Fresh air handling unit fault prediction method based on big data analysis

Abstract

The invention discloses a fresh air handling unit fault prediction method based on big data analysis, and relates to the technical field of fault prediction, and the method comprises the steps: collecting the equipment operation data of a fresh air handling unit and the state data of a sensor, and generating a target data set; performing feature extraction on the target data set, obtaining equipment operation features and sensor health features, constructing a sensor fault rule set and an equipment fault rule set, and generating a fault judgment logic rule base; and optimizing a trigger threshold condition of the fault judgment logic rule base, matching the trigger threshold condition with the preprocessed real-time sensor state data and real-time equipment operation data, representing a fault type based on a matching result, and representing a prediction result according to the fault type, so that accurate prediction is realized, false alarm and missing alarm are reduced, and the prediction efficiency is improved. Stable operation of the fresh air handling unit is guaranteed; and maintenance cost is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of fault prediction, and in particular to a method for predicting faults of fresh air units based on big data analysis. Background Art

[0002] Fresh air units are increasingly used in indoor locations such as data centers, large commercial complexes, and hospitals. Their stable operation is crucial to maintaining indoor air quality and energy efficiency. Because fresh air units require long-term, continuous operation in complex and changing environments, they are susceptible to issues such as filter clogs, motor aging, and control system failures, leading to problems such as insufficient air volume, increased energy consumption, and excessive noise levels.

[0003] At present, in the Chinese invention with publication number CN116821803A, a wind turbine fault prediction method based on sensor data is disclosed. The method uses sensors to collect initial data of wind turbines and preprocess them, uses factor analysis to reduce the dimension to obtain feature data, builds an initial model, and then optimizes the model parameters through the Big Bang algorithm to obtain an optimized model to predict faults and improve the accuracy of fault prediction. However, there are limitations in the relevant technology: it assumes that the data collected by the sensor is true and valid, and only processes abnormal data through simple preprocessing methods such as eliminating zero values ​​and duplicate values. The source of the abnormal data is not deeply analyzed, and it is impossible to distinguish whether the data anomaly is caused by the sensor. Is it caused by the fault of the sensor itself (such as disconnection, signal drift, electromagnetic interference) or the real fault of the unit equipment (such as gearbox wear, motor abnormality)? There is a lack of sensor status monitoring and fault self-diagnosis mechanism. When the sensor has a transient fault or performance degradation, the collected abnormal data will be directly used for model analysis, resulting in false positives or omissions in the prediction results. The data processing is in a passive state, and the model's ability to fit the existing data pattern can only be improved through algorithm optimization, but it cannot solve the confusion between sensor failure and equipment failure from the root, affecting the reliability of fault prediction and bringing potential risks to the safe operation of the fresh air unit. Summary of the Invention

[0004] The technical problem solved by the present invention is that the fresh air unit fault prediction method in the related technology cannot distinguish whether the data anomaly is caused by sensor failure or real equipment failure, and lacks sensor status monitoring and self-diagnosis mechanism, resulting in false positives and missed warnings in the prediction, affecting the operational reliability of the fresh air unit.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: The fresh air unit fault prediction method based on big data analysis includes the following steps: Step S1: Collect the equipment operation data of the fresh air unit and the sensor status data, and perform preprocessing to generate a target data set; Step S2: extracting features from the target data set to obtain device operation features and sensor health features; Step S3: Based on the equipment operation characteristics and sensor health characteristics, a sensor fault rule set and an equipment fault rule set are constructed to generate a fault judgment logic rule base; Step S4: Optimize the trigger threshold conditions of the fault judgment logic rule base, match the trigger threshold conditions with the preprocessed real-time sensor self-state data and real-time equipment operation data, represent the fault type based on the matching result, and represent the prediction result according to the fault type.

[0006] As a preferred solution of the method for predicting fresh air unit faults based on big data analysis described in the present invention, the equipment operation data includes power system data, mechanical structure data, environmental load data and control system data; The power system data includes impeller speed data, motor power data, stator current data and input terminal voltage data; The mechanical structure data includes gearbox oil temperature data, bearing temperature data, vibration peak data and pitch angle data; The environmental load data includes wind speed data, wind direction data, inlet air temperature data and load rate data; The control system data includes hydraulic system pressure data, pitch drive current data and brake pad pressure data; The sensor's own state data includes the sensor's operating voltage data, communication response time data, and environmental interference intensity data; The frequency of collecting the sensor's own status data is x times that of the equipment's operating data, which is used to capture instantaneous fault signals; The instantaneous fault signals include peak voltage signals, sudden voltage drop signals, calibration error signals, sensor measurement value jump signals, intermittent signals caused by poor contact of hardware interfaces, and instantaneous noise signals caused by electromagnetic interference.

