An intelligent monitoring method for argon recovery system

Abstract

The application discloses an intelligent monitoring method for an argon recovery system, and particularly relates to the technical field of gas purification adsorbent state monitoring, and is used for solving the problem of adsorbent hidden failure early warning lag caused by the dependence of the existing argon recovery system on tail gas average concentration monitoring; through real-time collection of regeneration tail gas concentration time series data, the slope change rate of the desorption rate rising section and the area integral value of the desorption peak decay section are extracted; the area integral value sequence is constructed and the symmetric deviation degree is calculated; the pressure pulse disturbance period is identified to obtain effective period data; the slope change rate offset and the fluctuation abnormality degree of the symmetric deviation degree are calculated based on the effective data; when both of them exceed the threshold value at the same time and the condition of three continuous regeneration cycles is met, an adsorbent replacement early warning signal is generated, and early and accurate early warning of adsorbent performance decay is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of gas purification adsorbent state monitoring, and more particularly to an intelligent monitoring method for an argon recovery system. Background Art

[0002] In metal powder production systems, argon recovery systems circulate and purify impure gases (such as oxygen, nitrogen, and water vapor) through adsorption towers to maintain argon purity and recovery efficiency. Existing technologies use average concentration monitoring or peak threshold alarms in regenerated exhaust gas to determine adsorbent status. This method relies on macroscopic statistical values ​​of gas composition and cannot detect progressive deterioration of the adsorbent's microstructure (such as molecular sieve pore blockage and activated alumina crack growth), resulting in delayed maintenance decisions.

[0003] Existing monitoring methods have defects in identifying the degradation of adsorbent performance: the dynamic time series characteristics in the regenerated exhaust gas that are strongly correlated with the physical state of the adsorbent (such as desorption rate fluctuations and concentration waveform tailing effects) are not effectively extracted and utilized. This causes the system to trigger an alarm only when the argon purity drops significantly, and it is unable to provide early warning of hidden failure of the adsorbent, resulting in sudden shutdown and waste of argon. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides an intelligent monitoring method for an argon recovery system to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] An intelligent monitoring method for an argon recovery system comprises the following steps:

[0007] S1. During the regeneration stage of the adsorption tower, real-time collection of concentration time series data of the regenerated tail gas;

[0008] S2. Extracting the slope change rate of the desorption rate rising segment and the area integral value of the desorption peak decay segment from the concentration time series data;

[0009] S3, continuously extracting the area integral values ​​of the current regeneration cycle and a preset number of previous regeneration cycles to construct an area integral value sequence;

[0010] S4. Calculate the symmetric deviation between the area integral value of the current regeneration cycle and the mean of the area integral value sequence;

[0011] S5. Identify the pressure pulse disturbance period in the regeneration phase, remove the slope change rate and area integral value in the pressure pulse disturbance period, and obtain effective period data;

[0012] S6. Calculate the slope change rate offset and the fluctuation anomaly of the symmetric deviation based on the valid period data;

[0013] S7. When the slope change rate offset exceeds a first threshold and the fluctuation abnormality exceeds a second threshold, an adsorbent replacement warning signal is generated.

[0014] Furthermore, during the regeneration phase of the adsorption tower, the concentration time series data of the regenerated tail gas is collected in real time, including:

[0015] A mid-infrared laser gas analyzer is used to obtain the original signal of the regenerated exhaust gas concentration at a sampling frequency of not less than once per second;

[0016] Synchronously receive the metal dust concentration detection results and pressure sensor data of the regeneration pipeline;

[0017] Perform sliding mean filtering on the original signal of regenerated exhaust gas concentration to eliminate instantaneous noise interference;

[0018] The filtered concentration signal, metal dust concentration detection results and pressure sensor data are associated with timestamps and stored as concentration time series data.

[0019] Furthermore, the slope change rate of the desorption rate rising section and the area integral value of the desorption peak decay section are extracted from the concentration time series data, including:

[0020] Identify the starting and ending points of the desorption rate increase segment in the concentration time series data;

[0021] The desorption rate rising section is divided into multiple sub-intervals at fixed time intervals, and the linear regression slope of the concentration change in each sub-interval is calculated. The standard deviation of the slopes of all sub-intervals is used as the slope change rate;

[0022] Determine the start and end time of the desorption peak decay segment;

[0023] The concentration values ​​and time intervals of all adjacent sampling points in the desorption peak attenuation section are accumulated in a trapezoidal manner, and the accumulated results are used as the area integral value.

[0024] Furthermore, the starting point is the moment when the concentration value exceeds the average concentration value for the first time, and the end point is the moment when the concentration value reaches the peak concentration for the first time; the starting time is the moment corresponding to the peak concentration, and the ending time is the moment when the concentration value falls back to the average concentration value.

[0025] Furthermore, the area integral values ​​of the current regeneration cycle and the previous preset number of regeneration cycles are continuously extracted to construct an area integral value sequence, including:

[0026] Get the end time of the current regeneration cycle and the historical end time of the previous preset number of regeneration cycles;

[0027] Retrieving the area integral value corresponding to the end time of each regeneration cycle from the stored concentration time series database;

[0028] Verify the integrity of the area integral value, and arrange the qualified area integral values ​​in chronological order of the end time of the regeneration cycle to form an area integral value sequence.

[0029] Furthermore, the integrity of the area integral value is verified as follows: if the area integral value of the current regeneration cycle does not contain pressure pulse disturbance period data, and the area integral values ​​of the historical regeneration cycles are continuous and complete, then it is determined to be qualified data.

