A data quality monitoring and storage system based on equipment electrical condition assessment

Abstract

This invention relates to the field of data processing technology for industrial equipment condition monitoring, and discloses a data quality monitoring and storage system based on equipment electrical condition assessment. The system includes: a physical correlation verification engine that uses a baseline library linking electrical condition and vibration characteristics to assess and label the diagnostic effectiveness of vibration data streams; and a baseline adaptive calibration module that can dynamically and adaptively calibrate the baseline library based on high-reliability data output by the engine itself. This invention solves the problem of long-term reliability decay caused by cognitive aging in static diagnostic models by constructing a self-evolving closed loop for the core knowledge base of the diagnostic system, enabling it to iterate synchronously with the long-term drift of equipment physical characteristics. This ensures the continuous accuracy of data quality assessment throughout the entire equipment lifecycle.

Description

Technical Field

[0001] This invention relates to a data quality monitoring and storage system based on equipment electrical condition assessment, belonging to the field of data processing technology for industrial equipment condition monitoring. Background Technology

[0002] Currently, in the monitoring systems of large steam turbine generator sets and other rotating power units, vibration sensors continuously collect vibration signals during equipment operation and record them completely in a high-fidelity manner. The purpose is to preserve comprehensive raw data for subsequent fault analysis or long-term trend assessment. However, when the generator set performs deep peak shaving tasks with significant short-term power output changes to meet grid requirements, the aforementioned technical approach faces a practical engineering problem at the data processing level: a conflict between the completeness of data recording and the effectiveness of data analysis. During such drastic transient operations as deep peak shaving, the unit's mechanical structure generates significant vibration responses, which falls under the specific operational conditions... Normal physical processes, not equipment failures, can lead to alarms being triggered by a comprehensive monitoring system that records all signal fluctuations. This can reduce operators' sensitivity to alarm signals. Furthermore, the indiscriminate storage of these transient data, which lack stable fault characteristics, in a historical database can interfere with the training and evaluation of automated diagnostic models that rely on that database. To address this issue, simply raising the alarm threshold may miss genuine early fault signals. Adding complex signal processing algorithms to analyze a single vibration data stream limits the accuracy of the judgment due to the lack of a reference to the actual operating conditions of the equipment at that time.

[0003] In view of this, although existing technologies have attempted to introduce more complex algorithm models to improve the intelligence level of diagnosis, these methods often focus on backend data analysis and pattern recognition. However, in the front-end stage of data collection and storage, namely, how to define the root cause of data validity from the physical mechanism level, there are still significant cognitive blind spots. For example, Chinese invention patent with authorization announcement number CN119622294B discloses an online fault diagnosis method for wind turbine transmission chains. Although this method uses a SCADA system to collect multi-source data such as environmental, electrical, and vibration data, and constructs a teacher-student network framework to achieve online continuous learning and diagnosis, However, its core approach still relies on complex backend algorithms (such as positive and negative joint learning strategies) to passively adapt and correct the fused features. The fundamental flaw of this method is that it fails to address the distinction between legitimate transient responses and genuine physical anomalies at the data source. Instead, it directly uses raw data containing a large number of legitimate transient processes as learning samples. This forces its complex online learning model to expend significant computing power to learn and adapt to pseudo-label noise caused by drastic changes in normal operating conditions—nothing that should not be considered fault characteristics. This not only affects the efficiency of model training and the accuracy of diagnosis but also fails to fundamentally avoid false alarms caused by drastic changes in operating conditions. Therefore, this type of technical approach does not solve the core problem of data contamination before being stored in historical databases, resulting in subsequent intelligent analysis always being built on a data foundation with inconsistent information quality.

[0004] Existing technologies have the following main shortcomings in this regard: 1. The monitoring system lacks a step in the data processing flow to distinguish the source of the collected vibration signals based on the actual operating conditions of the equipment. That is, it cannot effectively identify whether the signal is generated by a normal transient operating condition response or by a physical anomaly occurring under stable operating conditions; 2. Data storage does not include reference information characterizing the generating operating conditions, resulting in subsequent analysis applications being unable to prioritize data segments with higher diagnostic value collected under stable operating conditions when processing massive amounts of historical data. Therefore, how to establish a data quality monitoring and storage method that can pre-assess and differentiate the diagnostic validity of vibration monitoring data based on the electrical operating status of the equipment during the data acquisition and storage stage is the technical problem this invention aims to solve. Summary of the Invention

[0005] This invention provides a data quality monitoring and storage system based on equipment electrical condition assessment. Its main purpose is to solve the problem that existing monitoring systems cannot conduct pre-assessment of the diagnostic effectiveness of monitoring data based on equipment operating conditions, resulting in false alarms and interference with historical data under drastic changes in operating conditions.

[0006] To achieve the above objectives, this invention provides a data quality monitoring and storage system based on equipment electrical condition assessment, comprising a synchronous data acquisition and timing alignment module, a physical correlation verification engine, a data quality metadata injector, and a baseline adaptive calibration module. The synchronous data acquisition and timing alignment module is used to acquire the vibration data stream and the electrical status data stream characterizing the operating conditions of the monitored equipment in parallel and synchronously. The physical correlation verification engine contains a baseline library, which stores the correlation between the electrical state and vibration characteristics of the equipment under healthy conditions. The physical correlation verification engine is used to compare the real-time collected electrical state data stream and vibration data stream with the correlation in the baseline library to generate data quality labels. The data quality labels include at least a valid stability label, which is used to characterize the conformity of the data with the baseline under stable operating conditions. A data quality metadata injector is used to store data quality tags as metadata along with the corresponding vibration data stream; The baseline adaptive calibration module is used to continuously monitor the data quality labels generated by the physical correlation verification engine; it is used to identify high-confidence stable operation windows where the proportion of vibration data streams marked as valid stable labels exceeds a preset statistical threshold within a preset time period; when a high-confidence stable operation window is identified, it incrementally updates the correlation relationships in the baseline library associated with the operating conditions corresponding to the high-confidence stable operation window based solely on the statistical characteristics of the vibration data streams within that high-confidence stable operation window, according to a preset weighting coefficient.

