Real-time monitoring and safety warning method for anti-seismic structure of oil engine unit

Abstract

The application discloses a kind of real-time monitoring and safety early warning method of fuel unit anti-seismic structure, belong to structural health monitoring and safety control technical field, to solve the problem of passive, lag, rely on artificial experience of existing monitoring method, cannot automatically, quickly, objectively evaluate the health condition of anti-seismic system after earthquake.The real-time monitoring vibration acceleration is carried out using triaxial acceleration sensor, edge computing gateway triggers earthquake event when vibration peak exceeds the first preset threshold, and controls sensor array to switch to high-frequency recording mode;The high-frequency data collected is uploaded to the data processing platform, and the anti-seismic design threshold is compared and the machine learning model is analyzed, and the output component damage probability level and health evaluation result are fused;Based on the evaluation result, generate a targeted inspection checklist through the preset mapping relationship library and push it to the mobile terminal of operation and maintenance personnel.The fuel unit anti-seismic structure monitoring and early warning can realize fast and reliable evaluation and accurate operation and maintenance guidance after earthquake, and ensure the power supply recovery of key facilities.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of structural health monitoring and safety control, in particular to a kind of real-time monitoring and safety warning method of fuel unit anti-seismic structure based on edge computing and data fusion, it belongs to general control or regulating system, and monitoring or testing device for such system. BACKGROUND

[0002] Fuel unit (such as diesel generator set) is the emergency power supply guarantee of important facilities such as data center, hospital, communication hub. Its reliability under extreme natural disasters such as earthquake is crucial. In order to ensure that the unit is not damaged and can run normally during earthquake, a complex anti-seismic system will be designed for it, including inertia base, anchoring component, shock absorber, etc.

[0003] At present, the state management of the anti-seismic system mainly depends on periodic visual inspection and manual measurement (such as checking anchor bolt tightening force with torque wrench). This method has obvious disadvantages: passive lag, unable to real-time perceive the hidden damage caused by earthquake event to anti-seismic system, long inspection period, unable to quickly evaluate the availability of unit after disaster. The accuracy of inspection results is highly dependent on the experience of operation and maintenance personnel, and lacks objective and quantitative data support. Each inspection needs to be fully checked, cannot highlight the key points, consumes a lot of manpower and time, and delays the power recovery of critical facilities in emergency situations. It is difficult to track and analyze the performance degradation of anti-seismic system in the whole life cycle. Therefore, there is an urgent need in the field for an automatic processing method that can actively warn, intelligently evaluate and accurately guide operation and maintenance. SUMMARY

[0004] It is an object of the present application to solve at least the above problems and to provide at least the advantages stated further below.

[0005] In order to achieve these objects and other advantages in accordance with the present application, a real-time monitoring and safety warning method of fuel unit anti-seismic structure is provided, comprising the following steps:

[0006] S1, real-time monitor vibration acceleration by three-axis acceleration sensor arranged on fuel unit rigid base and building foundation; when edge computing gateway monitors that the peak value of vibration acceleration lasts more than first preset threshold value, determine that earthquake event occurs, and control three-axis acceleration sensor, and micro-strain sensor and torque sensor arranged on anti-seismic anchoring component to switch from low-frequency monitoring mode to preset high-frequency recording mode for data acquisition;

[0007] S2, upload high-frequency acceleration data, strain data and pretightening force data collected during earthquake event to data processing platform, data processing platform compares acceleration data, strain data and pretightening force data with preset anti-seismic design threshold value respectively, to obtain threshold comparison result;

[0008] Meanwhile, the acceleration data, the strain data and the pretightening force data are input as input features into a pre-trained machine learning classification model, the machine learning classification model taking historical seismic events or vibration data of the unit in a simulated shaking table test and confirmed structural damage states as training samples, and outputting damage probability levels of different components of the seismic structure;

[0009] The threshold comparison result and the damage probability level are fused to generate a health assessment result including specific damaged components, damage types and comprehensive risk levels;

[0010] S3, based on the health assessment result, if a component at risk is identified, the damaged component and the damage type are mapped to corresponding specific inspection actions according to a pre-set mapping relationship library, and a targeted inspection list is generated, the inspection list listing components and inspection items that need to be inspected first, and pre-warning information containing the electronic inspection list is pushed to a mobile terminal of an operation and maintenance personnel.

[0011] Preferably, in step S2, the data processing platform stores the data of the current seismic event, the health assessment result and the subsequent on-site inspection feedback of the operation and maintenance personnel, forms a case library, and periodically re-trains the machine learning evaluation model using the updated case library.

[0012] Preferably, the damage level includes at least four levels of normal, attention, warning and danger, and corresponds to different pre-warning colors and disposal suggestions, and when the level is warning or danger, the automatic start function of the fuel unit is locked.

[0013] Preferably, the training samples of the machine learning classification model are obtained and labeled in the following way:

[0014] The vibration data, strain data and pretightening force data of the fuel unit or similar units in a simulated shaking table test under seismic wave excitation of different acceleration peaks are obtained;

[0015] Based on the results of non-destructive testing of the seismic anchoring components after the test, the non-destructive testing includes at least one of magnetic powder testing, penetration testing or ultrasonic testing, and the damage state is labeled as no damage, plastic deformation, crack or loose bolt.

[0016] Preferably, before the acceleration data, the strain data and the pretightening force data are input as input features into the machine learning classification model, the data is pre-processed by feature standardization, and the feature standardization includes normalizing the data of various sensors to the same numerical interval.

[0017] Preferably, in step S2, a multi-parameter correlation analysis is further included, which comprises extracting the acceleration data sequence and the strain data sequence synchronously collected within the time history of the seismic event for the same seismic anchoring component, calculating the Pearson correlation coefficient between the acceleration data sequence and the strain data sequence, and comparing the Pearson correlation coefficient with a preset correlation coefficient threshold value, wherein if the Pearson correlation coefficient is lower than the correlation coefficient threshold value, it is determined that the seismic anchoring component has the risk of stiffness degradation, component connection failure, or anchoring loosening.

[0018] The preset correlation coefficient threshold value is determined by analyzing the Pearson correlation coefficients of the acceleration and strain data sequences of the fuel engine unit or similar units in the simulated shaking table test under normal undamaged state for multiple times, and taking the lower limit of the statistical confidence interval of the series of Pearson correlation coefficients.

[0019] Preferably, the mapping relationship library is constructed in the form of a decision tree or a production rule, and maps the damaged component and the damage type to the corresponding specific inspection action, including:

[0020] The damaged component identification, damage type, and comprehensive risk level in the health assessment result are taken as the input conditions for reasoning;

[0021] According to the rule matching and reasoning of the mapping relationship library, an inspection action sequence is dynamically generated, wherein the priority of the inspection action sequence is determined based on the cost and the dependency relationship: low-cost inspection including visual inspection and size measurement is given priority over high-cost inspection including non-destructive testing; at the same time, the inspection item corresponding to the highest risk level is given the highest priority.

[0022] Preferably, in step S1, the edge computing gateway determines that the seismic event occurs on the premise that the peak value of the vibration acceleration lasts more than a first preset threshold value, and further analyzes the vibration signal in the frequency domain and the time domain;

[0023] The frequency domain analysis specifically calculates the power spectral density of the vibration signal, and determines whether the signal energy is concentrated in a preset first frequency range; the time domain analysis specifically determines whether the cumulative duration of the vibration acceleration exceeding a second preset threshold value reaches a preset duration threshold value;

[0024] The first frequency range is determined according to the main energy distribution range of the seismic motion acceleration power spectral density in the historical earthquake record; the duration threshold value is determined according to the typical duration of the strong earthquake stage in the historical earthquake record.

[0025] Preferably, in step S1, after the edge computing gateway controls the sensor array to switch to the high-frequency recording mode, the following steps are further performed to determine the termination time of data acquisition:

[0026] An integral value of acceleration square of a vibration signal collected by a three-axis acceleration sensor is continuously calculated as an index representing cumulative vibration energy, and based on this, an average change rate of the integral value is calculated within a preset evaluation time window to obtain an energy release rate;

[0027] When the energy release rate is monitored to be lower than or equal to a preset energy release rate threshold and lasts for a preset stable time window, it is determined that the main energy of the earthquake has been released, and the sensor array is controlled to switch back to the low-frequency monitoring mode from the high-frequency recording mode; wherein the energy release rate threshold is determined by statistical analysis of the energy release rate after the strong shock segment in the historical earthquake record.