[0007] As a preferred solution of the method for predicting fresh air unit failure based on big data analysis according to the present invention, step S1 includes the following sub-steps: Step S11, collecting equipment operation data of the fresh air unit and sensor status data; Step S12, setting a health threshold range group for the sensor's own status data, and constructing a dynamic historical normal operating condition interval through a sliding window statistical method based on the device operation data; Step S13: performing correlation verification on the sensor's own state data and the device operation data. The correlation verification specifically includes: If one of the data in the sensor's own status data is outside the health threshold range group, it means that the device operation data collected by the sensor is sensor fault data, the marking information is the fault sensor number and the specific abnormal dimension, and the device operation data is marked as invalid data; If all the sensor status data are within the healthy threshold range, but the device operation data deviates from the historical normal operating range, it is represented as device abnormal data, and the marking information is the abnormality type and deviation magnitude, and the device operation data is marked as valid abnormal data; If the sensor's own status data is normal and the equipment operation data is within the historical normal operating range, it is considered normal equipment data and marked as valid normal data; Step S14: integrating the valid normal data, valid abnormal data and marking information to generate a target data set including a timestamp, the normal device data, abnormal device data and marking information.

[0008] As a preferred solution of the fresh air unit fault prediction method based on big data analysis described in the present invention, wherein: the health threshold range group includes an operating voltage threshold range, a communication response time threshold range, and an environmental interference intensity threshold range; The sliding window statistical method uses the wind speed data and load rate data as the basis for division, and the confidence interval threshold of the statistical equipment operation data in each window is used as the normal range; The historical normal operating condition interval includes the matching curve cluster of the impeller speed data and the motor power data, the threshold of the gearbox oil temperature data and the time gradient, the pitch angle data adjustment amplitude range and the fluctuation range of the hydraulic system pressure data.

[0009] As a preferred solution of the method for predicting fresh air unit failure based on big data analysis according to the present invention, step S2 includes the following sub-steps: Step S21, extracting the mean, peak value, and number of fluctuations of the device operation data within a first unit time, and using the mean, peak value, and number of fluctuations as device operation features; Step S22 , extracting the numerical drift and fluctuation frequency of the sensor status data within a continuous preset acquisition period, and using the numerical drift and fluctuation frequency as sensor health features.

[0010] As a preferred solution of the method for predicting fresh air unit failure based on big data analysis according to the present invention, step S3 includes the following sub-steps: Step 31: constructing the sensor fault rule set based on the sensor fault data and the sensor health characteristics, wherein the sensor fault rule set includes a power supply anomaly rule, a communication anomaly rule, a measurement drift rule, a hardware failure rule, and a redundancy mismatch rule; Step 32: constructing the equipment fault rule set based on the equipment abnormality data and the equipment operation characteristics, wherein the equipment fault rule set includes a gearbox fault rule, a generator fault rule, a bearing fault rule, a pitch system fault rule, and a hydraulic system fault rule; Step 33: Generate a fault judgment logic rule base based on the sensor fault rule set and the device fault rule set. Establish a two-layer rule architecture and priority logic based on the fault judgment logic rule base. The two-layer rule architecture includes upper-layer rules and lower-layer rules. The upper-layer rules are the sensor fault rule set, and the lower-layer rules are the device fault rule set. Before triggering the two-layer rule architecture, perform a sensor status check. The sensor status check specifically includes: If the sensor status is marked as abnormal, the upper layer rule is triggered first; If the sensor status is marked as normal, enter the lower layer rules.

[0011] As a preferred solution of the fresh air unit fault prediction method based on big data analysis described in the present invention, wherein: the power supply abnormality rule includes that the sensor operating voltage deviates from the rated value within a first threshold range and lasts for a first preset collection range, indicating that the sensor fault type is a sensor power supply fault; The communication abnormality rule includes that the communication response time is greater than the second threshold range and the packet loss rate is greater than the third threshold range, or there is no signal feedback for a second consecutive unit period, indicating that the sensor fault type is a sensor communication fault; The packet loss rate includes the ratio of the number of lost packets to the total number of sent packets, and the number of data packets that are lost to the total number of data packets. The measurement drift rule includes that the numerical drift amount is greater than a fourth threshold range and the fluctuation frequency is greater than a fifth threshold range, indicating that the sensor fault type is a sensor measurement drift fault; The hardware failure rule includes that the hardware self-test result is abnormal and the environmental interference intensity is less than a sixth threshold range, indicating that the sensor fault type is a sensor hardware failure fault; The redundancy mismatch rule includes that the difference between the primary sensor data and the redundant sensor data of the same monitoring point is greater than a seventh threshold range, and the primary sensor data and the redundant sensor data are normal, indicating that the sensor fault type is a sensor data mismatch fault; The gearbox fault rule includes that the gearbox temperature meets the first threshold temperature and the speed is stable, or the vibration peak meets the eighth threshold range and lasts for a third unit time, indicating that the equipment fault type is a gearbox abnormality; The generator fault rule includes that the wind speed meets the ninth threshold range and the pitch angle is adjusted normally, or the stator temperature meets the second threshold temperature, indicating that the equipment fault type is abnormal generator efficiency; The bearing fault rule includes that the bearing vibration peak value meets the tenth threshold range, indicating that the equipment fault type is a bearing wear fault; The pitch system fault rule includes that the pitch angle adjustment amplitude meets the eleventh threshold range, or the pitch motor current meets the twelfth threshold range, indicating that the equipment fault type is pitch drive abnormality; The hydraulic system failure rule includes that the hydraulic oil pressure fluctuation range meets the thirteenth threshold range, or the oil temperature meets the third threshold temperature, indicating that the equipment failure type is a hydraulic system failure.