[0030] Furthermore, the symmetric deviation between the area integral value of the current regeneration cycle and the mean value of the area integral value sequence is calculated, including:

[0031] Calculate the arithmetic mean of the area integral value sequence as the mean of the area integral value sequence;

[0032] Determine the absolute difference between the area integral value of the current regeneration cycle and the mean of the area integral value sequence;

[0033] Obtain the standard deviation of the area integral value sequence as the dispersion benchmark;

[0034] Divide the absolute difference by the dispersion benchmark to obtain the preliminary deviation;

[0035] The time decay factor of the cumulative running time of the adsorbent was introduced to correct the initial deviation and obtain a symmetrical deviation.

[0036] Furthermore, the pressure pulse disturbance period in the regeneration phase is identified, and the slope change rate and area integral value in the pressure pulse disturbance period are eliminated to obtain the effective period data, including:

[0037] Obtaining the time series of pressure sensor data during the regeneration phase;

[0038] Calculate the pressure change rate time series of pressure sensor data;

[0039] identifying consecutive periods in the pressure change rate time series where the pressure change rate threshold is exceeded and the pressure sensor data exceeds the pressure absolute value threshold;

[0040] When the duration of a continuous period is greater than the minimum duration threshold, it is marked as a pressure pulse disturbance period;

[0041] In the time interval of the desorption rate rising section and the desorption peak decay section, the slope change rate and area integral value of the period overlapping with the pressure pulse disturbance period are eliminated;

[0042] The uneliminated slope change rate and the uneliminated area integral value are combined into the effective period data.

[0043] Furthermore, the slope change rate offset and the fluctuation anomaly of the symmetric deviation are calculated based on the valid period data, including:

[0044] Extract the slope change rate of the current regeneration cycle from the valid period data;

[0045] Obtain the moving average of the slope change rate under historical normal operating conditions as a reference value for historical normal operating conditions;

[0046] Calculate the absolute difference between the slope change rate of the current regeneration cycle and the historical normal operating condition reference value, and divide it by the historical normal operating condition reference value to obtain the slope change rate offset;

[0047] Continuously extracting the symmetry deviation of the current regeneration cycle and the previous preset number of regeneration cycles;

[0048] Calculate the standard deviation of the symmetric deviation sequence for a preset number of consecutive regeneration cycles;

[0049] The standard deviation is divided by the arithmetic mean of the symmetric deviation degree sequence of a preset number of consecutive regeneration cycles to obtain the fluctuation abnormality degree of the symmetric deviation degree.

[0050] Furthermore, when the slope change rate offset exceeds a first threshold and the fluctuation abnormality exceeds a second threshold, an adsorbent replacement warning signal is generated, including:

[0051] Get the current values ​​of the slope change rate offset and fluctuation anomaly;

[0052] Reading a preset first threshold and a second threshold;

[0053] Determine whether the slope change rate offset is greater than a first threshold, and simultaneously determine whether the fluctuation abnormality is greater than a second threshold;

[0054] When the slope change rate offset is greater than the first threshold and the fluctuation abnormality is greater than the second threshold, the warning judgment condition is triggered;

[0055] Repeated detection of warning determination conditions within three consecutive regeneration cycles;

[0056] If the warning judgment conditions are met for three consecutive regeneration cycles, an adsorbent replacement warning signal is generated;

[0057] The adsorbent replacement warning signal is associated with the end time of the current regeneration cycle and stored.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] First, by deeply analyzing the dynamic time series characteristics of the regenerated exhaust, early detection of adsorbent microstructural degradation is achieved. Extracting the slope change rate of the desorption rate rise segment can sensitively reflect desorption kinetic anomalies caused by molecular sieve micropore blockage. Calculating the area integral of the desorption peak decay segment can quantify the concentration decay hysteresis effect caused by crack propagation in the activated alumina. These two microscopic characteristics synergistically construct a two-dimensional assessment system for the physical state of the adsorbent, breaking through the perception bottleneck of traditional macroscopic statistical monitoring. Characteristic sequence analysis based on continuous regeneration cycles further reveals the gradual change trajectory of the adsorbent performance, enabling the system to identify hidden failure risks before the argon purity has significantly decreased.

[0060] Second, the elimination of pressure pulse disturbances ensures the purity of characteristic data and avoids misjudgment caused by process fluctuations; the slope change rate offset reveals the degree of deviation of the current adsorption kinetics from the historical benchmark, and the fluctuation anomaly of the symmetry deviation characterizes the consistency degradation trend of the regeneration process; the combined trigger conditions of the two combined with continuous cycle verification can not only avoid single abnormal false alarms, but also capture the cumulative effect of progressive failures; the adsorbent replacement warning can be advanced to the critical performance attenuation period, effectively preventing sudden shutdowns, and optimizing the balance between argon recovery efficiency and adsorbent service life. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 The present invention is a flow chart of an intelligent monitoring method for an argon recovery system. DETAILED DESCRIPTION

[0062] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0063] Example: Figure 1 The present invention provides an intelligent monitoring method for an argon recovery system, which includes the following steps:

[0064] S1. During the regeneration stage of the adsorption tower, real-time collection of concentration time series data of the regenerated tail gas;

[0065] S2. Extracting the slope change rate of the desorption rate rising segment and the area integral value of the desorption peak decay segment from the concentration time series data;

[0066] S3, continuously extracting the area integral values ​​of the current regeneration cycle and a preset number of previous regeneration cycles to construct an area integral value sequence;

[0067] S4. Calculate the symmetric deviation between the area integral value of the current regeneration cycle and the mean of the area integral value sequence;

[0068] S5. Identify the pressure pulse disturbance period in the regeneration phase, remove the slope change rate and area integral value in the pressure pulse disturbance period, and obtain effective period data;

[0069] S6. Calculate the slope change rate offset and the fluctuation anomaly of the symmetric deviation based on the valid period data;

[0070] S7. When the slope change rate offset exceeds a first threshold and the fluctuation abnormality exceeds a second threshold, an adsorbent replacement warning signal is generated.

