Earthquake monitoring equipment fault diagnosis method and system based on machine learning

Abstract

The invention discloses an earthquake monitoring equipment fault diagnosis method and system based on machine learning, and belongs to the technical field of earthquake service, and the method comprises the following steps: monitoring the output frequency of each accelerometer in real time, and carrying out the self-inspection deviation adjustment after receiving a self-inspection signal, and outputting a fault score and a dynamic prediction trend of the accelerometer through a machine learning model, thereby determining a fault level of the accelerometer, and after comprehensive adjustment, carrying out periodic fault level analysis again to ensure normal operation of equipment and timely respond to potential faults. According to the earthquake monitoring equipment fault diagnosis method and system based on machine learning provided by the invention, accurate prediction and adjustment are carried out at the initial stage of the fault of the earthquake monitoring equipment, corresponding measures are automatically taken during fault upgrading, equipment damage is reduced to the maximum extent, and the stability and reliability of system operation are improved.

Description

Technical Field

[0001] This invention relates to the field of earthquake service technology, and in particular to a fault diagnosis method and system for earthquake monitoring equipment based on machine learning. Background Technology

[0002] Frequency domain and wavelet analysis methods are commonly used in fault diagnosis of earthquake monitoring equipment. The core of these methods is to convert the output signal of the accelerometer or seismometer from the time domain to the frequency domain or time-frequency domain. By analyzing the power spectral density, characteristic frequency drift, or changes in wavelet coefficients, abnormal features can be identified. Under normal operating conditions, the seismometer signal has stable frequency characteristics and amplitude distribution. However, when the equipment malfunctions, decreased sensor sensitivity, abnormal mass block movement, or increased electronic noise can lead to abnormal energy distribution or abrupt changes in wavelet coefficients in specific frequency bands. Through quantitative analysis of these frequency domain or time-frequency domain characteristics, minute performance deviations can be detected early, enabling accurate identification and early warning of potential equipment faults.

[0003] For example, Chinese invention patent CN102971645B discloses a seismic survey communication system and method, including an embodiment that utilizes a hardwired communication infrastructure by attaching a wireless unit to the hardwired communication. This allows, for example, a recording vehicle to control a remote seismic source via wireless communication within an exploration array. Another embodiment includes a communication system for field service equipment. The embodiment provides a fully or semi-automatic process for communicating equipment malfunctions between survey personnel (e.g., recording vehicle operators and line observers). The embodiment, for example, establishes an end-to-end channel between the recording vehicle and members of the field work group.

[0004] For example, Chinese invention patent CN114935779B discloses a master-slave switching system for automatic connection between equipment at an earthquake rescue site, comprising: a monitoring and alarm module: used to retrieve master and slave status information after the earthquake monitoring and alarm equipment is activated, determine the real-time monitoring master, and perform process-oriented monitoring of the real-time monitoring master, and generate monitoring information based on the process-oriented monitoring; a scheduling module: used to determine whether the real-time monitoring master is in an abnormal state based on the monitoring information, and generate a master-slave switching command when the state is abnormal; and a master-slave switching module: used to switch the real-time monitoring master to a slave according to the master-slave switching command, and reset the real-time monitoring master.

[0005] The above-mentioned technology has at least the following technical problems: In earthquake monitoring equipment, accelerometers, as the core components of seismometers, can experience deviations in the amplitude, frequency, and phase of their output signals due to mechanical structure elastic decay or external factors. This can lead to abnormal vibrations of the mass block, and prolonged abnormal vibrations can generate continuous stress on the springs and suspension structures of each accelerometer. Under repeated loading, the spring material gradually accumulates microcracks or undergoes permanent deformation, and the elasticity of the suspension connection points also decreases. This results in a drift in structural stiffness and damping characteristics, causing distortion of the measurement signals used to reflect earthquake acceleration. The accuracy of amplitude, frequency, and phase all decrease, thus affecting the identification of earthquake events, amplitude estimation, and source analysis, and reducing the data reliability and accuracy of the entire earthquake monitoring system. Summary of the Invention

[0006] On the one hand, a machine learning-based fault diagnosis method for earthquake monitoring equipment is provided, which includes: The output frequency of each accelerometer in the seismic monitoring equipment is monitored in real time. After receiving the self-test signal, the self-test and sensitivity detection procedures are executed to determine the output deviation index of each accelerometer and analyze the self-test deviation calibration strategy.

[0007] After completing the self-test and sensitivity detection procedures, multi-dimensional mechanical health characteristics are obtained. Using these characteristics as input, the accelerometer's fault score and dynamic prediction trend are output based on the machine learning model, thereby classifying the fault level and making preliminary interventions.

[0008] Calculate the variance of each fault level, determine the fault level of each accelerometer, and trigger the graded response strategy. When the accelerometer fault level is level two, perform temperature drift detection and trigger temperature control compensation. After comprehensive adjustment, re-perform the periodic fault level analysis.

[0009] On the other hand, a machine learning-based fault diagnosis system for earthquake monitoring equipment is provided. This system applies a machine learning-based fault diagnosis method for earthquake monitoring equipment and includes: The accelerometer self-test module is used to monitor the output frequency of each accelerometer in the earthquake monitoring equipment in real time. After receiving the self-test signal, it executes the self-test and sensitivity detection program to determine the output deviation index of each accelerometer and analyze the self-test deviation calibration strategy.

[0010] The fault level classification module is used to obtain multi-dimensional mechanical health characteristics after completing the self-test and sensitivity detection procedures. Using these characteristics as input, it outputs the fault score and dynamic prediction trend of the accelerometer based on the machine learning model, thereby classifying the fault level and making preliminary interventions.

[0011] The accelerometer diagnostic module is used to calculate the variance of each fault level, determine the fault level of each accelerometer, and trigger a graded response strategy. When the accelerometer fault level is level two, temperature drift detection is performed and temperature control compensation is triggered. After comprehensive adjustment, the periodic fault level analysis is performed again.

[0012] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following: 1. This invention enables accurate prediction and adjustment of accelerometer faults in earthquake monitoring equipment at their initial stage, and automatically takes corresponding measures when the fault escalates, minimizing equipment damage and improving the stability and reliability of system operation. It monitors each accelerometer signal in real time through a set sampling frequency and ensures data quality and accuracy through filtering, noise reduction, and standardization. During self-testing and sensitivity detection, it accurately calculates and compares zero-point drift values ​​and sensitivity deviations, effectively identifying accelerometer deviations and determining whether they are abnormal or causing interference. Through multi-source data fusion and mean square error analysis, it ensures the consistency of data sources, avoids misjudgments caused by environmental factors, and improves the reliability of fault diagnosis. Through a weighted fault scoring function, it accurately assesses different fault levels and adjusts the accelerometer gain coefficient and temperature control compensation based on real-time data, ensuring that the equipment can respond quickly and make effective adjustments under different conditions.

[0013] 2. This invention monitors the accelerometer output signal in real time by setting a sampling frequency, and checks the accelerometer performance in each cycle through self-testing and sensitivity detection programs. It can promptly identify any deviations, detect minute changes in the equipment in real time, ensure that the monitoring data always meets expectations, and detect any potential abnormal signals in the system to promptly identify problems and avoid the accumulation of errors during long-term operation of the equipment, thereby ensuring high-quality and accurate data.

[0014] 3. This invention introduces multi-source data fusion verification to analyze and compare data from various accelerometers and external vibration sensors, ensuring the consistency and reliability of each data source, thereby improving the accuracy of fault diagnosis. By calculating the mean square error, it determines whether there is a device fault or whether the accelerometer deviation is caused by changes in the external environment. It does not rely on a single data source, but makes a more accurate judgment by comprehensively analyzing multi-source data, reducing the risk of false alarms and missed alarms, and ensuring high accuracy of fault diagnosis.