[0012] As a preferred solution of the method for predicting fresh air unit failure based on big data analysis according to the present invention, step S4 specifically includes: Optimizing the trigger threshold conditions of the fault judgment logic rule base through a dynamic threshold algorithm, matching the trigger threshold conditions with pre-processed real-time sensor status data and real-time device operation data, determining the matching results based on the two-layer rule architecture and priority logic, directly indicating the fault type based on the matching results, and indicating the prediction results based on the fault type; The trigger threshold conditions of the dynamic threshold algorithm to optimize the fault judgment logic rule base include: Based on the accumulated operating time of the wind turbine, the historical normal operating condition interval is updated once in the second unit period, and the triggering threshold of the equipment fault rule set is adjusted synchronously; Presetting the threshold value of the sensor health feature according to the sensor model and installation location; The trigger threshold condition is matched with the pre-processed real-time sensor status data and real-time device operation data, and the matching result is determined according to the two-layer rule architecture and priority logic. The fault type is indicated based on the matching result. The matching process specifically includes: Setting the priority of the sensor fault rule set to be higher than that of the device fault rule set, when the upper-level rule is triggered, immediately terminating the matching process of the lower-level rule, indicating that the fault type is a sensor fault type; If the upper layer rule is not triggered and there is a matching item in the lower layer rule, it indicates that the fault type is a device fault type; If neither the upper-layer rule nor the lower-layer rule is triggered, it indicates that the device is operating normally.

[0013] As a preferred solution of the method for predicting fresh air unit failure based on big data analysis described in the present invention, the prediction results include: If it is the sensor fault type, it indicates the installation location of the faulty sensor and the fault manifestation; If it is the equipment failure type, it indicates the specific values ​​of the faulty components and abnormal parameters.

[0014] As a preferred solution of the method for predicting fresh air unit faults based on big data analysis described in the present invention, the fault judgment logic rule library has a self-updating mechanism, including: Comparing the actual maintenance records with the prediction results at a preset time, and correcting the trigger threshold conditions of the fault judgment logic rule base that have misjudgments; When a new fault type appears, the sensor fault rule set or device fault rule set corresponding to the new fault type is directly added to the fault judgment logic rule base without adjusting the two-layer rule architecture and priority logic.

[0015] The beneficial effects of the present invention are as follows: by distinguishing between sensor failures and real equipment failures, solving the problem of confusion over the source of data anomalies, building a two-layer rule architecture and priority logic, combining dynamic threshold algorithms to optimize trigger conditions, prioritizing the verification of sensor status, ensuring data reliability, introducing sensor status monitoring and self-diagnosis mechanisms, avoiding false positives and omissions due to sensor failures, and the fault judgment logic rule base has the ability to self-update, which can supplement new fault types and adapt to long-term equipment operation changes, thereby improving the accuracy and reliability of fresh air unit fault prediction, ensuring its stable operation, and reducing maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of the basic flow of a fresh air unit fault prediction method based on big data analysis provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0018] Example 1, reference Figure 1 , which is an embodiment of the present invention, provides a fresh air unit fault prediction method based on big data analysis, including: Step S1: Collect the equipment operation data of the fresh air unit and the sensor status data, and perform preprocessing to generate a target data set; Step S2: Extract features from the target data set to obtain device operation features and sensor health features; Step S3: Based on the equipment operation characteristics and sensor health characteristics, a sensor fault rule set and an equipment fault rule set are constructed to generate a fault judgment logic rule base; Step S4: Optimize the trigger threshold conditions of the fault judgment logic rule base, match the trigger threshold conditions with the preprocessed real-time sensor status data and real-time equipment operation data, represent the fault type based on the matching result, and represent the prediction result according to the fault type.

[0019] In one embodiment, the equipment operation data (power system data, mechanical structure data, environmental load data and control system data) and the sensor's own status data (sensor's working voltage stability, communication response time, hardware self-test results and environmental interference intensity) of the fresh air unit are collected to achieve comprehensive perception of the fresh air unit's operating status and build a data foundation; the equipment operation data and the sensor's own status data are pre-processed to generate a target data set, achieve data noise reduction and purification and preliminary distinction of abnormal sources, and provide high-quality data support for subsequent analysis; feature extraction is performed on the target data set to obtain equipment operation characteristics and sensor health characteristics, and realize the conversion from raw data to key information. ization, accurately capturing the core indicators that reflect faults; based on the equipment operation characteristics and sensor health characteristics, construct sensor fault rule sets and equipment fault rule sets, generate a fault judgment logic rule base, realize the systematization and structuring of fault judgment standards, and establish a framework for distinguishing sensor and equipment faults; optimize the trigger threshold conditions of the fault judgment logic rule base through the dynamic threshold algorithm, match the trigger threshold conditions with the pre-processed real-time sensor status data and real-time equipment operation data, directly indicate the fault type based on the matching result, and indicate the prediction result according to the fault type, realize accurate fault identification, type determination and efficient early warning, and improve the accuracy and reliability of fresh air unit fault prediction.

