Fault diagnosis method for desulfurized flue gas on-line monitoring device

Abstract

The present application relates to the field of desulfurization flue gas monitoring, and more particularly to a desulfurization flue gas online monitoring device fault diagnosis method, comprising the following steps: comparing target parameters collected by the monitoring device with preset working boundaries to obtain boundary pre-classification results, and performing abnormality determination based on the boundary pre-classification results. The data of the boundary pre-classification results determined as abnormal are preprocessed, and parallel detection of multiple fault modes is performed based on the preprocessed data. According to the results of abnormality determination and parallel detection, it is determined whether to issue a warning in combination with the current working condition. The present application preliminarily judges the daily working state of the monitoring device by adopting the working boundary mode, and performs initial classification according to the preliminary judgment, and then judges the abnormal state of the data according to the initial classification and parallel detection of multiple fault modes, so that the working state of the monitoring device can be automatically determined.

Description

Technical Field

[0001] This invention relates to the field of desulfurization flue gas monitoring, and more particularly to a fault diagnosis method for online desulfurization flue gas monitoring devices. Background Technology

[0002] Boiler systems are medium to large-scale combustion systems in thermal power plants that supply high-temperature steam to steam turbines. The main heat-generating components of a boiler system are coal mills, furnace chambers, and other combustion equipment. During operation, these systems produce a large amount of sulfur-containing flue gas, which requires post-treatment. During this treatment, various monitoring devices are used to monitor parameters in real time. Various gas sensors are typically used for detecting sulfur-containing flue gas, and their data is a crucial source of automated control parameters for the system. However, unlike traditional industrial applications, detecting sulfur-containing flue gas and treated desulfurized flue gas requires full contact with the gas. These gases often contain various impurities, and after a period of use, impurities accumulate on the sensor surface, affecting detection accuracy. Since the operating conditions and intensity of the boiler system may vary depending on different needs, it is impossible to set a fixed cleaning period. However, excessively short cleaning cycles incur excessive costs. Therefore, a fault diagnosis method for online monitoring devices of desulfurized flue gas is needed. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides the following technical solution: The fault diagnosis method for online monitoring devices for desulfurized flue gas includes the following steps: The target parameters collected by the monitoring device are compared with the preset working boundary to obtain the boundary pre-classification result, and anomaly determination is made based on the boundary pre-classification result.

[0004] The data from the boundary pre-classification results that pass the anomaly detection are preprocessed, and parallel detection of multiple fault modes is performed based on the preprocessed data.

[0005] Based on the results of anomaly detection and parallel detection, and combined with the current operating status, a decision is made on whether to issue an early warning.

[0006] As an improvement to the above technical solution, the working boundary is a preset working condition, which includes at least the physical absolute boundary, the working condition limit boundary, and the normal operation boundary.

[0007] The physical absolute boundary is the maximum boundary range determined based on the physical possible values ​​of the target parameters. The operating condition limit boundary is the maximum fluctuation range determined based on the extreme parameters of the desulfurization system parameters, process parameters, and / or historical operating conditions. The normal operating boundary is the normal fluctuation range determined based on historical data of the target operating stably with parameters.

[0008] As an improvement to the above technical solution, when the boundary pre-classification result is the normal operation boundary or exceeds the normal operation boundary but does not exceed the operating condition limit boundary, it is considered to pass the anomaly judgment.

[0009] When the boundary pre-classification result exceeds the physical absolute boundary, it is identified as an abnormal working state of the monitoring device, and an equipment failure warning is issued.

[0010] When the boundary pre-classification result exceeds the working condition limit boundary, the monitoring device is deemed abnormal and its status cannot be determined, and a warning prompt for manual intervention is issued.

[0011] As an improvement to the above technical solution, the preprocessing includes at least one of missing data processing, outlier processing, and filtering processing.

[0012] As an improvement to the above technical solution, the missing data processing uses the nearest neighbor method to handle occasional missing data and issues a data interruption warning for continuous missing data; the outlier processing includes identifying outliers based on the absolute deviation of the median within a sliding window, and replacing the data identified as outliers with the median in the window in the fault detection input, while retaining the original value and outlier label; the filtering process uses a first-order hysteresis filter to filter the data.