[0015] 4. This invention extracts multi-dimensional mechanical health characteristics, calculates fault scores, and determines fault levels. Based on the fault scores, the system dynamically adjusts the accelerometer gain coefficient or implements temperature control compensation and other measures. Corresponding handling schemes are adopted under different fault levels. Level 1 faults indicate that the equipment is operating stably and the system only performs routine monitoring. Level 2 and Level 3 faults will trigger preliminary intervention, temperature control compensation, damping adjustment and other measures. The most appropriate response strategy is automatically selected according to the severity of the fault to avoid unnecessary intervention, while ensuring the long-term stable operation of the equipment and the high efficiency of fault handling. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a machine learning-based fault diagnosis method for earthquake monitoring equipment provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a machine learning-based fault diagnosis system for earthquake monitoring equipment provided in an embodiment of the present invention; Figure 3 This is a strategy diagram of a machine learning-based fault diagnosis method for earthquake monitoring equipment provided in an embodiment of the present invention; Figure 4 This is a mind map of a machine learning-based fault diagnosis system for earthquake monitoring equipment provided in an embodiment of the present invention. Figure 5 This is a machine learning model diagram of a fault diagnosis system for earthquake monitoring equipment based on machine learning, provided in an embodiment of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0019] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0020] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0021] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0022] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0023] like Figure 1 The flowchart shown illustrates a machine learning-based fault diagnosis method for earthquake monitoring equipment. The process includes the following steps: monitoring the output frequency of the accelerometer, identifying zero-point drift and sensitivity deviation; if the accelerometer's output deviation exceeds a threshold and persists, it is considered a genuine anomaly; if the accelerometer's output deviation is less than the threshold and shows an upward trend, a self-check deviation adjustment is performed, and the equipment status is further determined through multi-source data comparison and analysis. After completing the self-check and sensitivity detection procedures, the system enters the fault analysis stage, extracting mechanical health characteristics and determining the fault level. If the fault level is level two or three, the system will perform initial intervention, followed by temperature control compensation or damping adjustment, and automatically adjust the equipment until the fault is controlled. If the fault remains unresolved, the system will enter a level three response strategy, triggering a shutdown warning.

[0024] Common earthquake monitoring equipment consists of accelerometers, velocity sensors, displacement sensors, and earthquake signal processors.

[0025] This invention provides an in-depth analysis of accelerometers in earthquake monitoring equipment. The core principle of an accelerometer is based on the relative motion between a mass block and a spring system. When seismic waves propagate to the ground, the vibrations caused by the seismic waves will result in relative motion between the ground and the fixed part of the accelerometer, causing the mass block to shift. The accelerometer measures the acceleration generated by this relative motion and converts it into an electrical signal.

[0026] like Figure 3This invention provides a strategy diagram for a machine learning-based fault diagnosis method for earthquake monitoring equipment. The method includes real-time monitoring of the output frequency of each accelerometer, executing self-test and sensitivity detection procedures upon receiving a self-test signal to determine the accelerometer's output deviation index, adjusting the self-test deviation, extracting multi-dimensional mechanical health features, and outputting the accelerometer's fault score and dynamic prediction trend through a machine learning model. Fault levels are then classified, and preliminary intervention operations are performed. This determines the accelerometer's fault level. When the accelerometer fault level is level two, temperature control compensation is performed to correct the impact of temperature changes on the accelerometer. After comprehensive adjustment, a new periodic fault level analysis is conducted to ensure the normal operation of the equipment and timely response to potential faults.

[0027] Furthermore, the output deviation index of each accelerometer is determined, and the specific analysis method is as follows: The frequency of the output signal of each accelerometer is monitored in real time at a set sampling frequency to detect frequency stability. When the frequency stability is greater than or equal to the preset frequency stability threshold in the database, a self-test signal is generated. When the frequency stability is less than the preset frequency stability threshold, a fault level classification signal is generated.

[0028] It should be noted that the output signals of each accelerometer are acquired in real time at a set sampling frequency, and the main frequency component is calculated by fast Fourier transform within each sampling window. The main frequency sequence of the sampling window within the sampling period is recorded, and the standard deviation of these main frequencies is calculated as the frequency stability. The calculated frequency stability is compared with a preset threshold in the database. When the frequency stability is greater than or equal to the threshold, the device frequency operation is determined to be stable, and a self-test signal is generated to record the normal state or trigger internal calibration. When the stability is lower than the threshold, a potential fault risk is determined, and a fault level classification signal is generated for subsequent real-time monitoring and status diagnosis of accelerometer faults.

[0029] It should be explained that the preset frequency stability threshold is usually based on the frequency fluctuation characteristics of the accelerometer under normal operating conditions. The reference range is obtained by statistically analyzing historical operating data, and a reasonable threshold is determined in combination with the equipment's technical specifications, environmental conditions, and operating accuracy requirements. In this embodiment, the stable signals of the equipment under no external interference are sampled first, and the standard deviation and average value of the main frequency over time are calculated. Based on the principle of three times the standard deviation, the threshold is initially set as the average value minus three times the standard deviation, so as to ensure that there are no false alarms when the equipment experiences normal slight drift, and at the same time, it can promptly identify frequency anomalies or potential faults, and realize the accurate triggering of self-test signals and fault level classification signals.

[0030] Upon receiving the self-test signal, the self-test and sensitivity detection program is executed. Based on the applied control signal, the measured mass block offset and actual sensitivity value are obtained and compared with the calculated expected normal mass block offset and standard sensitivity value to obtain the mass block offset difference and sensitivity difference, which are recorded as the zero-point drift value and sensitivity deviation. The frequency drift deviation is obtained by comparing the frequency of the acquired accelerometer output signal with the reference value.

[0031] It should be noted that the expected normal mass offset, zero-point drift reference value, standard sensitivity value, and frequency reference value are all determined through the equipment calibration model or historical benchmark data. Specifically, during the equipment's factory or periodic calibration phase, the system applies a standard control signal with known amplitude and frequency to the accelerometer, measures the relationship between the mass's response displacement and the output voltage or current signal, and establishes a sensitivity calibration curve. Based on this curve, the theoretical response can be calculated under any input conditions to obtain the expected normal offset, and the slope of the curve or the output amplitude corresponding to a unit input is the standard sensitivity. Simultaneously, during frequency calibration, a series of standard frequency signals are applied to measure the equipment's main output frequency and establish an input-output correspondence. The frequency band with the most stable output frequency and the most consistent amplitude response is selected as the reference benchmark, and the frequency reference value is determined by combining damping characteristics and environmental correction factors. During operation, the system can obtain the expected normal offset, zero-point drift reference value, standard sensitivity, and frequency reference value by calling the aforementioned calibration model and reference database.

[0032] It should be explained that after applying the preset control signal, the displacement sensing unit collects the motion response signal of the mass block in real time and simultaneously records the voltage or current signal output by the accelerometer. Based on the known input amplitude and frequency of the control signal, the system uses the accelerometer's mechanical response model to convert the output signal into the corresponding mass block displacement, thereby calculating the measured mass block offset. The sensitivity value can be obtained by the ratio of the output signal amplitude to the input control signal amplitude, i.e., the output signal amplitude divided by the input control signal amplitude. The measured mass block offset divided by the expected normal mass block offset yields the mass block offset difference, which is the zero-point drift value. The sensitivity difference obtained by subtracting the standard sensitivity value from the actual sensitivity value and then dividing by the standard sensitivity value is the sensitivity deviation. The frequency drift deviation is obtained by dividing the frequency of the collected accelerometer output signal by the frequency reference value.