[0007] Furthermore, the baseline library includes a series of lookup tables and polynomial functions; the lookup tables are used to define the expected amplitude range of the vibration signal under healthy conditions within a discrete stable power output range; the polynomial functions are used to describe the continuous relationship between the dominant frequency of the vibration signal spectrum under healthy conditions and the power variation within a continuously varying power range.

[0008] Furthermore, the incremental update operation rules for the baseline adaptive calibration module are defined as follows: based on the weighted moving average algorithm and the statistical center value of the vibration data stream within the high-confidence stable operating window, the existing correlations in the baseline library are adjusted. The statistical center value is assigned a weighting factor α less than 0.1, and the adjusted new correlations... Follow these rules:

[0009] in, For the old relationships in the baseline library before the update, The statistical center value is calculated based on the vibration data stream within a high-reliability stable operating window.

[0010] Furthermore, the data quality labels also include valid transient labels and abnormal disconnection labels. Valid transient labels are used to characterize whether the data conforms to the expected pattern of such changes in the baseline library during periods of drastic changes in operating conditions. Abnormal disconnection labels are used to characterize whether the data deviates significantly from the expected pattern in the baseline library under stable operating conditions. The data quality monitoring and storage system also includes an upper-layer application interface, which allows fault diagnosis applications to filter or weight the stored vibration data stream based on valid stable labels, valid transient labels, and abnormal disconnection labels.

[0011] Furthermore, the data quality monitoring and storage system also includes a time-domain coherence analysis module, which is used to detect electrical transient events in the electrical state data stream in real time. These events are characterized by the derivative value exceeding a preset slope threshold. When an electrical transient event is detected, the time delay between the electrical transient event and its corresponding response in the vibration data stream is calculated. Based on this time delay, structural health metadata is generated. This structural health metadata is used to characterize the structural health status of the equipment and is stored together with the corresponding vibration data stream through a data quality metadata injector.

[0012] Furthermore, the data quality monitoring and storage system also includes a baseline information calibration module, which is used to generate a baseline library during the system initialization phase. Specifically, under the premise that the equipment is in a healthy state, electrical status data streams and vibration data streams covering multiple stable operating conditions of the equipment are collected; and a baseline library is generated based on the collected data streams.

[0013] Furthermore, the data quality monitoring and storage system also includes a sensor health self-calibration module. This module is used to identify the shutdown free decay process of the monitored equipment based on events where the active power value drops to zero in the electrical status data stream. Specifically: The vibration data stream during the shutdown free decay process is collected, compared with the built-in standard template, and the residual signal is calculated. The standard template characterizes the vibration signal during the free decay process under healthy conditions. Based on the energy and spectral distribution characteristics of the residual signal, sensor health metadata characterizing the sensor's health status is generated; The sensor health metadata is appended to all subsequent monitoring data via the data quality metadata injector.

[0014] Furthermore, the data quality monitoring and storage system also includes a sensor installation status diagnostic module. This module is used to extract electromagnetic harmonic components with defined frequencies from the electrical status data stream and use them as probe signals. Specifically: Extract the vibration response signal with the same frequency as the probe signal from the vibration data stream; Based on the transmission gain between the vibration response signal and the probe signal, determine the installation status metadata characterizing the sensor's installation and fastening status; The installation status metadata is stored together with the corresponding vibration data stream via the data quality metadata injector.

[0015] Furthermore, the physical correlation verification engine is also used to calculate the residual signal between the vibration data stream and the corresponding correlation in the baseline library; the data quality monitoring and storage system also includes a residual pattern analysis module, which is used to perform time-series pattern analysis on the residual signal and generate fault mode metadata characterizing the fault evolution mode based on the time-series pattern analysis results; the fault mode metadata is stored together with the corresponding vibration data stream through the data quality metadata injector.

[0016] Furthermore, the data quality monitoring and storage system also includes an electrical entropy monitor and a decision confidence modifier; the electrical entropy monitor is used to calculate the sample entropy of the electrical state data stream in real time; the decision confidence modifier is used to perform two operations when the sample entropy exceeds a preset entropy threshold: Suppress the generation of abnormal disconnection tags by the physical association verification engine; Generate a status out-of-limit data quality label, which is used to characterize the equipment entering an unknown operating condition.

[0017] Compared with the prior art, the beneficial effects of the present invention are: This invention provides a data quality monitoring and storage system based on equipment electrical condition assessment. By synchronously acquiring and timing-aligning high-frequency vibration data streams with electrical condition data streams characterizing equipment operating conditions, and utilizing a baseline library storing the correlation between the two under healthy conditions, the system can perform real-time verification of the generation background of each set of vibration data streams before storing the data on the storage medium. This method allows transient vibrations with large amplitudes caused by drastic changes in normal equipment operating conditions, such as the deep peak shaving process of generator sets, to be identified as valid transients due to their high correlation with electrical conditions. Only vibrations that deviate significantly from the baseline expectation under stable electrical conditions are identified as high-value anomalies. Thus, the system's data storage is no longer an indiscriminate recording of physical signals, but rather an information archiving process that has pre-assessed and marked the diagnostic value of the data. This avoids the subsequent fault diagnosis system being interfered with by a large amount of legitimate but meaningless transient data, ensuring the objectivity of automated diagnostic conclusions.