[0028] Preferably, after the sensor array is controlled to switch back to the low-frequency monitoring mode from the high-frequency recording mode, the edge computing gateway starts a preset duration of a lock period, during which the edge computing gateway temporarily increases the acceleration peak threshold for event determination from the first preset threshold to a predetermined lock period threshold, and suspends the frequency domain and time domain analysis functions, and only determines whether a new earthquake event occurs based on whether the vibration acceleration peak value exceeds the lock period threshold;

[0029] After the lock period expires, the edge computing gateway restores the acceleration peak threshold for event determination to the first preset threshold, and re-enables the frequency domain and time domain analysis functions;

[0030] Wherein, the lock period threshold is determined by statistical analysis of the acceleration peak ratio of the main shock and the largest aftershock in the historical earthquake record, and the duration of the lock period is determined by statistical analysis of the time interval distribution between the main shock and the significant aftershock in the historical earthquake sequence, and the significant aftershock is an aftershock with an acceleration peak value reaching more than 30% of the main shock peak value.

[0031] The present application at least includes the following beneficial effects:

[0032] Firstly, the present application fundamentally changes the passive mode of relying on regular manual inspection by implementing automatic event triggering and high-frequency data acquisition mechanism by the edge computing gateway, realizes active perception and response to earthquake events, and solves the problem of monitoring lag. Combined with the fusion of physical threshold comparison and machine learning model evaluation by the data processing platform, this method discards the limitations of single reliance on empirical criteria, provides quantitative and objective structural health state evaluation results, and significantly improves the scientificity and reliability of the evaluation results. Further, based on the risk evaluation results, a priority inspection checklist is dynamically generated to guide the operation and maintenance personnel to accurately focus on high-risk points, greatly optimizing the post-earthquake inspection process and reducing unnecessary comprehensive inspection, so as to quickly judge the availability of the unit, gain valuable time for power restoration of critical facilities, and realize intelligentization of the whole process from monitoring, evaluation to operation and maintenance guidance.

[0033] Secondly, the application forms a self-improving closed-loop optimization system by constructing a case library containing field inspection feedback results and regularly retraining the model. This mechanism enables the machine learning model to continuously learn and adapt to complex and newly emerging damage patterns in the actual operating environment, effectively overcoming the evaluation bias or insufficient generalization ability of the initial model due to limited training data. As the system runs over time, the accuracy and reliability of health status assessment are continuously improved, enabling the system to evolve over the long term, enhancing the effectiveness and adaptability of the method throughout the life cycle.

[0034] Thirdly, the application adds an independent criterion based on the consistency of component dynamic responses to damage identification through multi-parameter correlation analysis based on Pearson correlation coefficients. This method is particularly sensitive to early-stage hidden damage such as stiffness degradation, microscopic crack initiation, or slight connection loosening, which may not have caused a significant peak in a single parameter but has disrupted the linear correlation between acceleration excitation and strain response. This correlation analysis based on physical mechanisms complements and cross- validates threshold comparison and machine learning models, forming a multi-angle, multi-level evaluation system that significantly enhances the identification ability and timeliness of early-stage hidden damage, especially for complex damage patterns.

[0035] Fourthly, the application determines seismic events by combining multiple criteria such as vibration peak, frequency domain features, and duration, determines the end of recording based on vibration energy decay, and introduces a lock-in period mechanism to form a precise event management strategy. This strategy can effectively distinguish between seismic events and common industrial disturbances, significantly reducing false positives; it can also adapt to the actual duration of seismic motion, optimizing data recording and avoiding resource waste; the lock-in period mechanism prevents aftershocks from being misjudged as independent new events after the main shock, ensuring stable operation of the system under complex seismic sequences. These measures significantly improve the robustness, reliability, and economy of the entire monitoring system in real industrial environments.

[0036] Fifthly, the application intelligently maps abstract health assessment results to specific and actionable inspection action sequences through a pre-set mapping relationship library. This approach overcomes the rigidity of traditional fixed inspection lists, enabling on-demand customization and precise delivery of operation and maintenance guidance. By following the principles of low-cost inspection priority and high-risk project priority, the generated inspection list can scientifically guide the allocation of operation and maintenance resources, prioritizing rapid troubleshooting and focusing on the most critical hidden dangers, thereby significantly improving the efficiency and economy of post-earthquake inspection while ensuring inspection effectiveness and reducing over-reliance on individual experience of operation and maintenance personnel.

[0037] Other advantages, objectives, and features of the application will be apparent from the following description, and will be understood by those skilled in the art through study and practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 Figure 1 is a flow chart of a real-time monitoring and safety warning method according to an embodiment of the present application. DETAILED DESCRIPTION

[0039] The present application will be further described in conjunction with examples, so that those skilled in the art can implement the present application according to the description.

[0040] It should be noted that the experimental methods described in the following embodiments are conventional methods unless otherwise specified, and the reagents and materials can be obtained commercially unless otherwise specified.

[0041] As shown in Figure 1 The present application provides a real-time monitoring and safety warning method for a fuel unit anti-seismic structure, which comprises the following steps:

[0042] The triaxial acceleration sensor is a commercially available ICP type piezoelectric triaxial acceleration sensor, which has a measurement range of ±50g and a frequency response range of 0.5-10 kHz. It is rigidly fixed and installed on the joint surface between the rigid base of the fuel unit and the building foundation by bolts. Preferably, at least two sensors are symmetrically arranged near the four corner points of the rectangular base to ensure that the vibration response in different directions can be effectively captured. The micro-strain sensor is a resistance strain gauge type sensor, which has a range of ±5000µm / m. It is directly pasted on the surface of the key anti-seismic anchoring component (such as a foundation bolt or an anchor rod), and the sensitive grid direction is consistent with the main stress direction of the component (usually axial). At least one micro-strain sensor is installed on each monitored anchoring component. The torque sensor is a flange type or sleeve type torque sensor, which has a range that can be selected according to the design preload, for example, 0-1000 . It is installed between the fastening nut of the anchoring component and the supporting surface, or a smart bolt with measurement function is used to replace the original fastening component for real-time monitoring of the preload state.

[0043] The edge computing gateway is an industrial-grade edge computing gateway device, which has the functions of multi-channel data acquisition, real-time signal processing, logical judgment and communication. The edge computing gateway is usually installed in the electric control cabinet matched with the fuel unit, connected with the above-mentioned various sensors through shielded cables, and is responsible for initial data processing and event triggering.

[0044] The data processing platform can be deployed on a local server or a cloud infrastructure. The platform software can be built by using Python programming language combined with Scikit-learn machine learning library to construct data processing and classification model. The data processing platform communicates with the edge computing gateway through wired network (such as Ethernet) or wireless network (such as 4G / 5G).

[0045] The first preset threshold is used for the initial judgment of the vibration acceleration peak of the seismic event. A typical value of the first preset threshold can be set to 0.1g (about 0.98m / s 2 ). Wherein, g is the acceleration of gravity, and the value is about 9.8m / s 2 . According to the seismic background noise level and seismic fortification requirements of the specific installation site, the first preset threshold can be adjusted in the range of 0.05-0.2g (about 0.49-1.96m / s 2 ). The determination of the first preset threshold can be based on the historical seismic observation data or environmental vibration monitoring data of the installation site, and the statistical distribution of the acceleration peak is analyzed. Usually, a certain percentile value (for example, 95% percentile) slightly higher than the background vibration level under normal working conditions is selected to balance the triggering sensitivity and anti-interference ability.

[0046] The damage probability level division divides the health state of the component into four discrete damage probability levels, and associates different warning colors and disposal suggestions:

[0047] Normal: Damage probability less than 0.3, corresponding to green identification, no special disposal is needed;

[0048] Attention: Damage probability between 0.3 (inclusive) and 0.6, corresponding to yellow identification, it is suggested to pay more attention;

[0049] Warning: Damage probability between 0.6 (inclusive) and 0.8, corresponding to orange identification, the automatic start function of the fuel engine unit is automatically locked, and an inspection list is generated;

[0050] Danger: Damage probability greater than or equal to 0.8, corresponding to red identification, the start function is immediately locked, and a high risk is prompted, which needs to be disposed of in priority;

[0051] When the evaluation level is "warning" or "danger", the edge computing gateway or the application layer of the data processing platform sends a chain signal to the control system (such as PLC) of the unit to prohibit the automatic start of the unit until the on-site inspection of the operation and maintenance personnel is confirmed and manually reset.

