Real-time monitoring and fault prediction methods for industrial waste gas treatment systems

Abstract

This invention discloses a real-time monitoring and fault prediction method for industrial waste gas treatment systems, relating to the field of intelligent operation and maintenance technology for industrial waste gas treatment and equipment. The method includes the following steps: A high-frequency dynamic pressure sensor is installed on the inlet / outlet pipe of each regenerator chamber. Using the periodic valve switching event of the regenerator as a trigger source, the high-frequency pressure waveforms at the inlet and outlet of each regenerator chamber before and after the switching, as well as the combustion chamber temperature and waste gas load rate at that moment, are simultaneously collected. This invention, by collecting high-frequency pressure waveform data triggered by valve switching events in the regenerator and combining it with combustion chamber temperature and waste gas load rate data, effectively captures abnormal airflow disturbances caused by micro-cracks in the regenerator. By establishing an individualized dynamic baseline for each regenerator chamber, inherent deviations caused by manufacturing and installation differences are eliminated, improving the sensitivity of anomaly detection.

Description

Technical Field

[0001] This invention relates to the field of industrial waste gas treatment and intelligent operation and maintenance technology, specifically to a method for real-time monitoring and fault diagnosis based on an industrial waste gas treatment system. Background Technology

[0002] In industrial waste gas treatment systems, the ceramic regenerators inside regenerative thermal oxidizers (RTOs) or regenerative catalytic oxidation (RCOs) operate for extended periods in high-temperature alternating and corrosive waste gas environments. Due to their high heat recovery efficiency and thorough purification, they are widely used in the treatment of volatile organic compounds (VOCs) in industries such as chemical, spraying, and pharmaceutical. Multi-stage series waste gas treatment systems used in industry typically include a pretreatment dust removal unit, an acid gas absorption tower, a regenerative oxidizer, and a continuous online monitoring system for the final stage of flue gas. The regenerative oxidizer is filled with a large number of ceramic regenerators. By periodically switching the airflow direction, the regenerators alternate between heat absorption and release states, achieving efficient heat recovery. During normal operation of the RTO, the combustion chamber temperature is maintained above 800 degrees Celsius, with switching cycles typically on the order of several minutes. The system is equipped with a conventional sensor network capable of real-time collection of parameters such as pressure, temperature, and flow rate at each node.

[0003] Currently, monitoring of regenerative thermal oxidizers involves monitoring the total pressure difference at the inlet and outlet of each regenerative chamber. When the pressure difference exceeds a preset threshold, blockage or abnormality is identified. Alternatively, monitoring the combustion chamber temperature and outlet pollutant concentration triggers an alarm when there are abnormal temperature fluctuations or emissions exceed standards. Another approach is to periodically shut down the unit and manually inspect the regenerative body using endoscopy or infrared thermal imaging. All these existing monitoring schemes are based on a common implicit assumption: that structural degradation of the regenerative body will inevitably cause changes in observable macroscopic parameters, and that these changes can be captured by sensors with conventional sampling frequencies. However, practical operating experience shows that this assumption does not hold true in the early stages of the regenerative body's gradual collapse. This is because total pressure difference monitoring only reflects the overall flow resistance of the regenerative chamber, while the cross-sectional area of ​​abnormal airflow channels caused by micro-cracks is extremely small in the initial stages. Their impact on the total pressure difference is completely overwhelmed by the normal pressure fluctuations caused by fluctuating exhaust gas concentration and flow rate. Furthermore, the control response of the combustion chamber temperature has significant inertia, and the minute heat loss changes caused by local structural damage cannot be reflected in the temperature readings in a timely manner.

[0004] Traditional monitoring systems for RTOs typically collect the average pressure difference between the inlet and outlet of each regenerator at a frequency below one Hz. At the same time, they treat the regular pressure pulses generated by the periodic valve switching operations inherent in the regenerator as interference signals and filter or smooth them out. This process discards the millisecond-level dynamic information that best reflects the health status of the regenerator structure. Furthermore, existing solutions lack the ability to compare and analyze the consistency of dynamic responses between different regenerators. When fluctuations in operating conditions cause all parameters to change synchronously, the slight anomalies of a single regenerator are masked by the overall drift.

[0005] Therefore, there is an urgent need to provide a real-time monitoring and fault diagnosis method based on industrial waste gas treatment systems, which can use existing sensor networks to capture early micro-motion characteristics of the heat storage body from periodic switching operations, and issue early warnings when cracks are still in an extremely small stage, thereby avoiding sudden collapse accidents.

