data in real time through a sensor network deployed in each section of the automated astragalus extract production line. The system then performs feature extraction and model input preprocessing on the data. For the vacuum pump in the concentration section, for example, the system obtains real-time readings from the vacuum pump outlet pressure sensor at a frequency of seconds, while simultaneously collecting drive motor winding temperature data and cooling water circulation flow counter values. For the extraction tank temperature parameter, the system inputs the most recent 30-minute liquid phase temperature sequence into a sliding window Fourier transform module, generating a frequency domain feature vector containing the fundamental amplitude and harmonic components.

[0070] At the same time, the equipment status data management system calls the wavelet packet decomposition algorithm to process the real-time waveform data uploaded by the centrifuge vibration sensor, extracts the energy proportion characteristics of the 200-500Hz frequency band, forms a current equipment status feature set with a timestamp, and uses the data verification module to ensure that the feature dimensions strictly match the input layer structure of the astragalus extract production line status prediction model.

[0071] Step 220: When the astragalus extract production line status prediction model predicts that the abnormal probability of the target equipment exceeds a dynamic threshold, a real-time alarm is triggered and the maintenance priority ranking in the equipment management policy set is updated.

[0072] In step 220, the equipment status data management system inputs the current equipment status characteristics into the trained Astragalus Extract production line status prediction model and calculates the abnormality probability index for each device in real time. When the abnormal probability value of the centrifuge bowl bearing vibration energy reaches 0.87, exceeding the dynamic threshold of 0.85, the equipment status data management system immediately triggers a level 3 alarm command, pushes the alarm information to the central control room human-machine interface, and raises the centrifuge maintenance level from P2 to P0 in the maintenance priority queue.

[0073] At the same time, the device status data management system automatically scans the topological network of associated devices, identifies the concentrate tank vacuum pump, which is strongly coupled to the centrifuge, and dynamically adjusts its abnormal probability threshold from 0.80 to 0.75, thereby simultaneously updating the risk level of associated devices. When an alarm event is triggered, the maintenance work order generation module within the device management policy set immediately interrupts the current low-priority task and recalculates the equipment maintenance sequence.

[0074] Step 230: Dynamically adjust the maintenance node planning scheme being executed according to the updated maintenance priority ranking, and reallocate unused maintenance resources to the target device.

[0075] In step 230, the equipment status data management system dynamically adjusts the maintenance resource allocation plan based on the updated maintenance priority ranking. When the centrifuge is marked as a P0 maintenance target, the equipment status data management system automatically allocates the spare lubricant inventory originally planned for the extraction tank agitator maintenance to the centrifuge maintenance point in Workshop 3 and reschedules the maintenance personnel.

[0076] As you can understand, for the initiation of the concentrate tank seal replacement operation but not yet completed, the equipment status data management system uses a progress assessment algorithm to calculate the remaining time for the task. If it determines that it can be completed within 1.5 hours, the current task is maintained; otherwise, the task is suspended and the vacuum testing equipment is transferred to the centrifuge section. Furthermore, the equipment status data management system simultaneously updates the maintenance resource status database, marking the assigned electric torque wrench and vibration analyzer as occupied to prevent duplicate calls for other work orders.

[0077] Step 240: Synchronously calibrate the adjusted maintenance node planning scheme with the current production parameter adjustment instruction so that the production line throughput after parameter adjustment is not lower than the preset guarantee value.

[0078] In step 240, the equipment status data management system performs a coordinated calibration operation of the maintenance plan and production parameters, including: when the centrifuge maintenance plan requires a reduced speed, calling the CTL-12 instruction in the production parameter adjustment rule library to reduce the feed pump speed from 300 rpm to 220 rpm, and at the same time extending the stirring time of the alcohol precipitation process by 25 minutes to compensate for the loss of mixing efficiency.

[0079] After adjusting the parameters, the equipment status data management system simulated the production line's throughput over the next eight hours using a digital twin model to verify whether the adjusted configuration could maintain the minimum output of 135 kg of Astragalus extract per hour. When the simulation results showed that the throughput dropped to 128 kg during a certain period, the equipment status data management system automatically activated a backup plan: while maintaining the centrifuge speed, it temporarily activated a spare concentrator to resume production. A dynamic load balancing algorithm then redistributed the material flow to ensure that overall output did not fall below the preset guaranteed value.

[0080] In an optional embodiment, the method for determining the dynamic threshold includes:

[0081] Step 221: Count the distribution of device status characteristics in the target time period before the occurrence of the historical device abnormal event.

[0082] In an embodiment of the present application, the equipment status data management system collects statistical characteristic distribution data of historical equipment abnormal events and establishes an association model between characteristic values and abnormal events. Taking the vacuum leakage event of the concentration tank as an example, the equipment status data management system retrieves the operating data of the 72 hours before the 8 leakage events that occurred in the past 12 months, and extracts characteristic parameters such as the vacuum drop rate, the cooling water temperature difference expansion value, and the motor current fluctuation amplitude. The probability distribution curve of each feature in the abnormal precursor period is constructed by the kernel density estimation algorithm, and the average value of the vacuum drop rate feature 6 hours before the event is recorded as 0.015MPa / h, which is significantly higher than the maximum allowable offset of 0.008MPa / h in the normal production cycle.

[0083] Step 222: Calculate the maximum allowable offset between the characteristic value offset of each device status characteristic in the device status characteristic distribution before the historical device abnormal event occurs and the normal production cycle.

[0084] In an embodiment of the present application, the equipment status data management system calculates a quantitative index of the characteristic value offset. For the centrifuge vibration energy feature, the maximum allowable offset of the feature within the normal production cycle can be calculated as ±12% of the reference value. When the actual offset reaches +18% 2 hours before the historical abnormal event occurs, its relative offset coefficient is calculated to be 1.5 (18% / 12%). By traversing all associated features, an offset coefficient matrix can be established, in which the mean value of the vacuum pump outlet pressure offset coefficient is 1.3, and the peak value of the extraction tank temperature fluctuation coefficient is 2.1, providing data support for subsequent threshold setting.

[0085] Step 223: Determine abnormal probability critical values at different confidence levels based on the cumulative distribution function of the eigenvalue offset.

[0086] In an embodiment of the present application, the equipment status data management system determines the abnormal probability critical value based on the cumulative distribution function of the characteristic offset. Taking the current characteristic of the alcohol precipitation tank agitator as an example, the equipment status data management system analyzes the current offset data of 1200 normal production periods and draws a cumulative probability distribution curve. When the confidence level is set to 95%, the corresponding current offset critical value is +14.7% of the baseline value, and this threshold is mapped to an abnormal probability of 0.82. Furthermore, the equipment status data management system establishes a confidence-critical value comparison table for each feature. For example, the vacuum drop rate corresponds to an abnormal probability of 0.91 at a confidence level of 99%, and corresponds to 0.85 at a confidence level of 90%.