[0020] Step S1 includes the following sub-steps: Step S11, collecting equipment operation data of the fresh air unit and sensor status data; Step S12: setting a health threshold range group for the sensor's own status data, and constructing a dynamic historical normal operating condition interval through a sliding window statistical method based on the equipment operation data; Step S13: performing correlation verification on the sensor's own status data and the device operation data. The correlation verification specifically includes: If any of the sensor's status data is outside the health threshold range, it indicates that the device operation data collected by the sensor is sensor fault data. The marking information is the fault sensor number and the specific abnormal dimension, and the device operation data is marked as invalid data. If all sensor status data are within the healthy threshold range, but the equipment operation data deviates from the historical normal operating range, it is represented as equipment abnormal data, and the marking information is the abnormality type and deviation magnitude, marking the equipment operation data as valid abnormal data; If the sensor's own status data is normal and the equipment operation data is within the historical normal operating range, it is considered normal equipment data and marked as valid normal data; Step S14: Integrate the valid normal data, valid abnormal data and marking information to generate a target data set including a timestamp, normal device data, abnormal device data and marking information.

[0021] In one embodiment, a 16-channel intelligent acquisition module (model: ADAM-4017, sampling accuracy 16 bits) is used to collect equipment operation data, and each sensor is connected through a shielded cable. The data is transmitted to a local server (storage capacity 1TB, sampling frequency 10Hz) via industrial Ethernet. A status monitoring unit (model: SMU-200, supporting 4-channel analog + 8-channel digital input) is provided for each device sensor to collect the sensor's own status data. It is independently powered (DC24V) and has an acquisition frequency of 50Hz (5 times that of the equipment operation data, used to capture transient fault signals). The data is summarized through the RS485 bus. The collected equipment operation data includes power system data (impeller speed data, motor power indication data, etc.). The collected sensor status data include working voltage data, communication response time data, hardware self-test result data and environmental interference intensity data, and set the health threshold range group of the sensor status data (the working voltage threshold range is 24V±10%, the communication response time threshold is greater than 200ms and the environmental interference intensity threshold is less than 100ms). Greater than -70dBm), based on equipment operation data including power system data (impeller speed data, motor power data, stator current data and input voltage data), mechanical structure data (gearbox oil temperature data, bearing temperature data, vibration peak data and pitch angle data), environmental load data (wind speed data, wind direction data, inlet temperature data and load rate data) and control system data (hydraulic system pressure data, pitch drive current data and brake pad pressure data (non-braking state)), through the sliding window statistical method (with 15 minutes as the window and 10 minutes as the step size, divided into 8 working condition intervals according to wind speed data and load rate data, and the 95% confidence interval of the equipment operation data in each interval is used as the normal range), Construct a dynamic historical normal operating condition range (matching curve clusters for impeller speed data and motor power data (e.g., 1200 rpm corresponds to 50 kW), gearbox oil temperature data and time gradient thresholds less than 2°C / 10 min, pitch angle data adjustment range of ±10°, hydraulic system pressure data fluctuation range of 1.2-1.8 MPa); perform correlation verification between sensor status data and equipment operation data, a priori verifying sensor reliability before determining equipment status, avoiding misidentification of sensor faults as equipment problems, improving fault location accuracy, and reducing ineffective maintenance costs. If any of the sensor status data is greater than the healthy threshold range (the operating voltage of sensor G1 (gearbox oil temperature) is 27.0V (exceeds the operating voltage threshold), response time 120ms and interference -82dBm are normal), it is indicated as sensor fault data, and the marking information is "fault sensor number G1, abnormal dimension: power supply voltage", marked as invalid data. If the sensor's own status data are all within the healthy threshold range group (the status data of sensor Y1 (hydraulic pressure) including voltage 23.5V, response time 100ms and interference -85dBm are all within the healthy threshold), but the equipment operation data deviates from the historical normal operating range (the pressure value 1.8MPa deviates from the historical normal range (1.4-1.6MPa)), it is indicated as equipment abnormal data, and the marking information is "abnormal type: pressure exceeds the limit, deviation amplitude 12.5%", marked as valid abnormal data; if the sensor status data are all within the healthy threshold range group (the status data of sensor Y1 (hydraulic pressure) including voltage 23.5V, response time 100ms and interference -85dBm are all within the healthy threshold), but the equipment operation data deviates from the historical normal operating range (the pressure value 1.8MPa deviates from the historical normal range (1.4-1.6MPa)), it is indicated as equipment abnormal data, and the marking information is "abnormal type: pressure exceeds the limit, deviation amplitude 12.5%", marked as valid abnormal data; If the sensor's status data is normal and the equipment's operating data falls within the historical normal operating range (e.g., sensor N1 (impeller speed) status data is normal, and the collected speed of 1100 r / min falls within the speed-power matching curve cluster corresponding to the wind speed and load rate data conditions), this data is considered normal and marked as valid. This data, along with the valid normal data, valid abnormal data, and the marking information, is integrated to generate a target dataset containing timestamps, normal equipment data, abnormal equipment data, and marking information (e.g., 14:00:00 gearbox oil temperature 70°C (invalid data, marked as G1 power supply abnormality)). The timestamp is retained to reflect data temporality. The data type and marking information clarify data quality and the cause of the abnormality, laying the foundation for subsequent feature extraction and rule construction.