[0052] Before inputting the acceleration data, strain data and pretightening force data into the machine learning classification model, feature standardization preprocessing is needed to normalize the data from different physical dimension sensors to the same numerical interval (for example, [0, 1] interval). Specifically, the minimum-maximum normalization method can be used, and the formula is as follows: Wherein, x norm is the sensor data value after the minimum-maximum normalization processing; x is the original sensor reading, x min and x maxThe minimum and maximum values of the sensor within the effective range (e.g. -50g and +50g for an acceleration sensor, -5000 pm / m and +5000 pm / m for a micro-strain sensor).

[0053] The specific steps include:

[0054] Step S1, seismic event trigger and data acquisition mode switching. The edge computing gateway continuously reads the vibration acceleration data of the three-axis acceleration sensor at a preset sampling frequency (e.g. 100 Hz), and calculates the acceleration peak value in real time.

[0055] The event determination condition is that the monitored vibration acceleration peak value continuously exceeds the first preset threshold value (e.g. 0.1g) for a certain length of time (e.g. 5 consecutive sampling points).

[0056] Once the event determination condition is met, the edge computing gateway immediately issues an instruction to the sensor array to control the three-axis acceleration sensor, micro-strain sensor and torque sensor to automatically switch from the low-frequency monitoring mode (e.g. sampling rate 100 Hz) to the preset high-frequency recording mode (e.g. sampling rate 1 kHz), and start recording high-frequency acceleration data, strain data and pre-tightening force data during the seismic event.

[0057] Step S2: data fusion and health state assessment. After the seismic event ends (or through streaming), the edge computing gateway uploads the high-frequency data collected during the event to the data processing platform.

[0058] The data processing platform performs the following analysis in parallel:

[0059] The received acceleration data peak value, strain data peak value and pre-tightening force data instantaneous value are compared with the preset seismic resistance design threshold values. These design threshold values are determined based on the unit seismic resistance design specification (e.g. the acceleration design threshold value can be 2.5 m / s 2 , the strain design threshold value can be 3000 pm / m, and the pre-tightening force design threshold value can be 800 ). A simple "overrun" or "normal" threshold comparison result is generated.

[0060] The acceleration data, strain data and pre-tightening force data after feature standardization preprocessing are input as input features into a pre-trained machine learning classification model (such as a random forest classifier). The model is trained using historical seismic event data or simulated shaking table test data (containing unit vibration data, strain data, pre-tightening force data and their corresponding structure damage state labels confirmed by non-destructive testing, such as no damage, plastic deformation, crack, bolt loosening). The model output is the damage probability value of each key component of the seismic resistant structure.

[0061] The data processing platform fuses the threshold comparison results and the damage probability level output by the machine learning model. For example, if the strain data of a certain anchoring component exceeds the design threshold, and the model gives a damage probability of 0.85, the fusion generates a health assessment result of the component in the "dangerous" state, indicating the damaged component, the possible damage type (such as plastic deformation), and the comprehensive risk level.

[0062] Step S3: Intelligent early warning and inspection list generation. If the health assessment result identifies a component with risk (in the "attention", "warning" or "danger" state), reasoning is performed according to the pre-set mapping relationship library (the mapping relationship library can be constructed in the form of a decision tree).

[0063] The mapping relationship library takes the specific damaged component identifier, damage type and comprehensive risk level in the health assessment result as input conditions, matches and dynamically generates a series of specific inspection action sequences. The priority setting principle of the inspection action is: low-cost inspection (such as visual inspection, basic size measurement) is preferred to high-cost inspection (such as ultrasonic detection, magnetic powder detection); at the same time, the inspection item corresponding to the highest risk level is given the highest execution priority.

[0064] Finally, a targeted electronic inspection list is generated, and the early warning information containing the list is sent to the mobile terminal of the designated operation and maintenance personnel through the push module (which can integrate channels such as SMS, WeChat for enterprises, and special APP). For "warning" and above levels, the unit start lock is executed synchronously.

[0065] The embodiment automatically triggers high-frequency monitoring through real-time sensor data, overcoming the passivity and lag of traditional periodic manual inspection. Combined with physical threshold comparison and data-driven machine learning models, it provides quantitative and objective structural health state assessment, reducing the dependence on manual experience. Based on the risk assessment results, a focused inspection list is dynamically generated, significantly improving the efficiency and pertinence of post-earthquake inspection, and facilitating the rapid recovery of power supply for key facilities. The modular hardware selection and parameterized configuration make the scheme easy to implement and expand on different types and installation environments of oil-fired units.

[0066] In another embodiment of the present application, in step S2, after generating the health assessment result, the following key operations are included:

[0067] The data processing platform correlates the high-frequency acceleration data, strain data and pre-tightening force data collected after the earthquake event, the health assessment results generated after platform analysis (including specific damaged components, damage types and comprehensive risk levels), and the on-site inspection feedback results fed back by the operation and maintenance personnel through their mobile terminals (such as smart phones or tablet computers installed with a special APP) after on-site inspection according to the inspection checklist pushed by the system. The on-site inspection feedback results should at least include: verification of the health assessment results (such as confirming whether the damage exists and its specific form), actual inspection measurement data (such as crack length, bolt torque value), measures taken (such as no need to handle, tighten, repair or replace) and related image records.

[0068] These correlated data constitute a complete case with clear input (monitoring data) and output (real damage state) verified by practice. The data processing platform stores each case in a structured form (for example, in JSON or XML format) in the case library specially set up in the data storage module of the platform. The case library can be implemented based on common database systems, for example, open source MySQL relational database or MongoDB document database, deployed on a local server or cloud platform (such as Ali Cloud RDS instance).

[0069] To continuously improve the accuracy of the assessment, a model retraining trigger mechanism is set. This mechanism can be based on one or a combination of the following two conditions:

[0070] Time period trigger, set a fixed time interval, for example, automatically start the retraining process once every 3 months (about 90 days) or every 6 months;

[0071] Case quantity trigger, set a threshold for the number of new cases in the case library, for example, trigger retraining when the number of new valid cases reaches 50 or 100;

[0072] When the retraining condition is met, the data processing platform automatically extracts all case data from the case library to form a new training data set. The input features of this data set are the acceleration, strain, and pretension data collected during historical seismic events (which need to be preprocessed by feature standardization), and the corresponding labels are the real structure damage states confirmed by on-site inspection feedback results (such as "no damage", "plastic deformation", "crack", and "loose bolt"). Subsequently, the machine learning classification model (for example, using the random forest classifier or support vector machine model in the Scikit-learn library) is retrained using this updated data set. The retraining process includes relearning and optimization of model parameters. After training is complete, the performance indicators (such as accuracy, precision, and recall) of the new model are evaluated using a reserved test set or cross-validation method. If the performance of the new model is better than or equal to the current online running model, the new model is updated and deployed to the multi-source data fusion and intelligent decision engine of the platform, replacing the old model for subsequent seismic event health assessment.

[0073] The present embodiment can utilize the continuously accumulated real case data verified by on-site practice, enabling the machine learning model to continuously learn and adapt to complex damage patterns and unit characteristics in the actual operating environment, thereby gradually improving the accuracy and reliability of health state assessment. The case library covers different seismic event characteristics and unit responses, which helps the model better understand various potential risks, enhances its judgment ability for scenarios that have not been experienced, and improves its robustness in different application environments. The monitoring-evaluation-early warning-on-site inspection-feedback links are connected in a closed loop, so that the evaluation results can be tested in the field, and the test results can be fed back for model optimization, forming an intelligent system that continuously improves itself. The structured case library not only serves as a data basis for model optimization, but also provides valuable empirical data support for subsequent in-depth analysis of the evolution of the seismic performance of the oil-fired unit and optimization of the seismic design. The case library construction and model retraining mechanism greatly enhance the long-term effectiveness and intelligence level of the present method, making it an intelligent monitoring and early warning system that can evolve and improve with the accumulation of practical experience.