[0087] Step 224: Based on the key indicator requirements of the current production stage, dynamically select a confidence level that matches the production line stability requirements, and set the abnormal probability critical value corresponding to the matching confidence level as the dynamic threshold.

[0088] In an embodiment of the present application, the equipment status data management system dynamically selects the confidence level according to the current production demand. When the Astragalus Extract production line performs high-purity product production tasks, the equipment status data management system increases the production line stability demand index to Level-4 and automatically selects the abnormal probability threshold corresponding to the 99% confidence level. At this time, the centrifuge vibration energy threshold is adjusted from the conventional 0.83 to 0.78, significantly improving the risk identification sensitivity. When the production line switches to normal batch production mode, the equipment status data management system restores the 90% confidence level, balancing the false alarm rate and detection efficiency. The dynamic threshold adjustment module polls the production planning system every 15 minutes to ensure that the threshold setting is synchronized with the process requirements in real time.

[0089] In an optional embodiment, the method further includes:

[0090] Step 310: Generate a device status management knowledge graph, which includes the relationship between device structure attributes, historical maintenance records and fault solutions.

[0091] In step 310, the equipment status data management system constructs an equipment status management knowledge graph, and forms a multi-dimensional knowledge network by integrating the structured data and unstructured documents of the equipment in the automated astragalus extract production line.

[0092] Specifically, the equipment status data management system extracts the centrifuge component structure tree from the equipment asset management module, clarifying the assembly relationships and functional dependencies between subcomponents such as the bowl, spindle, and seals. For example, seal failure will directly lead to a decrease in the vacuum level in the bowl chamber. Simultaneously, the system analyzes text records from maintenance work orders from the past three years and uses a bidirectional long-short-term memory network model to identify key entities in fault descriptions. For example, from the description of work order number CT-2023-045, "abnormal centrifuge vibration with unusual noise," the system extracts the faulty equipment component "spindle bearing" and the corresponding repair action "replace SKF 6318 bearing." The system maps these entity relationships into a knowledge graph triple structure, forming association paths such as "centrifuge-occurrence-spindle wear" and "spindle wear-repair method-bearing replacement." It also adds attribute fields for the repair action's average repair time (e.g., seal replacement takes an average of 2.3 hours) and spare part consumption (e.g., two seals are consumed per replacement).

[0093] Step 320: Match and search the output result of the astragalus extract production line state prediction model with the equipment state management knowledge graph to obtain a set of candidate maintenance solutions related to the current predicted equipment state.

[0094] In step 320, the equipment status data management system matches the output of the Astragalus Extract production line status prediction model with the knowledge graph for a search. When the model predicts a centrifuge main shaft bearing abnormality probability of 0.91, the equipment status data management system uses "centrifuge bearing" as the query subject and performs a multi-condition search within the knowledge graph. First, it matches the "bearing wear" node with a failure probability greater than 0.85. Then, along the "repair method" edge, it retrieves three candidate solutions: "immediately shut down and replace the bearing," "reduced speed and grease injection," and "load adjustment and monitoring operation." Furthermore, the equipment status data management system uses a graph neural network to calculate the degree of match between each solution and the current equipment status. The "immediate replacement" solution has the highest match score (0.92), due to its historical application scenario where the bearing abnormality probability exceeds 0.9 and a failure conversion rate of 98%.

[0095] Step 330: Sort candidate maintenance plans based on historical evaluation data of maintenance plan execution effects, and give priority to maintenance plans with execution efficiency higher than a preset benchmark and resource consumption lower than a preset upper limit.

[0096] In step 330, the equipment status data management system optimizes the ranking of candidate solutions based on historical maintenance plan execution data. Specifically, it accesses 1,208 execution records from the maintenance work order database and calculates the comprehensive performance index of each solution. For the "replace centrifuge bearing" solution, its average repair time of 2.3 hours corresponds to a performance index of 75, and its spare part cost of 2,850 yuan corresponds to a resource consumption index of 60. The comprehensive score = (75 × 0.6) + (60 × 0.4) = 69. Meanwhile, the "grease injection" solution, while only requiring 0.5 hours (with a performance index of 90), suffers a 42% recurrence rate, resulting in a lower quality factor and a final score of 68. For example, the equipment status data management system prioritizes solutions with a comprehensive score above 70. If no other solution meets the criteria, it initiates a cross-solution combination optimization process, recommending, for example, the combined solution of "grease injection + vibration monitoring," which improves its comprehensive score to 73.

[0097] Step 340: Fusing the ranked candidate maintenance plans with the equipment management policy set to generate an enhanced equipment management policy containing multi-dimensional decision-making basis.

[0098] In step 340, the equipment status data management system integrates the optimized maintenance plan with the real-time equipment management strategy. For the centrifuge bearing warning event, the "replace immediately" solution recommended by the knowledge graph and the "reduce speed to 220 rpm" parameter adjustment instruction generated by the predictive model can be combined to form an enhanced strategy. This strategy inserts an emergency maintenance work order into the equipment management strategy set, specifying that the bearing should be replaced within two hours. Simultaneously, a production parameter instruction is issued to gradually reduce the centrifuge speed from 300 rpm to 220 rpm to relieve bearing load.

[0099] As expected, the equipment status data management system verified the feasibility of the strategy using digital twins. The simulation showed that while the production line throughput decreased by 7% during the reduced speed period, the capacity gap could be offset by starting a backup centrifuge in parallel, ensuring that overall output met targets. The resulting enhanced strategy included five parameter adjustment instructions, three maintenance work orders, and two resource scheduling plans, which were issued for execution after logical verification.

[0100] In an optional embodiment, generating the device state management knowledge graph in step 310 includes:

[0101] Step 311: Extract the equipment component structure tree and the functional dependency relationships between components from the production line equipment database.

[0102] In step 311, the equipment status data management system extracts equipment hierarchical structure data from the production line equipment database and constructs a component structure tree with topological relationships. For example, for the extraction tank in an automated astragalus extract production line, the root node of the component structure tree can be defined as the extraction tank itself. Child nodes include secondary components such as the tank jacket, agitator, and temperature sensor. The agitator is further broken down into tertiary components such as blades, drive shaft, and coupling.

[0103] Furthermore, the equipment status data management system uses the bill of materials to correlate the functional dependencies between components. For example, when a temperature sensor fails, the directly affected upper-layer component is automatically marked as the temperature control system. A dependency propagation algorithm then identifies the affected process as the liquid extraction step. For critical equipment such as centrifuges, a dynamic dependency relationship between the drum assembly and the power system is established, recording that when the drive motor power drops to 85% of the rated value, the drum speed will deviate by 3.2%.