[0013] As an improvement to the above technical solution, the parallel detection of multiple fault modes includes at least two of the following: freeze detection, drift and bias detection, physical consistency detection, noise and accuracy degradation detection, and intermittent fault detection.

[0014] As an improvement to the above technical solution, the freeze detection is based on the fluctuation of the target parameter within the sliding time window and the fluctuation state of the associated parameter; the drift and bias detection is based on the residual between the measured value of the target parameter and the soft measurement estimate; the accuracy degradation detection is based on the variance or coefficient of variation within the sliding time window; and the intermittent fault detection is based on the frequency of outlier occurrence.

[0015] As an improvement to the above technical solution, the drift and bias detection includes: Perform a sequential probability ratio test on the residuals and obtain at least one of the following.

[0016] When the mean residual changes abruptly and meets the preset detection conditions, it is determined to be a bias fault.

[0017] When the residual shows a continuous unidirectional shift and meets the preset detection conditions, it is determined to be a drift fault.

[0018] As an improvement to the above technical solution, the physical consistency verification includes at least one of the following: The amount of sulfur dioxide removed from the flue gas side is calculated based on the inlet sulfur dioxide concentration, outlet sulfur dioxide concentration, and flue gas flow rate, and the deviation is measured with the amount of sulfur dioxide absorbed from the slurry side calculated based on the slurry parameters.

[0019] Obtain the deviation between the inlet flue gas flow rate and the outlet flue gas flow rate of the desulfurization system.

[0020] Obtain the temperature relationship between the inlet flue gas temperature and the outlet flue gas temperature of the desulfurization system.

[0021] A physical inconsistency index is generated based on the calculated deviation and temperature relationship.

[0022] As an improvement to the above technical solution, a method for suppressing false alarms is also provided, comprising the following steps: Set the corresponding confirmation delay according to the fault type.

[0023] Set alarm trigger thresholds and alarm recovery thresholds to form a hysteresis zone.

[0024] Under periodic or sudden operation, alarm lockout or threshold adjustment shall be performed on the corresponding fault detection results.

[0025] During the alarm interlocking period, fault detection calculations continue to be performed, and after the interlocking is released, it is determined whether to output fault diagnosis results based on the current fault detection results.

[0026] The beneficial effects of this invention are: By employing a working boundary approach, the daily working status of the monitoring device is initially assessed and classified. Based on this assessment, the abnormal status of the data is determined using parallel detection of multiple fault modes. The system then determines whether to issue an early warning based on the current operating conditions. Unlike traditional working modes, this approach automatically determines the working status of the monitoring device and combines it with other data to determine whether the device is still in a normal working state. Finally, it comprehensively assesses whether the device's operation is within the normal boundary state, thereby determining whether cleaning of the monitoring device is necessary. Detailed Implementation

[0027] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention.

[0028] Unlike traditional industrial applications, the detection of sulfur-containing flue gas and treated desulfurized flue gas requires full contact with the gas. These gases usually contain various impurities. After a period of use, impurities will adhere to the sensor surface, affecting the sensor's detection accuracy. Since the working state and intensity of the boiler system may change according to different needs, it is not possible to directly set a fixed period for cleaning. However, too short a cleaning cycle will generate excessive costs.

[0029] Based on the above problems, this embodiment proposes a fault diagnosis method for an online monitoring device for desulfurized flue gas.

[0030] The fault diagnosis method for online monitoring devices for desulfurized flue gas includes the following steps: S10: Compare the target parameters collected by the monitoring device with the preset working boundary to obtain the boundary pre-classification result, and make anomaly judgment based on the boundary pre-classification result.

[0031] The system starts operating the moment the desulfurized flue gas is detected by the sensors of the monitoring device. To avoid the influence of too much data, the data is classified before data processing. Since our data analysis needs to be based on the working state, the classification is done by the working state. Therefore, we take the working boundary as the preset working condition, and this working condition includes at least the physical absolute boundary, the working condition limit boundary, and the normal operation boundary.