[0022] Step S2 includes the following sub-steps: Step S21, extracting the mean, peak value, and number of fluctuations of the equipment operation data within the first unit time, and using the mean, peak value, and number of fluctuations as equipment operation features; Step S22 , extracting the numerical drift and fluctuation frequency of the sensor status data within a continuous preset acquisition period, and using the numerical drift and fluctuation frequency as sensor health features.

[0023] In one embodiment, the extracted gearbox oil temperature within 10 minutes is 55°C, the peak value is 62°C, and the number of fluctuations is 4 times; the hydraulic system pressure within 10 minutes is 1.5MPa, the peak value is 1.7MPa, and the number of fluctuations is 2 times; and the bearing vibration within 10 minutes is 0.3mm / s, the peak value is 0.45mm / s, and the number of fluctuations is 3 times as the equipment operation characteristics; for 5 consecutive acquisition cycles (each cycle is 10 minutes), the temperature sensor voltage drifts from 24V to 23.2V, the numerical drift is 0.8V, the fluctuation frequency is 2 times in 1 cycle, and the communication response time drift of the vibration sensor is 15ms, and the fluctuation frequency is 1 time in 1 cycle as the sensor health characteristics.

[0024] Step S3 includes the following sub-steps: Step 31: construct a sensor fault rule set based on the sensor fault data and sensor health characteristics. The sensor fault rule set includes power supply anomaly rules, communication anomaly rules, measurement drift rules, hardware failure rules, and redundancy mismatch rules. Step 32: constructing an equipment fault rule set based on the equipment abnormality data and equipment operation characteristics. The equipment fault rule set includes gearbox fault rules, generator fault rules, bearing fault rules, pitch system fault rules, and hydraulic system fault rules. Step 33: Generate a fault judgment logic rule base based on the sensor fault rule set and the device fault rule set. Establish a two-layer rule architecture and priority logic based on the fault judgment logic rule base. The two-layer rule architecture includes upper-layer rules and lower-layer rules. The upper-layer rules are the sensor fault rule set, and the lower-layer rules are the device fault rule set. Before triggering the two-layer rule architecture, perform a sensor status check. The sensor status check specifically includes: If the sensor status is marked as abnormal, the upper-level rules will be triggered first; If the sensor status is marked as normal, enter the lower level rules.

[0025] In one embodiment, a sensor fault rule set is constructed based on sensor fault data and sensor health characteristics. The sensor fault rule set specifically includes a power supply abnormality rule (the sensor operating voltage deviates from the rated value first threshold range (set to ±15% of the rated value, i.e., 24V±3.6V) and lasts for a first preset acquisition range (set to 30 seconds), indicating that the sensor fault type is a sensor power supply failure), a communication abnormality rule (the communication response time is greater than the second threshold range (set to 200ms) and the packet loss rate is greater than the third threshold range (set to 5%), or for a second consecutive unit cycle (set to 3 acquisition cycles, each cycle is 10 seconds) ) no signal feedback, indicating that the sensor fault type is sensor communication failure), measurement drift rule (the value drift is greater than the fourth threshold range (set to be greater than 1.0V) and the fluctuation frequency is greater than the fifth threshold range (set to be greater than 3 times / 10min), indicating that the sensor fault type is sensor measurement drift fault), hardware failure rule (the hardware self-test result is abnormal and the environmental interference intensity is less than the sixth threshold range (set to be less than -75dBm) (excluding external interference), indicating that the sensor fault type is sensor hardware failure) and redundancy mismatch rule (the difference between the main sensor data and the redundant sensor data at the same monitoring point is greater than the seventh threshold range (set to be greater than 5% rated value, such as the temperature sensor difference is greater than 3°C, the pressure sensor difference is greater than 0.1MPa), and the main sensor data and redundant sensor data are normal, indicating that the sensor fault type is a sensor data mismatch fault); based on the equipment abnormal data and equipment operation characteristics, a set of equipment fault rules is constructed. The equipment fault rule set specifically includes gearbox fault rules (the gearbox temperature meets the first threshold temperature (set to 70°C) and the speed is stable, or the vibration peak meets the eighth threshold range (set to 0.6mm / s) and lasts for the third unit time (set to 5 minutes), indicating that the equipment fault type is a gearbox abnormality), generator fault rules (the wind speed meets the ninth threshold range), and generator fault rules (the wind speed meets the ninth threshold range). The following are the equipment fault rules: (1) the pitch angle adjustment range is within the range of ±0.3 MPa (set to ±0.3 MPa) and the pitch angle adjustment is normal, or the stator temperature is greater than the second threshold temperature (set to 120°C), indicating that the equipment fault type is abnormal generator efficiency); (2) the bearing fault rule (including the bearing vibration peak value is greater than the tenth threshold range (set to 0.5 mm / s), indicating that the equipment fault type is bearing wear fault); (3) the pitch angle adjustment range is within the range of ±15° (set to ±15°), or the pitch motor current is greater than the twelfth threshold range (set to 15A), indicating that the equipment fault type is abnormal pitch drive); and (4) the hydraulic system fault rule (including the hydraulic oil pressure fluctuation range is within the range of ±0.13 threshold range (set to ±0.13 threshold range).3MPa), or the oil temperature is greater than the third threshold temperature (set to 55°C), indicating that the equipment fault type is a hydraulic system fault. Based on the sensor fault rule set and the equipment fault rule set, a fault judgment logic rule base is generated. Based on the fault judgment logic rule base, a two-layer rule architecture and priority logic are established. The two-layer rule architecture includes upper-layer rules and lower-layer rules. The upper-layer rules are the sensor fault rule set, and the lower-layer rules are the equipment fault rule set. Before triggering the two-layer rule architecture, a sensor status check is performed. Specifically, the sensor status check includes: if the sensor status is marked as abnormal, the upper-layer rules are triggered first; if the sensor status is marked as normal, the lower-layer rules are triggered.