[0033] It should be added that the mechanical response model of an accelerometer is essentially a classic second-order mass-spring-damped system. The sensitive mass is connected to the outer shell through an elastic cantilever beam, and its motion follows Newton's second law, that is, the inertial force of the mass is balanced by the external acceleration, the elastic restoring force generated by the spring deformation, and the viscous damping force caused by the damping medium. This physical relationship can be accurately described by a second-order ordinary differential equation. By solving this equation, it can be found that the system's response to impact in the time domain is reflected in the damping decay time and the coefficient of restitution, and in the frequency domain, it is reflected in the inherent resonant frequency and bandwidth. Any mechanical failure that leads to performance degradation, such as structural fatigue, component adhesion, or damping changes, can be mapped to changes in stiffness, damping, or mass parameters in the model, thereby directly affecting its output characteristics.

[0034] It should be added that the amplitude of the input control signal can be obtained by acquiring the amplitude of the preset control signal, while the amplitude of the output signal can be obtained directly from the electrical signal read from the accelerometer itself.

[0035] The output deviation index of each accelerometer is obtained by combining the frequency drift deviation, zero drift value, and sensitivity deviation.

[0036] Frequency drift reflects the change in the accelerometer's natural frequency during long-term operation. When the operating frequency drifts, the dynamic response characteristics of the system will shift, causing high-frequency or low-frequency components to be amplified or attenuated, thus distorting the output signal. Zero-point drift represents the deviation of the output acceleration under no-input conditions. Its accumulation will cause a systematic shift in the acceleration baseline, resulting in errors in the detection results of static or weak vibrations. When the sensitivity is reduced or unbalanced, ground vibrations of the same amplitude will be misjudged as signals of different intensities. Under the combined effect of these three factors, the output signal will deviate in multiple dimensions in the amplitude, phase, and frequency domains, thus affecting the accuracy of the accelerometer in truly reflecting ground vibrations.

[0037] Frequency drift deviation can cause changes in the resonant characteristics of the accelerometer, resulting in nonlinear fluctuations in the system's response sensitivity to signals in different frequency bands. This amplifies the impact of sensitivity deviation. When the sensitivity changes unevenly, the dynamic range and amplification ratio of the output signal will also change, further exacerbating the cumulative effect of zero drift at the output. Increased zero drift can also interfere with the system's automatic calibration process, causing sensitivity correction distortion and indirectly affecting the stability of the frequency response.

[0038] , Where D represents the output deviation index of the accelerometer, B represents the sensitivity deviation, G represents the frequency drift deviation, E represents the zero drift value, and E0 represents the zero drift reference value.

[0039] Furthermore, the self-test deviation calibration strategy is analyzed, and the specific analysis method is as follows: Extract the output deviation index threshold and confirmation window size from the database.

[0040] It should be noted that the preset output deviation threshold and confirmation window size are usually determined based on the equipment's operating characteristics, historical data statistics, and requirements for fault sensitivity. By long-term acquisition of the output signals of accelerometers or seismometers under normal operating conditions, the amplitude fluctuation range, instantaneous interference characteristics, and short-term trend changes are analyzed. Combined with the equipment's technical specifications and allowable error range, an output deviation threshold that can distinguish normal minor fluctuations and promptly identify anomalies is determined. Based on the signal sampling frequency and fault response time requirements, the confirmation window size is set so that the system only recognizes a real anomaly when the deviation persists beyond the window, thereby avoiding false alarms caused by brief interference or occasional noise and achieving reliable judgment and early warning of the equipment's output status.

[0041] For an accelerometer, if the output deviation index of the accelerometer is greater than or equal to the output deviation index threshold, it is detected whether the output deviation index of the accelerometer persists within the confirmation window. If the output deviation persists within the confirmation window, the output deviation is identified as a real anomaly, the self-test deviation calibration strategy is recorded as no adjustment is performed, and a fault level classification signal is generated.

[0042] It should be noted that for an accelerometer, if the output deviation index is greater than or equal to the output deviation index threshold, it means that the current output signal of the accelerometer has exceeded the allowable fluctuation range under normal operating conditions, indicating a potential anomaly or performance deviation. This may be due to accelerometer problems or environmental interference. Therefore, it is necessary to check whether the output deviation index of the accelerometer persists within the confirmation window. If the output deviation persists within the confirmation window, it means that the output deviation index of the accelerometer is not an occasional or transient interference, but a persistent real anomaly, reflecting a real performance deviation or potential fault in the equipment. Fault level classification and further diagnosis are required to ensure the reliability of monitoring data and the safe operation of the equipment. Therefore, the output deviation is identified as a real anomaly, the self-test deviation calibration strategy is recorded as no adjustment is performed, and a fault level classification signal is generated.

[0043] If the detected output deviation does not persist within the confirmation window, the output deviation is considered as transient interference, the self-test deviation calibration strategy is recorded as no adjustment is made, and no fault level classification signal is generated.

[0044] It should be noted that if the detected output deviation does not persist within the confirmation window, it means that the accelerometer deviation is only a short-term or instantaneous disturbance with a small amplitude, and is not a real fault or long-term performance deviation. The overall operating status of the equipment is still normal. Therefore, there is no need to perform self-test adjustment or fault level classification. This event can be regarded as an occasional fluctuation and ignored. Therefore, the self-test deviation calibration strategy is recorded as no adjustment.

[0045] Obtain the output deviation sequence of the accelerometer within the confirmation window, thereby obtaining the slope sequence of the accelerometer, and analyze the average slope of the output deviation index of the accelerometer within the confirmation window.

[0046] It should be noted that, firstly, the output deviation index sequence of the accelerometer is collected in real time within the confirmation window, and the deviation value corresponding to each sampling point is recorded. By calculating the deviation change of adjacent sampling points and the sampling time interval, the deviation slope sequence of each time period is obtained. The slope sequence is averaged over the entire confirmation window to obtain the mean value of the output deviation slope.

[0047] If the accelerometer's output deviation index is less than the output deviation index threshold and the average slope of the accelerometer's output deviation index is greater than or equal to the slope threshold within the confirmation window, the self-test deviation calibration strategy is recorded as self-test deviation adjustment, and this accelerometer is recorded as an accelerometer to be verified. No fault level classification signal is generated.

[0048] It should be noted that if the accelerometer's output deviation index is less than the output deviation index threshold and the average slope of the accelerometer's output deviation index is greater than or equal to the slope threshold within the confirmation window, it means that although the deviation amplitude has not yet exceeded the allowable range, its changing trend is relatively fast, and the accelerometer's output is showing a trend of gradually deviating from the normal state. This suggests that the accelerometer may have slight drift or early performance degradation, which is a warning signal of potential abnormality. It is necessary to trigger self-test deviation adjustment for calibration or compensation, but it is not enough to determine a fault or classify the fault level.

[0049] If the accelerometer's output deviation index is less than the output deviation index threshold and the slope of the accelerometer's output deviation index is less than the slope threshold within the confirmation window, then the self-test deviation calibration strategy will be recorded as maintaining routine monitoring, the data of this stage will be stored in the historical database, and no fault level classification signal will be generated.

[0050] It should be noted that if the accelerometer's output deviation index is less than the output deviation index threshold and the slope of the accelerometer's output deviation index is less than the slope threshold within the confirmation window, it means that the accelerometer's output signal is neither outside the allowable deviation range nor has a stable trend, indicating that the equipment is operating normally without significant drift or potential anomalies. This indicates that the current measurement results are reliable, and no self-test adjustment or fault level classification is required. It can be considered as a normal operating state, and the self-test deviation calibration strategy should be recorded as maintaining routine monitoring.