[0104] Step 312: Parse the fault description text in the historical work order data, and use entity recognition technology to mark the faulty equipment components and the maintenance measures adopted for the faulty equipment components.

[0105] In step 312, the equipment status data management system performs natural language processing on historical maintenance work orders to extract structured knowledge elements. For example, a named entity recognition model enhanced by a domain dictionary is used to accurately locate equipment components and maintenance actions from unstructured text. For another example, when processing work order number VAC-20230417, the equipment status data management system identified the faulty equipment component as the "vacuum pump exhaust valve spring" and the maintenance measure as "replacement of the high-temperature resistant alloy spring" from the fault description of "vacuum pump outlet pressure fluctuation exceeds the allowable range, and upon inspection, it is found to be fatigue fracture of the exhaust valve spring." Furthermore, the equipment status data management system simultaneously establishes a mapping relationship library between failure modes and solutions, annotating the corresponding relationships between 1,235 failure phenomena and 892 maintenance operations, of which the coverage rate for vacuum pump-related failure modes reaches 96%.

[0106] Step 313: Generate a heterogeneous relationship network of equipment components, failure modes, and maintenance measures, and add the mean time to repair and spare parts consumption attributes corresponding to each maintenance measure to the heterogeneous relationship network.

[0107] In step 313, the equipment status data management system generates a heterogeneous relationship network of equipment components, failure modes, and repair measures, adding quantitative attribute dimensions. Specifically, failure modes such as a broken extraction tank impeller and a failed centrifuge seal can be used as intermediate nodes, connecting the equipment component nodes on the left with the repair measure nodes on the right. For example, the "Centrifuge Drum Seal" component node is connected to the "Sealing Surface Wear" fault node through the "Occurrence" relationship, which in turn connects to the "Replace PTFE Seal" action node through the "Repair Method" relationship. Furthermore, the equipment status data management system adds attributes such as mean time to repair, spare parts cost, and tool requirements to each repair measure node. For example, the standard operation time for replacing the concentrator sight glass is 45 minutes, requiring specialized tools such as vacuum silicone grease and a torque wrench.

[0108] Step 314: Perform embedding representation learning on the heterogeneous relational network through a graph neural network to obtain a set of embedding vectors that reflect the relevance of device status management knowledge.

[0109] In step 314, the device status data management system uses a graph neural network to learn features from the heterogeneous relational network and generate a low-dimensional embedding vector representation. For example, the device status data management system designs a heterogeneous graph attention network model to map three types of nodes: device components, failure modes, and maintenance measures, into a unified vector space.

[0110] During training, the heterogeneous graph attention network model learned that the embedding vector for the fault node "Vacuum Pump Exhaust Valve Spring Fracture" had a cosine similarity of 0.89 with the vector of its solution, "Spring Replacement," while the similarity with the irrelevant node, "Agitator Speed Control," was only 0.12. After 50 rounds of iterative training, the embedding vector generated by the equipment status data management system accurately reflected the strong correlation between centrifuge spindle wear and the bearing replacement solution. The quantitative correlation strength value was 0.93, significantly higher than the 0.45-0.67 range for other candidate solutions.

[0111] Step 315: Generate a knowledge graph query interface that supports multi-hop reasoning based on the embedding vector set.

[0112] In step 315, the equipment status data management system constructs a knowledge graph query interface that supports multi-hop reasoning. When a user queries "abnormal centrifuge vibration," the system performs a two-hop reasoning algorithm using a graph traversal algorithm. First, it locates the "abnormal centrifuge vibration" fault node and retrieves directly associated first-level fault modes such as "spindle imbalance" and "bearing wear." It then extends this information to potential second-level causes such as "coupling offset" and "dynamic balancing failure." The query interface returns three repair paths: Path 1: "abnormal vibration → bearing wear → bearing replacement + dynamic balancing correction," with a historical average repair time of 2.5 hours; Path 2: "abnormal vibration → coupling offset → coaxiality adjustment," with an average repair time of 1.2 hours. Furthermore, the equipment status data management system provides semantic search capabilities based on embedded vectors. Entering the keyword "seal leakage" expands the query to related faults across equipment categories, such as "centrifuge shaft seal aging" and "extraction tank flange gasket damage."

[0113] In an optional embodiment, the method further includes:

[0114] Step 410: Perform sensor monitoring on at least part of the equipment corresponding to the automated astragalus extract production line through a multimodal sensor array, and collect multimodal sensor data corresponding to at least part of the equipment in real time, wherein the multimodal sensor data includes equipment vibration spectrum data, infrared thermal imaging data, and voiceprint signal data.

[0115] In this embodiment, the equipment status data management system deploys a multimodal sensor array to comprehensively monitor key equipment in the automated Astragalus Extract production line. For the centrifuge equipment, a triaxial vibration sensor is installed on the drum bearing seat to collect 0-10kHz vibration spectrum data. Simultaneously, an infrared thermal imaging camera is positioned on the equipment housing to capture temperature field distribution images at a rate of 5 frames per second.

[0116] In the extraction tank area, the equipment status data management system deploys a directional acoustic sensor array to capture the equipment's operating soundprint signals. Its sampling frequency is set to 48kHz to cover the audible frequency band and ultrasonic components. For example, during the third production cycle, the system simultaneously captured the centrifuge's X-axis vibration acceleration spectrum, infrared thermal images of the extraction tank jacket area, and the operating soundprint waveform of the alcohol precipitation tank agitator, forming a multimodal sensor data stream with strictly aligned timestamps.

[0117] Step 420: Perform a joint time-frequency analysis on the multimodal sensor data to extract sensitive features of the equipment mechanical state.

[0118] In step 420, the equipment status data management system performs a joint time-frequency analysis on the multimodal sensor data. Discrete wavelet packet transform is used to decompose the centrifuge vibration spectrum data into 16 frequency bands, and the Shannon entropy of the energy distribution of each frequency band is calculated. When the drum bearing shows early signs of wear, the system detects an increase in the energy entropy value in the 3.2-4.8 kHz frequency band from a baseline of 0.45 to 0.68, marking it as a characteristic of mechanical wear.

[0119] In infrared thermal imaging analysis, the equipment status data management system modeled the temperature gradient field in the flange connection area of the concentrate tank. When a seal failure caused heat leakage, the system measured an abnormal area growth rate of 12 square centimeters per minute, exceeding the 4 square centimeters threshold under normal operating conditions. For the agitator soundprint data of the alcohol precipitation tank, the system extracted 128 Vimel cepstral coefficient feature vectors and identified the blade scratching noise pattern using a pre-trained hidden Markov model, with a severity score of 8.7 out of 10.