[0032] The physical absolute boundary reflects the physical limit. For example, water is 100°C under standard atmospheric pressure. This limit temperature changes with the natural environment. However, under non-laboratory conditions, a limit value under normal conditions is usually generated according to different altitudes and natural environments. This limit value is the physical absolute boundary we define. Therefore, we define the physical absolute boundary as the maximum boundary range determined by the physical possible values ​​of the target parameters.

[0033] Operating condition limit boundaries are the maximum fluctuation range determined based on the process, system, and historical conditions. For example, the system determines that the operating temperature is 500℃, with fluctuations not exceeding 10℃. However, during the testing process, an operating temperature of 520℃ is found at a certain time. This situation may be due to process changes, sensor malfunctions, or short-term anomalies caused by extreme environments at a certain moment in the historical records. Therefore, in this case, it is impossible to determine the specific problem range. When this problem occurs, the system cannot directly determine the problem location, and an alarm must be triggered to remind manual maintenance.

[0034] The normal operating boundary refers to the fluctuation range within a defined range of the system that is in a normal state. This is because any system will generate a certain range of operational fluctuations during operation due to the influence of environmental, system, and other factors. Based on this, we need to collect the fluctuation range under normal operating conditions, determine this range, and provide a certain amount of redundancy. Therefore, we define the normal operating boundary as the normal fluctuation range determined based on historical data of the target's stable operation under parameters.

[0035] Based on the definitions of the various boundaries above, we propose the following judgment scheme: When the boundary pre-classification result is the normal operation boundary or exceeds the normal operation boundary but does not exceed the operating condition limit boundary, it is considered to pass the anomaly judgment.

[0036] Being identified as abnormal means that the behavior is abnormal, but the system can analyze and determine the specific problem. A state that exceeds the normal operating boundary and the operating condition limit boundary is usually considered to be able to analyze the problem.

[0037] When the boundary pre-classification result exceeds the physical absolute boundary, it is identified as an abnormal working state of the monitoring device, and an equipment failure warning is issued.

[0038] If the range of the physical absolute boundary is defined at the beginning, then subsequent detection should not detect situations that exceed the absolute boundary. Once such situations occur, it can be directly identified as a malfunction of the monitoring device, and an equipment malfunction warning will be issued to remind manual replacement or repair.

[0039] When the boundary pre-classification result exceeds the working condition limit boundary, the monitoring device is deemed abnormal and its status cannot be determined, and a warning prompt for manual intervention is issued.

[0040] This part has already been explained earlier, so I won't go into too much detail here. After determining the boundaries, we have completed the boundary pre-classification and obtained the analysis results. Based on these results, we will continue to the next step.

[0041] S20: Preprocess the data of the boundary pre-classification results that pass the anomaly determination, and perform parallel detection of multiple fault modes based on the preprocessed data.

[0042] We have already explained the concept of anomaly detection. Here, in order to ensure the accuracy of the data analysis results, we provide a data preprocessing scheme, which should include at least one of missing data processing, outlier processing, and filtering processing. Although only three are given in this embodiment, in reality, depending on the complexity of the data and the different environments, other data preprocessing schemes may be included, such as data augmentation, data standardization, and other processing methods.

[0043] In this embodiment, the missing data processing uses the nearest neighbor method to handle occasional missing data, and issues a data interruption warning for continuous missing data.

[0044] Intermittent data loss refers to the loss of data from individual sampling points during the detection process due to rare system fluctuations or transient effects. In this case, the sensor is usually not faulty, but we need to address the data loss issue. In this embodiment, we use neighbor-value filling for interpolation. Of course, the interpolated values ​​need to be marked to distinguish them from the actual sampled data. However, continuous data loss indicates that there is a fluctuation over a long period of time. Normal operation does not produce such continuous data loss. Therefore, in this case, a data interruption warning needs to be issued to indicate that the monitoring equipment may have a malfunction and further investigation is required.