[0026] Step S4 includes: The trigger threshold conditions of the fault judgment logic rule base are optimized through the dynamic threshold algorithm, and the trigger threshold conditions are matched with the preprocessed real-time sensor status data and real-time equipment operation data. The matching results are determined according to the two-layer rule architecture and priority logic. The fault type is directly indicated based on the matching result, and the prediction result is indicated according to the fault type.

[0027] In one embodiment, a dynamic threshold algorithm is used to optimize the trigger threshold conditions of the fault judgment logic rule base. Based on the cumulative operating time of the wind turbine exceeding 1000 hours, the historical normal operating condition interval is updated once within the second unit period (set to be within 24 hours), and the trigger threshold of the equipment fault rule set is adjusted synchronously (the gearbox temperature threshold is raised from greater than 70°C to greater than 72°C, and the hydraulic pressure fluctuation threshold is relaxed from greater than ±0.3MPa to greater than ±0.35MPa). The thresholds of sensor health characteristics are preset according to the sensor model and installation location (temperature sensor (installed in the gearbox, model PT100): voltage drift threshold greater than 1.0V, fluctuation frequency threshold greater than 3 times / 10 minutes; pressure sensor (installed in the hydraulic Pressure pipeline, model MPM480): numerical drift threshold greater than 0.1MPa, fluctuation frequency threshold greater than 2 times / 10 minutes; vibration sensor (installed on the bearing seat, model YD-12): environmental interference intensity threshold <-75dBm (for hardware failure rules)); matching the trigger threshold condition with the pre-processed real-time sensor status data and real-time equipment operation data, setting the priority of the sensor fault rule set higher than the equipment fault rule set, and determining the matching result based on the two-layer rule architecture and priority logic. Scenario 1: Upper-layer rule triggering, real-time sensor status data: gearbox temperature sensor (number T1) operating voltage 28V (exceeding 24V±10%, i.e., greater than the 26.4V threshold), duration 4 0s (exceeding the first preset collection range by 30s), the matching result is to trigger the power supply abnormality rule, immediately terminate the lower-level device rule matching, indicating that the fault type is a sensor fault type; Scenario 2: The lower-level rule is triggered, the real-time sensor status data: the hydraulic pressure sensor (No. Y1) is in normal status (voltage 23.5V, response time 180ms, etc. are all within the threshold), the real-time device operation data: the hydraulic pressure fluctuation is ±0.4MPa (exceeding the thirteenth threshold range by more than ±0.3MPa), the matching result is that the upper-level rule is not triggered, the hydraulic system fault rule is triggered, indicating that the fault type is a device fault type; Scenario 3: No rule is triggered, the real-time sensor status data are all normal, the equipment operation data (impeller speed 1150r / m in, gearbox oil temperature 58°C) are both within the historical normal range, and the matching result is that neither the upper-level rules nor the lower-level rules are triggered, indicating that the equipment is operating normally. The fault type is directly indicated based on the matching result, and the prediction result is indicated according to the fault type. The prediction result includes, if it is a sensor fault type, the installation position and fault manifestation of the fault sensor. Sensor fault type: indicates the installation position of the fault sensor: top of the gearbox (number T1), fault manifestation: working voltage continuously 28V (exceeding the upper limit of 26.4V); if it is an equipment fault type, the specific values ​​of the faulty component and abnormal parameters are indicated. Equipment fault type: indicates the faulty component: hydraulic system, abnormal parameters: pressure fluctuation ±0.4MPa (exceeding the upper limit of ±0.3MPa).

[0028] The fault judgment logic rule library has a self-update mechanism, including: Compare the actual maintenance records with the predicted results at the preset time, and correct the trigger threshold conditions of the fault judgment logic rule base that have misjudgments; When a new fault type appears, the sensor fault rule set or device fault rule set corresponding to the new fault type is directly added to the fault judgment logic rule base without adjusting the two-layer rule architecture and priority logic.