[0018] This invention further utilizes the effective and stable data quality tags generated during its own operation to construct a baseline dynamic adaptive calibration mechanism. This mechanism identifies long-term continuous stable operating windows and uses the vibration data streams within these windows that are recognized as effective and stable by the system itself as high-confidence samples. It incrementally fine-tunes and updates the correlation between the corresponding operating conditions in the baseline library. This process enables the baseline library, which serves as the judgment benchmark, to automatically and slowly evolve along with the minor physical wear or characteristic drift of the equipment caused by long-term operation. This avoids the problem that the initially calibrated static baseline degenerates from an accurate benchmark into a source of false alarms because it cannot match the state changes throughout the equipment's life cycle. This ensures the long-term accuracy of data quality assessment and the continuous availability of the entire monitoring system.

[0019] This invention also utilizes real-time analysis of electrical state data streams, treating stable electromagnetic harmonic components, such as the twice-power-frequency vibration generated by a motor stator, as a continuous detection signal penetrating the equipment structure. By synchronously analyzing the intensity of the response signal received by the vibration sensor at the same frequency and calculating the transmission gain between them, this system can monitor the physical installation and tightness between the vibration sensor and the equipment housing online. When the sensor becomes slightly loose, its ability to transmit high-frequency vibrations will decrease, directly reflected in the trend of attenuation of the transmission gain. This allows the system to first confirm the reliability of the physical connection of the sensor itself, which is the source of information, before processing the vibration data stream to diagnose equipment faults. This distinguishes between two completely different situations: equipment abnormality and sensing path abnormality, providing a fundamental guarantee for the integrity and reliability of the entire monitoring and diagnostic information chain. Attached Figure Description

[0020] Figure 1 This is a flowchart of the data processing based on physical correlation verification and adaptive calibration of the present invention; Figure 2 This is a schematic diagram of the sensor state diagnosis based on the attenuation trend of transfer gain according to the present invention; Figure 3 This is a schematic diagram of the three-layer logical architecture and cross-layer data flow of the system of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0022] This invention provides a data quality monitoring and storage system based on equipment electrical condition assessment. Its system architecture includes a synchronous data acquisition and timing alignment module, a physical correlation verification engine, a data quality metadata injector, and a baseline adaptive calibration module. During operation, the system uses the electrical condition data stream characterizing the operating conditions of a steam turbine generator set as a reference to perform a preliminary, online evaluation and labeling of the diagnostic effectiveness of the synchronously acquired vibration data stream based on a physical correlation model. The evaluated and labeled data is then stored in a computer storage system, providing effective data for subsequent computer-aided fault diagnosis or equipment repair decisions. In the computerized equipment condition monitoring process, to ensure the causal relationship of subsequent analyses... The synchronization data acquisition and timing alignment module uses a unified high-precision clock source to add a synchronization timestamp with a resolution of no less than 1 millisecond to each data point in the high-frequency vibration data stream from the vibration sensor and the electrical status data stream from the data acquisition system (DCS) or monitoring and data acquisition (SCADA) system. The electrical status data stream may include parameters that can characterize the core operating status of the equipment, such as the generator's active power, reactive power, or frequency. After processing by this module, a unified data structure is formed within the system, in which any timestamp uniquely corresponds to a set of vibration data streams and electrical status data streams. This constitutes the necessary data prerequisite for the physical correlation verification engine to accurately determine the real-time correlation between the two.

[0023] To determine the diagnostic validity of the vibration data stream, the physical correlation verification engine, a computer processing unit, internally stores a working condition-vibration correlation baseline library. This baseline library is established during system initialization by running a benchmark calibration module. After confirming that the monitored equipment is in a healthy state under multiple stable operating conditions, it collects electrical state data streams and vibration data streams. This generates a series of lookup tables defining the expected amplitude range of healthy vibration signals within discrete stable power output ranges, and a polynomial function describing the continuous relationship between the dominant frequency of the healthy vibration signal spectrum and power variation within continuously changing power ranges. During system operation, the engine uses the real-time acquired electrical state data stream as input, queries the baseline library to obtain the expected vibration characteristic range under the current working condition, and then applies the synchronously acquired real-time data stream. The engine compares the actual vibration data stream with this expected range. For example, if the active power collected at a certain moment is 500MW, the expected range of healthy vibration amplitude is 0.8mm / s to 1.2mm / s obtained by querying the lookup table. If the actual measured vibration amplitude is 1.0mm / s, the engine determines that the data is consistent with the baseline and generates a valid and stable data quality label. Conversely, if the actual vibration amplitude is 2.5mm / s, an abnormal disconnection label is generated to characterize that the data deviates from the baseline expectation under stable operating conditions. When the electrical status data stream shows that the unit is performing operations with drastic power changes such as deep peak shaving, the engine calls the corresponding polynomial function model in the baseline library. If the dynamic response mode of the vibration data stream is consistent with the expected mode of the model, a valid transient label is generated to characterize that the vibration is caused by changes in normal operating conditions.