[0071] When performing concentration time series data collection during the regeneration phase of the adsorption tower, a mid-infrared laser gas analyzer is used to continuously acquire the original signal of the regenerated exhaust gas concentration at a sampling frequency of twice per second. The mid-infrared laser gas analyzer transmits a laser beam of a specific wavelength through the regeneration pipe to detect the absorption intensity of the laser by impurity gas molecules in the argon gas. The laser wavelength is set to 4.26 microns to match the characteristic absorption peak of carbon dioxide, and is equipped with an automatic gain control circuit to ensure stable light intensity reception in a metal dust environment. During each sampling, the photodetector inside the analyzer converts the optical signal into a voltage signal, which is then converted into a digital concentration original signal by a 16-bit analog-to-digital converter. This process meets the performance standard requirements of the IEC 61207 gas analyzer.

[0072] Two auxiliary parameters from the regeneration duct are received synchronously via the Modbus communication protocol: metal dust concentration is measured by a beta-ray dust monitor installed on the duct sidewall. This monitor calculates dust concentration by detecting the attenuation of beta-rays as they pass through the airflow; and pressure sensor data is collected by a piezoresistive sensor with a measurement range of 0 to 500 kPa and an accuracy level of 0.5. The dust monitor outputs its results every 200 milliseconds, and the pressure sensor refreshes its data every 100 milliseconds. Both timestamps are synchronized to the central processor's system clock.

[0073] When performing sliding mean filtering on the raw signal of regenerated exhaust gas concentration, the sliding window width is set to 10 sampling points. The specific processing process is as follows: the concentration values ​​of the current moment and the previous nine consecutive sampling points are taken to form a data window. The arithmetic mean of all concentration values ​​within this window is calculated, and this mean is used as the filtered concentration value at the current moment. Each time a new sampling point arrives, the oldest data point in the window is removed, the new point is added, and the arithmetic mean is recalculated. This continuous filtering process effectively suppresses transient impulse noise caused by metal powder splashing. For example, when a single sampling value deviates abnormally from the adjacent value by more than 30%, the deviation can be reduced to less than 5% after filtering.

[0074] The processed concentration signal, metal dust concentration test results, and pressure sensor data are linked and stored with millisecond-level time accuracy. This storage utilizes the hierarchical data structure of a time series database: the absolute timestamp is the primary key, and each record contains three fields: filtered concentration value (floating point, unit: ppm), dust concentration value (floating point, unit: mg / m³), and pressure value (floating point, unit: kPa). Timestamp synchronization accuracy is controlled within ±10 milliseconds, ensuring strict time alignment among the three data types. This concentration time series data is written to solid-state storage in real time, retaining the last 30 days of complete data for subsequent analysis.

[0075] Metal dust concentration measurement results are obtained as follows: The radioactive source built into the beta-ray dust monitor emits an electron beam. A Geiger counter on the opposite side of the pipe measures the beam's intensity decay and calculates the dust mass concentration based on the Beer-Lambert law. If the dust concentration exceeds 50 mg / m³, a compressed air backflush cleaning process is automatically triggered to prevent dust accumulation from affecting the laser analyzer's accuracy. Pressure sensor data is collected using a temperature compensation algorithm. A built-in thermistor monitors the sensor's temperature in real time. If the temperature changes by more than 5°C, the pressure reading is automatically corrected by 0.05% / °C to eliminate thermal drift errors.

[0076] The sliding mean filter's window width is set based on analyzing the noise spectrum characteristics of historical data, determining that the primary interference frequency is in the 5 Hz to 20 Hz range. Based on Shannon's sampling theorem, a 10-point window width (corresponding to a 0.5-second window width) effectively filters out noise in this frequency range. During system initialization, a noise baseline test is automatically performed: a 30-second background signal is collected without regeneration gas flow, the standard deviation of the background noise is calculated, and the window width is dynamically adjusted to cover at least three times the standard deviation.

[0077] Concentration time series data is stored using a ring buffer mechanism, allocating 200 megabytes of memory space. When the data volume reaches the storage limit, the oldest data block is automatically overwritten, and data older than 24 hours is compressed and transferred to the hard disk. Network time protocol synchronization is used for data association. The main controller synchronizes with the timing server every 10 minutes to ensure that the timestamp error between multiple devices is less than 1 millisecond. A timeout retransmission mechanism is set for the reception of dust concentration and pressure data: if no new data is received within 500 milliseconds, a retransmission request is automatically sent to the sensor, and the data for that period is marked as pending in the database.

[0078] The calibration process for the mid-infrared laser gas analyzer includes daily automated zero and span calibrations. Zero calibration involves introducing high-purity nitrogen into the optical path to reset the measured value to zero. Span calibration involves introducing a 1000 ppm concentration of carbon dioxide standard gas, and adjusting the gain factor to ensure the reading error is within ±2% of the reference value. Calibration data is stored separately in a calibration log. When light intensity attenuation exceeds 30% of the initial value, an optical window cleaning reminder is generated. The dust monitor's beta-ray source intensity is automatically checked monthly, triggering a warning to replace the radiation source when the source intensity drops below 80% of the initial value.