[0006] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to provide a real-time monitoring and fault diagnosis method for industrial waste gas treatment systems. This invention solves the problems in the background art by acquiring high-frequency pressure waveform data triggered by valve switching events, establishing individualized dynamic baselines for each heat storage chamber, calculating the micro-movement deviation by comparing the inter-chamber consistency through the geometric centroids of the three heat storage chambers, and further combining time-series trend decomposition, accumulation and monitoring.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a method for real-time monitoring and fault prediction of an industrial waste gas treatment system, comprising the following steps: S1. Install a high-frequency dynamic pressure sensor on the intake / exhaust pipe of each regenerator chamber. Take the periodic valve switching event of the regenerator as the trigger source, and synchronously collect the high-frequency pressure waveforms of the inlet and outlet of each regenerator chamber before and after the switching, as well as the combustion chamber temperature and exhaust gas load rate at that moment, to obtain the dynamic pressure data corresponding to each switching event. S2. Extract multi-dimensional feature parameters reflecting the dynamic response characteristics of the regenerator from the high-frequency pressure waveform of each switching event, and establish an individualized dynamic baseline for each regenerator based on its own feature vector statistical distribution during the initial stable operation phase of the regenerator. Calculate the degree of deviation of the current feature vector from its own individualized dynamic baseline to obtain a single-cycle anomaly marker. S3. Within the same switching cycle, the dynamic response feature vectors of the three heat storage chambers are compared, and the deviation vector and norm of each heat storage chamber relative to the centroid of the feature vectors of the three heat storage chambers are calculated as the micro-movement deviation degree to offset the common mode interference of the operating condition fluctuation and retain the difference mode reflecting the degradation of the heat storage structure. S4. Decompose the cumulative micro-displacement of each heat storage chamber over time, extract the long-term trend component and calculate the trend change rate. At the same time, use the cumulative and monitored cumulative offset to merge the trend change rate and cumulative offset into a collapse risk index. S5. When the collapse risk index continues to exceed the preset threshold, an early warning is triggered, and the remaining number of safe switching times is extrapolated based on the current trend to output a graded maintenance decision.

[0009] Furthermore, the steps for obtaining the dynamic pressure data corresponding to each switching event are as follows: Real-time monitoring of valve position feedback signals of switching valves in each heat storage chamber. When any heat storage chamber is detected to switch from intake state to exhaust state or from exhaust state to purging state, the rising or falling edge of the switching action is immediately captured as a time reference point. Based on the time reference point, a preset short time window is automatically captured forward and a preset long time window is captured backward to form the total acquisition time interval. During the total acquisition time interval, the high-frequency dynamic pressure sensor continuously acquires the instantaneous gas pressure values ​​at the inlet and outlet of each heat storage chamber at a fixed sampling frequency, thus obtaining a discrete pressure sequence that changes with time. Simultaneously, the combustion chamber temperature, exhaust gas inlet temperature, and exhaust gas load rate calculated from the normalized values ​​of exhaust gas flow rate and concentration, which characterize the treatment load, are recorded at each moment within the corresponding total collection time interval. After data collection, the total collection time interval, discrete pressure sequence, combustion chamber temperature, exhaust gas inlet temperature, and exhaust gas load rate are packaged into an event data block, which is used as the dynamic pressure data corresponding to the switching event, in combination with the type identifier of the switching event.

[0010] Furthermore, the sampling frequency range of the high-frequency pressure waveform is 100Hz-500Hz, and a preset short time window range of 0.2s-0.5s is taken forward and a preset long time window range of 1.0s-2.5s is taken backward, with the rising or falling edge of the switching valve action as the time reference point. The range of the high-frequency dynamic pressure sensor should cover more than twice the normal operating pressure of the regenerative thermal oxidation device, and its natural frequency should not be less than 500 Hz. The regenerative thermal oxidation unit comes with its own temperature sensors, including a combustion chamber thermocouple and an exhaust gas inlet resistance thermometer, as well as an exhaust gas flow meter.

[0011] Furthermore, the extraction steps for the multidimensional feature parameters are as follows: Based on the discrete pressure sequence that changes over time for each switching event, preprocessing is performed. Then, median filtering is used to remove spike noise. Five sampling points are taken from the window length of the total acquisition time interval. Then, a zero-phase bandpass filter with a passband frequency of 0.1 Hz to 50 Hz is used to retain the effective frequency band components that reflect the dynamic response, resulting in a clean waveform after preprocessing. Based on the preprocessed clean waveform, the average value of the pressure waveform in the window before switching is taken as the steady-state pressure reference value, and the maximum pressure value and its corresponding time are searched in the window after switching. Then, based on the pre-processed clean waveform, steady-state pressure reference value, maximum pressure value and its corresponding time, calculate five characteristics: pressure peak offset, maximum pressure change rate, oscillation decay time, pressure recovery curve integral and waveform symmetry skewness. The calculated pressure peak offset, maximum pressure change rate, oscillation decay time, pressure recovery curve integral, and waveform symmetry skewness are combined to form a feature vector, thus creating multidimensional feature parameters.

[0012] Furthermore, the steps for establishing the individualized dynamic baseline are as follows: During the initial stable operation phase of the regenerative oxidation unit, a multi-dimensional feature vector sequence is collected from the regenerative chamber itself, and the mean and covariance matrix of the multi-dimensional feature vector sequence are calculated to form an individualized dynamic baseline. For each subsequent new switching cycle, the deviation of the current multidimensional feature vector from its individualized dynamic baseline is measured using Mahalanobis distance.