[0074] In another embodiment of the present application, the real-time monitoring and safety warning method for the seismic structure of the oil-fired unit comprises:

[0075] The commercial large-scale seismic simulation shaking table system is used. The shaking table should have three-directional six-degree-of-freedom movement capability, the table size should be able to accommodate the fuel unit or its scaled model to be tested, the maximum load capacity should not be less than 10 tons, the frequency range should cover at least 0.1-50 Hz, and the maximum acceleration should not be less than 1.0g. The shaking table body is fixedly installed on a dedicated anti-seismic foundation and located inside the structural laboratory. The fuel unit (which can be a real unit or a full-scale / scaled model designed according to the similarity law) is rigidly fixed on the shaking table surface through its original anti-seismic base and anchoring components according to the actual installation specification.

[0076] On the fuel unit fixed on the shaking table, an array of sensors is installed, including:

[0077] Three-axis acceleration sensors are installed at key positions of the unit rigid base and upper structure to collect unit vibration data (acceleration response);

[0078] Micro-strain sensors are pasted on the surface of the anti-seismic anchoring components (anchor bolts, anchor rods);

[0079] Torque sensors are installed at the fastening positions of the anchoring components to monitor pre-tightening force data;

[0080] All sensors are connected to a high-speed data acquisition instrument through data lines, and the sampling rate of the acquisition instrument should be set to not less than 1 kHz to ensure that high-frequency dynamic response can be captured.

[0081] Commercial non-destructive testing equipment is used to confirm and mark the damage state of the anti-seismic anchoring components after the test, including: magnetic powder detection equipment suitable for ferromagnetic material surface crack detection; penetration detection equipment suitable for non-porous material surface defect detection; ultrasonic detection equipment suitable for internal defect and crack depth detection.

[0082] After the test, the anti-seismic anchoring components removed from the shaking table are detected in the laboratory environment.

[0083] The specific steps include:

[0084] A series of simulated seismic excitation is designed. Typical seismic wave records (such as El-Centro wave, Kobe wave, Wenchuan wave, etc.) with different acceleration peak values (PGA) are selected as the input waves of the shaking table. The PGA gradient setting should cover the expected seismic level, for example, set to 0.1g, 0.2g, 0.4g, 0.6g, 0.8g, etc. Multiple repeated tests can be performed at each PGA level to consider uncertainties.

[0085] During each shaking table test, the data acquisition instrument is started to record the vibration data (acceleration time history), strain data (strain time history) and pre-tightening force data (torque / force time history) of all sensors under each seismic wave excitation at a high frequency (e.g. 1 kHz).

[0086] After each shaking table test, the key anti-seismic anchoring components of the fuel unit (such as anchor bolts, anchor rods) are disassembled. These components are then subjected to detailed testing using selected non-destructive testing equipment. For example, visual inspection is first performed, then magnetic powder testing is used to check for surface cracks, and ultrasonic testing is used to detect internal defects or measure crack depth if necessary. For pre-tightening force changes, verification can be performed in combination with torque sensor readings and manual torque wrench re-inspections.

[0087] Based on the non-destructive testing results, the structural damage state of each monitored component is confirmed by experienced testing personnel or structural engineers. The damage state is labeled according to the detected physical defects, with the main categories including:

[0088] No damage: No plastic deformation, cracks or looseness are found;

[0089] Plastic deformation: The component has undergone irreversible deformation but no cracks have formed;

[0090] Cracks: Macroscopic or microscopic cracks have appeared on the surface or inside the component;

[0091] Loose bolt: The pre-tightening force of the anchor bolt has been significantly lost, exceeding the allowable range;

[0092] This labeling result is the true label corresponding to the test data of this test.

[0093] The sensor data collected during each test (as input features) and the corresponding damage state label confirmed by non-destructive testing are matched one-to-one to form a training sample. By aggregating data from all tests (different seismic waves, different PGAs), a training sample library covering multiple seismic excitation levels and damage states is constructed. Each sample contains complete time history data or feature vectors extracted therefrom, as well as its corresponding discrete damage state label.

[0094] The PGA of the seismic wave is selected in the range of 0.05-0.8g, covering the level from small earthquakes to strong earthquakes. The specific range can be adjusted according to the fortification intensity of the region where the fuel unit is located, and the sampling frequency should not be less than 1 kHz to meet the needs of high-frequency component acquisition and subsequent feature extraction. The determination of the damage state should be based on existing national or industry standards, such as GB / T9443 "Magnetic Particle Testing of Cast Steel" and GB / T18851 "Non-destructive Testing - Penetrant Testing", to ensure the objectivity and accuracy of the labeling.

[0095] The training sample obtained and labeled by the method of the embodiment can provide a high-quality and high-reliability learning basis for a machine learning classification model. The input features (vibration, strain, and pre-tightening force data) in the training sample and the output label (damage state) have a clear and experimentally verified physical causal relationship, ensuring the effectiveness of the model learning rule. By designing different intensity seismic input, various states from no damage to severe damage (plastic deformation, crack) can be induced, so that the model learned can recognize and predict these typical damage modes. Based on strict nondestructive testing technology for state confirmation and labeling, manual misjudgment is greatly reduced, providing accurate supervision signals for the model, which is beneficial to improve the accuracy and reliability of the classification model. The sample library contains responses under various seismic wave excitations, enabling the model to learn the relationship between different spectral characteristic seismic waves and damage, and improving the generalization ability of the model in response to different seismic characteristics in actual seismic events. The training sample acquisition and labeling method of the embodiment provides a solid and reliable data basis for the entire intelligent monitoring and early warning system, and is a key prerequisite to ensure the accuracy and credibility of the machine learning model evaluation results.

[0096] In another embodiment of the present application, the real-time monitoring and safety warning method of the anti-seismic structure of the oil-fired unit comprises:

[0097] The three-axis acceleration sensor, micro-strain sensor and the like are correctly installed on the oil-fired unit, and synchronous acceleration data sequence and strain data sequence are collected through the vibration table test or actual seismic event. These data are the basis for subsequent correlation analysis.

[0098] In step S2, while performing threshold comparison and machine learning model evaluation, the following multi-parameter correlation analysis steps are also performed in parallel, mainly for the same anti-seismic anchoring component (for example, a specific anchor bolt):

[0099] From the complete data collected within the time history of the seismic event, the acceleration data sequence (usually from one axis of the three-axis acceleration sensor installed on the nearby base) and the strain data sequence (from the micro-strain sensor pasted on the component) corresponding to the same monitored anti-seismic anchoring component are extracted. Ensure that the two data sequences are strictly synchronized in time, have the same time stamp and sampling frequency (for example, both are 1 kHz).

[0100] The data processing platform calls the built-in mathematical calculation library to calculate the Pearson correlation coefficient r of the extracted acceleration data sequence (denoted as X) and strain data sequence (denoted as Y) corresponding to the component. The calculation formula is:

[0101] , wherein n is the number of data points, X i and Y irespectively the acceleration value and strain value at the i th time instant, and respectively the mean values of the corresponding sequences. The calculated correlation coefficient r has a value range of [-1, 1].

[0102] The calculated Pearson correlation coefficient r is compared with a pre-set correlation coefficient threshold (denoted as r th ). If r ≥ r th , it is considered that the acceleration response and strain response of the component have strong linear correlation, and the component is in a normal cooperative working state. If r < r th , it is determined that the seismic anchoring component has the risk of stiffness degradation, component connection failure or anchoring loosening. Because when the component is damaged or the connection performance is reduced, the dynamic response characteristics will change, resulting in a decrease in the degree of linear correlation between the input acceleration excitation and the generated strain response.

[0103] The correlation coefficient threshold r th is not arbitrarily set, but is obtained based on the data obtained from multiple tests of the fuel engine unit or similar units in the simulated shaking table test under normal undamaged state.

[0104] In multiple (for example, 20) simulated seismic excitation tests, when it is confirmed that the seismic anchoring component is in a perfect undamaged state, the Pearson correlation coefficient of the acceleration sequence and the strain sequence of the same component in each test is calculated according to the above method, thereby obtaining a correlation coefficient sequence (for example, containing 20 r values). Statistical analysis is performed on the correlation coefficient sequence, and the mean (μ) and standard deviation (σ) are calculated. The lower limit of the statistical confidence interval of the sequence is taken as the threshold (r th ). For example, μ-2σ (corresponding to about 95% confidence) or μ-3σ (corresponding to about 99.7% confidence) can be taken as the threshold. It is intended to ensure that most of the calculated correlation coefficients should be higher than the threshold in the normal state, thereby reducing the risk of false positives.