[0024] To preserve all original information for subsequent analysis and provide data quality indexes for upper-level computer applications, a data quality metadata injector receives the aforementioned data quality tags generated by the physical correlation verification engine before writing the original vibration data stream to the computer storage medium. This tag is then stored as an additional metadata field in the same record of the database along with the vibration data stream corresponding to the timestamp. In this way, the data in the storage system is transformed from undifferentiated physical signal records into a high-value intelligence repository rich in contextual information and preliminarily evaluated. Based on this, upper-level application interfaces can allow fault diagnosis applications to filter or weight historical data according to tags such as effective stability, effective transients, and abnormal disconnection, thereby improving the accuracy of automated diagnostic conclusions.

[0025] Considering that the physical characteristics of equipment may slowly drift due to minor wear or material fatigue during long-term operation, causing the initially calibrated static baseline library to gradually become inaccurate, this system further incorporates a baseline adaptive calibration module. This module continuously monitors the data quality label stream generated by the physical correlation verification engine and uses an internal counter and timer to identify a high-confidence stable operating window where the proportion of vibration data streams marked with valid stable labels exceeds a preset statistical threshold (e.g., 99.5%) within a preset time period (e.g., 24 consecutive hours). When this window is identified, the module performs a batch of background statistical analysis based solely on the vibration data streams within that window, calculating the statistical center value of the current vibration characteristics under this specific operating condition. And based on a weighted moving average algorithm, the original old correlations in the baseline database are analyzed. Perform an incremental update, the update rule of which is to compare (1 minus the weight factor α) with the old association. Multiply by, then multiply by the weighting factor α and the statistical center value. Multiply them, and finally add the two products to get a new association. ;in, For the updated new relationship, This refers to the old relationship before the update. The weighting factor α is set to a value less than 0.1 to calculate the statistical center value based on the vibration data stream within a high-reliability stable operating window. To enhance the system's diagnostic capabilities and gain insight into the structural integrity of the internal force transmission path, the system may also include a time-domain coherence analysis module. This module is used to treat electrical transient events in the electrical state data stream caused by power grid disturbances or load switching, where the derivative value exceeds a preset slope threshold, as a detection signal penetrating the equipment structure. When the module's internal detector identifies such an event, it immediately uses the event's timestamp as a reference and, within a high-resolution time window (e.g., 50 milliseconds before and after) of the synchronized vibration data stream, uses a peak detection algorithm to find the corresponding vibration response peak and calculate the time delay between these two peaks. By comparing the real-time calculated delay with a baseline delay time representing the health status, the module can generate structural health metadata characterizing the equipment's structural health status. This metadata is stored along with the corresponding vibration data stream via a data quality metadata injector. A trending increase in response delay can serve as an early warning signal indicating early faults such as rotor microcracks or loose shaft connections.

[0026] To differentiate between equipment malfunctions and sensor path malfunctions in the computer diagnostic process, this system can further integrate a sensor health self-calibration module and a sensor installation status diagnostic module. The sensor health self-calibration module utilizes the equipment shutdown process to reverse-calibrate the sensor's operating status. It identifies the shutdown free decay process of the monitored equipment by monitoring events where the active power value drops to zero in the electrical status data stream. Then, it compares the vibration data stream collected during this period with a built-in standard template representing the vibration signal during the free decay process in a healthy state. It calculates the residual signal between the actual signal and the template, and based on the energy and spectral distribution characteristics of this residual signal, it generates sensor health metadata (e.g., health status, increased noise, or...) representing the sensor's health status. (Sensitivity drift), this metadata is attached to all subsequent monitoring data through the data quality metadata injector; while the sensor installation status diagnostic module uses a stable, frequency-determined electromagnetic harmonic component (such as twice the power frequency vibration generated by the motor stator) in the electrical status data stream as a continuous detection signal. It monitors the physical installation tightness between the vibration sensor and the equipment housing online by synchronously analyzing the vibration response signal with the same frequency as the detection signal in the vibration data stream and calculating the transfer gain between the two. A trend-decreasing transfer gain directly indicates a slight looseness in the sensor installation. Based on this, the module generates an installation status metadata characterizing the sensor installation tightness, providing a verification basis for the integrity of the entire monitoring and diagnostic information chain.

[0027] Furthermore, to utilize the diagnostic information contained in the abnormal signals, this system may also include a residual pattern analysis module. This module uses the residual signal between the actual vibration and the expected vibration calculated by the physical correlation verification engine as a purified abnormal signal. By performing time-series pattern analysis on the buffered continuous residual signals—for example, using autocorrelation functions to detect the presence of periodic components that are harmonicly related to the equipment rotation speed, or using peak detection and time interval statistics to determine whether it exhibits sparse random impact pulses—this module can identify the fault evolution pattern and generate fault mode metadata (e.g., periodic pattern or impact pattern) characterizing the fault evolution pattern. This metadata, along with the abnormal disconnection tag, is injected and stored, providing a deeper level of information for subsequent root cause analysis. The system provides the basis for this; finally, to address the risk of false alarms when equipment enters unknown operating conditions not covered by the baseline library, this system can also be equipped with an electrical entropy monitor and a decision confidence adjuster. The electrical entropy monitor uses a sliding window-based sample entropy algorithm to calculate the information complexity of the electrical state data stream in real time. When the equipment enters a rapidly fluctuating unknown transient process, the sample entropy of the data stream will instantly increase and exceed a preset entropy threshold. At this time, the decision confidence adjuster is triggered and actively suppresses the physical correlation verification engine from generating abnormal disconnection labels, instead generating a state exceeding data quality label to characterize the equipment entering an unknown operating condition. This ensures that the core decision logic only operates within its trusted knowledge boundaries, enhancing the applicability of the entire system across all operating conditions.