[0079] The pressure sensor is installed in the middle straight section of the regeneration pipeline, at least five pipe diameters away from the elbow to ensure a stable flow field. The sensor's pressure tapping hole has a diameter of 2 mm, with rounded edges to reduce eddy current interference. The axis of the pressure tapping hole is perpendicular to the airflow direction. Pressure is transmitted to the sensing diaphragm via a capillary tube filled with silicone oil as a pressure transmission medium to prevent dust from entering the sensing cavity. The sensor automatically performs zero drift calibration every 24 hours: the zero pressure value is recorded when the regeneration pipeline valve is closed, and this offset is deducted from subsequent measurements.

[0080] After obtaining the concentration time series data, the arithmetic mean of the exhaust gas concentration during the entire regeneration phase is first calculated as the baseline average concentration value. This average concentration value is calculated by extracting the concentration values ​​of all sampling points from the start to the end of the regeneration phase, adding these values, and dividing by the total number of sampling points. For example, if a regeneration phase lasts 1200 seconds and the sampling frequency is twice per second, the number of sampling points involved in the calculation is 2400. The calculated average concentration value serves as the baseline threshold for subsequent feature extraction.

[0081] When identifying the starting point of the desorption rate rise segment, the concentration time series data is scanned point by point in chronological order starting from the starting time point of the regeneration phase. The starting point judgment condition is: when the concentration value of a certain sampling point exceeds the benchmark average concentration value for the first time, and the concentration values ​​of the two subsequent consecutive sampling points remain above the benchmark average concentration value, the point is confirmed as the starting moment. This continuous verification mechanism can avoid misidentification caused by transient noise. The end point identification continues to scan backward from the starting moment. When it is detected that the concentration value reaches the global maximum value, and the subsequent three consecutive sampling points show a monotonically decreasing trend, the moment corresponding to the peak point is determined as the end point. The judgment range of the global maximum value is the data of the entire regeneration phase, which is achieved by comparing the concentration values ​​of all sampling points.

[0082] Within the determined desorption rate ramp-up period, multiple subintervals are divided at fixed time intervals. The time interval is set to 10 seconds, a value determined based on the adsorbent's desorption kinetics. Analysis of desorption rate test data for various molecular sieve materials indicates that a 10-second interval captures the main reaction trends while minimizing noise interference. Each subinterval contains 20 sampling points (calculated at a sampling frequency of two per second). A linear regression is performed on the concentration data within each subinterval, using time as the independent variable and concentration as the dependent variable, using the least squares principle to fit a straight line. The calculation process involves calculating the covariance of the time series and concentration series, dividing it by the variance of the time series, to obtain the linear regression slope for that subinterval. After calculating the slopes for all subintervals, the standard deviation of these slopes is calculated. The standard deviation is calculated using an unbiased estimator: first, sum the squares of the differences between each slope and the average slope, divide by the number of subintervals minus one, and then take the square root. This standard deviation represents the slope change rate, which indicates the stability of the desorption process.

[0083] When determining the desorption peak decay phase, the starting time is directly the time corresponding to the global maximum concentration. The end time is determined by scanning backward from the peak point. The end time is determined when the concentration value first falls below the baseline average concentration value, and when two subsequent sampling points are both below the baseline average concentration value. This continuous verification mechanism prevents premature termination due to data fluctuations.

[0084] When performing area integration on the desorption peak attenuation section, the trapezoidal method is used for cumulative calculation. The specific process is: traverse all adjacent pairs of sampling points in the section in chronological order. For any two adjacent points, calculate the area of ​​the trapezoid they constitute: take the average of the concentration values ​​of the two points, and multiply it by the time interval between the two points. For example, if the time of two adjacent points is t1 and t2, and the concentrations are c1 and c2, then the infinitesimal area is [(c1+c2) / 2]×(t2-t1). The infinitesimal areas of all adjacent point pairs in the attenuation section are accumulated to obtain the total area integral value. The time interval is the actual sampling interval. When the sampling frequency is twice per second, the fixed time interval is 0.5 seconds. Compared with the rectangular integration method, this trapezoidal integration method can more accurately reflect the nonlinear attenuation characteristics of concentration.

[0085] During feature extraction, an abnormal data handling mechanism is implemented: when missing concentration values ​​are detected within a subinterval, the concentration mean of the two preceding and following sampling points is automatically interpolated; if more than three consecutive sampling points are missing, the slope calculation for that subinterval is discarded. For the area integral of the decay segment, if the number of sampling points within the start and end time interval is less than 10, a data validity check is automatically triggered: the total concentration change for that segment is calculated. If the change is less than 20% of the average concentration value, the decay segment is deemed invalid, the area integral value extraction is abandoned, and a data anomaly log is generated.

[0086] The linear regression slope calculation includes a data validity check: if the standard deviation of the concentration values ​​within a subinterval is less than 1% of the average concentration value, the concentration change in that segment is considered insignificant, and the slope calculation is abandoned. The subinterval length is automatically extended to 20 seconds and the calculation is repeated. If the condition is still not met, the output slope is zero and the segment is marked as low activity. Baseline correction is added to the area integral calculation: the baseline average concentration value is deducted before integration. This means that the baseline average concentration value is subtracted from the concentration value of each sampling point before the area calculation is included in the calculation to eliminate the influence of background concentration.

[0087] The identification of the rising desorption rate phase is protected by a time boundary: if the starting point is within 5 seconds of the start of the regeneration phase, or the end point is within 5 seconds of the end, the time range is automatically expanded and the identification is repeated. If the boundary is still reached after the expansion, a backup identification algorithm is activated: a sequence of concentration change rates is calculated, starting with the moment the rate first exceeds 0.5 ppm / second and ending with the moment the rate falls below 0.2 ppm / second. This mechanism ensures that valid features can be obtained even when concentration data is abnormal.