[0105] For a rigidly connected component with a perfect structure, the Pearson correlation coefficient of the acceleration and strain response in the effective seismic excitation frequency band is usually high. The r th value determined by the above method may fall within the range of 0.7-0.9, and the specific value depends on the unit type, sensor position and seismic wave characteristics.

[0106] The embodiment provides another independent criterion based on physical mechanism for the evaluation results of the threshold comparison and the machine learning model from the perspective of consistency of component dynamic response, forms a multi-angle and cross-verification evaluation system, and improves the reliability of damage identification. Stiffness degradation, slight connection loosening and other damages may not be enough to cause the strain or acceleration peak to obviously exceed the physical threshold in the early stage, but can affect the correlation between responses. The method is sensitive to such implicit damage and is beneficial to early warning. The method is particularly suitable for judging the connection effectiveness between the anchoring component and the foundation, and can effectively identify the connection performance degradation problem caused by bolt loosening, gasket deformation and the like, which is difficult to directly achieve by single parameter threshold comparison. By comparing with historical normal state data to set the threshold, the judgment method has strong adaptability and can reduce the misjudgment caused by environmental noise or slight drift of individual sensors. The multi-parameter correlation analysis method of the embodiment enriches the technical means of the seismic structure health state evaluation of the fuel turbine unit, and enhances the identification ability of the whole monitoring and early warning system to the complex damage mode and the confidence of the evaluation results.

[0107] In another embodiment of the application, a specific inspection action sequence is dynamically generated according to the health evaluation result, mainly relying on an intelligent work order generation module in the data processing platform, which is a software module and can be deployed on the same server or cloud computing instance as the data processing platform. The core is a preconfigured mapping relationship library. The construction of the mapping relationship library does not depend on special hardware, but needs to be defined based on domain knowledge.

[0108] The mapping relationship library is constructed and maintained in the intelligent work order generation module in the form of a decision tree or a production rule. After the health evaluation result is generated in step S2, the inspection list generation process in step S3 is as follows:

[0109] The intelligent work order generation module receives the health evaluation result and extracts the key input conditions therefrom, mainly including:

[0110] Damage component identification: indicating which component has risks, such as "No. 1 anchor bolt", "southwest corner of rigid base", etc.;

[0111] Damage type: possible damage forms obtained by evaluation, such as "plastic deformation", "crack", "loose bolt";

[0112] Comprehensive risk level: for example, "attention", "warning", "danger".

[0113] The intelligent work order generation module matches the above input conditions with the pre-defined rules in the mapping relationship library. If production rules are used, the rule form can be as follows:

[0114] IF Damage Component Identifier LIKE 'Anchor Bolt' AND Damage Type = 'Crack' AND Overall Risk Level = 'Dangerous' THEN execute action sequence: ['Immediately visually inspect crack morphology', 'Measure depth using crack depth gauge', 'Perform magnetic particle inspection', 'Evaluate if immediate replacement is required'];

[0115] IF Damage Component Identifier LIKE 'Anchor Bolt' AND Damage Type = 'Loose Bolt' AND Overall Risk Level = 'Warning' THEN execute action sequence: ['Visually inspect nut position marker', 'Check pre-tightening force using calibrated torque wrench', 'Record torque value and compare with design value'];

[0116] If a decision tree is used, by a series of attribute judgments based on input conditions, eventually traverse to the leaf node, which corresponds to a check action sequence.

[0117] According to the rule matching result, an ordered check action sequence is dynamically generated. The generation of the check action sequence follows two priority principles:

[0118] Cost-based priority: low-cost inspection actions are prioritized over high-cost inspection actions. For example, in the generated sequence, low-cost actions such as "visual inspection" and "using a tape measure for dimension measurement" are arranged before high-cost non-destructive testing actions such as "ultrasonic detection" and "radiographic detection". This helps to quickly screen and avoid unnecessary complex detection.

[0119] Risk level-based priority: At the same time, the inspection items related to the highest risk level (such as "dangerous") in the sequence are given the highest execution priority. Even if a high-cost detection is necessary for a component with a "dangerous" risk level, it may be arranged at an early position in the sequence. For example, for a component with a risk level of "dangerous", even if "ultrasonic detection" is costly, it may be recommended to be performed first to quickly confirm internal damage.

[0120] Each "specific inspection action" in the check action sequence is defined based on existing, mature operation procedures and inspection standards (such as equipment manufacturer maintenance manuals, industry standards such as NB / T 47013 "Non-destructive Testing of Pressure Equipment"), ensuring its feasibility and effectiveness. The definition of "cost" in the rules (such as visual inspection as low cost, ultrasonic detection as high cost) and the impact of risk level on priority are based on historical operation data, expert experience, and comprehensive evaluation of the time, resources, and technology required for inspection work. These weight parameters can be adjusted in the system configuration file.

[0121] The embodiment overcomes the rigid drawbacks of the traditional fixed inspection list by generating the inspection list based on the intelligent inference of the rule base, can generate a highly targeted inspection scheme according to the specific damaged component, damage type and risk level, and enables the operation and maintenance personnel to 'treat the disease with the right medicine' and directly focus on the most likely problem and the most need to pay attention to the damage mode. Through the implementation of the principles of low-cost priority and high-risk priority, the operation and maintenance workflow is rationalized. Prioritizing the implementation of fast and simple inspections can quickly rule out or confirm most problems, avoiding the waste of resources caused by blindly starting high-cost detection, while ensuring that high-risk hazards are prioritized, improving operation and maintenance efficiency and economy. The expert knowledge is solidified in the rules of the mapping relationship base, so that even inexperienced operation and maintenance personnel can perform standardized and efficient inspections according to the list generated by the system, reducing inspection omissions or misjudgments caused by individual experience differences, and improving the objectivity and consistency of the evaluation results. After an emergency such as an earthquake, a highly instructive inspection list can be quickly generated to help the operation and maintenance team quickly develop an investigation plan and determine the priority, thereby speeding up the decision-making process for judging the availability of the unit and restoring power supply. The inspection list generation method based on the intelligent mapping relationship base is a key link to achieve the precise guidance of the operation and maintenance target of the embodiment, and significantly improves the intelligent level and execution efficiency of post-earthquake inspection work.

[0122] In another embodiment of the present application, the edge computing gateway accurately determines the earthquake event by combining frequency domain and time domain analysis, mainly relying on the enhanced signal processing capability of the edge computing gateway. The selection of the edge computing gateway needs to ensure that it has the ability to perform fast Fourier transform (FFT) and other frequency domain analysis algorithms and real-time logical judgment.

[0123] The edge computing gateway selects a high-performance industrial-grade edge computing gateway that needs to have strong floating-point operation capability and sufficient RAM to support real-time frequency domain and time domain signal analysis. The edge computing gateway is installed in the electric control cabinet matched with the oil-fired unit, and is connected with the three-axis acceleration sensor and other devices arranged on the rigid base and anti-seismic anchoring components of the unit through shielded cables.

[0124] The first frequency range is determined according to the main distribution characteristics of seismic energy in historical earthquake records. The typical seismic acceleration power spectral density mainly concentrates in the frequency band of 0.5-20Hz. Therefore, the first frequency range can be set in this interval, for example, the lower limit f L =0.5Hz, and the upper limit f H =20Hz. Statistical analysis of the acceleration power spectral density of a large number of historical strong earthquake records in the target area finds the frequency interval corresponding to the energy concentration degree (for example, the cumulative power reaches 80% or 90% of the total power), which is taken as the first frequency range.

[0125] The second preset threshold is used for time domain analysis, usually set to be higher than the first preset threshold, to screen out the stronger vibration phase. Its value can be set to 0.15-0.25g (about 1.47-2.45m / s 2 ), for example, 0.2g (about 1.96m / s 2 ), which can be selected based on the experience of earthquake engineering to effectively represent the acceleration level at the beginning of the strong vibration phase.

[0126] The duration threshold is used to judge the duration characteristics of strong vibration. According to historical earthquake records, the typical duration of a strong vibration phase is usually between 2-10 seconds. Therefore, the duration threshold T d can be set to 3s, 5s, or 8s, etc. Statistical analysis of the duration of the phase in which the acceleration exceeds a certain intensity (such as 0.15-0.25g) in historical earthquake records, such as the mean or a certain quantile, is used to determine the duration threshold.