[0028] Example 1: In a computer monitoring system of a thermal power plant equipped with a 1000MW ultra-supercritical steam turbine generator unit, the unit receives a grid dispatch instruction requiring it to reduce its active power output from 980MW to 250MW within 30 minutes to perform deep peak shaving. Under this condition, a monitoring system that relies solely on vibration signal amplitude for judgment will trigger a fault alarm if the vibration amplitude at a specific measuring point of the unit exceeds the alarm threshold set for stable operation, potentially leading to unplanned shutdown checks by operators. When the system employing the technical solution of this invention is deployed on the unit, its synchronous data acquisition and timing alignment module synchronously acquires, with a time resolution of 1 millisecond, both an electrical status data stream composed of active power output data from the generator and a vibration data stream composed of vibration sensor data from the same measuring point on the unit. When the unit's power begins to decrease as instructed, physical correlation verification... Upon receiving these two precisely aligned data streams, the engine identified that the electrical state data stream exhibited rapid, large-amplitude unidirectional changes. Simultaneously, the amplitude and spectral characteristics of the vibration data stream also fluctuated drastically, exceeding the healthy vibration characteristic range defined in the baseline library for any stable power range. At this point, the physical correlation verification engine used its internally stored polynomial function describing the unit's transient response pattern under healthy conditions to verify the correlation between the drastic changes in the current electrical state and the drastic changes in the vibration response. Since the dynamic pattern of the current vibration response matched the transient model pre-stored in the baseline library for deep peak shaving conditions, the engine determined all vibration data streams within this duration to be normal transient responses. The data quality metadata injector then attached valid transient data quality tags to each of these data points before storing them in the long-term historical database.

[0029] During this operation, the precise time-aligned data provided by the synchronous data acquisition and timing alignment module is a prerequisite for the physical correlation verification engine to make accurate causal judgments. The physical correlation verification engine uses this input to enable the monitoring system to maintain the sensitivity of monitoring small abnormal signals under stable operating conditions, while also being able to identify responses to violent but normal transient operating conditions. The system does not only analyze the vibration signal itself, but also introduces the electrical state as a reference dimension, shifting the focus of the monitoring task from judging whether the vibration signal is abnormal to judging whether the vibration response matches its physical cause, thereby avoiding false alarms caused by drastic changes in operating conditions. Finally, when the upper-level computer fault diagnosis application calls historical data for long-term trend analysis or model training, it can ignore or downweight the data during this period based on the metadata tag of effective transient state, avoiding the interference of this data without stable fault characteristics on the equipment's true health status assessment model. The data quality and decision support capabilities of the computer monitoring and diagnostic system are thus improved.

[0030] Example 2: To objectively verify the effectiveness of the technical solution of this invention in distinguishing between physical anomalies and normal operating condition changes in equipment, the following comparative experiment was designed and executed. The experiment employed an experimental platform capable of simulating the physical characteristics of a rotating electric motor. This platform included a variable frequency speed-regulating motor controlled by a programmable logic controller, an electromagnetic dynamometer coaxially connected to the motor, a piezoelectric accelerometer mounted on the motor bearing housing, and a data acquisition system for synchronously acquiring the motor's three-phase current and voltage, as well as the sensor's vibration signals. The accelerometer's measurement range was ±50g, its frequency response range was 0.5Hz to 10kHz, the data acquisition system's time synchronization accuracy was 0.1ms, and the sampling frequency was set to 20.48kHz. This sampling frequency setting was based on the Nyquist algorithm. The special sampling theorem is used to ensure that signals do not alias. Two systems were set up for comparative testing: a control group that judges based solely on vibration amplitude exceeding a fixed threshold, and an experimental group that adopted the technical solution of this invention. The test process simulated two operating states of the equipment: one was to simulate physical abnormalities caused by rotor imbalance by adding a 5g counterweight at a specific position on the motor rotor; the other was to simulate transient conditions by controlling the motor and dynamometer to complete a rapid power reduction process from high speed heavy load to low speed light load within 10 seconds in a healthy state without a counterweight. During the test, the two systems processed synchronous data streams from the same experimental platform in parallel. The data processing results and final alarm output under different operating conditions are shown in Table 1.

[0031] Table 1: Comparison of system performance test data under different operating conditions.

[0032]

[0033] As shown in Table 1, under the stable, fault-free operating conditions (numbers 1 and 2), the two systems behaved identically and did not generate any alarms. Under the power transient operating conditions (numbers 3, 4, and 5), the vibration amplitude increased due to transient impacts and exceeded the alarm threshold of the control group, causing false alarms in the control group. The physical correlation verification engine of the experimental group, however, identified that the vibration response pattern matched the change in electrical state, thus marking the data as a valid transient and suppressing the alarm output. Under the stable, faulty operating conditions (numbers 6, 7, 8, and 9), the vibration amplitude also exceeded the limit. The control group output an alarm, while the experimental group, finding that the vibration data stream deviated from the baseline library's expectations under stable electrical conditions, marked it as an abnormal disconnection and also output an alarm. The experimental results show that the system of this invention, by introducing the electrical state data stream as a reference and establishing a physical correlation verification mechanism, can identify and filter transient vibration responses caused by changes in normal operating conditions while maintaining the ability to alarm for physical anomalies, thus adding diagnostic information to the stored data.