[0013] Furthermore, the steps for obtaining the single-cycle anomaly marker are as follows: Calculate the Mahalanobis distance of the current multidimensional feature vectors based on the established individualized dynamic baseline; The Mahalanobis distance of the current multidimensional feature vector is compared with a pre-set control limit, where the pre-set control limit is based on the Mahalanobis distance distribution of the individualized dynamic baseline feature vector, taking the value corresponding to the 95th percentile of the chi-square distribution. If the Mahalanobis distance of the current multidimensional feature vector is greater than the preset control limit, it is considered that there is a dynamic response anomaly in the regenerator in the current cycle, and a single-cycle anomaly marker is generated.

[0014] Furthermore, the steps for extracting the long-term trend component and the rate of change of trend are as follows: For each heat storage chamber, its own micro-deviation time series is obtained. To extract the long-term trend component using adaptive noise complete ensemble empirical mode decomposition (ICEEMDAN), adaptive white noise is first added to the micro-deviation time series of each heat storage chamber to construct several replica sequences. Perform empirical mode decomposition on each replica sequence to extract the first intrinsic mode function (IMF). Calculate the average of all copies of the first IMF as the first IMF component; Subtract the first IMF from the original sequence to obtain the first residual; Repeat the above process for the residuals until the residuals become a monotonic function. At this point, the residuals are the extracted long-term trend components. The Theil-Sen robust slope estimation method is then used to calculate the rate of change of the long-term trend component. The overall rate of change is estimated by calculating the median slope between all point pairs in the long-term trend component sequence.

[0015] Furthermore, the calculation steps for the micro-motion deviation are as follows: Within the same switching cycle, the multi-dimensional feature vectors of the dynamic response of the three regenerators are recorded; Calculate the geometric centroids of the three eigenvectors based on the dynamic response eigenvectors of the three regenerators; Calculate the deviation vector of the dynamic response characteristic vector of each heat storage chamber relative to the geometric centroid. The Euclidean norm of the deviation vector is then taken as the micro-deviation degree of the heat storage chamber in the current switching cycle; The Theil-Sen robust slope estimation method is then used to calculate the rate of change of the long-term trend component. The overall rate of change is estimated by calculating the median slope between all point pairs in the long-term trend component sequence.

[0016] Furthermore, the steps for merging the collapse risk index are as follows: For the micro-deviation sequence, the cumulative offset is monitored using a cumulative sum control chart (CUSUM); The calculated trend change rate is fused with the current cumulative CUSUM to construct a collapse risk index. Then, a multiplicative normalization method is used to map the two to the [0,1] interval and multiply them, while limiting the upper limit to 1.

[0017] Furthermore, the step of extrapolating the remaining safe switching count is as follows: Extrapolation is made based on the assumption that the observed degradation trend will remain constant in the near future; Obtain the current collapse risk index and the collapse risk index sequence for the most recent switching cycles; To calculate the average rate of change of the collapse risk index with each switching cycle, linear regression is used to obtain the average increment of the collapse risk index for each switching cycle. Calculate the number of switching cycles required for the current collapse risk index to increase to the limit value of 1.0, and divide by the average rate of change to obtain the remaining safe switching cycles; The remaining number of switching cycles is converted into remaining safe operating time so that maintenance personnel can make reasonable arrangements for maintenance plans.

[0018] The technical effects and advantages provided by the present invention in the above technical solution are as follows: This invention acquires high-frequency pressure waveform data triggered by valve switching events in a regenerative thermal oxidation device, and combines this data with combustion chamber temperature and exhaust gas load rate data to effectively capture abnormal airflow disturbances caused by microcracks in the regenerator. By establishing individualized dynamic baselines for each regenerator, inherent deviations caused by manufacturing and installation differences are eliminated, improving the sensitivity of anomaly detection. Furthermore, by comparing the geometric centroids of the three regenerators to calculate the degree of micro-movement deviation, common-mode interference from global operating condition fluctuations such as exhaust gas concentration and temperature is automatically offset, making weak structural degradation signals stand out from strong background noise. In addition, by combining time-series trend decomposition, accumulation, and monitoring, it can distinguish between random fluctuations and genuine structural degradation, issuing early warning signals weeks in advance, transforming sudden collapse accidents into planned maintenance, and providing clear operational guidance to maintenance personnel through graded maintenance decision output, effectively reducing unplanned downtime and maintenance costs. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0020] Figure 1 This is a flowchart of the real-time monitoring and fault diagnosis method based on an industrial waste gas treatment system according to the present invention. Detailed Implementation