[0051] Further, the self-test deviation adjustment is analyzed using the following specific methods: Multi-source data fusion verification is introduced, and observation data from the accelerometer to be verified, other accelerometers in the seismometer, and external vibration sensors are collected. The amplitude, frequency, and phase of each element are paired and compared within a unified time window. Then, the mean square error between each element is calculated and compared with the set error threshold.

[0052] It should be noted that the accelerometer to be verified is synchronized with other accelerometers in the seismometer and external vibration sensors in time to ensure that all signals correspond to the same sampling point within the same time window. Subsequently, for each pair of components, their amplitude, frequency and phase are compared one by one, the square of the difference at each sampling point is calculated, and the average of the squared differences over the entire time window is calculated to obtain the mean square error of the pair of components. The mean square error among all components is calculated and recorded as the mean square error of the accelerometer to be verified.

[0053] When the mean square error is greater than or equal to the error threshold, the self-test deviation adjustment is recorded as the reduction factor of the output deviation index threshold and the shortening value of the sampling and self-test interval determined by the slope mean of the output deviation index of the accelerometer. The output deviation index threshold and the sampling and self-test interval are updated based on the reduction factor of the output deviation index threshold and the shortening value of the sampling and self-test interval.

[0054] It should be noted that when the mean square error is greater than or equal to the error threshold, it means that the output signal of the accelerometer to be verified deviates significantly from other reference accelerometers, reflecting that the equipment may have potential abnormalities or performance drift. In this case, the self-test deviation adjustment strategy is set to be a dynamic adjustment based on the average slope of the accelerometer output deviation. The reduction factor of the output deviation index threshold and the shortening value of the sampling and self-test interval are determined by the average slope value. The output deviation index threshold and the sampling and self-test interval are updated according to the reduction factor and the shortening value of the interval, thereby ensuring that potential abnormalities of the equipment can be captured more promptly, and realizing dynamic monitoring and preventive maintenance of the accelerometer's operating status.

[0055] In this embodiment, the preset process for the reduction coefficient of the output deviation index threshold and the shortening value of the sampling and self-test interval is based on a pre-set mapping rule. First, the mean slope of the output deviation of the accelerometer to be verified within the confirmation window is calculated to quantify the trend and rate of deviation change. The mapping rule is preset through statistical analysis of historical operating data, including the output deviation index threshold reduction coefficient, sampling frequency adjustment range, and self-test interval shortening value corresponding to different slope mean ranges. It is initially determined in combination with the analysis of equipment importance, fault tolerance requirements, and system sensitivity to thresholds. Continuous or discrete mapping functions are generated through the mean, weighted mean, or percentile methods, and a safety margin is added to cope with abnormal fluctuations. Based on the slope mean obtained in real time, the corresponding output deviation index threshold reduction coefficient and sampling and self-test interval shortening value are generated through the mapping rule: the larger the slope mean, the larger the reduction coefficient and the smaller the sampling and self-test interval shortening value, so as to improve the self-test sensitivity and frequency, thereby enhancing the early detection capability of potential deviation anomalies.

[0056] When the mean square error is less than the error threshold, the self-test deviation adjustment is recorded as not being adjusted.

[0057] It should be noted that when the mean square error is less than the error threshold, it means that the output signal of the accelerometer to be verified is in good consistency with other reference accelerometers, the equipment is operating stably, and there is no significant deviation or potential abnormality. The self-test deviation adjustment strategy is marked as not to be adjusted, and the current output deviation index threshold and sampling and self-test interval remain unchanged, thereby avoiding unnecessary self-test or calibration operations triggered by occasional fluctuations or small deviations.

[0058] like Figure 5 This is a machine learning model diagram of a fault diagnosis system for earthquake monitoring equipment based on machine learning, provided in an embodiment of the present invention. The machine learning model takes multi-dimensional mechanical health features, environmental sensor data, related circuit signals, and equipment status extracted in real time by the accelerometer as input. By deeply fusing and jointly analyzing these heterogeneous features that reflect the dynamic performance of the mechanical system, the model can accurately learn the complex mapping relationship between different fault modes and feature combinations. Through the a1 hidden layer and the a2 hidden layer, 32 neurons and 16 neurons are arranged in the box, respectively. Every 2 neurons in the a2 hidden layer are connected to all neurons in the output layer, and finally, an intuitive fault score is output to quantify the current degree of equipment health degradation. At the same time, a dynamic prediction trend is generated to predict the evolution of accelerometer performance in the future.

[0059] Furthermore, fault severity classification and preliminary intervention are carried out. The specific analysis methods are as follows: After receiving the fault level classification signal, multidimensional mechanical health features are extracted from the accelerometer data that generated the signal. These multidimensional mechanical health features are then used as input to output the accelerometer's fault score and dynamic prediction trend based on a machine learning model.

[0060] It should be noted that after receiving the fault level classification signal, the accelerometer data collected for generating the signal is subjected to denoising, normalization, and spatiotemporal alignment operations. Then, multidimensional mechanical health features are extracted, such as sensitivity deviation, mass block displacement change, damping decay time, recovery coefficient, and micro-displacement stability index. These features comprehensively reflect the dynamic performance and potential abnormal conditions of the accelerometer and mechanical structure. The extracted multidimensional mechanical health features are used as input to a pre-trained machine learning model. The model's prediction or classification capabilities are used to output the accelerometer's fault score and dynamic prediction trend.

[0061] It should be added that the mechanical structure refers to the microelectromechanical system mechanical sensing unit inside the accelerometer.

[0062] Extract the preset fault rating range from the database.

[0063] In this embodiment, the fault scoring interval logic preset in the database is determined based on statistical analysis of a large number of historical samples and model output results. The fault score value is divided into several continuous intervals, each interval corresponding to a different level of equipment health status. The division of fault scoring intervals usually follows a progressive logic from low to high: below the lower limit of the fault scoring interval corresponds to normal or slight deviation of equipment operation, within the fault scoring interval corresponds to potential abnormalities or performance degradation, and above the upper limit of the fault scoring interval corresponds to obvious fault or failure risk. The lower limit and upper limit of the fault scoring interval can be determined by empirical thresholds, statistical distribution quantiles or clustering results, and stored in the database in the form of parameters so that they can be directly called in subsequent diagnosis to realize automatic matching of scoring results and status levels.

[0064] An accelerometer fault level is defined as a Level 1 fault when the fault score is less than or equal to the lower limit of the fault score range.

[0065] It should be noted that when the accelerometer's fault score is less than or equal to the lower limit of the preset fault score range in the database, it means that the accelerometer's performance indicators deviate from the normal state to the minimum, its quality is good, the equipment is operating stably, and its sensitivity and response characteristics remain within the ideal range without significant drift or attenuation. Therefore, it is judged as a Level 1 fault.

[0066] Accelerometer faults with fault scores falling within the fault score range are classified as Level 2 faults.

[0067] It should be noted that a fault score within the fault score range means that the equipment performance has deviated to a certain extent or is slightly abnormal, but the overall system is still operational. At this time, the accelerometer's sensitivity, linearity, or response characteristics may drift slightly, but have not yet affected the core monitoring function. The accelerometer fault level is recorded as a level two fault.

[0068] An accelerometer fault level is defined as a level three fault when the fault score is greater than or equal to the upper limit of the fault score range.

[0069] It should be noted that when the fault score is greater than or equal to the upper limit of the fault score range, it indicates that the equipment performance has deteriorated significantly. The accelerometer has problems such as decreased sensitivity, output distortion, response lag, or signal drift, which has exceeded the allowable range for normal operation. Therefore, the accelerometer fault level is recorded as a level three fault.