[0034] Example 3: This embodiment combines Figures 1 to 3 A description of a data quality monitoring and storage system based on equipment electrical condition assessment is provided, such as... Figure 1 As shown, the externally input electrical status data stream and vibration data stream first enter the synchronous data acquisition and timing alignment module. This module adds a high-precision synchronization timestamp to the data and outputs the synchronous data stream to the physical correlation verification engine. Based on the health status correlation relationship read from the operating condition vibration correlation baseline library, this engine compares and verifies the real-time synchronous data stream in real time and generates the original data stream and data quality tags. This data stream is sent to the data quality metadata injector, which stores the quality tags as metadata along with the vibration data stream to form tagged high-value data for subsequent fault diagnosis and decision-making. At the same time, the effective and stable data tag stream generated by the physical correlation verification engine is sent to the baseline adaptive calibration module. This module identifies high-confidence data and issues incremental update instructions to the operating condition vibration correlation baseline library, thereby constructing a dynamic adaptive closed-loop calibration mechanism.

[0035] like Figure 2 As shown, the horizontal axis represents the continuous operating time from hour 1 to hour 24. The amplitude unit of the electromagnetic harmonic component of the probe signal, V, exhibits stable periodic fluctuations, while the amplitude unit of the synchronously acquired vibration response signal, mm / s, shows a corresponding dynamic response. The key diagnostic indicator, namely the transfer gain unit, dB, which characterizes the correlation strength between the two, shows a continuous and trending attenuation throughout the monitoring period, eventually crossing the preset lower limit of the normal fluctuation threshold around hour 22. This phenomenon precisely indicates that the physical fastening condition between the sensor and the equipment mounting surface may have deteriorated. Figure 3 As shown, it is logically divided into three parts: the field equipment layer, the data acquisition and processing layer, and the data storage and application layer. When the steam turbine generator set in the field equipment layer is running, the vibration data stream it generates is collected by vibration sensors, and its electrical status data stream is collected by the factory's DCS / SCADA system. These two data streams are sent in parallel to the data processing server in the data acquisition and processing layer. This server is equipped with core functional modules such as synchronous acquisition and alignment, physical correlation verification, metadata injection, and baseline adaptive calibration. The tagged data and baseline data generated after processing are sent to the long-term historical database of the data storage and application layer for archiving. The diagnostic analysis terminal can access this database through the historical data query interface to support subsequent analysis and decision-making. At the same time, the data processing server can also output real-time status and alarm signals based on the real-time processing results.

[0036] Example 4: To ensure that the system of this invention, when first deployed on a newly commissioned steam turbine generator set, accurately reflects the initial health status of that specific unit in its internal baseline library and that the operating parameters of its baseline adaptive calibration module are reasonably configured, a standardized system initialization and parameter calibration procedure needs to be executed. The initial state of this procedure is that the monitored steam turbine generator set is confirmed to be in a fault-free healthy state and can operate safely within a specified power range. The data quality monitoring and storage system of this invention has completed hardware installation and software deployment. Its synchronous data acquisition and timing alignment module can synchronously acquire electrical status data streams composed of generator active power and vibration data streams at specified measurement points at a sampling rate of not less than 1 kHz. The calibration process is executed by the reference information calibration module, and its steps are as follows: First, the control generator set... First, the generator set is continuously operated for no less than 30 minutes at four discrete stable operating points at 25%, 50%, 75%, and 100% of rated power, during which synchronous data is continuously collected. Second, for the vibration data stream collected at each stable operating point, the statistical average and standard deviation of the vibration amplitude are calculated, and the expected amplitude range of the healthy vibration signal at that operating point is set as the interval of the average value plus or minus three times the standard deviation. The results are used to fill the lookup table in the baseline library. Third, the generator set is controlled to perform a continuous power increase process from the minimum stable load to the rated load. The continuous change data of power and the corresponding vibration signal spectrum frequency during this process are recorded, and the least squares method is used to fit the data to a third-order polynomial function to obtain a polynomial function describing the continuous change relationship of the frequency and store it in the baseline library.

[0037] Subsequently, key parameters in the baseline adaptive calibration module were set. Among these, the preset duration and preset statistical threshold for identifying high-reliability stable operating windows were set based on the typical operating cycle of the monitored equipment. For equipment like this generator set, which typically operates with long-term stable cycles, the preset duration could be set to 24 hours, and the preset statistical threshold to 99.5%, thus avoiding the impact of short-term disturbances on baseline library updates. For incremental update rules, i.e., new correlations… By subtracting the weighting factor α from (1) and comparing it with the old association After multiplication, add the weighting factor α and the statistical center value. The product of these factors is used to obtain the weighting factor α in this rule. The value of α is based on the expected aging rate of the equipment. Given that the unit is in its initial operational phase and its physical characteristics are drifting slowly, to ensure the stability of the baseline library, the weighting factor α is set to 0.01. In the formula, For the updated new relationship, This refers to the old relationship before the update. The statistical center value is calculated based on the vibration data stream within a high-reliability stable operating window. By executing the above procedures, the system's baseline library is set with quantitative data that can characterize the initial health status of the specific unit, and the key parameters of its adaptive calibration mechanism also obtain clear engineering setting basis.

[0038] Example 5: To ensure that the sensor health self-calibration module and sensor installation status diagnosis module in the system have a judgment benchmark before being put into use, a calibration procedure for sensor diagnostic functions must be executed after the initial deployment of the system. This procedure first targets the sensor health self-calibration module. Under the premise that the monitored equipment is in a healthy state and the sensor is securely installed, the control equipment performs no less than five complete free decay processes from rated operating conditions to shutdown, and collects the vibration data stream during this period. Subsequently, the system normalizes the time axis of these five collected vibration data streams and calculates the average value point by point to generate a vibration signal curve that can characterize the standard decay characteristics of the measuring point in a healthy state. This curve is then stored as a standard template in the built-in database of the sensor health self-calibration module.