[0120] Step 430: Perform cross-modal feature alignment on the mechanical state sensitive feature and the production line equipment state feature set to obtain a cross-modal feature difference.

[0121] In this step, the equipment status data management system performs a cross-modal feature alignment analysis, temporally and spatially aligning the mechanical state-sensitive features extracted by multimodal sensors with the process parameters in the production line equipment state feature set. For example, a Pearson correlation analysis is performed between the centrifuge vibration entropy features and the vacuum level data of the concentrator during the same period. When the correlation coefficient falls below the historical baseline value of 0.6, the cross-modal feature difference D = 1-|r| = 0.55 is calculated, exceeding the preset tolerance of 0.4. The equipment status data management system locates the source of the anomaly using the feature difference matrix. When the difference between the vibration characteristics and the vacuum level characteristics continues to exceed the standard, it determines that there is a mismatch between the centrifuge's mechanical state and process parameters.

[0122] Step 440: When the cross-modal feature difference exceeds a preset tolerance, the equipment status review process is triggered and the input features of the astragalus extract production line status prediction model are corrected.

[0123] In step 440, the equipment status data management system triggers the equipment status review process. When the difference in the soundprint characteristics of the extraction tank agitator reaches 0.52 (threshold 0.45), a three-level review mechanism is initiated: first, a high-definition industrial endoscope is used to conduct a visual inspection of the agitator blade, then a vibration analyst is assigned to conduct on-site spectrum verification, and finally, a multi-physics field simulation is performed using the digital twin model. After confirming the presence of a 3mm crack in the agitator blade, the equipment status data management system corrects the input features of the astragalus extract production line status prediction model, adjusting the soundprint cepstral coefficient weight from 0.3 to 0.5 and adding crack growth rate as a new feature dimension.

[0124] Step 450: Regenerate a set of equipment management correction strategies based on the revised output results of the astragalus extract production line state prediction model, and enter the feature difference traceability labels corresponding to the cross-modal feature difference into the generated equipment state management knowledge graph.

[0125] In step 450, the equipment status data management system generates a revised equipment management strategy and updates the knowledge graph. Based on the revised predictive model output, the abnormality probability of the centrifuge bearing is recalculated from 0.82 to 0.91, and a preventative replacement strategy is immediately generated. Simultaneously, the equipment status data management system enters the cross-modal feature difference traceability results into the knowledge graph, establishing the association path from "vibration entropy anomaly - voiceprint cepstrum variation - blade crack," and annotating the action plan when the feature difference threshold is exceeded. As can be seen, the updated knowledge graph improves the efficiency of subsequent maintenance plan retrieval for similar operating conditions and shortens the mean time to fault location.

[0126] In an optional embodiment, the performing of a time-frequency joint analysis on the multimodal sensor data in step 420 to extract sensitive features of the equipment mechanical state includes:

[0127] Step 421: performing wavelet packet decomposition on the equipment vibration spectrum data, extracting energy distribution entropy of different frequency bands, and using the energy distribution entropy as an indicator of equipment mechanical wear.

[0128] In step 421, the equipment status data management system performs wavelet packet decomposition on the vibration spectrum data, decomposes the vibration signal of the centrifuge drive end bearing into 16 equal-width frequency bands, calculates the normalized energy value of each frequency band, and generates an energy distribution histogram. When pitting defects appear in the bearing raceway, the equipment status data management system detects that the energy proportion of the 6.4-7.2kHz frequency band suddenly increases from 3.8% in the normal state to 9.2%, and its corresponding energy distribution entropy value increases from 0.32 to 0.71, accurately reflecting the trend of increasing mechanical wear. Furthermore, the equipment status data management system establishes a mapping relationship between the energy entropy value of each frequency band and the remaining life of the equipment. For example, when the energy entropy of the high-frequency band (>5kHz) exceeds 0.65 for three consecutive hours, it is determined that the remaining service life of the bearing is less than 72 hours.

[0129] Step 422: Perform regional temperature gradient analysis on the infrared thermal imaging data to determine the area growth rate of the thermal anomaly region on the device surface.

[0130] In step 422, the equipment status data management system processes the thermodynamic characteristics of the infrared thermal imaging data and grids the vacuum pump motor housing. The temperature data for each grid cell forms a three-dimensional space-time tensor. When a local short circuit occurs in the motor winding, the equipment status data management system detects that the temperature gradient in region C3 reaches 8°C / cm, a four-fold increase compared to normal operating conditions, and the area of the abnormal region expands at a rate of 9.6 square centimeters per minute. The equipment status data management system then analyzes the thermal image sequence to determine the direction of fault propagation. For example, if the high-temperature area migrates from the rear end cover of the motor to the front bearing seat, it is determined to be a frictional heating mode caused by lubrication failure.

[0131] Step 423: extract the Mel-frequency cepstral coefficients of the voiceprint signal data, and identify the abnormal sound pattern of the device operation by combining the hidden Markov model.

[0132] In step 423, the equipment status data management system analyzes the pattern characteristics of the voiceprint signal: for the cavitation noise of the extraction tank agitator, 40 Mel-Meer cepstral coefficients are extracted to represent the acoustic fingerprint, and the similarity score with the standard voiceprint template is calculated using a Gaussian mixture model. When the agitator blade deforms, the equipment status data management system identifies an increase in the harmonic component in the 200-800Hz frequency band, and its cepstral coefficient variation index reaches 2.35, triggering an abnormal sound mode alarm. Then, combined with the state transition probability of the hidden Markov model, it is determined that the current acoustic feature belongs to a fault mode such as blade breakage (probability 0.78) or bearing oil shortage (probability 0.21).

[0133] Step 424: Combine the equipment mechanical wear index, the area growth rate, and the equipment operation abnormal noise mode to generate a joint health assessment function of multimodal sensitive features; wherein the joint health assessment function at least integrates the frequency band energy distribution entropy, the thermal anomaly growth rate, and the abnormal noise mode severity score.

[0134] In step 424, the equipment status data management system constructs a joint health evaluation function of multi-modal sensitive features. For example, a health function is defined for a centrifuge device:

[0135] H = 0.4 × E_vib + 0.3 × R_thermal + 0.3 × S_audio, where E_vib represents the vibration energy entropy, R_thermal represents the growth rate of the thermal anomaly area, and S_audio is the abnormal sound mode score.