[0088] Before outputting the slope rate of change, normalization is performed: the standard deviation is divided by the average slope value of the rising segment to convert it into a dimensionless parameter. The area integral value undergoes a dimensionless conversion: the integral result is multiplied by the sampling frequency to make the results comparable at different sampling rates. All eigenvalues ​​are stored with associated quality flags, including the number of subintervals for calculating the slope standard deviation of the rising segment, the number of sampling points for the decay segment integration, and an abnormal data handling flag. If the quality flag indicates that the percentage of valid data is less than 80%, the regeneration cycle eigenvalue is marked as suspect.

[0089] The feature extraction process monitors the computational load in real time. If the processing time of a single regeneration cycle exceeds 10 seconds, a simplified mode is automatically initiated: the subinterval length is extended to 20 seconds, and the trapezoidal integration sampling interval is increased to 1 second. Simplified mode triggers are recorded in the system log, and the feature values ​​are marked as fast calculation mode results. The system automatically generates a daily feature extraction quality report, which summarizes the slope change rate calculation failure rate and the area integral value abnormality rate. If the failure rate exceeds 15% for three consecutive days, a sensor calibration reminder is triggered.

[0090] Global peak concentration detection utilizes a local maximum validation method: a valid peak is identified when the concentration at a sampling point is higher than both the previous and next points, and the difference between this value and the previous point is greater than three times the noise threshold. The noise threshold is calculated using 1.5 times the standard deviation of the concentration during the 30 seconds before regeneration. This mechanism prevents small fluctuations from being mistaken for peaks. In the presence of multiple peaks, the highest peak is selected as the starting point of the desorption peak decay phase, and the remaining peaks are recorded as auxiliary features and stored in the database.

[0091] The optimization mechanism of the time interval parameter is that the system automatically analyzes historical data every quarter to calculate the characteristic stability index under different time intervals. The stability index is defined as the coefficient of variation of the characteristic value of the same adsorption tower for 10 consecutive cycles. The interval with the minimum coefficient of variation of the slope change rate is selected as the new parameter, and the update range is limited to between 5 seconds and 30 seconds. The time interval of area integration is fixed and bound to the sampling frequency, and does not undergo dynamic adjustment. All parameter changes are recorded in the system configuration file, and the last set of characteristic values before the change is saved as a reference.

[0092] At the end of the current regeneration cycle, the regeneration end signal trigger timestamp recorded by the adsorption tower controller is read. The timestamp has a precision of milliseconds and is generated by the programmable logic controller when the regeneration valve closing action is detected. The historical end time of the previous preset number of regeneration cycles is obtained by querying the historical operation log, and the preset number is initially defaulted to 8 cycles, which is determined according to the typical aging period of the adsorbent: by analyzing the continuous operation data of the adsorption tower, it is found that when the adsorbent performance decays, the area integral value usually shows a trend change within 8 consecutive cycles. The preset number can be dynamically adjusted in the system parameters, and the adjustment range is 5 to 12 cycles, which is automatically matched according to the type of adsorbent: 10 cycles for molecular sieve adsorbent and 6 cycles for activated carbon adsorbent.

[0093] When retrieving the area integral value from the concentration time series database, an accurate matching mechanism based on timestamps is established. The concentration time series database uses a time partition storage structure, and each regeneration cycle data is stored independently in a data partition named after the end time. When retrieving, first locate the target partition, then query the data record labeled "area integral value" in the partition. Each record contains three key fields: integral value (float type), calculation state flag (integer type), and time interval start and end timestamps (each 8 bytes). When multiple records with the same label are retrieved, the record with the latest timestamp is automatically selected as the valid value.

[0094] Two levels of verification are performed to verify the integrity of the area integral value. The first level of verification is the current regeneration cycle data: check the "pressure pulse disturbance flag" associated with the area integral value of this cycle. The flag is generated by S5 and stored in the database, and a flag value of 0 indicates that it does not contain pressure pulse disturbance period data, and a flag value of 1 indicates that it contains disturbance data. When the flag value is 0, it passes the verification. The second level of verification is the historical data: check the continuity of the area integral value within the previous preset number of cycles. The continuity criterion is that the interval between the end times of the two adjacent regeneration cycles is within ±15% of the standard regeneration cycle length, and there is a valid area integral value record for all cycles. The standard regeneration cycle length is the average regeneration length of the last 30 cycles. If the current cycle passes the first level of verification and the historical cycle passes the second level of verification, the area integral value is determined to be qualified data.

[0095] When data integrity requirements are not met, an automatic compensation mechanism is activated. If the current cycle is deemed unqualified due to the presence of disturbed data, the data is traced back one cycle: the current value is replaced with the area integral value of the previous cycle and marked as replacement data in the sequence. If the historical data is discontinuous, gap filling is performed: the moving average of the area integral values ​​of the two cycles before and after the gap is calculated as the filling value. If the gap exceeds three cycles, the sequence construction is terminated and a data anomaly alert is generated.

[0096] When constructing the area integral value sequence, qualified area integral values ​​are sorted by the end time of their regeneration cycles. Sorting is in ascending order, with the area integral value corresponding to the earliest end time placed at the beginning of the sequence. The sequence is stored as a dynamic array structure containing the following metadata: sequence generation timestamp, number of regeneration cycles included, and a mapping between the end time of each cycle and the area integral value. The sequence length dynamically adapts to the preset number and is automatically reconstructed when the preset number changes.