[0045] Outliers are isolated sampling points within a sliding window that deviate from the overall trend of the current data due to electromagnetic interference or transient signal distortion. In this embodiment, outliers are identified using a robust method based on the absolute deviation of the median, specifically including the following steps: Outliers are identified by replacing the mean with the median within the window as the reference center, and replacing the standard deviation with the median of the absolute deviations of each point from the median as the measure of dispersion.

[0046] Data identified as outliers are replaced with windowed values ​​in the input channel for fault detection to prevent contamination of subsequent cumulative statistical algorithms.

[0047] In this embodiment, a first-order hysteresis filter is used for data filtering. The first-order hysteresis filter smooths the signal by applying a weighted average of the current original sample value and the filtered output value from the previous time step. Its core parameter is the filter coefficient α. The smaller α is, the stronger the filtering and the smoother the output signal, but the greater the response hysteresis. It needs to be set according to the dynamic characteristics of the parameters. Smaller α values ​​are used for slowly changing concentration and temperature signals, while larger α values ​​are used for rapidly changing flow and pressure signals. It is important to note that the filtering only applies to the input channel of fault detection. Manual detection will be based on the original data to avoid the processed data affecting manual detection.

[0048] After data preprocessing is completed, fault detection is also required. This embodiment adopts a parallel detection scheme with multiple fault modes, and the following are some of the schemes provided in this embodiment: Freeze detection, drift and bias detection, physical consistency detection, noise and accuracy degradation detection, and intermittent fault detection.

[0049] For freeze detection, the standard deviation of the target parameter within a sliding time window is used to determine whether the sensor outputs a stable value. If the standard deviation within the window is lower than a preset threshold, while the standard deviations of other independent but related parameters (such as load, flow rate, temperature, etc.) remain normal within the same time period, then the sensor is considered to have experienced a freeze fault. If all parameters remain stable, it indicates that the system may be in a shutdown state, thus avoiding a false diagnosis of sensor failure.

[0050] Drift and bias detection analyzes the residuals between the physical sensor measurements and the software-estimated values ​​of the target parameters. The software-estimated model is trained using historical data from normal operation and uses other independent parameters to extrapolate the target parameters as a reference. When a systematic residual occurs between the physical reading and the software estimate, a slow, unidirectional deviation of the residual mean from zero is considered a drift fault, while a step jump in the residual mean is considered a bias fault. A sequential probability ratio test algorithm is used to statistically evaluate the residual sequence, triggering alarms promptly while controlling the false alarm and false negative rates.

[0051] Accuracy degradation detection determines whether sensor measurement noise has increased abnormally by monitoring the variance or coefficient of variation (the ratio of standard deviation to mean) of the target parameter within a sliding time window. When a sensor ages or has poor contact, the fluctuation amplitude of the output signal often becomes abnormal. The coefficient of variation of the current window is compared with a historical baseline under the same operating conditions. If the current value exceeds the baseline value by a certain degree, an accuracy degradation alarm is triggered. This "degree" needs to be determined based on different sensors and different environments, and is usually smaller than the actual range to allow for maintenance buffer. The detection must be performed under relatively stable operating conditions to avoid misinterpreting real parameter fluctuations caused by drastic changes in operating conditions as sensor accuracy problems.

[0052] Intermittent fault detection is based on the frequency of outliers occurring within a unit of time. Under normal circumstances, outliers occur with a low probability. When the frequency of an outlier for a certain parameter is significantly higher than the historical baseline level within a short time window and continues for multiple windows, an intermittent fault alarm is triggered.

[0053] The system will respond to an alarm and further distinguish the type of interference source: if the outlier frequencies of multiple physically unrelated parameters rise synchronously, it is highly likely to point to an external electromagnetic interference source; if the outlier frequency rise is concentrated on a single parameter, it is more likely to be caused by the parameter's own reasons such as loose wiring or intermittent failure of internal components. However, in either case, specific inspection actions need to be performed according to the specific problem.