[0136] When the centrifuge drum bearing health H dropped from the baseline value of 0.85 to 0.63, the equipment status data management system detected the vibration entropy value E_vib = 0.71 and the thermal anomaly rate R_thermal = 12cm 2 / min, abnormal sound score S_audio=7.9, comprehensive calculation H=0.4×0.71+0.3×(12 / 4)+0.3×(7.9 / 10)=0.63, which is lower than the preset threshold value of 0.70, triggering the mechanical state abnormality judgment.

[0137] Step 425: When the function value of the joint health evaluation function is lower than the preset health threshold, the equipment mechanical state is determined to be abnormal and a review task priority label is generated.

[0138] In step 425, the equipment status data management system generates a review task priority tag based on the health assessment results. When the health function value of the concentrate tank agitator is below the threshold for two consecutive hours, the equipment status data management system automatically increases the priority level of the equipment in the maintenance queue. For example, when the system detects that the agitator health level H = 0.65, it calculates the risk probability of shaft fracture based on historical fault data as 38%, generates a red priority tag, and assigns it to the emergency review queue. Similarly, the equipment status data management system synchronously associates multimodal feature data and stores evidence chains such as vibration entropy anomalies, thermal gradient exceeding the standard, and abnormal sound feature matching in the review task database.

[0139] As an optional but non-limiting embodiment, the step 123 of generating a production line topology network based on the physical connection relationship between devices and extracting inter-device association features based on the production line topology network includes:

[0140] Step 1231: Determine the physical connection relationship between the devices in the automated astragalus extract production line, the physical connection relationship includes the docking direction of the material transmission interface between the devices, the sensor signal transmission link and the power supply dependent path; generate a set of adjacent device pairs according to the connection status of the input and output ports of each device in the physical connection relationship, the adjacent device pair set includes a device group with a direct physical connection and the connection interface type of the device group; generate a production line topology network diagram based on the adjacent device pair set, the nodes of the production line topology network diagram represent device entities, the edges represent the physical connection relationship between devices, and the attributes of the edges include the interface type and the signal transmission direction.

[0141] Step 1232: Traverse the adjacent device nodes of each device node in the production line topology network diagram, intercept the operating status parameter sequence of the adjacent device pairs in the continuous production cycle, align the timestamps of the operating status parameter sequence and calculate the synchronization index of the parameter transmission between adjacent devices, the synchronization index includes the fluctuation phase difference between the upstream device output parameter and the downstream device input parameter in the same time window and the parameter change trend correlation coefficient.

[0142] Step 1233: Divide the equipment into groups according to the degree centrality of the equipment nodes in the production line topology network diagram, wherein the equipment groups include key central equipment and a set of sub-equipment directly connected thereto; extract the operating status parameters of each equipment in the same equipment group in the same production cycle, and calculate the coupling index of the operating status of the equipment in the group, wherein the coupling index includes a similarity measure of the fluctuation of the equipment parameters in the group and the co-occurrence frequency of abnormal events.

[0143] Step 1234: Classify the synchronization index and the coupling index according to devices and device groups to generate a set of device association features that includes the strength of association between devices and group operation dependencies; assign weights to the set of device association features according to the attributes of the edges in the production line topology network diagram, so that the association features corresponding to physical connection relationships with different interface types and signal transmission directions have differentiated contributions.

[0144] Step 1235: Based on the weighted inter-device association feature set, the attributes of the edges in the production line topology network diagram are updated to form a dynamic production line topology network that integrates the real-time operation status association; the association feature vector of each device node is extracted from the dynamic production line topology network, and the association feature vector includes the synchronization index of adjacent devices, the coupling index of the group to which it belongs, and the network edge weight distribution characteristics.

[0145] In step 1231, the equipment status data management system analyzes the physical connections between the equipment in the automated Astragalus Extract production line and constructs a precise production line topology network. For example, the system traverses the input and output port configuration tables for core equipment such as the extraction tank, centrifuge, and concentration tank. It identifies the material transmission path from the bottom outlet of the extraction tank to the centrifuge feed port via a DN80 stainless steel pipe. It also marks the signal transmission link of the electromagnetic flowmeter installed on the pipe as pointing to the central control system.

[0146] Regarding power supply dependencies, the device status data management system confirms that the centrifuge drive motor is powered by the third circuit of power cabinet No. 3 and shares the same power busbar as the vacuum pump. Based on this physical connection information, the device status data management system generates a set of adjacent device pairs, including physical connection groups such as the "extraction tank-centrifuge" pair (with a material pipeline interface) and the "centrifuge-concentrator" pair (with a pneumatic valve control signal interface). The connection attribute fields for each device pair record the interface size, media type, and signal transmission direction. For example, the pipe connection from the centrifuge outlet to the concentrator feed pump is marked as unidirectional.

[0147] In step 1232, the equipment status data management system calculates the synchronization index of the operating parameters between the equipment. For example, the time series data of the opening degree of the extraction tank outlet valve and the speed sequence of the centrifuge feed pump in the third production cycle can be intercepted, and the time axis is aligned through the dynamic time warping algorithm. It is found that when the opening degree of the extraction tank valve is increased to 65%, the speed of the centrifuge feed pump is synchronously increased to 285rpm after a delay of 2.3 seconds, and the Pearson correlation coefficient of its parameter change trend reaches 0.92. For the vacuum degree of the concentration tank and the cooling water circulation flow parameters, the equipment status data management system uses a window sliding cross-correlation analysis to detect that for every 0.01MPa drop in vacuum pressure, the cooling water flow rate increases by 5.2m after 1.8 seconds. 3 / h, with the phase difference remaining stable within ±0.3 seconds. The equipment status data management system then generates a synchronization index matrix for each adjacent equipment pair. For example, the trend correlation coefficient for the "centrifuge-concentrator" pair is 0.87, and the phase difference standard deviation is 0.41 seconds, indicating a strong linkage between the two devices.

[0148] In step 1233, the equipment status data management system divides the equipment into groups and calculates the coupling index. Based on the degree centrality analysis of the production line topology network nodes, it is determined that the centrifuge node has the highest connectivity (degree = 5). It is designated as the key central device and forms a core equipment group with directly connected devices such as the extraction tank, concentration tank, and lubrication system.

[0149] Furthermore, the equipment status data management system extracted the vibration, temperature, and pressure parameters of this group during the fifth production cycle and used a dynamic time warping algorithm to calculate the similarity of parameter fluctuations. The system found that when the centrifuge vibration amplitude increased by 15%, the concentration tank vacuum fluctuation similarity index also increased to 0.78. Abnormal event co-occurrence frequency statistics showed that when the centrifuge bearing temperature exceeded 85°C, the probability of abnormal current in the associated extraction tank agitator increased by 62%. Based on this, the equipment status data management system calculated the group coupling index as 0.69, reflecting the risk level of fault transmission between equipment.