[0097] Quality monitoring metrics are set during the sequence construction process. Key metrics include data integrity (the ratio of qualified data to total data volume) and historical data continuity (the longest consecutive valid cycles). When the data integrity rate falls below 90%, the preset number of qualified data is automatically reduced to the current qualified data volume, but with a minimum of five cycles. When the historical data continuity falls below 80% of the preset number, a data continuity check command is triggered: the operation logs of the previous 30 cycles are scanned, abnormal downtime events are identified, and associated tags are generated.

[0098] Sequence data storage utilizes a dual-backup mechanism. A primary copy is stored in an in-memory database, retaining the 10 most recently constructed sequences for real-time analysis. A secondary copy is stored in a disk-based database, archiving historical sequence data on a daily basis. Sequences in the in-memory database are persisted every 10 minutes to prevent data loss due to system power outages. The sequence access interface provides time range queries, allowing retrieval of historical sequences by the end of a regeneration cycle.

[0099] The optimization mechanism for the preset number of cycles is as follows: the system automatically analyzes the stability of the sequence monthly. The coefficient of variation (standard deviation divided by mean) of the sequence is calculated for different preset numbers, and the value with the lowest coefficient of variation is selected as the new preset value. The optimization process is constrained by the requirement that the preset number of cycles must not be less than 5 and not more than 20. The optimization results are recorded in the system configuration log, and the last set of sequences before the optimization is saved as a baseline.

[0100] Standardized unit processing for area integral values: Dimensional normalization is automatically performed during the sequence construction phase. Raw integral values ​​(unit: ppm × second) are converted to a percentage scale by taking the maximum area integral value from the last 100 cycles as a baseline, dividing the current value by the baseline value and multiplying it by 100. The converted values ​​range from 0 to 100, eliminating magnitude differences between adsorption towers. The conversion factor is stored in the device configuration file. If the area integral value exceeds the baseline value for 10 consecutive cycles, the baseline value is automatically updated and the historical data is reconverted.

[0101] After receiving a sequence of area integral values, the arithmetic mean of the sequence is calculated as the mean. This calculation is performed by summing all area integral values ​​in the sequence and dividing the resulting sum by the number of regeneration cycles in the sequence. For example, if the sequence contains data for eight cycles, the sum of the eight area integral values ​​is divided by 8. The arithmetic mean is calculated using double-precision floating-point arithmetic, and intermediate results are rounded to 16 significant digits to avoid roundoff errors. If the sequence is empty, the calculation terminates immediately and returns a null error code.

[0102] Strict data type checking is performed when determining the absolute difference between the area integral value of the current regeneration cycle and the mean of the area integral value sequence. The current area integral value is obtained from the most recent record generated in step S2, and the sequence mean is obtained using the above calculation. The absolute difference is the absolute value of the sequence mean minus the current value, ensuring that the result is always non-negative. Overflow protection is set during the calculation: when the absolute value of the difference exceeds 10 times the maximum value of the sequence, the raw data review process is automatically triggered to verify the dimensional consistency of the area integral value.

[0103] A statistically unbiased estimation method is used to obtain the standard deviation of a series of area integral values. The specific process is to first calculate the square of the difference between each area integral value in the series and the series mean, sum these squared values, divide them by the number of series periods minus one, and finally take the square root of the quotient. The standard deviation indicates the degree of dispersion of the series data and serves as a benchmark for subsequent normalization. To enhance stability in the calculation process, when the number of series periods is less than 5, the standard deviation is fixed at an empirical value of 0.15, which is derived from the 90th percentile of the distribution of the minimum standard deviation in historical data analysis.

[0104] Divide the absolute difference by the dispersion benchmark to obtain the preliminary deviation. Before this division, perform a non-zero divisor check: if the standard deviation is less than 0.0001, reset the dispersion benchmark to 5% of the series mean to avoid division by zero errors. The preliminary deviation is a dimensionless parameter, and the calculation result is rounded to four decimal places. If the preliminary deviation exceeds 10.0, it is automatically marked as an extreme outlier, triggering the data traceability mechanism: recheck the original data calculated for the area integral value of that period.

[0105] When the time decay factor is introduced to modify the preliminary deviation degree, the cumulative running time of the adsorbent is first obtained. The running time is obtained by reading the total running hours of the adsorption tower from the device control system, accurate to 0.1 hours. The calculation of the time decay factor uses an inverse proportional function model: the decay factor is equal to 1 divided by (1 plus the cumulative running time multiplied by the decay coefficient). The decay coefficient is initially set to 0.00015 per hour, which is determined according to the aging rate of the molecular sieve adsorbent: through accelerated life testing, when the adsorbent capacity decays by 20%, the deviation tolerance needs to be expanded by 1.5 times.

[0106] The specific process of the correction calculation is: multiply the preliminary deviation degree by the time decay factor to obtain the symmetric deviation degree. This calculation model realizes the aging compensation effect: as the running time increases, the same preliminary deviation degree will produce a smaller symmetric deviation degree, reflecting the expected impact of natural decay of adsorbent performance. The symmetric deviation degree result is stored as a percentage value, ranging from 0 to 100, and automatically resets to the boundary value when it exceeds the range.

[0107] The dynamic adjustment mechanism of the time decay factor is: the coefficient is calibrated every half year. The calibration method is: select symmetric deviation degree samples of the same type of adsorption tower in the new state and the scrap state, establish the relationship curve of running time and deviation degree, and update the decay coefficient according to the slope of the curve. The calibration process limits the change amplitude of the decay coefficient to not more than plus or minus 30% of the original value, avoiding parameter mutation.

[0108] Multiple data checks are set in the calculation process. Before obtaining the sequence mean, the time validity of each area integral value in the sequence is verified: check whether its corresponding regeneration cycle end time is before the current system time to prevent future data from being mixed in. When calculating the standard deviation, automatically exclude outliers whose deviation from the mean is more than 3 times the standard deviation, and replace the outliers with the moving average of the two previous and subsequent cycles.