[0039] Subsequently, for the calibration of the sensor installation status diagnostic module, the control equipment operates stably for no less than 1 hour at 80% of its rated power. During this period, the module continuously extracts an electromagnetic harmonic component with a stable frequency and clear amplitude from the electrical status data stream as a probe signal, and simultaneously extracts a vibration response signal with the same frequency as the probe signal from the vibration data stream. The system then calculates the statistical average value of the transfer gain between the vibration response signal energy and the probe signal energy within this 1-hour stable operating window, and sets this average value as the reference gain characterizing the sensor's installation tightness. At the same time, a normal fluctuation threshold range of ±15% is set around this reference gain. After completing this procedure, the sensor diagnostic function has a quantitative comparison basis that can be used for subsequent online monitoring.

[0040] Example 6: To complete the parameter tuning and logic rule definition for the advanced diagnostic functions of the system, after the initial calibration procedures of Examples 3 and 4 are completed, the standardized configuration process for the advanced diagnostic module continues to be executed. First, for the time-domain coherence analysis module, the system analyzes the 24-hour electrical status data stream covering all normal operating conditions collected during the initial calibration period, calculates the first derivative of its time series, and analyzes the statistical distribution of the derivative value. The 99.9% quantile of this distribution is set as the preset slope threshold for triggering transient event detection to distinguish between real electrical transients and background noise. Subsequently, using this threshold, no less than 100 transient events and their corresponding vibration responses are identified from the health status data. The electromechanical response delay of each event pair is calculated, and the statistical average of these 100 delay times is set as the benchmark delay time characterizing the health structural state of the equipment.

[0041] Subsequently, for the electrical entropy monitor, the system uses the same 24-hour health status electrical data stream and a sliding window sample entropy algorithm to calculate its full-time entropy value. The maximum value of the entropy calculated under all known normal operating conditions is added to twice the standard deviation of the entropy value sequence. The result is set as a preset entropy threshold for determining whether the equipment has entered an unknown operating condition. Finally, the specific judgment logic of the residual mode analysis module is quantitatively defined. The judgment rule for periodic mode is set as follows: when the peak value of the autocorrelation function of the residual signal at the delay corresponding to the equipment frequency and its harmonics exceeds five times the standard deviation of the function value at other delay positions, a periodic mode is determined to exist. The judgment rule for impulse mode is set as follows: when three or more pulses with amplitudes exceeding six times the root mean square value of the signal appear consecutively in the residual signal, and the standard deviation of the time interval between these pulses is greater than 50% of their average value, an impulse mode is determined to exist. Through this series of procedures, the advanced diagnostic functions of the system obtain reproducible parameters and logical basis.

[0042] To further verify the technical effectiveness of the present invention in distinguishing between physical anomalies and normal operating condition changes in equipment, this comparative example is set up.

[0043] Comparative Example 1: This comparative example uses the exact same experimental platform, sensors, data acquisition system, and parameter settings as Example 2. The only difference is that this comparative example uses a conventional data processing module that judges based solely on a fixed threshold of the vibration signal, replacing the physical correlation verification engine in Example 2. The operating logic of this conventional data processing module is as follows: During the system initialization phase, a single fixed alarm threshold is set based on the statistical maximum value of the vibration amplitude of the device under multiple stable and fault-free operating conditions (set to 2.0 mm / s in this example); During system operation, the module receives the vibration data stream in real time and compares the vibration amplitude of each sampling point with the fixed alarm threshold. Once the measured vibration amplitude exceeds the threshold, the system determines that the device has malfunctioned and outputs an alarm signal. Using this conventional technical approach, tests were conducted on two operating states that are exactly the same as in Example 2. The data processing results and the final alarm output are shown in Table 2.

[0044] Table 2: System performance test data of conventional technical solutions under different operating conditions.

[0045]

[0046] The test data in Table 2 show that under stable, fault-free operating conditions (numbers 1 and 2) and stable, faulty operating conditions (numbers 4 and 5), the judgment results of the conventional technical solution are consistent with the physical state of the equipment. However, under the power transient operating condition (number 3), the equipment itself is in a healthy state, but its vibration amplitude increases instantaneously to 4.5 mm / s due to normal transient impact, exceeding the alarm threshold of 2.0 mm / s set for stable operating conditions, causing the system to output an incorrect alarm signal. The test results of Comparative Example 1 show that a monitoring system that relies solely on a fixed threshold for vibration signal judgment, due to the lack of reference to the actual operating conditions of the equipment in its design principle, is indeed unable to handle transient vibration responses caused by drastic changes in normal operating conditions. Therefore, when performing tasks such as deep peak shaving, it will inevitably generate false alarms, interfering with the normal operation of the unit. Furthermore, by injecting a large amount of pseudo-fault data unrelated to real faults into the historical database, it has a negative impact on the subsequent long-term trend assessment of the equipment's health status.

[0047] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0048] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A data quality monitoring and storage system based on equipment electrical condition assessment, characterized in that, This includes a synchronous data acquisition and timing alignment module, a physical correlation verification engine, a data quality metadata injector, and a baseline adaptive calibration module. The synchronous data acquisition and timing alignment module is used to acquire the vibration data stream and the electrical status data stream characterizing the operating conditions of the monitored equipment in parallel and synchronously. The physical correlation verification engine contains a baseline library, which stores the correlation between the electrical state and vibration characteristics of the equipment under healthy conditions. The physical correlation verification engine is used to compare the real-time collected electrical state data stream and vibration data stream with the correlation in the baseline library to generate data quality labels. The data quality labels include at least a valid stability label, which is used to characterize the conformity of the data with the baseline under stable operating conditions. A data quality metadata injector is used to store data quality tags as metadata along with the corresponding vibration data stream; The baseline adaptive calibration module is used to continuously monitor the data quality labels generated by the physical correlation verification engine; Used to identify a high-confidence stable operation window in which the proportion of vibration data streams marked as valid stable labels exceeds a preset statistical threshold within a preset time period; When a high-confidence stable operating window is identified, the correlation between the operating conditions corresponding to the high-confidence stable operating window in the baseline library is incrementally updated based solely on the statistical characteristics of the vibration data stream within that high-confidence stable operating window, according to a preset weighting coefficient.