[0127] In step S1, the edge computing gateway determines that a seismic event has occurred, which is a multi-condition judgment process, and the workflow is as follows:

[0128] The edge computing gateway monitors the vibration acceleration signal from the triaxial acceleration sensor in real time and calculates its peak value. When it is monitored that the vibration acceleration peak value lasts for more than a first preset threshold (for example, 0.1g) for a short stable time (for example, 3-5 consecutive sampling points, corresponding to tens of milliseconds), further detailed signal analysis is triggered.

[0129] The edge computing gateway performs fast Fourier transform (FFT) on the current collected vibration acceleration signal segment (for example, the data window of the last 2 seconds) to calculate its power spectral density. The calculated power spectral density is analyzed to determine whether the signal energy is concentrated in the preset first frequency range. The specific judgment condition can be that whether the proportion of the energy in the first frequency range to the total energy exceeds a set proportion threshold (for example, 60% or 70%). If it does, it is considered that the signal has the typical frequency domain characteristics of seismic vibration.

[0130] In the same analysis time window, the edge computing gateway simultaneously performs time domain analysis. Specifically, it is determined whether the cumulative duration of the vibration acceleration value exceeding a second preset threshold (for example, 0.2g) reaches a preset duration threshold (for example, 5 seconds). This cumulative duration does not require continuous exceeding, but the sum of the durations of all signal segments exceeding the second preset threshold within the analysis window.

[0131] The edge computing gateway performs logical "and" judgment, and only when the following three conditions are met at the same time, it is finally determined that a seismic event has occurred:

[0132] a. The peak value of the vibration acceleration exceeds a first preset threshold value (preliminary triggering condition) for more than a first preset time interval;

[0133] b. The frequency domain analysis result shows that the signal energy is concentrated in a first frequency range;

[0134] c. The time domain analysis result shows that the cumulative duration of the acceleration exceeding a second preset threshold value reaches a duration threshold value;

[0135] Once the earthquake event is comprehensively determined, the edge computing gateway immediately performs an operation of controlling the sensor array to switch to a high-frequency recording mode.

[0136] The embodiment can effectively distinguish the earthquake event from common industrial vibrations, vehicle passing, mechanical start-stop and other non-earthquake disturbances by combining three types of criteria, i.e., vibration intensity (time domain peak value), frequency characteristics (frequency domain distribution) and duration (time domain persistence), thereby greatly reducing the false alarm rate of the system. The frequency domain analysis ensures that the triggered event has typical frequency spectrum characteristics of seismic motion, the time domain analysis ensures that the event has a certain intensity and duration, which conforms to the engineering characteristics of the earthquake, and the triggering mechanism is more in line with the needs of structural seismic analysis. The multi-criteria joint decision avoids the false action caused by accidental impact or sensor noise of a single threshold criterion, and enhances the reliability and robustness of the entire monitoring system in a complex industrial environment. The earthquake event determination method based on frequency domain and time domain analysis significantly improves the accuracy of event detection, and lays a solid foundation for the reliability of subsequent data acquisition and health assessment.

[0137] In another embodiment of the application, the edge computing gateway determines the end of the earthquake event according to the vibration energy attenuation, and controls the sensor array to switch back to the low-frequency monitoring mode. The edge computing gateway can be an industrial-grade edge computing gateway, which ensures that it has sufficient computing power to process acceleration data and perform integral operation in real time.

[0138] The evaluation time window is used to calculate the vibration energy attenuation rate, and its length should effectively reflect the attenuation trend of seismic motion energy, and is usually much larger than the vibration period. The evaluation time window T w may be selected as 10-30s, for example, 15s or 20s. The intensity duration concept commonly used in earthquake engineering can be referred to to select a typical time length that can cover the attenuation process after the main energy release of the earthquake.

[0139] The energy attenuation rate threshold value is a dimensionless ratio threshold value for judging the speed of energy attenuation. Its value needs to be determined through statistical analysis, and represents the lower limit of normal attenuation of seismic energy. A large number of acceleration time history data of historical strong earthquakes, especially the data segment after the strong earthquake segment (i.e., after the peak acceleration), are collected. For each record, the curve of the integral value of the acceleration square (i.e., the calculation basis part of the Arias Intensity) with respect to time is calculated.

[0140] Within a time window T w , which is equivalent to the evaluation time window T th , the rate of decrease of the acceleration square integral value is calculated. The rate of decrease can be defined as (initial value - final value) / initial value or using the logarithmic decay rate.

[0141] The statistics of all the calculated rates of decrease are taken, and a lower quantile value (e.g., 5% or 10% quantile) of the distribution is taken as the energy decay rate threshold η th . This means that when the earthquake is normally decaying, there is a high probability (e.g., 90% or 95%) that its energy decrease rate is higher than this threshold. If the measured decrease rate is lower than this threshold, it means that the energy decay is abnormally slow, which may mean that the shaking has essentially stopped or entered a very weak aftershock phase.

[0142] A stabilization time window T s is used to observe whether the state of the decay rate being lower than the threshold persists, avoiding false positives due to transient fluctuations. The stabilization time window T s should be shorter than the evaluation time window, and can typically be chosen to be 3-10 s, e.g., 5 s. Based on observations of the transition period when strong earthquakes end and the shaking stabilizes, a time period sufficient to confirm that the decay trend is stable is chosen.

[0143] In step S1, after the edge computing gateway controls the sensor array to switch to the high-frequency recording mode, the following steps are performed in parallel to determine the termination time of data acquisition, the core of which is to determine whether the main shock phase of the earthquake has ended by monitoring whether the release of vibration energy tends to be stable, including:

[0144] The edge computing gateway continuously reads the vibration acceleration signals a(t) (typically the combined acceleration of the three axes or the single-axis acceleration with the most energy) collected by the three-axis acceleration sensor and calculates an index representing the cumulative vibration energy, the integral of the acceleration square I(t), in real time. The calculation is implemented in the discrete time domain using numerical integration methods (such as the trapezoidal method):

[0145] where a(t) represents the vibration acceleration value, with units of m / s 2 . t0 represents the starting time of the integral, which is typically set to the starting time of the earthquake event determination. I(t) represents the cumulative acceleration square integral from t0 to the current time t, which is proportional to the Arias Intensity, a classic index for measuring the total energy of ground motion, with units of m 2 / s 3 ; a(τ) is a time-varying acceleration function, and τ is a continuously changing integral variable within the time interval [t0, t].

[0146] To quantify the slowdown of energy accumulation, an energy release rate indicator R(t) is calculated. The edge computing gateway sets an evaluation time window of length T w (e.g. 15 seconds) and calculates the average growth rate R(t) of the accumulated energy I(t) in this time window:

[0147] where T w denotes the length of the evaluation time window in seconds (s). Its value should reflect the macroscopic trend of seismic energy decay, usually referencing the significant duration concept in earthquake engineering. R(t) denotes the average growth rate of the accumulated energy I(t) in the most recent T w time period, in m 2 / s 4 . R(t) is large when the earthquake is in the strong shaking phase, and tends to zero when the main shock energy is released and the earthquake enters the weak shaking or calm period.

[0148] The edge computing gateway compares the real-time calculated energy release rate R(t) with a pre-set energy release rate threshold R th .

[0149] The decision condition is: when R(t)≤R th is monitored for more than a pre-set stable time window T s (e.g. 5 seconds), it is determined that the main energy of the earthquake has been released and the shaking enters the insignificant phase.

[0150] Once this condition is met, the edge computing gateway immediately issues an instruction to the sensor array to control all sensors to switch back from high-frequency recording mode to low-frequency monitoring mode.

[0151] The energy release rate threshold R th is determined by statistical analysis of historical strong earthquake records. Calculate the R(t) value of these records after the main shock segment and enter the decay segment, and take a lower quantile value (such as 10% quantile) as the threshold. The threshold is a positive number close to zero, representing the critical state of the system that the energy release has basically stopped.

[0152] The stable time window T s is used to avoid misjudgment caused by transient fluctuations, and its length is determined by observing the transition time when the energy tends to be stable after the end of the strong earthquake in the historical earthquake records.