[0109] Smooth processing is performed before the symmetric deviation degree is output: take the moving average of the last three calculation results as the final output. When it is detected that the difference between a single calculation result and the previous result exceeds 20%, start the review calculation: reacquire the original sequence data, and record the review log throughout the process. The final result is associated with the following metadata: calculation timestamp, number of sequence cycles used, time decay factor version number, and data check status flag.

[0110] An exception handling mechanism is set for the acquisition of the cumulative running time of the adsorbent. When reading from the device control system fails, start the backup timer: the cumulative running time is equal to the average length of the last 200 regeneration cycles multiplied by the total number of completed regeneration cycles. The backup timer is synchronized with the main system every 24 hours, and a clock synchronization alarm is triggered when the error exceeds 2 hours. The running time is stored as an unsigned long integer, with a unit of 0.1 hours, and a maximum of 1 million hours.

[0111] The historical data storage of symmetry deviation degree adopts differential compression technology. A full value is stored every 10 cycles, and only the difference value with the previous cycle is stored in the middle cycle. When reading, the data sequence is reconstructed by accumulating the difference values, saving more than 50% of the storage space. When storing, the temperature compensation coefficient is associated: when the ambient temperature exceeds 40 degrees Celsius, the symmetry deviation degree value is automatically reduced by 0.1% per degree Celsius, compensating for the influence of temperature on adsorption performance.

[0112] When obtaining the time sequence of the pressure sensor data in the regeneration stage, the data stream labeled as "pressure value" is extracted from the concentration time sequence database established in step S1. The pressure value is in kilopascals, the sampling frequency is 10 times per second, and the timestamp accuracy is millisecond level. The data extraction range is limited to the time interval from the start signal to the end signal of the regeneration stage, the start signal is the time when the regeneration valve opening instruction is issued, and the end signal is the time when the regeneration valve closing confirmation is made. When continuous missing of pressure data is detected, the data is automatically completed using the cubic spline interpolation method, with a maximum allowed completion time of 2 seconds. If the time exceeds, the pressure data of the regeneration cycle is marked as invalid.

[0113] When calculating the pressure change rate time sequence, the central difference method is used for real-time processing. The specific calculation method is: for the i-th sampling point in the pressure sequence, the pressure change rate is equal to the pressure value of the i+1-th sampling point minus the pressure value of the i-1-th sampling point, and then divided by twice the time interval. The time interval is fixed at 0.1 seconds (corresponding to a sampling frequency of 10 Hz), and the calculation result is in kilopascals per second. In the boundary area of 1 second at the beginning and end of the sequence, forward difference or backward difference is used: the first point uses the second point minus the first point divided by the time interval, and the last point uses the last point minus the second last point divided by the time interval. All change rate values are stored as a double-precision floating-point number sequence, strictly aligned with the original pressure value timestamp.

[0114] When identifying the pressure pulse disturbance period, three threshold conditions are applied synchronously. The pressure change rate threshold is set to 50 kilopascals per second, which is determined based on historical disturbance event statistical analysis: taking 80% of the minimum value of the change rate in 100 typical disturbance events as the safety threshold. The pressure absolute value threshold is set to 300 kilopascals, corresponding to 75% of the design pressure of the regeneration pipeline. The minimum duration threshold is set to 100 milliseconds, which can filter transient interference. The identification process uses a sliding window mechanism: when the pressure change rate of three consecutive sampling points is detected to be above the threshold and the pressure absolute value is above the threshold, it is marked as the start point of the potential disturbance period; when the subsequent two consecutive sampling points do not meet the conditions, it is marked as the end point. The period between the start point and the end point is recorded as the candidate disturbance period.

[0115] The effectiveness verification is performed on the candidate perturbation period. When the candidate period duration is greater than the minimum duration threshold, the pressure variation characteristics within the period are further checked: the integral value of the pressure rate of change within the period is calculated, and if the integral value exceeds 500 kPa, it is confirmed as an effective pressure pulse perturbation period. The perturbation period information is stored as a binary tuple data (start time stamp, end time stamp), associated with the perturbation intensity index (the ratio of the maximum pressure rate of change to the average pressure within the period).

[0116] The data rejection is performed within the time interval of the desorption rate rising segment and the desorption peak decay segment. First, the accurate time boundaries of the two characteristic segments are obtained from the S2 step: the desorption rate rising segment is the start time to the end time, and the desorption peak decay segment is the start time to the end time. The time interval overlap detection algorithm is established: for each characteristic segment and the pressure pulse perturbation period, the duration of the time intersection is calculated. When the intersection duration is greater than 10 milliseconds, it is determined that there is an overlap. In the overlapping period, the slope change rate and the area integral value rejection operation is performed: if the desorption rate rising segment overlaps with the perturbation period, the corresponding slope change rate in the rising segment is marked as invalid; if the desorption peak decay segment overlaps with the perturbation period, the area integral value of the decay segment is marked as invalid.

[0117] The un-rejected slope change rate and the un-rejected area integral value are combined into effective period data. The effective period data adopts a structured storage format, including three core fields: effective slope change rate (floating point type), effective area integral value (floating point type), and characteristic segment state marker (integer type). The state marker coding rule is: 00 indicates that both characteristic segments are valid, 01 indicates that the rising segment is invalid, 10 indicates that the decay segment is invalid, and 11 indicates that both segments are invalid. When the 11 state occurs, the supplementary data acquisition mechanism is automatically triggered: high-precision monitoring is preferentially performed in the next regeneration cycle.