[0029] In one embodiment, the actual maintenance records (of the three warnings last month, two vibration peaks of 0.62 mm / s were detected but no abnormalities were found during disassembly inspection (false positives), and one vibration of 0.75 mm / s was confirmed as bearing wear (true positives)) were compared with the predicted results (the gearbox vibration peak was greater than 0.6 mm / s and lasted for 5 minutes, indicating a gearbox abnormality) at a preset time (set to 30 days). The trigger threshold conditions of the fault judgment logic rule base with false positives were modified (the vibration peak threshold was changed from greater than 0.6 mm / s to Adjust to greater than 0.7mm / s, maintain the duration unchanged for 5 minutes, and reduce the false alarm rate; when a new fault type occurs (pitch motor overheating during operation, manifested as pitch drive current greater than 18A and motor housing temperature greater than 85°C, and the sensor status is normal), a new pitch motor overheating rule is added to the equipment fault rule set: when the pitch drive current is greater than 18A and the motor temperature is greater than 85°C, it is judged as a pitch motor overheating fault. There is no need to adjust the priority of the upper and lower rules. The new rule is automatically included in the matching process, and the fault is accurately identified when it first occurs, maintaining the stability and scalability of the two-layer architecture.

[0030] 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 the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A new air unit fault prediction method based on big data analysis is characterized by: The following steps are involved: Step S1: Collect the equipment operation data of the fresh air unit and the sensor status data, and perform preprocessing to generate a target data set; Step S2: extracting features from the target data set to obtain device operation features and sensor health features; Step S3: Based on the equipment operation characteristics and sensor health characteristics, a sensor fault rule set and an equipment fault rule set are constructed to generate a fault judgment logic rule base; Step S4: Optimize the trigger threshold conditions of the fault judgment logic rule base, match the trigger threshold conditions with the preprocessed real-time sensor self-state data and real-time equipment operation data, represent the fault type based on the matching result, and represent the prediction result according to the fault type.

2. The method for predicting failure of a fresh air unit based on big data analysis according to claim 1, characterized in that: The equipment operation data includes power system data, mechanical structure data, environmental load data and control system data; The power system data includes impeller speed data, motor power data, stator current data and input terminal voltage data; The mechanical structure data includes gearbox oil temperature data, bearing temperature data, vibration peak data and pitch angle data; The environmental load data includes wind speed data, wind direction data, inlet air temperature data and load rate data; The control system data includes hydraulic system pressure data, pitch drive current data and brake pad pressure data; The sensor's own state data includes the sensor's operating voltage data, communication response time data, and environmental interference intensity data; The frequency of collecting the sensor's own status data is x times that of the equipment's operating data, which is used to capture instantaneous fault signals; The instantaneous fault signals include peak voltage signals, sudden voltage drop signals, calibration error signals, sensor measurement value jump signals, intermittent signals caused by poor contact of hardware interfaces, and instantaneous noise signals caused by electromagnetic interference.

3. The method for predicting failure of a fresh air unit based on big data analysis according to claim 2, characterized in that: The step S1 includes the following sub-steps: Step S11, collecting equipment operation data of the fresh air unit and sensor status data; Step S12, setting a health threshold range group for the sensor's own status data, and constructing a dynamic historical normal operating condition interval through a sliding window statistical method based on the device operation data; Step S13: performing correlation verification on the sensor's own state data and the device operation data. The correlation verification specifically includes: If one of the data in the sensor's own status data is outside the health threshold range group, it means that the device operation data collected by the sensor is sensor fault data, the marking information is the fault sensor number and the specific abnormal dimension, and the device operation data is marked as invalid data; If all the sensor status data are within the healthy threshold range, but the device operation data deviates from the historical normal operating range, it is represented as device abnormal data, and the marking information is the abnormality type and deviation magnitude, and the device operation data is marked as valid abnormal data; If the sensor's own status data is normal and the equipment operation data is within the historical normal operating range, it is considered normal equipment data and marked as valid normal data; Step S14: integrating the valid normal data, valid abnormal data and marking information to generate a target data set including a timestamp, the normal device data, abnormal device data and marking information.

4. The method for predicting failure of a fresh air unit based on big data analysis according to claim 3, characterized in that: The health threshold range group includes an operating voltage threshold range, a communication response time threshold range, and an environmental interference intensity threshold range; The sliding window statistical method uses the wind speed data and load rate data as the basis for division, and the confidence interval threshold of the statistical equipment operation data in each window is used as the normal range; The historical normal operating condition interval includes the matching curve cluster of the impeller speed data and the motor power data, the threshold of the gearbox oil temperature data and the time gradient, the pitch angle data adjustment amplitude range and the fluctuation range of the hydraulic system pressure data.

5. The method for predicting failure of a fresh air unit based on big data analysis according to claim 1, characterized in that: The step S2 includes the following sub-steps: Step S21, extracting the mean, peak value, and number of fluctuations of the device operation data within a first unit time, and using the mean, peak value, and number of fluctuations as device operation features; Step S22 , extracting the numerical drift and fluctuation frequency of the sensor's own state data within a continuous preset acquisition period, and using the numerical drift and fluctuation frequency as sensor health characteristics.