[0153] The embodiment determines the data collection termination opportunity by monitoring vibration energy attenuation, avoids the problems of data redundancy (early end of the earthquake) or incomplete recording (long duration of the earthquake) caused by fixed time recording, can adapt to the actual duration of the earthquake vibration, and optimizes data storage. Based on the vibration energy attenuation index with clear physical meaning, the end of the main stage of the earthquake can be more reliably determined than simply relying on the peak value return of the acceleration, and the recorded data is ensured to cover the strong vibration stage with engineering significance. Timely switching back to the low-frequency monitoring mode reduces unnecessary energy consumption and data storage space occupation, which is particularly beneficial to long-term deployment and battery-powered scenarios. The clear recording termination point helps the data processing platform to more accurately locate and analyze the effective data segment of the earthquake event, and improves the efficiency and quality of subsequent health assessment. The data collection termination method based on energy attenuation judgment realizes the adaptive management of the high-frequency recording mode, and enhances the intelligent level and economy of the system operation.

[0154] In another embodiment of the application, the lock period threshold (temporary event judgment threshold) is determined by statistical analysis of the acceleration peak value ratio of the main shock and the largest aftershock in historical earthquake records. For example, based on the records of the same type of site in the global earthquake database (such as USGS), the acceleration peak value ratio of the main shock and the largest aftershock is usually (1:0.3)-(1:0.5). Therefore, the lock period threshold can be set to 1.3-1.5 times of the first preset threshold, i.e. 0.13-0.15g (about 1.27-1.47m / s 2 ). The specific value needs to be adjusted according to the site seismicity to ensure that aftershocks are not misjudged as new events.

[0155] The lock period duration is determined by statistical analysis of the time interval distribution of the main shock and significant aftershocks (aftershocks with acceleration peak values reaching more than 30% of the main shock peak value) in historical earthquake sequences. For example, based on typical earthquake sequence data (such as the Wenchuan earthquake record), significant aftershocks often occur within 10-30 minutes after the main shock. Therefore, the lock period duration can be set to 20 minutes (1200 seconds), and the range can be selected as 10-60 minutes. The specific duration is the 90th percentile of the historical time interval to cover most aftershock occurrence windows. The ratio of 30% is based on statistical analysis of historical earthquake sequences, which is used to distinguish significant aftershocks from general microseisms.

[0156] After the edge computing gateway controls the sensor array to switch back to the low-frequency monitoring mode from the high-frequency recording mode, the lock period mechanism is immediately started, and the specific process is as follows:

[0157] When the edge computing gateway confirms that the main energy release of the earthquake is complete through energy attenuation judgment, and switches the sensor to the low-frequency monitoring mode (sampling rate 100 Hz), a lock period of a preset duration (for example, 20 minutes) is automatically started.

[0158] During the lock-in period, the edge computing gateway temporarily raises the acceleration peak threshold for event determination from the first preset threshold (0.1g) to the lock-in period threshold (0.15g). At the same time, the functions of frequency domain analysis (such as power spectral density calculation) and time domain analysis (such as duration accumulation judgment) are suspended, and whether a new earthquake event occurs is determined only based on whether the vibration acceleration peak value collected in real time by the three-axis acceleration sensor exceeds the lock-in period threshold.

[0159] If the vibration acceleration peak value continuously exceeds the lock-in period threshold (for example, for 5 consecutive sampling points), the edge computing gateway determines that a new earthquake event has occurred, and immediately controls the sensor array to switch to a high-frequency recording mode (sampling rate 1kHz). Otherwise, the low-frequency monitoring mode is maintained.

[0160] When the lock-in period expires (for example, after 20 minutes), the edge computing gateway automatically restores the event determination threshold to the first preset threshold (0.1g), and re-enables the frequency domain and time domain analysis functions, restoring the normal multi-criteria event detection mode.

[0161] The present embodiment effectively avoids the misjudgment of aftershocks as independent new events after the main earthquake by temporarily raising the determination threshold and simplifying the detection logic, reducing unnecessary system triggering and resource consumption. Based on historical earthquake statistical data, the parameters are set to adapt the lock-in period threshold and duration to the characteristics of actual earthquake sequences, enhancing the robustness of the system in complex seismic environments. During the lock-in period, the complex signal analysis functions are suspended, reducing the computational load of the edge computing gateway and prolonging the service life of the device, which is particularly suitable for long-term deployment scenarios.

[0162] The present application provides a real-time monitoring and safety warning system for the anti-seismic structure of a fuel unit, comprising:

[0163] The perception layer is responsible for real-time acquisition of physical parameters of the anti-seismic structure of the fuel unit, including vibration acceleration, strain and pretension data.

[0164] The three-axis acceleration sensor uses an ICP type piezoelectric three-axis acceleration sensor. The sensor is rigidly fixed to the joint surface between the rigid base of the fuel unit and the building foundation through M8 high-strength stainless steel bolts, and at least two sensors are preferably arranged symmetrically at the four corners of the base to capture multi-directional vibration responses.

[0165] The micro-strain sensor uses a resistance strain gauge type sensor, which is directly pasted on the surface of the anti-seismic anchoring component (such as a foundation bolt or an anchor rod), with the sensitive grid direction consistent with the main stress direction (axial direction) of the component. At least one sensor is installed for each key anchoring component, and the surface of the component needs to be cleaned before pasting to ensure the bonding strength.

[0166] The torque sensor adopts a flange type torque sensor, which is installed between the fastening nut of the anchoring component and the supporting surface, or is replaced by an intelligent bolt, for real-time monitoring of the pre-tightening force state. When installing, the torque sensor needs to be perpendicular to the fastening surface to avoid partial load.

[0167] The network layer is responsible for data collection, event triggering and mode switching, and the core device is an edge computing gateway.

[0168] The edge computing gateway selects an industrial-grade edge computing gateway device, which has multi-channel data acquisition, floating point operation capability and real-time logic judgment function. The gateway is connected with the perception layer sensor through a shielded cable and is installed in the electric control cabinet matched with the fuel engine unit and fixed on the guide rail in the cabinet. The working temperature range of the gateway is -40°C to +70°C, which is suitable for industrial environment.

[0169] The platform application layer is deployed on a local server or cloud infrastructure, including the following modules:

[0170] The data storage module uses a relational database or a document database to store monitoring data, seismic design parameters and case library data. The database is deployed on the local storage or cloud storage of the server, supporting JSON / XML format data storage.

[0171] The multi-source data fusion and intelligent decision engine is built based on Python programming language, integrating Scikit-learn machine learning library, running pre-trained random forest classification model or support vector machine model, for generating health assessment results. The engine is deployed on the CPU or GPU accelerated environment of the server.

[0172] The intelligent work order generation module constructs a mapping relationship library in the form of decision tree or production rule, and the rules are pre-defined based on industry standards and deployed on the same server instance as the data processing platform.

[0173] The push module integrates SMS gateway, enterprise WeChat API or dedicated APP push channel, for sending early warning information to the mobile terminal of the operation and maintenance personnel.

[0174] The sensor sampling rate is 100Hz for low-frequency monitoring mode and 1kHz for high-frequency recording mode; the seismic design threshold is set based on the unit seismic design specification, with an acceleration design threshold of 2.5m / s 2 , a strain design threshold of 3000µm / m and a pre-tightening force design threshold of 800 .

[0175] Edge computing gateway event decision parameters, first preset threshold 0.1g, adjusted according to the 95% quantile of the site background noise, range 0.05-0.2g. Frequency domain analysis first frequency range 0.5-20Hz, determined based on the energy distribution of historical seismic records. Time domain analysis second preset threshold 0.2g, duration threshold 5s.

[0176] Data standardization adopts the minimum-maximum normalization method to map sensor data to the [0, 1] interval.

[0177] The working process of the system is as follows:

[0178] The edge computing gateway continuously reads the three-axis acceleration sensor data at a sampling rate of 100Hz, and when the peak value of the vibration acceleration continuously exceeds the first preset threshold (0.1g) and meets the frequency / time domain analysis conditions, it is determined as a seismic event;

[0179] The edge computing gateway immediately controls the sensor array to switch to a 1kHz high-frequency recording mode to collect acceleration, strain, and pretension data.