[0021] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0022] Example 1 This invention provides, for example Figure 1 The real-time monitoring and fault diagnosis method based on an industrial waste gas treatment system, as shown, includes the following steps: S1. Install a high-frequency dynamic pressure sensor on the intake / exhaust pipe of each regenerator chamber. Take the periodic valve switching event of the regenerator as the trigger source, and synchronously collect the high-frequency pressure waveforms of the inlet and outlet of each regenerator chamber before and after the switching, as well as the combustion chamber temperature and exhaust gas load rate at that moment, to obtain the dynamic pressure data corresponding to each switching event. Specifically, the steps for obtaining dynamic stress data for each switching event are as follows: The system monitors the valve position feedback signals of each regenerator's switching valve in real time. When any regenerator switches from intake to exhaust or from exhaust to purging, the system immediately captures the rising or falling edge of this switching action as a time reference point. Centered on this time reference point, a preset short time window is automatically extracted forward, and a preset long time window is extracted backward, forming a total acquisition time interval. The expression for the total acquisition time interval is as follows: In the formula, Represented as a time reference point, This is represented as a preset short time window. Represented as a preset long-term window; the expression for each sampling point within the total acquisition time interval is: In the formula, This represents the sampling point number; negative values ​​indicate sampling points before the time reference point, and positive values ​​indicate sampling points after the time reference point. , , This represents the number of sample points collected before the time reference point. This represents the number of sample points collected after the time reference point. This is expressed as the sampling frequency of the dynamic pressure sensor; Within the total acquisition time interval, the high-frequency dynamic pressure sensor continuously acquires the instantaneous gas pressure values ​​at the inlet and outlet of each regenerator chamber at a fixed sampling frequency, obtaining a discrete pressure sequence that varies with time. The expression for the instantaneous gas pressure value is as follows: In the formula, Represented as the first The first heat storage chamber In the 1st switching event, the 2nd Pressure values ​​at sampling points Represented as an analog-to-digital conversion function, it converts analog quantities into digital quantities. This is represented as a dynamic pressure sensor at time [time]. The output analog signal, This is expressed as the sampling frequency of the dynamic pressure sensor; Simultaneously, the combustion chamber temperature, exhaust gas inlet temperature, and exhaust gas load rate calculated from the normalized values ​​of exhaust gas flow rate and concentration, which characterize the treatment load, are recorded at each moment within the corresponding total collection time interval. After data collection, the total collection time interval, discrete pressure sequence, combustion chamber temperature, exhaust gas inlet temperature, and exhaust gas load rate are packaged into an event data block, which is used as the dynamic pressure data corresponding to the switching event, in combination with the type identifier of the switching event.

[0023] Specifically, the sampling frequency range of the high-frequency pressure waveform is 100Hz-500Hz. The rising or falling edge of the switching valve action is used as the time reference point. A preset short time window range of 0.2s-0.5s is taken forward, and a preset long time window range of 1.0s-2.5s is taken backward. The range of the high-frequency dynamic pressure sensor should cover more than twice the normal operating pressure of the regenerative thermal oxidation device, and the natural frequency should be no less than 500 Hz to ensure that it can capture millisecond-level pressure transients. The regenerative thermal oxidation unit comes with its own temperature sensors, including a combustion chamber thermocouple and an exhaust gas inlet resistance thermometer, as well as an exhaust gas flow meter.

[0024] S2. Extract multi-dimensional feature parameters reflecting the dynamic response characteristics of the regenerator from the high-frequency pressure waveform of each switching event, and establish an individualized dynamic baseline for each regenerator based on its own feature vector statistical distribution during the initial stable operation phase of the regenerator. Calculate the degree of deviation of the current feature vector from its own individualized dynamic baseline to obtain a single-cycle anomaly marker. Specifically, the steps for extracting multidimensional feature parameters are as follows: Based on the discrete pressure sequence that changes over time obtained for each switching event, preprocessing is performed. Then, median filtering is used to remove spike noise caused by electromagnetic interference or vibration. Five sampling points are taken from the window length in the total acquisition time interval. Then, a zero-phase bandpass filter with a passband frequency of 0.1 Hz to 50 Hz is used to retain the effective frequency band components that reflect the dynamic response, resulting in a clean waveform after preprocessing. Based on the preprocessed clean waveform, the average value of the pressure waveform in the window before switching is taken as the steady-state pressure reference value, and the maximum pressure value and its corresponding time are searched in the window after switching. Based on the preprocessed clean waveform, steady-state pressure reference value, maximum pressure value and its corresponding time, five characteristics are calculated: pressure peak offset, maximum pressure change rate, oscillation decay time, pressure recovery curve integral, and waveform symmetry skewness. The formula for calculating the pressure peak offset is as follows: ,and ,as well as , In the formula, This is expressed as the pressure peak offset. This is represented as the maximum pressure. Represented as the steady-state pressure reference value, Represented as the first The first heat storage chamber In each switching event, after preprocessing... Clean waveform at any moment This represents the clean waveform at the sampling interval of the five sampling points before the time reference point after preprocessing. This is represented as the sampling interval; The formula for calculating the maximum pressure change rate is: In the formula, Expressed as the maximum rate of change of pressure, This represents the time corresponding to the maximum search pressure. Represented as the first Each discrete sampling time; The formula for calculating the oscillation decay time is: ,and In the formula, This is represented as the oscillation decay time. This represents the time it takes for the waveform to first enter the range of ±5% near its steady-state value. Represented as within the time length time, Represented as the first The first heat storage chamber In each switching event, after preprocessing... A clean waveform at any given moment; The formula for calculating the integral of the pressure recovery curve is as follows: In the formula, Represented as the first The first heat storage chamber The integral of the stress recovery curve for each switching event; The formula for calculating waveform symmetry skewness is as follows: ,and , ,as well as In the formula, Represented as the first The first heat storage chamber The waveform symmetry skewness of each switching event Indicated as rise time, This is represented as the descent time; The calculated pressure peak offset, maximum pressure change rate, oscillation decay time, pressure recovery curve integral, and waveform symmetry skewness are combined to form a feature vector, creating multidimensional feature parameters. The expression for the feature vector is as follows: In the formula, Represented as the first The heat storage chamber is in the first Multidimensional feature parameter vectors in a switching event This is represented as a vector transpose.