[0150] In step 1234, the device status data management system assigns weights to the associated features and groups them together. For example, the device status data management system classifies the synchronization index of the "centrifuge-concentrator" pair (correlation coefficient 0.87, phase difference 0.41 seconds) into the material transfer associated feature subset and assigns a weight coefficient of 0.8 based on the interface type being DN100 stainless steel pipe.

[0151] For the centrifuge-control cabinet signal transmission connection, its synchronization indicator, with a trend correlation coefficient of 0.65, was categorized as the control signal association subset, with a weight coefficient of 0.5. The equipment status data management system then established a weighting rule base, specifying a baseline contribution value of 0.7 for material transmission connections, 0.6 for power supply connections, and 0.4 for control signal connections. This ensures that the association characteristics of different physical connection types have different impact weights in subsequent analysis.

[0152] In step 1235, the equipment status data management system constructs a dynamic production line topology network and extracts feature vectors. For example, the equipment status data management system updates network edge weights based on real-time operating data. When the centrifuge feed pump speed increases to 300 rpm, the material transfer edge weight with the extraction tank is dynamically adjusted from 0.8 to 0.9, reflecting the strong correlation under the current operating conditions. The associated feature vector of the centrifuge node is extracted from the updated network. This vector includes distribution features such as a synchronization index of 0.92 with the extraction tank, a coupling degree of 0.78 with its core group, a material transfer edge weight of 0.9, and a control signal edge weight of 0.6. This feature vector is encoded into a 128-dimensional embedding vector using a graph convolutional neural network and input into the Astragalus Extract production line status prediction model. The model accurately predicts that when the centrifuge vibration synchronization index decreases by 10%, the probability of a cascading failure of the associated group equipment within 24 hours increases to 73%.

[0153] As can be understood, step 1231 systematically analyzes the physical connections between devices to construct a topological network that accurately reflects the production line structure, enabling visual modeling of the entire link for material transmission, signal control, and power supply. This step converts the connection status of device input and output ports into a standardized set of adjacent device pairs, providing a structured data foundation for subsequent correlation analysis. This effectively resolves the issue of ambiguous connection relationship descriptions in traditional device management, allowing for precise annotation of key attributes such as the physical interface type and signal transmission direction between devices, establishing a spatial correlation framework for dynamic analysis of device collaborative operation.

[0154] The dynamic feature extraction system formed by steps 1232-1235 realizes the in-depth mining of device association status from static topology to real-time operation characteristics. The time domain correlation law of parameter transmission between devices is revealed by calculating the synchronization index, and the fault conduction effect of devices in the group is quantified by combining the coupling index to construct a multi-dimensional device operation status association model. The weight distribution mechanism distinguishes the impact level of different types of physical connections on the system, and the dynamically updated network edge attributes realize the feature fusion of real-time operation data. The resulting association feature vector transforms complex device relationships into a computable feature space, providing the intelligent prediction model with a multi-dimensional input that can both characterize the device space topology and reflect the real-time operation status, significantly improving the spatiotemporal correlation of device status analysis.

[0155] As an optional but non-limiting embodiment, the ranked candidate maintenance solutions are integrated with the set of equipment management policies in step 340 to generate an enhanced equipment management policy containing multi-dimensional decision-making basis, including:

[0156] Step 341: Obtain the sorted candidate maintenance plan priority list, and extract the maintenance time window and equipment identifier corresponding to the maintenance node planning plan generated in the equipment management policy set; establish a mapping relationship between the candidate maintenance plan and the maintenance time window in the maintenance node planning plan based on the maintenance plan execution efficiency and resource consumption indicators in the maintenance plan priority list.

[0157] Step 342: Based on the mapping relationship, analyze the matching degree between the resource requirements of each candidate maintenance plan and the allocated maintenance resources in the maintenance node planning plan, and generate a resource matching analysis result; detect the maintenance instruction conflict between the candidate maintenance plan and the same equipment in the maintenance node planning plan within the same time window according to the resource matching analysis result, and output the conflict detection result.

[0158] Step 343: Based on the conflict detection result, dynamically adjust the candidate maintenance plans that have resource conflicts or time overlaps, and the dynamic adjustment includes reallocating the maintenance time window or splitting the maintenance resources to different equipment identifiers; perform a joint logical check on the adjusted candidate maintenance plan and the production parameter adjustment instructions in the equipment management policy set to verify whether the adjusted maintenance time window causes the throughput after the production parameter adjustment to be lower than the preset guarantee value; based on the verification result, screen the candidate maintenance plans that meet the throughput guarantee and have no resource conflicts, and generate an optimized set of maintenance plans.

[0159] Step 344: Extract the equipment structure attributes and historical maintenance records corresponding to each maintenance plan in the maintenance plan optimization set, and perform correlation matching with the fault solution in the equipment status management knowledge graph; based on the correlation matching results, perform secondary sorting on the plans in the maintenance plan optimization set, and give priority to retaining maintenance plans that are compatible with the current equipment structure attributes and have a historical repair success rate higher than a preset threshold; perform time series interpolation and fusion on the secondary sorted maintenance plan optimization set and the maintenance node planning plans in the equipment management strategy set to generate an enhanced equipment management strategy that includes time dimension, resource dimension and equipment compatibility dimension.

[0160] In step 341, the equipment status data management system obtains a prioritized list of candidate maintenance plans and extracts information about planned maintenance nodes from the equipment management policy set. For the centrifuge bearing replacement plan (priority P0) as an example, the equipment status data management system matches its execution efficiency indicator (estimated time: 2.3 hours) with the equipment idle time periods in the planned maintenance node.

[0161] By analyzing resource availability during the equipment cleaning period at the end of the fifth production cycle (02:00-04:00 on May 18), the equipment status data management system mapped this plan to the time window: three maintenance personnel, specialized hydraulic puller tools, and inventory of SKF 6318 bearings were assigned. Simultaneously, the extraction tank seal maintenance plan (priority P2) was mapped to the suboptimal time window of 10:00-11:30 on May 19. During this time, lubricant inventory remained sufficient, but the vibration analyzer was already occupied by centrifuge maintenance. The equipment status data management system automatically noted that equipment resources needed to be coordinated.

[0162] In step 342, the equipment status data management system performs resource conflict detection and matching analysis. For example, the equipment status data management system detects that both the vacuum pump maintenance plan (priority P1) and the concentration tank maintenance plan require the use of a vacuum detector between 14:00 and 16:00 on May 18, generating a resource conflict warning code C-087. Calculated by the resource matching algorithm, the vacuum pump maintenance plan matches the spare parts inventory for the current period by 92%, while the concentration tank plan only matches by 65%. The equipment status data management system outputs the conflict detection result and recommends that vacuum pump maintenance be performed first. For the centrifuge spindle calibration plan, the equipment status data management system verifies the reservation status of the required dynamic balancing equipment and finds that the equipment has been occupied in the alcohol precipitation section maintenance task, generating a resource mismatch indicator R-112 and triggering the dynamic adjustment process.