[0180] After the seismic event ends, the edge computing gateway uploads the high-frequency data to the platform application layer through 4G / 5G or Ethernet;

[0181] The data storage module receives and stores the data, and the multi-source data fusion and intelligent decision engine performs threshold comparison and machine learning model evaluation in parallel to generate health assessment results (including damaged components, types, and risk levels);

[0182] If the assessment result identifies a risk component, the intelligent work order generation module dynamically generates an inspection checklist based on the mapping relationship library, prioritizing low-cost inspections (such as visual inspections) and high-risk items;

[0183] The push module sends the electronic inspection checklist to the mobile terminal of the maintenance personnel and triggers the unit start lock for "warning" or "danger" levels;

[0184] The on-site inspection feedback results are stored in the case library, and the machine learning model is retrained regularly to improve the accuracy of the assessment.

[0185] The system overcomes the lag of traditional manual inspection through real-time monitoring by the perception layer and intelligent triggering by the network layer. The platform application layer integrates multi-source data and machine learning models to provide quantitative and reliable health status assessments, reducing reliance on human experience. Based on the rule library, the inspection checklist is dynamically generated, focusing on high-risk components to improve post-earthquake inspection efficiency and accelerate power restoration. Using commercially available hardware and parameterized configuration, it is easy to deploy and optimize in different types of units and environments.

[0186] While embodiments of the application have been disclosed in connection with the above specification, it will be evident to those skilled in the art that many modifications, substitutions, and alterations to the embodiments of the application can be made and that many specifically protected details shown can be substituted by other specifically protected details. Accordingly, it is intended that the application not be limited to the specific details shown and described above but that it be given broadest scope indicated by the following claims and their equivalents.

Claims

1. A method for real-time monitoring and safety warning of anti-seismic structure of a fuel engine unit, characterized in that, The method comprises the following steps: S1, real-time monitoring of vibration acceleration through a three-axis acceleration sensor arranged on a rigid base of a fuel engine unit and a building foundation; When the edge computing gateway monitors that the peak value of the vibration acceleration continuously exceeds a first preset threshold value, it is determined that an earthquake event occurs, and the three-axis acceleration sensor, as well as the micro-strain sensor and the torque sensor arranged on the seismic anchoring component, are controlled to switch from a low-frequency monitoring mode to a preset high-frequency recording mode for data acquisition; S2, uploading the high-frequency acceleration data, strain data and pre-tightening force data collected during the earthquake event to a data processing platform, and comparing the acceleration data, strain data and pre-tightening force data with preset seismic design thresholds respectively to obtain threshold comparison results; Meanwhile, the acceleration data, strain data and pre-tightening force data are input as input features into a pre-trained machine learning classification model, and the machine learning classification model takes the unit vibration data in historical earthquake events or simulated shaking table tests and the confirmed structural damage state as training samples, and outputs the damage probability level of different components of the seismic structure; Fusing the threshold comparison results and the damage probability level, a health assessment result containing specific damaged components, damage types and comprehensive risk levels is generated; S3, based on the health assessment result, if a component at risk is identified, the damaged component and the damage type are mapped to corresponding specific inspection actions according to a preset mapping relationship library, and a targeted inspection list is generated, which lists the components and inspection items that need to be inspected first, and the pre-warning information containing the electronic inspection list is pushed to the mobile terminal of the operation and maintenance personnel; In step S1, the edge computing gateway determines that an earthquake event occurs on the premise that the peak value of the vibration acceleration continuously exceeds a first preset threshold value, and further analyzes the vibration signal in the frequency domain and the time domain; The frequency domain analysis specifically calculates the power spectral density of the vibration signal and determines whether the signal energy is concentrated in a preset first frequency range, and the time domain analysis specifically determines whether the cumulative duration of the vibration acceleration exceeding a second preset threshold value reaches a preset duration threshold value; The first frequency range is determined according to the main energy distribution range of the seismic acceleration power spectral density in historical earthquake records; the duration threshold value is determined according to the typical duration of the strong earthquake stage in historical earthquake records; In step S1, after the edge computing gateway controls the sensor array to switch to the high-frequency recording mode, the following steps are further performed to determine the termination time of data acquisition: The integral value of the acceleration square of the vibration signal collected by the three-axis acceleration sensor is continuously calculated as an index representing the cumulative vibration energy, and based on this, the average change rate of the integral value is calculated in a preset evaluation time window to obtain the energy release rate; When the energy release rate is monitored to be lower than or equal to a preset energy release rate threshold, and lasts for a preset stable time window, it is determined that the main energy of the earthquake has been released, and the sensor array is controlled to switch back to the low-frequency monitoring mode from the high-frequency recording mode; wherein the energy release rate threshold is determined by statistical analysis of the energy release rate after the strong shock segment in the historical earthquake record; After the sensor array is controlled to switch back to the low-frequency monitoring mode from the high-frequency recording mode, the edge computing gateway starts a preset duration of a lock-in period, during which the edge computing gateway temporarily increases the acceleration peak threshold for event determination from the first preset threshold to a predetermined lock-in period threshold, and suspends the frequency domain and time domain analysis functions, and only determines whether a new earthquake event occurs based on whether the vibration acceleration peak value exceeds the lock-in period threshold; After the lock-in period expires, the edge computing gateway restores the acceleration peak threshold for event determination to the first preset threshold, and re-enables the frequency domain and time domain analysis functions; Wherein the lock-in period threshold is determined by statistical analysis of the acceleration peak value ratio of the main shock to the largest aftershock in the historical earthquake record, and the duration of the lock-in period is determined by statistical analysis of the time interval distribution between the main shock and the significant aftershock in the historical earthquake sequence, and the significant aftershock is an aftershock with an acceleration peak value ratio to the main shock peak value exceeding 30%.

2. The method of claim 1, wherein the method further comprises: In step S2, the data processing platform stores the data of the current earthquake event, the health assessment result and the on-site inspection feedback result of the subsequent operation and maintenance personnel, forms a case library, and periodically re-trains the machine learning evaluation model using the updated case library.

3. The method of claim 1, wherein the method further comprises: The damage probability level includes at least four levels of normal, attention, warning and danger, and corresponds to different warning colors and disposal suggestions, and when the level is warning or danger, the automatic start function of the fuel unit is locked.

4. The method of claim 2, wherein the method further comprises: The training samples of the machine learning classification model are obtained and labeled as follows: Obtain the unit vibration data, strain data and pretightening force data of the fuel unit or similar unit under the excitation of seismic waves with different acceleration peak values in the simulated shaking table test; Based on the results of non-destructive testing of the aseismic anchoring components after the test, at least one of magnetic powder testing, penetration testing or ultrasonic testing is determined, and the damage state is labeled as no damage, plastic deformation, crack or bolt loosening.

5. The method of claim 1, wherein the method further comprises: Before inputting the acceleration data, strain data and pretightening force data as input features into the machine learning classification model, the data is preprocessed by feature standardization, which includes normalizing the data of various sensors to the same numerical interval.

6. The method of claim 4, wherein the method further comprises: In step S2, it also includes multi-parameter correlation analysis, which includes extracting the acceleration data sequence and strain data sequence synchronously collected within the time sequence of the earthquake event for the same aseismic anchoring component, calculating the Pearson correlation coefficient between the acceleration data sequence and the strain data sequence, and comparing the Pearson correlation coefficient with a preset correlation coefficient threshold, wherein if the Pearson correlation coefficient is lower than the correlation coefficient threshold, it is determined that the aseismic anchoring component has the risk of stiffness degradation, component connection failure or anchoring loosening; The preset correlation coefficient threshold is determined by analyzing the Pearson correlation coefficients of acceleration and strain data sequences of the fuel engine unit or similar units in the simulated vibration table test under normal non-damaged state for multiple times, and taking the lower limit of the statistical confidence interval of the Pearson correlation coefficients.

7. The method of claim 1, wherein the method further comprises: The mapping relationship library is constructed in the form of a decision tree or a production rule, and maps the damaged component and the damage type to corresponding specific inspection actions, including: The damaged component identification, damage type and comprehensive risk level in the health assessment result are taken as input conditions for reasoning; According to the rule matching and reasoning of the mapping relationship library, the inspection action sequence is dynamically generated, wherein the priority of the inspection action sequence is determined based on the cost and the dependency relationship: the low-cost inspection including visual inspection and size measurement is prior to the high-cost inspection including non-destructive testing; meanwhile, the inspection item corresponding to the highest risk level is given the highest priority.

Images