[0025] Specifically, the multidimensional characteristic parameters include pressure peak offset, maximum pressure change rate, oscillation decay time, pressure recovery curve integral, and waveform symmetry skewness. Specifically, the steps for establishing an individualized dynamic baseline are as follows: During the initial stable operation phase of the regenerative oxidation device, multi-dimensional feature vector sequences were collected from the regenerative chambers themselves, and the mean and covariance matrix of the multi-dimensional feature vector sequences were calculated to form an individualized dynamic baseline, which was used to record the dynamic fingerprint of each regenerative chamber under healthy conditions. For each subsequent new switching cycle, the deviation of the current multidimensional feature vector from its individualized dynamic baseline is measured using Mahalanobis distance.

[0026] Specifically, the steps for obtaining single-cycle anomaly markers are as follows: Calculate the Mahalanobis distance of the current multidimensional feature vectors based on the established individualized dynamic baseline; The Mahalanobis distance of the current multidimensional feature vector is compared with a pre-set control limit, where the pre-set control limit is based on the Mahalanobis distance distribution of the individualized dynamic baseline feature vector, taking the value corresponding to the 95th percentile of the chi-square distribution. If the Mahalanobis distance of the current multidimensional feature vector is greater than the preset control limit, it is considered that there is a dynamic response anomaly in the regenerator in the current cycle, and a single-cycle anomaly marker is generated.

[0027] S3. Within the same switching cycle, the dynamic response feature vectors of the three heat storage chambers are compared, and the deviation vector and norm of each heat storage chamber relative to the centroid of the feature vectors of the three heat storage chambers are calculated as the micro-movement deviation degree to offset the common mode interference of the operating condition fluctuation and retain the difference mode reflecting the degradation of the heat storage structure. Specifically, the calculation steps for the micro-motion deviation are as follows: Within the same switching cycle, the multi-dimensional feature vectors of the dynamic responses of the three regenerators are recorded and calibrated as follows: , , ; The geometric centroids of the three eigenvectors are calculated based on the dynamic response eigenvectors of the three regenerators. The formula for calculating the geometric centroids is as follows: In the formula, Represented as the first The geometric centroid of the three regenerator eigenvectors within a switching cycle is a vector of the same dimension as the multidimensional eigenvector. Calculate the deviation vector of the dynamic response characteristic vector of each regenerator relative to its geometric centroid, where the formula for calculating the deviation vector is: ,and In the formula, Represented as the first The heat storage chamber is in the first The deviation vector of the multi-dimensional feature vector relative to the geometric centroid in a switching event. Represented as the first The heat storage chamber is in the first Multidimensional feature parameter vectors in a switching event; The Euclidean norm of the deviation vector is then taken as the micro-deviation degree of the heat storage chamber in the current switching cycle, where the formula for calculating the micro-deviation degree is: In the formula, Represented as the first The heat storage chamber is in the first The micro-deviation degree per switching cycle is a non-negative scalar; the larger the value, the greater the degree to which the dynamic response characteristics of this regenerator deviate from the other two regenerators. Represented as deviation vector The Euclidean norm, Represented as the component index in the multidimensional feature vector. Represented as deviation vector The The component 5 represents a multidimensional feature vector consisting of the corresponding pressure peak offset, maximum pressure change rate, oscillation decay time, pressure recovery curve integral, and waveform symmetry skewness.