[0070] Accelerometer fault levels are collected within the sampling period. When a level 2 or level 3 fault is detected, the corresponding accelerometer is marked as an abnormal accelerometer. The abnormal accelerometers are statistically analyzed, and preliminary intervention is carried out.

[0071] The initial intervention, specifically the following steps: For an abnormal accelerometer, the increase in accelerometer gain coefficient is determined based on the dynamic prediction trend score. The updated value of accelerometer gain coefficient is obtained based on the increase in accelerometer gain coefficient and the accelerometer gain coefficient. The updated value of accelerometer gain coefficient is then applied to the accelerometer.

[0072] It should be noted that the development trend of performance deviation is evaluated based on dynamic prediction trend score. The increase of the accelerometer gain coefficient is determined according to the trend score to compensate for sensitivity attenuation or signal attenuation effect. The calculated increase is superimposed with the current accelerometer gain coefficient to obtain the updated accelerometer gain coefficient value. The updated gain coefficient parameter is written into the accelerometer control module so that the signal is amplified with the corrected gain in subsequent acquisition, thereby realizing adaptive compensation and performance stabilization adjustment of the accelerometer output.

[0073] In this embodiment, the process of determining the increase in accelerometer gain coefficient based on dynamic prediction trend score is based on a pre-set mapping rule. First, the dynamic prediction trend score of the abnormal accelerometer is calculated using historical operating data and current monitoring results to reflect the development trend of sensor performance degradation or signal drift. The mapping rule is preset through statistical analysis of a large number of historical calibration and degradation samples, including the mean, extreme values ​​and typical change ranges of gain adjustment amplitude under different trend score ranges. It also comprehensively considers factors such as sensor sensitivity attenuation rate, signal stability and system response sensitivity to gain changes to form a continuous or discrete mapping function. Based on the current trend score of the accelerometer, the corresponding increase in gain coefficient is generated through this mapping function: the higher the trend score, the more obvious the performance degradation, and the larger the corresponding gain adjustment amplitude, so as to achieve adaptive compensation and sensitivity recovery of the output signal.

[0074] Furthermore, the fault level of each accelerometer was determined, and the specific analysis method is as follows: For an accelerometer, calculate the variance of level 1, level 2, and level 3 faults within a sampling period to obtain the total variance of faults within the period, and thus obtain the variance contribution rate of each fault level.

[0075] It should be noted that within a sampling period, the corresponding Level 1, Level 2, and Level 3 fault score data sequences are extracted separately, and the variance of each fault level score is calculated to reflect the dispersion of fault fluctuations at different levels. The variances of each level are summed to obtain the total fault variance within the sampling period, which serves as a measure of the overall fault fluctuation intensity. The variance contribution rate is obtained by dividing the variance of each level by the total variance, which measures the proportion of each fault level in the overall fluctuation. The higher the variance contribution rate, the more significant the impact of the fluctuation of that level on the overall fault instability of the accelerometer, providing a quantitative basis for subsequent fault feature identification and priority intervention.

[0076] Extract the preset contribution rate threshold from the database.

[0077] In this embodiment, the preset logic of the contribution rate threshold is mainly used to distinguish the significance of different fault levels in the overall fluctuation. By statistically analyzing the variance contribution rate distribution of historical operating data and multiple batches of equipment samples, the typical range and boundary point of the contribution rate of each level are determined. Based on these statistical results, combined with equipment stability requirements, fault sensitivity analysis and misjudgment tolerance, the contribution rate threshold is preset.

[0078] When the variance contribution rate of a certain fault level is greater than or equal to the contribution rate threshold within the sampling period, the fault level is taken as the fault level of the accelerometer.

[0079] It should be noted that when the variance contribution rate of a certain level of fault is greater than or equal to the contribution rate threshold within the sampling period, it indicates that the fault of that level dominates the overall fluctuation in this period, and its corresponding fault characteristics have the most significant impact on the accelerometer's operating status. In this case, the fluctuation of this level is considered to dominate the device's performance deviation and output anomaly, reflecting the accelerometer's main fault modes and health characteristics within this period. Based on this judgment, this level is identified as the representative fault level of the accelerometer within this sampling period.

[0080] If the variance contribution rate of all three fault levels is less than the contribution rate threshold within the sampling period, calculate the difference between the variances of each level and extract the preset variance difference threshold from the database.

[0081] It should be noted that when the variance contribution rates of the three fault levels calculated within a sampling period are all lower than the contribution rate threshold, it indicates that no single fault level plays a significant dominant role in that period. That is, the overall fault fluctuation is relatively weak or the distribution is relatively balanced. In order to further determine the subtle state differences of the accelerometer, the difference between the variances of each fault level is calculated to compare the relative strength of the fluctuation amplitude between different levels. The preset variance difference threshold is extracted from the database.

[0082] The preset logic of the variance difference threshold is determined based on the statistical analysis and model training results of a large amount of historical sampling period data. First, the variance distribution characteristics of the accelerometer under different health states and fault levels are collected, and the difference range and typical change patterns between the variances of each level are calculated. By comparing the distribution differences of variance differences between normal operation and abnormal state, the critical value range that can effectively distinguish the significance of fault level fluctuations is extracted. Combined with the equipment sensitivity, environmental noise level and system judgment stability requirements, the preliminary threshold is smoothed and corrected with safety margin to form the final variance difference threshold.

[0083] When the maximum variance difference exceeds the variance difference threshold, the fault level corresponding to the maximum variance difference is determined as the fault level of the accelerometer.

[0084] It should be noted that when the maximum variance difference exceeds the variance difference threshold, it means that the fluctuation difference between different fault levels within the sampling period has reached a significant level. Among them, the fault level corresponding to the maximum variance difference is the most prominent in the overall fluctuation, reflecting that the fault characteristics corresponding to this level have the strongest impact on the accelerometer's operating status. Based on this, it is judged that this level represents the main abnormal performance of the accelerometer within this period, and the fault level corresponding to the maximum variance difference is determined as the fault level of the accelerometer.

[0085] If the maximum variance difference does not exceed the variance difference threshold, the highest fault level occurring within that period is taken as the fault level of the accelerometer.

[0086] It should be noted that when the maximum variance difference does not exceed the preset variance difference threshold, it means that the fluctuation difference between each fault level is not significant enough, and the dominant fault level cannot be clearly distinguished by the variance change characteristics. This indicates that the overall fluctuation of the accelerometer is relatively balanced within the sampling period and there are no obvious prominent fault characteristics. At this time, in order to avoid misjudgment or ignoring potential risks, a conservative judgment strategy is adopted, and the highest fault level that occurs within the period is taken as the fault level of the accelerometer. This ensures that the diagnostic results still have safety redundancy and protection under uncertain conditions, thereby avoiding the system underestimating the actual fault degree of the equipment.

[0087] Furthermore, the tiered response strategy is triggered, and the specific analysis method is as follows: If the accelerometer fault level is Level 1, the graded response strategy is recorded as data verification. The integrity and consistency of the collected data are verified to confirm that the accelerometer output and communication link are normal. Then, the monitoring indicators and characteristic parameters within this period are recorded.

[0088] It should be noted that if the accelerometer's fault level is Level 1, it indicates that its operating status is good and there are no obvious performance abnormalities. The graded response strategy is set to data verification to ensure the reliability of monitoring results and the stability of the data link. First, the integrity of the original data collected by the accelerometer in this sampling period is checked to confirm that there are no frame drops, breaks, or abnormal interruptions in the data sampling. Then, a consistency verification is performed to compare the synchronous data of multiple channels or adjacent accelerometers to verify the consistency of the accelerometer's output amplitude, phase, and trend, ensuring that the signal transmission and communication link are in normal condition. After the verification is completed, the key monitoring indicators and characteristic parameters in this period, including but not limited to sensitivity deviation, noise level, and micro-displacement stability, are recorded and archived to provide data support for long-term performance tracking and baseline updates.