[0054] Physical consistency verification utilizes deterministic relationships in the desulfurization system that are independent of sensor calibration accuracy and based on mass conservation and stoichiometry. These relationships provide a verification path for sensor fault detection that is completely independent of the data-driven model. Based on this, in this embodiment, we define physical consistency verification as the following steps: The amount of sulfur dioxide removed from the flue gas side is calculated based on the inlet sulfur dioxide concentration, outlet sulfur dioxide concentration, and flue gas flow rate, and the deviation is measured with the amount of sulfur dioxide absorbed from the slurry side calculated based on the slurry parameters.

[0055] Obtain the deviation between the inlet flue gas flow rate and the outlet flue gas flow rate of the desulfurization system.

[0056] Obtain the temperature relationship between the inlet flue gas temperature and the outlet flue gas temperature of the desulfurization system.

[0057] A physical inconsistency index is generated based on the calculated deviation and temperature relationship.

[0058] By calculating data such as flue gas flow rate and concentration, and analyzing the deviations and relationships of the data, the authenticity of the data can be verified indirectly. This avoids false data appearing within the monitoring range due to system fluctuations or sensor malfunctions, and prevents the data from affecting the accuracy of fault diagnosis.

[0059] S30: Based on the results of anomaly detection and parallel detection, determine whether to issue an early warning in conjunction with the current operating status.

[0060] We have already provided various fault diagnosis methods for different states, parameters, and schemes. If an abnormality is confirmed at any point, it can be considered that the detection results of the monitoring equipment are abnormal and that the equipment has a fault.

[0061] Of course, we also provide solutions to limit false alarms, which include the following steps: The corresponding confirmation delay is set according to the fault type. Different fault modes have different speeds, so a uniform confirmation time cannot be used. Different confirmation times are set according to different fault modes, and each fault mode detection is designed according to the actual situation.

[0062] Alarm trigger thresholds and alarm recovery thresholds are set to form a hysteresis zone. Since the data fluctuates, we create a certain distance between the alarm trigger threshold and the recovery threshold. This distance stabilizes the alarm status, preventing repeated triggering and cancellation due to parameter fluctuations. Of course, the distance must be determined based on the specific process conditions to avoid excessive distance affecting analysis and judgment.

[0063] Under periodic or sudden operation, alarm lockout or threshold adjustment shall be performed on the corresponding fault detection results.

[0064] During system operation, periodic maintenance or sudden operations may occasionally be performed. These operations may coincide with sensor readings, causing the sensors to determine that the system has abnormal problems. These operations may be data changes under conditions such as backflushing, calibration, unit start-up and shutdown, drastic load changes, or coal quality switching. Therefore, it is necessary to set thresholds separately or disable alarms.

[0065] During the alarm lockout period, fault detection calculations continue to be performed, and after the lockout is released, it is determined whether to output fault diagnosis results based on the current fault detection results.

[0066] The alarm output is locked, not the detection algorithm itself. All detection metrics continue to update during the lockout period. This design is based on the following principle: if the anomaly is solely caused by a change in operating conditions, the detection metrics should automatically return to normal once the operating conditions return to normal, without issuing any alarms. However, if a sensor actually malfunctions during the lockout period, the anomaly in the detection metrics will not disappear as the operating conditions recover. Upon release of the lockout, the system will immediately detect the fault and trigger an alarm without any confirmation delay. This design avoids false alarms caused by changes in operating conditions while also ensuring that genuine sensor malfunctions caused by the lockout are not missed.

[0067] The above embodiments are merely illustrative of the technical solutions of the present invention and are not intended to limit it. Anyone skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.

Claims

1. A fault diagnosis method for an online monitoring device for desulfurized flue gas, characterized in that, Includes the following steps: The target parameters collected by the monitoring device are compared with the preset working boundary to obtain the boundary pre-classification result, and anomaly determination is made based on the boundary pre-classification result; The data of the boundary pre-classification results that pass the anomaly detection are preprocessed, and parallel detection of multiple fault modes is performed based on the preprocessed data; Based on the results of anomaly detection and parallel detection, and combined with the current operating status, a decision is made on whether to issue an early warning.