[0028] S4. Decompose the cumulative micro-displacement of each heat storage chamber over time, extract the long-term trend component and calculate the trend change rate. At the same time, use the cumulative and monitored cumulative offset to merge the trend change rate and cumulative offset into a collapse risk index. Specifically, the calculation steps for extracting the long-term trend component and the rate of change of trend are as follows: For each heat storage chamber, its own micro-deviation time series is obtained. To extract the long-term trend component using adaptive noise complete ensemble empirical mode decomposition (ICEEMDAN), adaptive white noise is first added to the micro-deviation time series of each heat storage chamber to construct several replica sequences. Perform empirical mode decomposition on each replica sequence to extract the first intrinsic mode function (IMF). Calculate the average of all copies of the first IMF as the first IMF component; Subtract the first IMF from the original sequence to obtain the first residual; Repeat the above process for the residuals until they become a monotonic function. At this point, the residuals represent the extracted long-term trend component. This long-term trend component represents the irreversible component of the micro-deviation that monotonically changes with the number of periods, increasing / decreasing. Other high-frequency IMF components correspond to random noise or periodic load fluctuations. The expression for the ICEEMDAN decomposition is: ,and ,as well as ,and In the formula, Represented as the first component of the high-frequency IMF. This is expressed as the number of integrations, typically 100. This is represented as an operator that generates a first-order IMF. Represented as the first The total number of cycles observed so far for each regenerator. The sequence of micro-movement deviations in a switching event. This is represented as the noise amplitude coefficient for stage 0. Represented as the first The white noise sequence added next time. This is represented as the residual after the first decomposition. Represented as the second component of the high-frequency IMF. This is represented as the noise amplitude coefficient for stage 1. Represented as the residual after the second decomposition; The final expression for the extracted long-term trend component is: In the formula, This is represented by the final extracted long-term trend component. This is expressed as until the residual becomes a monotonic function; The Theil-Sen robust slope estimation method is then used to calculate the rate of change of the long-term trend component. This method estimates the overall rate of change by calculating the median slope between all point pairs in the long-term trend component sequence, and is insensitive to outliers. The expression for the Theil-Sen slope estimation is as follows: ,and , ,as well as In the formula, Represented as the slope between two points. Represented as any symbol, Represented as the first The heat storage chamber is in the first The long-term trend component in each switching event. Represented as the first The heat storage chamber is in the first The long-term trend component in each switching event. Represented as the first The rate of change of the long-term trend component of the micro-displacement deviation of each heat storage chamber This is represented as the median function.

[0029] Specifically, the steps for integrating the collapse risk index are as follows: For the micro-deviation sequence, the cumulative offset is monitored using a Cumulative Sum Control Chart (CUSUM), where the expression for monitoring the cumulative offset using CUSUM is: ,and ,as well as In the formula, This is represented as the cumulative offset on the upper side after the 0th switching cycle. Represented as the first Cumulative offset on the upper side after each switching cycle Represented as the first Cumulative offset on the upper side after each switching cycle It represents the mean of the individual's dynamic baseline deviation. This is represented as the allowed offset. Represented as the first The heat storage chamber is in the first Micro-movement deviation per switching cycle; The calculated trend change rate is fused with the current cumulative CUSUM to construct a collapse risk index. The fusion principle is that the larger the trend change rate and the larger the cumulative offset, the higher the risk index. A multiplicative normalization method is used, mapping both to the [0,1] interval before multiplication, with an upper limit of 1. The fusion expression for the collapse risk index is as follows: ,and In the formula, Represented as the first The heat storage chamber is in the first The collapse risk index for each switching cycle, with a value range of [0,1]. Represented as CUSUM control limit, when When the cumulative offset is determined to be significant, Expressed as the rate of change over a long-term trend. This is expressed as the trend change rate threshold, taken as one-tenth of the standard deviation of the individual's dynamic baseline period micro-movement. This indicates that the risk index must not exceed 1.

[0030] S5. When the collapse risk index continues to exceed the preset threshold, an early warning is triggered, and the remaining number of safe switching times is extrapolated based on the current trend to output a graded maintenance decision.

[0031] Specifically, the steps for extrapolating the remaining safe switching times based on the current trend are as follows: Extrapolation is made based on the assumption that the observed degradation trend will remain constant in the near future; Obtain the current collapse risk index and the collapse risk index sequence for the most recent switching cycles, where the current collapse risk index is... The collapse risk index for each switching cycle is The expression for the collapse risk index sequence of the most recent switching cycles is as follows: In the formula, Represented as the first Collapse risk index for each switching cycle Represented as the first Collapse risk index for each switching cycle This is expressed as the window length for calculating the rate of change over the most recent switching cycle; To calculate the average rate of change of the collapse risk index over the switching cycle, linear regression is used to obtain the average increment of the collapse risk index for each switching cycle. The formula for calculating the average rate of change is as follows: In the formula, This is expressed as the average rate of change of the collapse risk index per switching cycle; The number of switching cycles required for the current collapse risk index to increase to the limit value of 1.0 is calculated, and then divided by the average rate of change to obtain the remaining safe switching cycles. The formula for calculating the remaining safe switching cycles is as follows: In the formula, This represents the number of remaining safe switching cycles required to reach a collapse risk index of 1.0 from the current cycle. Represent it as a very small positive number; The remaining number of switching cycles is converted into remaining safe operating time so that maintenance personnel can make reasonable arrangements for maintenance plans.