[0163] In step 343, the equipment status data management system implements dynamic scheduling optimization for the conflicting plans. In view of the instrument conflict between the concentration tank overhaul and the vacuum pump maintenance, the equipment status data management system divides the concentration tank plan into two stages: 14:00-15:00 to perform mechanical component replacement without vacuum detector, and 15:30-16:30 to complete the sealing test after the instrument is released. The adjusted plan was verified by the digital twin system, confirming that the parameter adjustment instruction (increasing the concentration tank vacuum setting value from -0.092MPa to -0.089MPa) can maintain the production line throughput at 133kg / h, which is higher than the preset guarantee value of 130kg / h. The system screened out 6 conflict-free plans to form an optimization set, among which the centrifuge bearing replacement plan was marked as executable because the resources were fully matched and the simulation verification passed.

[0164] In step 344, the equipment status data management system combines the knowledge graph with solution optimization and strategy integration. For example, the equipment status data management system searches the equipment status management knowledge graph and finds that the centrifuge SKF 6318 bearing replacement solution has a 96% success rate in 23 applications over the past two years and is fully compatible with the current equipment model. It ranks it first in the optimization set.

[0165] For the extraction tank sealing ring solution, the knowledge graph shows that the average service life of polytetrafluoroethylene seals under high-temperature conditions is 38% longer than that of fluororubber seals. Based on this, the equipment status data management system generates material upgrade recommendations and updates the solution parameters.

[0166] Ultimately, the equipment status data management system incorporated eight optimization plans into the maintenance plan in a time-sequential manner: A centrifuge bearing replacement was performed between 2:00 AM and 4:00 AM on May 18th, accompanied by a simultaneous adjustment of the feed pump speed reduction instruction; an extraction tank seal upgrade was performed at 10:00 AM on May 19th, accompanied by adjustments to the agitation parameters in the alcohol precipitation process. The resulting enhanced strategy included 14 maintenance instructions with minute-level timestamps, nine production parameter compensation rules, and three cross-shift resource allocation plans. This strategy was then distributed to the shop floor execution terminal via the OPC UA protocol.

[0167] Thus, the maintenance strategy optimization mechanism constructed in steps 341-344 achieves intelligent coordination between equipment maintenance plans and production scheduling. By establishing a dynamic mapping between candidate plans and maintenance time windows, the system can accurately identify the matching relationship between resource demand and supply, and utilize conflict detection algorithms to proactively mitigate the risk of resource competition during equipment maintenance. While ensuring production continuity, the dynamic adjustment mechanism maximizes maintenance efficiency through time window reallocation and resource splitting. Combined with digital twin verification, it ensures stable production capacity after parameter adjustments, forming a set of optimized solutions that balance maintenance effectiveness and production assurance.

[0168] It is worth mentioning that step 344 significantly improves the engineering adaptability of the maintenance strategy through secondary optimization driven by the knowledge graph. The correlation and matching of equipment structure attributes and historical maintenance records ensures that the preferred solution has equipment compatibility and high success rate verification. Time series interpolation and fusion technology converts discrete maintenance instructions into a time-series operation blueprint. The generated enhanced strategy integrates multi-dimensional decision factors, realizes the deep coupling of maintenance resource scheduling, production process compensation and equipment structure characteristics, and forms a dynamic management solution that can accurately guide on-site execution. It effectively solves the pain points of frequent resource conflicts and low solution adaptability in traditional maintenance planning, and greatly improves the systematization level of equipment life cycle management.

[0169] Based on the same inventive concept, the present application also provides a device status data management system. Figure 2 As shown, it is a structural diagram of a possible device status data management system provided in an embodiment of the present application. Figure 2 In the embodiment, the device status data management system 200 includes a processor 210 and a memory 220. The memory 220 stores a computer program executable by the processor 210. The processor 210 executes the instructions stored in the memory 220 to perform the steps of the device status data management method applied to the automated astragalus extract production line.

[0170] Based on the same inventive concept, an embodiment of the present application provides a computer-readable storage medium, which includes a computer program. When the computer program is run on a device status data management system, the computer program is used to enable the device status data management system to execute the steps of the device status data management method applied to the automated astragalus extract production line. In some possible implementations, various aspects of the device status data management method applied to the automated astragalus extract production line provided by the present application can also be implemented in the form of a program product, which includes a computer program. When the program product is run on a device status data management system, the computer program is used to enable the device status data management system to execute the steps of the device status data management method applied to the automated astragalus extract production line. For example, the device status data management system can execute the following steps: Figure 1 Follow the steps shown in .

Claims

1. A device status data management method for an automated Astragalus extract production line, characterized in that: The method is executed by a device status data management system, and the method includes: Acquire a set of production line equipment status data, wherein the set of production line equipment status data includes production parameters, operating status parameters, and environmental monitoring parameters of each device in multiple continuous production cycles of the automated astragalus extract production line; Extracting a production line equipment status feature set from the production line equipment status data set, wherein the production line equipment status feature set includes multi-dimensional time series features reflecting equipment operation trends and topological features of association relationships between devices; Training an Astragalus Extract production line state prediction model based on the production line equipment state feature set, wherein the Astragalus Extract production line state prediction model is used to predict the probability of equipment abnormality and performance degradation trend within a selected time period based on the current equipment state features; An equipment management strategy set is generated according to the output result of the astragalus extract production line status prediction model, and the equipment management strategy set includes production parameter adjustment instructions and maintenance node planning schemes for different equipment maintenance priorities.

2. The method according to claim 1, wherein The extracting of a production line equipment status feature set from the production line equipment status data set includes: Segmenting the production line equipment status data set to obtain equipment status data subsets corresponding to each production cycle; Extracting equipment operation trend characteristics for each equipment status data subset, wherein the equipment operation trend characteristics include the fluctuation amplitude of the production parameters within a continuous time window, the parameter change rate, and the cumulative duration of the parameter deviation from the standard threshold; Generate a production line topology network based on the physical connection relationship between devices, and extract inter-device association features based on the production line topology network. The inter-device association features include a synchronization index for parameter transmission between adjacent devices and a coupling index for the operating status of a device group. The equipment operation trend characteristics are integrated with the inter-equipment correlation characteristics to obtain the production line equipment status feature set.