2. The data quality monitoring and storage system based on equipment electrical condition assessment according to claim 1, characterized in that, The baseline library includes a series of lookup tables and polynomial functions; the lookup tables are used to define the expected amplitude range of the vibration signal under healthy conditions within a discrete range of stable power output; the polynomial functions are used to describe the continuous relationship between the dominant frequency of the vibration signal spectrum under healthy conditions and the power variation within a continuously varying power range.

3. The data quality monitoring and storage system based on equipment electrical condition assessment according to claim 1, characterized in that, The incremental update operation rules for the baseline adaptive calibration module are defined as follows: based on a weighted moving average algorithm and the statistical center value of the vibration data stream within a high-reliability stable operating window, the existing correlations in the baseline library are adjusted. The statistical center value is assigned a weighting factor α less than 0.1, and the adjusted new correlations... Follow these rules: in, For the old relationships in the baseline library before the update, The statistical center value is calculated based on the vibration data stream within a high-reliability stable operating window.

4. The data quality monitoring and storage system based on equipment electrical condition assessment according to claim 1, characterized in that, The data quality labels also include valid transient labels and abnormal disconnection labels. Valid transient labels are used to characterize whether the data conforms to the expected pattern of such changes in the baseline library during periods of drastic changes in operating conditions. Abnormal disconnection labels are used to characterize whether the data deviates significantly from the expected pattern in the baseline library under stable operating conditions. The data quality monitoring and storage system also includes an upper-layer application interface, which allows fault diagnosis applications to filter or weight the stored vibration data stream based on valid stable labels, valid transient labels, and abnormal disconnection labels.

5. A data quality monitoring and storage system based on equipment electrical condition assessment according to claim 4, characterized in that, The data quality monitoring and storage system also includes an electrical entropy monitor and a decision confidence modifier; the electrical entropy monitor is used to calculate the sample entropy of the electrical state data stream in real time; the decision confidence modifier is used to perform two operations when the sample entropy exceeds a preset entropy threshold: Suppress the generation of abnormal disconnection tags by the physical association verification engine; Generate a status out-of-limit data quality label, which is used to characterize the equipment entering an unknown operating condition.

6. A data quality monitoring and storage system based on equipment electrical condition assessment according to claim 1, characterized in that, The data quality monitoring and storage system also includes a time-domain coherence analysis module, which is used to detect electrical transient events in the electrical state data stream in real time. These events are characterized by the derivative value exceeding a preset slope threshold. When an electrical transient event is detected, the time delay between the electrical transient event and its corresponding response in the vibration data stream is calculated. Based on this time delay, structural health metadata is generated. This structural health metadata is used to characterize the structural health status of the equipment and is stored together with the corresponding vibration data stream through the data quality metadata injector.

7. A data quality monitoring and storage system based on equipment electrical condition assessment according to claim 1, characterized in that, The data quality monitoring and storage system also includes a baseline information calibration module, which is used to generate a baseline library during the system initialization phase. Specifically, under the premise that the equipment is in a healthy state, electrical status data streams and vibration data streams covering multiple stable operating conditions of the equipment are collected; and a baseline library is generated based on the collected data streams.

8. A data quality monitoring and storage system based on equipment electrical condition assessment according to claim 1, characterized in that, The data quality monitoring and storage system also includes a sensor health self-calibration module. This module is used to identify the shutdown free decay process of the monitored equipment based on events where the active power value drops to zero in the electrical status data stream. Specifically: The vibration data stream during the shutdown free decay process is collected, compared with the built-in standard template, and the residual signal is calculated. The standard template characterizes the vibration signal during the free decay process under healthy conditions. Based on the energy and spectral distribution characteristics of the residual signal, sensor health metadata characterizing the sensor's health status is generated; The sensor health metadata is appended to all subsequent monitoring data via the data quality metadata injector.

9. A data quality monitoring and storage system based on equipment electrical condition assessment according to claim 1, characterized in that, The data quality monitoring and storage system also includes a sensor installation status diagnostic module. This module is used to extract electromagnetic harmonic components with defined frequencies from the electrical status data stream and use them as probe signals. Specifically: Extract the vibration response signal with the same frequency as the probe signal from the vibration data stream; Based on the transmission gain between the vibration response signal and the probe signal, determine the installation status metadata characterizing the sensor's installation and fastening status; The installation status metadata is stored together with the corresponding vibration data stream via the data quality metadata injector.

10. A data quality monitoring and storage system based on equipment electrical condition assessment according to claim 1, characterized in that, The physical correlation verification engine is also used to calculate the residual signal between the vibration data stream and the corresponding correlation in the baseline library; the data quality monitoring and storage system also includes a residual pattern analysis module, which is used to perform time-series pattern analysis on the residual signal and generate fault mode metadata characterizing the fault evolution mode based on the time-series pattern analysis results. The fault mode metadata is stored together with the corresponding vibration data stream via the data quality metadata injector.

Images