[0089] If the accelerometer fault level is level two, the graded response strategy will be recorded as integrated adjustment.

[0090] It should be noted that if the accelerometer fault level is level two, it means that the accelerometer's performance has a slight deviation or instability, but it has not yet affected the overall monitoring function. For this moderate deviation, the graded response strategy is recorded as comprehensive adjustment, so as to dynamically optimize the accelerometer status without interrupting operation.

[0091] If the accelerometer fault level is level three, the graded response strategy will be recorded as a shutdown alarm. The system will automatically shut down, and an alarm will generate maintenance task information, prompting that the accelerometer replacement operation needs to be performed.

[0092] It should be noted that if the accelerometer fault level is three, it indicates that the accelerometer has experienced a serious performance abnormality and can no longer guarantee the reliability and security of the monitoring data. To prevent the fault from escalating further or affecting the normal operation of the overall monitoring system, the tiered response strategy is recorded as a shutdown alarm. Upon confirming the level three fault status, a shutdown command is immediately sent to the control module to terminate the accelerometer's data acquisition and signal output process. Simultaneously, an alarm message is generated, and the time, location, and key characteristic parameters of the fault are recorded. Subsequently, the system automatically generates maintenance task information, marking it as requiring accelerometer replacement, and pushes the task to the maintenance management platform or the on-duty personnel's terminal, prompting on-site personnel to perform accelerometer replacement or in-depth maintenance.

[0093] In this embodiment, the alarm message could be: "Attention! Accelerometer fault level is three." Further comprehensive adjustments and specific analysis methods are as follows: Accelerometers with level 2 faults are categorized as demand-regulated accelerometers.

[0094] For a demand-adjustable accelerometer, temperature drift detection is performed. The specific steps are as follows: collect ambient temperature data, calculate the correlation coefficient between temperature and fault score, perform linear regression fitting on fault score and temperature data, and obtain the regression slope and goodness-of-fit value.

[0095] It should be noted that for accelerometers requiring adjustment, the process first involves temperature drift detection to assess the impact of temperature changes on their fault scores. Temperature data of the accelerometer's environment is collected synchronously during the sampling period, and fault score sequences for the corresponding time periods are extracted to form a temperature-scoring sample set. The correlation coefficient between temperature and fault scores is calculated, and then linear regression fitting is performed on the fault scores and temperature data to obtain the regression slope and goodness-of-fit value. The regression slope reflects the sensitivity of the score shift to temperature changes, while the goodness-of-fit value assesses the model's interpretability of the actual data. This result will serve as an important basis for subsequent temperature compensation and gain adjustment, correcting the accelerometer's output deviation under different temperature conditions.

[0096] Extract the preset regression slope threshold and confidence threshold from the database.

[0097] In this embodiment, the preset logic for the regression slope threshold and the confidence threshold is determined based on the statistical regularity of historical sample data and operational experience. The system first performs linear regression analysis on a large number of accelerometer operating samples under normal and abnormal conditions, statistically analyzing the regression slope distribution and the corresponding goodness-of-fit value distribution between temperature changes and fault scores. By comparing the differences between normal and abnormal samples on these two indicators, the critical interval with the highest discrimination is extracted. Combined with the accelerometer's sensitivity level, the range of ambient temperature fluctuations, and the system's tolerance to false alarms, preliminary thresholds are set. These two thresholds are dynamically corrected during multiple runs and verifications, so that they can effectively identify performance deviations caused by temperature drift while avoiding misjudging random noise as temperature coupling anomalies, ultimately forming a stable and reliable regression slope threshold and confidence threshold mapping rule.

[0098] When the regression slope exceeds the preset regression slope threshold and the goodness of fit is higher than the confidence threshold, temperature control compensation measures are triggered. The change in temperature compensation resistance is obtained by using the regression slope and temperature change value. The updated value of temperature compensation resistance is obtained by using the change in temperature compensation resistance and the temperature compensation resistance in the system log. The compensation resistance in the bridge circuit is automatically adjusted by the digital control device in the circuit.

[0099] It should be noted that when the regression slope exceeds the preset regression slope threshold and the goodness-of-fit value is higher than the confidence threshold, it means that the accelerometer's fault score exhibits a significant linear dependence on changes in ambient temperature, and this linear relationship has high confidence. The impact of temperature changes on the accelerometer output has exceeded the normal fluctuation range, and this impact is not caused by accidental noise or random fluctuations, but rather by a stable and repeatable offset trend. This indicates that the accelerometer has a significant temperature drift effect, and intervention is needed through measures such as temperature compensation or gain adjustment to restore the accuracy of the accelerometer output and the reliability of the monitoring system.

[0100] Based on the temperature data and accelerometer fault score sequence within the sampling period, the regression slope is calculated using a linear regression method. Combined with the actual temperature change within the same period, the sensitivity of temperature to the accelerometer output is quantified. The regression slope and temperature change are input into a pre-set mapping formula or mapping function. This mapping formula, based on historical sample analysis and experimental calibration, defines the linear or nonlinear adjustment relationship of temperature deviation on the bridge compensation resistor. The corresponding temperature compensation resistor change is calculated through the mapping formula, thereby providing a precise adjustment amount for the bridge circuit, enabling the accelerometer output to remain stable and accurately achieve temperature drift compensation under different temperature conditions.

[0101] If the regression slope does not exceed the preset regression slope threshold or the goodness-of-fit value is not higher than the confidence threshold, the system will perform controlled damping adjustment.

[0102] It should be noted that if the regression slope does not exceed the preset regression slope threshold or the goodness of fit is not higher than the confidence threshold, it means that the temperature has no significant effect on the accelerometer output. That is, the temperature drift effect is not enough to be intervened by the temperature compensation resistor. In this case, controlled damping adjustment should be carried out to optimize the accelerometer performance through mechanical adjustment.

[0103] The initial damping adjustment value is determined based on the fault score, and the damper is adjusted based on the calculated initial damping adjustment value.

[0104] It should be noted that the initial damping adjustment value is calculated based on the fault score of the current sampling period to quantify the relationship between the accelerometer output deviation and the damping requirement. This initial damping adjustment value is applied to the damper, and by adjusting the damping coefficient or friction of the damper, the mechanical vibration characteristics of the accelerometer can be adjusted online, thereby reducing output fluctuations, optimizing sensitivity response, and maintaining the accelerometer within an acceptable performance range.

[0105] The process of determining the initial damping adjustment value based on the fault score is based on a pre-set mapping rule. First, the deviation between the fault score of the accelerometer and the standard reference value in the current sampling period is calculated to obtain the deviation of each key performance indicator. The mapping rule is preset through statistical analysis of historical accelerometer operating data, including the mean, extreme values, and typical fluctuation range of the initial damping adjustment value under different fault score deviation ranges. It is also preliminarily determined by combining the accelerometer sensitivity, mechanical characteristics, and the system's sensitivity to the adjustment threshold. Then, a continuous or discrete mapping function is generated through the mean, weighted mean, or percentile method, while a safety margin is added to deal with abnormal deviations. Based on the fault score deviations in the current sampling period, the corresponding initial damping adjustment value is generated through this mapping rule: the larger the fault score deviation, the higher the adjustment value, so as to enhance the damping adjustment effect, thereby achieving control and performance optimization of accelerometer output fluctuations.