3. The method according to claim 2, wherein The training of the Astragalus Extract production line state prediction model based on the production line equipment state feature set includes: Dividing the production line equipment status feature set into a training data set and a validation data set, wherein the training data set includes historical equipment status features and corresponding equipment maintenance records; Generate a multi-layer neural network model that includes a temporal attention mechanism to capture the weight distribution of device state features at different time steps; A dynamic graph convolution module is used to process the inter-device association features, and the dynamic graph convolution module is used to update the node connection weights of the production line topology network according to the real-time changes in the equipment operation status; Iteratively adjust the model parameters of the multi-layer neural network model by jointly optimizing the prediction loss function and the association constraint function, wherein the prediction loss function is used to measure the prediction error of the device abnormality probability, and the association constraint function is used to constrain the contribution of the correlation features between devices to the prediction results to meet the preset threshold; When the prediction accuracy of the verification data set reaches the convergence condition, the training of the multi-layer neural network model is stopped and the astragalus extract production line status prediction model is output.

4. The method according to claim 3, wherein Generating a set of equipment management strategies according to the output results of the Astragalus Extract production line status prediction model includes: Determine the maintenance priority of each device based on the distribution of equipment abnormality probability output by the astragalus extract production line status prediction model; Classifying equipment maintenance types based on the analysis results of the performance degradation trend, wherein the maintenance types include emergency shutdown maintenance, production interval maintenance, and preventive maintenance; Match different equipment maintenance types with the preset production parameter adjustment rule library to generate parameter adjustment instructions that are adapted to the current production line operation status; Combining the maintenance priority ranking with production plan data, allocating maintenance resources within a preset maintenance time window and generating a maintenance node planning scheme; The parameter adjustment instruction is logically verified with the maintenance node planning scheme, and the device management policy set is output after excluding policies with resource conflicts or parameter contradictions.

5. The method according to claim 4, wherein The method further comprises: The latest equipment status data of the automated astragalus extract production line is obtained in real time, and the current equipment status features corresponding to the latest equipment status data are extracted and input into the astragalus extract production line status prediction model; When the astragalus extract production line status prediction model predicts that the abnormal probability of the target equipment exceeds a dynamic threshold, a real-time alarm is triggered and the maintenance priority ranking in the equipment management policy set is updated; Dynamically adjust the ongoing maintenance node planning scheme based on the updated maintenance priority ranking and reallocate unused maintenance resources to target equipment; The adjusted maintenance node planning scheme is synchronously calibrated with the current production parameter adjustment instructions to ensure that the production line throughput after parameter adjustment is not lower than the preset guarantee value.

6. The method according to claim 5, wherein The method for determining the dynamic threshold includes: Collect statistics on the distribution of equipment status characteristics in the target time period before the occurrence of historical equipment abnormal events; Calculate the maximum allowable offset between the characteristic value offset of each device status feature in the device status feature distribution before the historical device abnormal event occurs and the normal production cycle; Determining abnormal probability critical values at different confidence levels based on the cumulative distribution function of the eigenvalue offset; According to the key indicator requirements of the current production stage, a confidence level that matches the production line stability requirements is dynamically selected, and the abnormal probability critical value corresponding to the matching confidence level is set as the dynamic threshold.

7. The method according to claim 1, wherein The method further comprises: Generate a device status management knowledge graph, which includes the relationship between device structure attributes, historical maintenance records, and fault solutions; Matching and searching the output result of the Astragalus Extract production line state prediction model with the equipment state management knowledge graph to obtain a set of candidate maintenance solutions related to the current predicted equipment state; Based on the historical evaluation data of the maintenance plan execution effect, candidate maintenance plans are ranked, and maintenance plans with execution efficiency higher than the preset benchmark and resource consumption lower than the preset upper limit are given priority; The ranked candidate maintenance plans are integrated with the equipment management strategy set to generate an enhanced equipment management strategy containing multi-dimensional decision-making basis.

8. The method according to claim 7, wherein Generating the device state management knowledge graph includes: Extract the equipment component structure tree and the functional dependencies between components from the production line equipment database; Parse the fault description text in historical work order data and use entity recognition technology to mark the faulty equipment components and the maintenance measures taken for the faulty equipment components; Generate a heterogeneous relationship network of equipment components, failure modes, and maintenance measures, and add the mean time to repair and spare parts consumption attributes corresponding to each maintenance measure in the heterogeneous relationship network; Performing embedding representation learning on the heterogeneous relational network through a graph neural network to obtain an embedding vector set reflecting the relevance of device status management knowledge; A knowledge graph query interface supporting multi-hop reasoning is generated based on the embedding vector set.

9. The method according to claim 1, wherein The method further comprises: Performing sensor monitoring on at least a portion of the equipment corresponding to the automated astragalus extract production line using a multimodal sensor array, and collecting multimodal sensor data corresponding to at least a portion of the equipment in real time, the multimodal sensor data including equipment vibration spectrum data, infrared thermal imaging data, and voiceprint signal data; Performing a joint time-frequency analysis on the multimodal sensor data to extract sensitive features of the equipment's mechanical state; Performing cross-modal feature alignment on the mechanical state sensitive feature and the production line equipment state feature set to obtain a cross-modal feature difference; When the cross-modal feature difference exceeds a preset tolerance, triggering an equipment status review process and correcting the input features of the astragalus extract production line status prediction model; Regenerate a set of equipment management correction strategies based on the revised output results of the Astragalus Extract production line status prediction model, and enter the feature difference traceability labels corresponding to the cross-modal feature difference into the generated equipment status management knowledge graph; The performing a time-frequency joint analysis on the multimodal sensor data to extract sensitive features of the equipment mechanical state includes: Performing wavelet packet decomposition on the vibration spectrum data of the equipment to extract energy distribution entropy of different frequency bands, and using the energy distribution entropy as an indicator of mechanical wear of the equipment; Performing regional temperature gradient analysis on the infrared thermal imaging data to determine the area growth rate of the thermal anomaly area on the surface of the equipment; Extract the Mel-frequency cepstral coefficients from the voiceprint signal data and use the hidden Markov model to identify abnormal noise patterns during device operation. Combining the equipment mechanical wear index, the area growth rate, and the equipment operation abnormal noise pattern, generating a joint health assessment function of multimodal sensitive features; wherein the joint health assessment function at least integrates the frequency band energy distribution entropy, the thermal anomaly growth rate, and the abnormal noise pattern severity score; When the function value of the joint health evaluation function is lower than a preset health threshold, it is determined that the mechanical state of the equipment is abnormal and a review task priority label is generated.

10. A device status data management system, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor is enabled to perform the steps of any one of the methods of claims 1 to 9.

Images