6. The method for predicting failure of a fresh air unit based on big data analysis according to claim 3, characterized in that: The step S3 includes the following sub-steps: Step 31: constructing the sensor fault rule set based on the sensor fault data and the sensor health characteristics, wherein the sensor fault rule set includes a power supply anomaly rule, a communication anomaly rule, a measurement drift rule, a hardware failure rule, and a redundancy mismatch rule; Step 32: constructing the equipment fault rule set based on the equipment abnormality data and the equipment operation characteristics, wherein the equipment fault rule set includes a gearbox fault rule, a generator fault rule, a bearing fault rule, a pitch system fault rule, and a hydraulic system fault rule; Step 33: Generate a fault judgment logic rule base based on the sensor fault rule set and the device fault rule set. Establish a two-layer rule architecture and priority logic based on the fault judgment logic rule base. The two-layer rule architecture includes upper-layer rules and lower-layer rules. The upper-layer rules are the sensor fault rule set, and the lower-layer rules are the device fault rule set. Before triggering the two-layer rule architecture, perform a sensor status check. The sensor status check specifically includes: If the sensor status is marked as abnormal, the upper layer rule is triggered first; If the sensor status is marked as normal, enter the lower layer rules.

7. The method for predicting failure of a fresh air unit based on big data analysis according to claim 6, characterized in that: The power supply anomaly rule includes that the sensor operating voltage deviates from the rated value within a first threshold range and lasts for a first preset acquisition range, indicating that the sensor fault type is a sensor power supply fault; The communication abnormality rule includes that the communication response time is greater than the second threshold range and the packet loss rate is greater than the third threshold range, or there is no signal feedback for a second consecutive unit period, indicating that the sensor fault type is a sensor communication fault; The packet loss rate includes the ratio of the number of lost packets to the total number of sent packets, and the number of data packets that are lost to the total number of data packets. The measurement drift rule includes that the numerical drift amount is greater than a fourth threshold range and the fluctuation frequency is greater than a fifth threshold range, indicating that the sensor fault type is a sensor measurement drift fault; The hardware failure rule includes that the hardware self-test result is abnormal and the environmental interference intensity is less than a sixth threshold range, indicating that the sensor fault type is a sensor hardware failure fault; The redundancy mismatch rule includes that the difference between the primary sensor data and the redundant sensor data of the same monitoring point is greater than a seventh threshold range, and the primary sensor data and the redundant sensor data are normal, indicating that the sensor fault type is a sensor data mismatch fault; The gearbox fault rule includes that the gearbox temperature meets the first threshold temperature and the speed is stable, or the vibration peak meets the eighth threshold range and lasts for a third unit time, indicating that the equipment fault type is a gearbox abnormality; The generator fault rule includes that the wind speed meets the ninth threshold range and the pitch angle is adjusted normally, or the stator temperature meets the second threshold temperature, indicating that the equipment fault type is abnormal generator efficiency; The bearing fault rule includes that the bearing vibration peak value meets the tenth threshold range, indicating that the equipment fault type is a bearing wear fault; The pitch system fault rule includes that the pitch angle adjustment amplitude meets the eleventh threshold range, or the pitch motor current meets the twelfth threshold range, indicating that the equipment fault type is pitch drive abnormality; The hydraulic system failure rule includes that the hydraulic oil pressure fluctuation range meets the thirteenth threshold range, or the oil temperature meets the third threshold temperature, indicating that the equipment failure type is a hydraulic system failure.

8. The method for predicting failure of a fresh air unit based on big data analysis according to claim 7, characterized in that: The step S4 specifically includes: Optimizing the trigger threshold conditions of the fault judgment logic rule base through a dynamic threshold algorithm, matching the trigger threshold conditions with pre-processed real-time sensor status data and real-time device operation data, determining the matching results based on the two-layer rule architecture and priority logic, directly indicating the fault type based on the matching results, and indicating the prediction results based on the fault type; The trigger threshold conditions of the dynamic threshold algorithm to optimize the fault judgment logic rule base include: Based on the accumulated operating time of the wind turbine, the historical normal operating condition interval is updated once in the second unit period, and the triggering threshold of the equipment fault rule set is adjusted synchronously; Presetting the threshold value of the sensor health feature according to the sensor model and installation location; The trigger threshold condition is matched with the pre-processed real-time sensor status data and real-time device operation data, and the matching result is determined according to the two-layer rule architecture and priority logic. The fault type is indicated based on the matching result. The matching process specifically includes: Setting the priority of the sensor fault rule set to be higher than that of the device fault rule set, when the upper-level rule is triggered, immediately terminating the matching process of the lower-level rule, indicating that the fault type is a sensor fault type; If the upper layer rule is not triggered and there is a matching item in the lower layer rule, it indicates that the fault type is a device fault type; If neither the upper-layer rule nor the lower-layer rule is triggered, it indicates that the device is operating normally.

9. The method for predicting failure of a fresh air unit based on big data analysis according to claim 8, characterized in that: The prediction results include: If it is the sensor fault type, it indicates the installation location of the faulty sensor and the fault manifestation; If it is the equipment failure type, it indicates the specific values ​​of the faulty components and abnormal parameters.

10. The method for predicting failure of a fresh air unit based on big data analysis according to claim 1, characterized in that: The fault judgment logic rule library has a self-update mechanism, including: Comparing the actual maintenance records with the prediction results at a preset time, and correcting the trigger threshold conditions of the fault judgment logic rule base that have misjudgments; When a new fault type appears, the sensor fault rule set or device fault rule set corresponding to the new fault type is directly added to the fault judgment logic rule base without adjusting the two-layer rule architecture and priority logic.

Images