[0106] Furthermore, after comprehensive adjustments, the accelerometer fault level analysis was re-performed. The specific analysis method is as follows: After comprehensive adjustment, the accelerometer fault level analysis is performed again. If the accelerometer is in a level one fault state after comprehensive adjustment, a compensation completion signal is generated.

[0107] It should be noted that after collecting and adjusting the accelerometer output data, the accelerometer fault level analysis is re-performed. The fault score obtained after adjustment is compared with the preset fault score range to re-determine the fault level of the accelerometer. If the analysis results show that the accelerometer has recovered to the first-level fault state, it indicates that the accelerometer performance has recovered to an excellent state, the output signal is stable and reliable, and a compensation completion signal is generated.

[0108] If the accelerometer is in a level 2 fault state after comprehensive adjustment, the secondary damping adjustment value is determined based on the fault score and the average value of the temperature compensation resistor change, and the damper is adjusted according to the calculated secondary damping adjustment value.

[0109] It should be noted that if the accelerometer is in a level 2 fault state after comprehensive adjustment, it means that although the system has taken adjustment measures, the performance of the accelerometer has not yet returned to the normal range, and there is a certain degree of deviation or instability. Although the fault has not reached a serious level, further optimization and adjustment are still needed, including damping adjustment.

[0110] The process of determining the secondary damping adjustment value based on the average of the fault score and the temperature compensation resistor change is based on preset mapping rules. The system first calculates the average of the fault score and the temperature compensation resistor change within the current sampling period. These two indicators reflect the accelerometer's performance deviation and the impact of temperature on the accelerometer output, respectively. Through statistical analysis of historical data, damping adjustment values ​​are preset for different fault score and resistance change ranges, including their average, extreme values, and typical fluctuation ranges. Combined with the system's sensitivity to adjustment, fault tolerance requirements, and the importance of adjustment, preliminary secondary damping adjustment rules are formulated. Based on these rules, the system generates the corresponding secondary damping adjustment value according to the calculated average of the fault score and resistance change. When the fault score deviation is large or the temperature compensation resistor change is high, the system increases the damping adjustment value to enhance the damping effect and reduce accelerometer output fluctuations.

[0111] If the accelerometer is in a level 3 fault state after comprehensive adjustment, roll back the operation and jump to the level 3 fault response strategy.

[0112] It should be noted that if the accelerometer is in a level 3 fault state after comprehensive adjustment, it means that the accelerometer's performance has been severely compromised. The adjustment has not been able to restore it to the normal operating range and cannot meet the system's requirements for monitoring accuracy and reliability. In this case, a rollback operation is triggered, and the system switches to the level 3 fault response strategy. First, the rollback operation is executed to cancel the previous adjustment measures, including temperature compensation and damping adjustment, and restore the sensor to its original or preset safe state. Then, the accelerometer's fault level is recorded as level 3 and marked as a serious fault state. This information is ensured to be stored in a timely manner and pushed to maintenance personnel or the control center to prevent further misoperation or damage to the equipment.

[0113] The embodiments of the present invention provide, as follows Figure 2 The diagram shown is a structural schematic of a machine learning-based fault diagnosis system for earthquake monitoring equipment. The system's processing flow may include the following steps: an accelerometer self-test module, a fault level classification module, and an accelerometer diagnosis module.

[0114] Among them, the accelerometer self-test module is used to monitor the output frequency of each accelerometer of the earthquake monitoring equipment in real time. After receiving the self-test signal, it executes the self-test and sensitivity detection program to determine the output deviation index of each accelerometer and analyze the self-test deviation calibration strategy.

[0115] The fault level classification module is used to obtain multi-dimensional mechanical health characteristics after completing the self-test and sensitivity detection procedures. Using these characteristics as input, it outputs the fault score and dynamic prediction trend of the accelerometer based on the machine learning model, thereby classifying the fault level and making preliminary interventions.

[0116] The accelerometer diagnostic module is used to calculate the variance of each fault level, determine the fault level of each accelerometer, and trigger a graded response strategy. When the accelerometer fault level is level two, temperature drift detection is performed and temperature control compensation is triggered. After comprehensive adjustment, the periodic fault level analysis is performed again.

[0117] See Figure 4 The diagram shown is a mind map of a somatosensory interactive cardiovascular nursing operation training system provided by an embodiment of the present invention. The system monitors the output signal of the accelerometer, executes self-test and sensitivity detection procedures, calculates the accelerometer deviation, and performs self-test deviation calibration strategy analysis. In the fault analysis stage, multidimensional mechanical health characteristics are extracted, and fault scores are calculated and evaluated to determine the accelerometer fault level. For level two or three faults, preliminary intervention is performed first, and then measures such as temperature compensation and damping adjustment are determined according to the accelerometer fault level to adjust the equipment status and ensure stable equipment operation.

[0118] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0119] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0120] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0121] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0122] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0123] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0124] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for diagnosing a fault of a seismic monitoring device based on machine learning, characterized by, The method comprises: Real-time monitoring of the output frequency of each accelerometer of the seismic monitoring device, performing a self-checking and sensitivity detection procedure after receiving a self-checking signal, thereby determining the output deviation index of each accelerometer, analyzing the self-checking deviation calibration strategy; After completing the self-checking and sensitivity detection procedure, obtaining multi-dimensional mechanical health features, taking them as input, and outputting the fault score and dynamic prediction trend of the accelerometer based on a machine learning model, thereby performing fault level classification and preliminary intervention; Calculating the variance of each fault level, determining the fault level of each accelerometer, and triggering a hierarchical response strategy, when the fault level of the accelerometer is a secondary fault, performing temperature drift detection, and triggering temperature control compensation, and after comprehensive adjustment, reanalyzing the fault level of the accelerometer. 2.The machine learning based fault diagnosis method for seismic monitoring equipment according to claim 1, wherein, The specific analysis method for determining the output deviation index of each accelerometer is as follows: Real-time monitoring of the frequency of the output signal of each accelerometer at a set sampling frequency, detecting the frequency stability, generating a self-checking signal when the frequency stability is greater than or equal to a preset frequency stability threshold, and generating a fault level classification signal when the frequency stability is less than the preset frequency stability threshold; After receiving the self-checking signal, performing a self-checking and sensitivity detection procedure, obtaining the actual mass block offset and actual sensitivity value according to the applied control signal, and comparing them with the calculated expected normal mass block offset and standard sensitivity value to obtain the mass block offset difference and sensitivity difference, denoted as zero-point drift value and sensitivity deviation, and comparing the frequency of the collected accelerometer output signal with the reference value to obtain the frequency drift deviation; The frequency drift deviation, zero-point drift value and sensitivity deviation are integrated to obtain the output deviation index of each accelerometer. 3.The machine learning based fault diagnosis method of a seismic monitoring device according to claim 1, characterized in that, The specific analysis method for analyzing the self-checking deviation calibration strategy is as follows: Extract the output deviation index threshold and confirmation window size from the database; For an accelerometer, if the output deviation index of the accelerometer is greater than or equal to the output deviation index threshold, detect whether the output deviation index of the accelerometer persists within the confirmation window, if the output deviation persists within the confirmation window, it is identified as a real anomaly, the self-checking deviation calibration strategy is recorded as no adjustment, and a fault level classification signal is generated; If the output deviation does not persist within the confirmation window, it is identified as transient interference, the self-checking deviation calibration strategy is recorded as no adjustment, and no fault level classification signal is generated; Obtain the output deviation sequence of the accelerometer within the confirmation window, thereby obtaining the slope sequence of the accelerometer, and analyze the average slope of the output deviation index of the accelerometer within the confirmation window; If the output deviation index of the accelerometer is less than the output deviation index threshold and the average slope of the output deviation index of the accelerometer within the confirmation window is greater than or equal to the slope threshold, the self-checking deviation calibration strategy is recorded as self-checking deviation adjustment, and this accelerometer is recorded as a to-be-verified accelerometer, and no fault level classification signal is generated; If the output deviation index of the accelerometer is less than the output deviation index threshold value and the slope of the output deviation index of the accelerometer within the confirmation window is less than the slope threshold value, the self-checking deviation calibration strategy is recorded as keeping regular monitoring, the stage data is stored in the historical database, and no fault level division signal is generated.