[0032] Specifically, the output steps of the hierarchical maintenance decision are as follows: When the collapse risk index is in the first threshold range If a yellow warning signal is issued, it is recommended to schedule an endoscopy during the next planned shutdown; there is no need to shut down the machine immediately. When the collapse risk index is in the second threshold range If an abnormality is confirmed through variable operating condition testing, an orange warning signal will be issued, and it is recommended to arrange maintenance within seven days and reduce the load to 80% of the rated value to delay crack propagation. When the collapse risk index is in the third threshold range ≥ If an abnormality is confirmed through variable operating condition testing and the vibration cross-verification shows significant abnormality, a red warning signal will be output, suggesting an emergency shutdown and inspection instruction, and a maintenance work order containing recommended maintenance windows and a list of required spare parts will be automatically generated. If the collapse risk index does not reach any threshold range, i.e., <0.7, there will be no early warning output, and monitoring can continue.

[0033] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0034] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0035] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0036] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0037] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for real-time monitoring and fault prediction of industrial waste gas treatment systems, characterized in that, Includes the following steps: S1. Install a high-frequency dynamic pressure sensor on the intake / exhaust pipe of each regenerator chamber. Take the periodic valve switching event of the regenerator as the trigger source, and synchronously collect the high-frequency pressure waveforms of the inlet and outlet of each regenerator chamber before and after the switching, as well as the combustion chamber temperature and exhaust gas load rate at that moment, to obtain the dynamic pressure data corresponding to each switching event. S2. Extract multi-dimensional feature parameters reflecting the dynamic response characteristics of the regenerator from the high-frequency pressure waveform of each switching event, and establish an individualized dynamic baseline for each regenerator based on its own feature vector statistical distribution during the initial stable operation phase of the regenerator. Calculate the degree of deviation of the current feature vector from its own individualized dynamic baseline to obtain a single-cycle anomaly marker. S3. Within the same switching cycle, the dynamic response feature vectors of the three heat storage chambers are compared, and the deviation vector and norm of each heat storage chamber relative to the centroid of the feature vectors of the three heat storage chambers are calculated as the micro-movement deviation degree to offset the common mode interference of the operating condition fluctuation and retain the difference mode reflecting the degradation of the heat storage structure. S4. Decompose the cumulative micro-displacement of each heat storage chamber over time, extract the long-term trend component and calculate the trend change rate. At the same time, use the cumulative and monitored cumulative offset to merge the trend change rate and cumulative offset into a collapse risk index. S5. When the collapse risk index continues to exceed the preset threshold, an early warning is triggered, and the remaining number of safe switching times is extrapolated based on the current trend to output a graded maintenance decision.

2. The real-time monitoring and fault diagnosis method based on an industrial waste gas treatment system according to claim 1, characterized in that, The steps for obtaining the dynamic pressure data corresponding to each switching event are as follows: The valve position feedback signal of each heat storage chamber switching valve is monitored in real time. When any heat storage chamber is detected to switch from the air intake state to the exhaust state or from the exhaust state to the purging state, the rising edge or falling edge of the switching action is immediately captured as the time reference point. With the time reference point as the center, a preset short time window is automatically intercepted forward and a preset long time window is intercepted backward to form the total acquisition time interval. During the total acquisition time interval, the high-frequency dynamic pressure sensor continuously acquires the instantaneous gas pressure values ​​at the inlet and outlet of each heat storage chamber at a fixed sampling frequency, thus obtaining a discrete pressure sequence that changes with time. Simultaneously, the combustion chamber temperature, exhaust gas inlet temperature, and exhaust gas load rate calculated from the normalized values ​​of exhaust gas flow rate and concentration, which characterize the treatment load, are recorded at each moment within the corresponding total collection time interval. After data collection, the total collection time interval, discrete pressure sequence, combustion chamber temperature, exhaust gas inlet temperature, and exhaust gas load rate are packaged into an event data block, which is used as the dynamic pressure data corresponding to the switching event, in combination with the type identifier of the switching event.

3. The real-time monitoring and fault diagnosis method based on an industrial waste gas treatment system according to claim 2, characterized in that, The sampling frequency range of the high-frequency pressure waveform is 100Hz-500Hz. Taking the rising or falling edge of the switching valve action as the time reference point, a preset short time window range of 0.2s-0.5s is extracted forward, and a preset long time window range of 1.0s-2.5s is extracted backward. The range of the high-frequency dynamic pressure sensor should cover more than twice the normal operating pressure of the regenerative thermal oxidation device, and its natural frequency should not be less than 500 Hz. The regenerative thermal oxidation unit comes with its own temperature sensors, including a combustion chamber thermocouple and an exhaust gas inlet resistance thermometer, as well as an exhaust gas flow meter.