2. The fault diagnosis method for the online monitoring device for desulfurized flue gas according to claim 1, characterized in that: The working boundary is a preset working condition, which includes at least the physical absolute boundary, the working condition limit boundary, and the normal operation boundary. The physical absolute boundary is the maximum boundary range determined based on the physical possible values ​​of the target parameters. The operating condition limit boundary is the maximum fluctuation range determined based on the extreme parameters of the desulfurization system parameters, process parameters, and / or historical operating conditions. The normal operating boundary is the normal fluctuation range determined based on historical data of the target operating stably with parameters.

3. The fault diagnosis method for the online monitoring device for desulfurized flue gas according to claim 2, characterized in that: When the boundary pre-classification result is the normal operation boundary or exceeds the normal operation boundary but does not exceed the operating condition limit boundary, it is considered to pass the anomaly judgment; When the boundary pre-classification result exceeds the physical absolute boundary, it is determined that the monitoring device is in an abnormal working state and a device failure warning is issued. When the boundary pre-classification result exceeds the working condition limit boundary, the monitoring device is deemed abnormal and its status cannot be determined, and a warning prompt for manual intervention is issued.

4. The fault diagnosis method for the online monitoring device for desulfurized flue gas according to claim 1, characterized in that: The preprocessing includes at least one of missing data processing, outlier processing, and filtering processing.

5. The fault diagnosis method for the online monitoring device for desulfurized flue gas according to claim 4, characterized in that: The missing data processing uses a nearest neighbor method to handle occasional missing data, and issues a data interruption warning for continuous missing data; the outlier processing includes identifying outliers based on the absolute deviation of the median within a sliding window, and replacing the identified outliers with the median in the fault detection input while retaining the original values ​​and outlier markers; the filtering process uses a first-order hysteresis filter to filter the data.

6. The fault diagnosis method for the online monitoring device for desulfurized flue gas according to claim 1, characterized in that: The parallel detection of the multiple fault modes includes at least two of the following: freeze detection, drift and bias detection, physical consistency detection, noise and accuracy degradation detection, and intermittent fault detection.

7. The fault diagnosis method for the online monitoring device for desulfurized flue gas according to claim 6, characterized in that: The freeze detection is based on the fluctuation of the target parameter within the sliding time window and the fluctuation state of the associated parameters. The drift and bias detection is based on the residual between the measured value of the target parameter and the soft measurement estimate. The accuracy degradation detection is based on the variance or coefficient of variation within the sliding time window. The intermittent fault detection is based on the frequency of outlier occurrence.

8. The fault diagnosis method for the online monitoring device for desulfurized flue gas according to claim 7, characterized in that: The drift and bias detection includes: Perform a sequential probability ratio test on the residuals and obtain at least one of the following; When the mean residual changes abruptly and meets the preset detection conditions, it is determined to be a bias fault; When the residual shows a continuous unidirectional shift and meets the preset detection conditions, it is determined to be a drift fault.

9. The fault diagnosis method for the online monitoring device for desulfurized flue gas according to claim 6, characterized in that: The physical consistency verification includes at least one of the following: The amount of sulfur dioxide removed from the flue gas side is calculated based on the inlet sulfur dioxide concentration, outlet sulfur dioxide concentration and flue gas flow rate, and the deviation is calculated with the amount of sulfur dioxide absorbed from the slurry side based on the slurry parameters. Obtain the deviation between the inlet flue gas flow rate and the outlet flue gas flow rate of the desulfurization system; Obtain the temperature relationship between the inlet flue gas temperature and the outlet flue gas temperature of the desulfurization system; A physical inconsistency index is generated based on the calculated deviation and temperature relationship.

10. The fault diagnosis method for the online monitoring device for desulfurized flue gas according to any one of claims 1-9, characterized in that: A method for suppressing false alarms is also provided, comprising the following steps: Set the corresponding confirmation delay according to the fault type; Set alarm trigger thresholds and alarm recovery thresholds to form a hysteresis zone; Under periodic or sudden operation, alarm lockout or threshold adjustment shall be performed on the corresponding fault detection results; During the alarm lockout period, fault detection calculations continue to be performed, and after the lockout is released, it is determined whether to output fault diagnosis results based on the current fault detection results.