4. The machine learning based fault diagnosis method of a seismic monitoring device according to claim 3, characterized in that, The self-checking deviation adjustment, and the specific analysis method is as follows: Multi-source data fusion verification is introduced, and the observation data of the accelerometer to be verified and other accelerometers in the seismometer and external vibration sensors are collected. The amplitude, frequency and phase of each element are paired and compared in a unified time window. Then, the mean square error between each element is calculated and compared with the set error threshold value. When the mean square error is greater than or equal to the error threshold value, the output deviation index threshold value and the sampling and self-checking interval are updated based on the reduction coefficient of the output deviation index threshold value and the sampling and self-checking interval reduction value determined by the average value of the slope of the output deviation index of the accelerometer. When the mean square error is less than the error threshold value, the self-checking deviation adjustment is not performed. 5.The machine learning based fault diagnosis method for seismic monitoring equipment according to claim 1, wherein, The fault level division and preliminary intervention, and the specific analysis method is as follows: After receiving the fault level division signal, multi-dimensional mechanical health features are extracted from the data collected by the accelerometer generating the signal, and the multi-dimensional mechanical health features are obtained. The fault score and dynamic prediction trend of the accelerometer are output based on the machine learning model with the multi-dimensional mechanical health features as input. The preset fault score interval in the database is extracted. When the fault score is less than or equal to the lower limit value of the fault score interval, the fault level of the accelerometer is recorded as level one fault. When the fault score is within the fault score interval, the fault level of the accelerometer is recorded as level two fault. When the fault score is greater than or equal to the upper limit value of the fault score interval, the fault level of the accelerometer is recorded as level three fault. The fault levels of the accelerometer within the sampling period are collected. When level two fault and level three fault are detected, the corresponding accelerometer is marked as an abnormal accelerometer. The preliminary intervention is performed on the abnormal accelerometers. The preliminary intervention, and the specific steps are as follows: For an abnormal accelerometer, the acceleration gain coefficient increase amplitude is determined according to the dynamic prediction trend score, and the acceleration gain coefficient update value is obtained according to the acceleration gain coefficient increase amplitude and the acceleration gain coefficient. The acceleration gain coefficient update value is applied to the accelerometer. 6.The machine learning based fault diagnosis method of a seismic monitoring device according to claim 1, wherein, The specific analysis method for determining the fault level of each accelerometer is as follows: For an accelerometer, the variances of level one fault, level two fault and level three fault within a sampling period are calculated to obtain the total fault variance within the period, and the variance contribution rate of each fault level is obtained. The preset contribution rate threshold value in the database is extracted. When the variance contribution rate of a certain fault level is greater than or equal to the contribution rate threshold value within the sampling period, the fault level is taken as the fault level of the accelerometer. If the variance contribution rates of the three fault levels within the sampling period are all less than the contribution rate threshold value, the difference between the variances of each level is calculated, and the preset variance difference threshold value in the database is extracted. When the maximum variance difference exceeds the variance difference threshold, the fault level corresponding to the maximum variance difference is determined as the fault level of the accelerometer; If the maximum variance difference does not exceed the variance difference threshold, the highest fault level appearing in the period is taken as the fault level of the accelerometer.

7. The machine learning based fault diagnosis method of a seismic monitoring device according to claim 1, characterized in that, The trigger hierarchical response strategy is specifically analyzed as follows: If the fault level of the accelerometer is a first-level fault, the hierarchical response strategy is recorded as data review, the integrity and consistency of the collected data are reviewed, it is confirmed that the output of the accelerometer and the communication link are normal, and then the monitoring indicators and characteristic parameters in the period are recorded; If the fault level of the accelerometer is a second-level fault, the hierarchical response strategy is recorded as comprehensive adjustment; If the fault level of the accelerometer is a third-level fault, the hierarchical response strategy is recorded as shutdown alarm, the system is automatically shut down, alarm generates maintenance task information, and it is prompted that the accelerometer replacement operation needs to be performed.

8. The machine learning based fault diagnosis method of a seismic monitoring device according to claim 7, characterized in that, The comprehensive adjustment is specifically analyzed as follows: The accelerometer with a second-level fault is recorded as a demand adjustment accelerometer; For a demand adjustment accelerometer, a temperature drift detection is performed, and the specific steps are as follows: environmental temperature data are collected, a correlation coefficient of temperature and fault score is calculated, a linear regression fitting is performed on the fault score and the temperature data, a regression slope and a goodness of fit value are obtained; A preset regression slope threshold and a credibility threshold in the database are extracted; When the regression slope exceeds the preset regression slope threshold and the goodness of fit value is higher than the credibility threshold, a temperature control compensation measure is triggered, a temperature compensation resistance change amount is obtained by using the regression slope and a temperature change value, a temperature compensation resistance update value is obtained by using the temperature compensation resistance change amount and a temperature compensation resistance in the system log, and the compensation resistance in the bridge circuit is automatically adjusted by a digital control device in the circuit; If the regression slope does not exceed the preset regression slope threshold or the goodness of fit value is not higher than the credibility threshold, the system performs controlled damping adjustment; A damping initial adjustment value is determined according to the fault score, and a damper is adjusted according to the calculated damping initial adjustment value. 9.The machine learning based fault diagnosis method of a seismic monitoring device according to claim 1, wherein, After the comprehensive adjustment, the accelerometer fault level analysis is performed again, and the specific analysis method is as follows: After the comprehensive adjustment, the accelerometer fault level analysis is performed again, if the current accelerometer is in a first-level fault state after the comprehensive adjustment, a compensation completion signal is generated; If the current accelerometer is in a second-level fault state after the comprehensive adjustment, a damping secondary adjustment value is determined according to the mean value of the fault score and the temperature compensation resistance change amount, and the damper is adjusted according to the calculated damping secondary adjustment value; If the current accelerometer is in a third-level fault state after the comprehensive adjustment, a rollback operation is performed and jumps to the third-level fault response strategy.

10. A machine learning based seismic monitoring equipment fault diagnosis system, applying a machine learning based seismic monitoring equipment fault diagnosis method according to any one of claims 1-9, characterized in that, The system comprises an accelerometer self-checking module, a fault level division module and an accelerometer diagnosis module. The accelerometer self-checking module is used for monitoring the output frequency of each accelerometer of the seismic monitoring equipment in real time, performing a self-checking and sensitivity detection program after receiving a self-checking signal, determining the output deviation index of each accelerometer, and analyzing a self-checking deviation calibration strategy. The fault level classification module is configured to obtain multi-dimensional mechanical health features after completing the self-checking and sensitivity detection procedures, take the multi-dimensional mechanical health features as inputs, output fault scores and dynamic prediction trends of the accelerometers based on a machine learning model, and thereby perform fault level classification and preliminary intervention; The accelerometer diagnosis module is configured to calculate variances of each fault level, determine fault levels of each accelerometer, and trigger a hierarchical response strategy. When the fault level of the accelerometer is a secondary fault, the accelerometer diagnosis module is configured to perform temperature drift detection, trigger temperature control compensation, and perform periodic fault level analysis again after comprehensive adjustment.

Images