4. The real-time monitoring and fault diagnosis method based on an industrial waste gas treatment system according to claim 1, characterized in that, The steps for extracting the multidimensional feature parameters are as follows: Based on the discrete pressure sequence that changes over time for each switching event, preprocessing is performed. Then, median filtering is used to remove spike noise. Five sampling points are taken from the window length of the total acquisition time interval. Then, a zero-phase bandpass filter with a passband frequency of 0.1 Hz to 50 Hz is used to retain the effective frequency band components that reflect the dynamic response, resulting in a clean waveform after preprocessing. Based on the preprocessed clean waveform, the average value of the pressure waveform in the window before switching is taken as the steady-state pressure reference value, and the maximum pressure value and its corresponding time are searched in the window after switching. Then, based on the pre-processed clean waveform, steady-state pressure reference value, maximum pressure value and its corresponding time, calculate five characteristics: pressure peak offset, maximum pressure change rate, oscillation decay time, pressure recovery curve integral and waveform symmetry skewness. The calculated pressure peak offset, maximum pressure change rate, oscillation decay time, pressure recovery curve integral, and waveform symmetry skewness are combined to form a feature vector, thus creating multidimensional feature parameters.

5. The real-time monitoring and fault diagnosis method based on an industrial waste gas treatment system according to claim 1, characterized in that, The steps for establishing the individualized dynamic baseline are as follows: During the initial stable operation phase of the regenerative oxidation unit, a multi-dimensional feature vector sequence is collected from the regenerative chamber itself, and the mean and covariance matrix of the multi-dimensional feature vector sequence are calculated to form an individualized dynamic baseline. For each subsequent new switching cycle, the deviation of the current multidimensional feature vector from its individualized dynamic baseline is measured using Mahalanobis distance.

6. The real-time monitoring and fault diagnosis method based on an industrial waste gas treatment system according to claim 1, characterized in that, The steps for obtaining the single-cycle anomaly marker are as follows: Calculate the Mahalanobis distance of the current multidimensional feature vectors based on the established individualized dynamic baseline; The Mahalanobis distance of the current multidimensional feature vector is compared with a pre-set control limit, where the pre-set control limit is based on the Mahalanobis distance distribution of the individualized dynamic baseline feature vector, taking the value corresponding to the 95th percentile of the chi-square distribution. If the Mahalanobis distance of the current multidimensional feature vector is greater than the preset control limit, it is considered that there is a dynamic response anomaly in the regenerator in the current cycle, and a single-cycle anomaly marker is generated.

7. The real-time monitoring and fault diagnosis method based on an industrial waste gas treatment system according to claim 1, characterized in that, The steps for extracting the long-term trend component and the rate of change of trend are as follows: For each heat storage chamber, its own micro-deviation time series is obtained. To extract the long-term trend component using adaptive noise complete ensemble empirical mode decomposition (ICEEMDAN), adaptive white noise is first added to the micro-deviation time series of each heat storage chamber to construct several replica sequences. Perform empirical mode decomposition on each replica sequence to extract the first intrinsic mode function (IMF). Calculate the average of all copies of the first IMF as the first IMF component; Subtract the first IMF from the original sequence to obtain the first residual; Repeat the above process for the residuals until the residuals become a monotonic function. At this point, the residuals are the extracted long-term trend components. The Theil-Sen robust slope estimation method is then used to calculate the rate of change of the long-term trend component. The overall rate of change is estimated by calculating the median slope between all point pairs in the long-term trend component sequence.

8. The real-time monitoring and fault diagnosis method based on an industrial waste gas treatment system according to claim 1, characterized in that, The calculation steps for the micro-motion deviation are as follows: Within the same switching cycle, the multi-dimensional feature vectors of the dynamic response of the three regenerators are recorded; Calculate the geometric centroids of the three eigenvectors based on the dynamic response eigenvectors of the three regenerators; Calculate the deviation vector of the dynamic response characteristic vector of each heat storage chamber relative to the geometric centroid. The Euclidean norm of the deviation vector is then taken as the micro-deviation degree of the heat storage chamber in the current switching cycle; The Theil-Sen robust slope estimation method is then used to calculate the rate of change of the long-term trend component. The overall rate of change is estimated by calculating the median slope between all point pairs in the long-term trend component sequence.

9. The real-time monitoring and fault diagnosis method based on an industrial waste gas treatment system according to claim 1, characterized in that, The steps for merging the collapse risk index are as follows: For the micro-deviation sequence, the cumulative offset is monitored using a cumulative sum control chart (CUSUM); The calculated trend change rate is fused with the current cumulative CUSUM to construct a collapse risk index. Then, a multiplicative normalization method is used to map the two to the [0,1] interval and multiply them, while limiting the upper limit to 1.

10. The real-time monitoring and fault diagnosis method based on an industrial waste gas treatment system according to claim 1, characterized in that, The steps for extrapolating the remaining safe switching counts are as follows: Extrapolation is made based on the assumption that the observed degradation trend will remain constant in the near future; Obtain the current collapse risk index and the collapse risk index sequence for the most recent switching cycles; To calculate the average rate of change of the collapse risk index with each switching cycle, linear regression is used to obtain the average increment of the collapse risk index for each switching cycle. Calculate the number of switching cycles required for the current collapse risk index to increase to the limit value of 1.0, and divide by the average rate of change to obtain the remaining safe switching cycles; The remaining number of switching cycles is converted into remaining safe operating time so that maintenance personnel can make reasonable arrangements for maintenance plans.

Images