[0118] The dynamic optimization mechanism of the pressure threshold parameter is: statistical analysis of pressure pulse perturbation events is performed every month. The maximum pressure value, maximum rate of change and duration of each event are recorded, and the 95th percentile value of these parameters is taken as the new threshold value. The optimization process sets the following constraints: the pressure rate of change threshold adjustment amplitude does not exceed ±20% of the original value, and the pressure absolute value threshold cannot be higher than 85% of the pipeline design pressure.

[0119] The data rejection process sets protection measures: when the rejection proportion of a single characteristic segment exceeds 50%, the segment reservation mechanism is started. The characteristic segment is divided into 10 equal length subintervals, only the subintervals completely overlapping with the perturbation period are rejected, and the non-overlapping part of the partially overlapping subintervals is reserved. For example, for the desorption rate rising segment, the average slope of the non-overlapping subintervals is calculated as the replacement value.

[0120] Pressure change rate calculation adds noise filter: When the pressure value is detected to jump more than 100 kPa between adjacent sampling points, the median filter preprocessing is automatically enabled. A 3-point sliding window median filter is used, that is, the middle value of the current point and the previous and next points is taken as the effective pressure value. The filter processing record is marked for subsequent data analysis reference.

[0121] After the effective period data is generated, integrity verification is performed. The verification rules include: the effective slope change rate must be within the historical normal range (0.1 to 5.0 ppm / s), and the effective area integral value must be greater than 50 ppm·s. When the data is out of range, the metal dust concentration data of the regeneration cycle is automatically associated, and if the dust concentration exceeds the threshold, it is marked as a dust interference event, otherwise it is marked as an equipment anomaly.

[0122] When extracting the slope change rate of the current regeneration cycle from the effective period data, the structured data record generated in step S5 is accessed. According to the feature segment state flag, the effective data source is selected: when it is marked as 00 or 10, the effective slope change rate field is directly read; when it is marked as 01 or 11, the backup data acquisition mechanism is enabled - the moving average value of the slope change rate of the last three undisturbed regeneration cycles is retrieved as a substitute value. The extracted data performs unit consistency verification: ensure that the unit of the slope change rate is ppm / s, if the dimension is detected to be inconsistent, automatically multiply by 60 to convert to the standard unit of ppm / min, and record the conversion operation in the data log.

[0123] When obtaining the historical normal working condition reference value, the historical normal working condition is defined as a set of periods with a symmetry deviation of less than 1.0 in the first 100 regeneration cycles. The moving average calculation uses a rolling window mechanism: the window size is fixed at 20 cycles, and each time it is rolled forward by one cycle to recalculate. The specific calculation process is: in the normal working condition period set, take the slope change rate of the latest 20 cycles in reverse time order, and calculate the arithmetic mean as the reference value. The reference value is updated every hour, and the data of the period being regenerated is excluded when updating. When the number of normal working condition periods is less than 20, the window size is automatically reduced to the actual number but not less than 5 cycles, and when it is less than 5, a historical data expansion alarm is triggered.

[0124] When calculating the slope change rate offset, a three-step operation procedure is performed. First, the absolute difference between the current slope change rate and the historical normal condition reference value is calculated, and the absolute value is taken to ensure that the result is non-negative. Second, divisor protection is performed: when the historical normal condition reference value is less than 0.01, the reference value is reset to 1.1 times the historical minimum slope change rate of the adsorption tower. Finally, the absolute difference is divided by the historical normal condition reference value, and the result is expressed in percentage form. For example, if the current value is 0.5 ppm / s and the reference value is 0.45 ppm / s, the offset is (|0.5-0.45| / 0.45) x 100 = 11.11%. The calculation result is limited to the range of 0 to 1000%, and when it exceeds, it is automatically truncated and marked as an extreme value.

[0125] When continuously extracting the symmetry deviation degree, the preset number is consistent with the sequence construction parameter of S3 step. The search range is limited to the current regeneration cycle and the previous continuous preset number of cycles, and the search condition is that the cycle state is marked as "complete calculation". When extracting, time sequence verification is performed: it is checked whether the end time of each symmetry deviation degree corresponding to the regeneration cycle is strictly continuous. When an abnormal time interval is detected, the latest continuous cycle group is automatically traced back to ensure that the extracted symmetry deviation degree sequence meets the time continuity requirement.

[0126] When calculating the standard deviation of the symmetry deviation degree sequence, two-stage data processing is adopted. In the first stage, outliers are excluded: the preliminary mean and standard deviation of the sequence are calculated, and data points deviating from the mean by more than 3 times the standard deviation are removed. In the second stage, the standard deviation is recalculated on the purified sequence using the unbiased estimation formula: the sum of the squares of the differences between each symmetry deviation degree and the mean is divided by the number of data points minus one, and then the square root is taken. To ensure calculation stability, when the number of sequence cycles is less than 3, the standard deviation is output as zero and marked as a low confidence result.

[0127] When obtaining the fluctuation abnormality degree, the arithmetic mean of the symmetry deviation degree sequence is first calculated: the sum of all values in the sequence is divided by the number of cycles. Then, divisor check is performed: when the arithmetic mean is less than 0.01, it is reset to 0.01 to avoid division by zero error. Finally, the fluctuation abnormality degree is equal to the standard deviation divided by the arithmetic mean, which is a dimensionless parameter representing the relative fluctuation intensity. The calculation result is converted to percentage and stored, for example, an original value of 0.15 represents a fluctuation level of 15%.

[0128] The update mechanism of the historical normal condition reference value includes a self-learning function. A reference benchmark evaluation is performed every month: the difference between the current reference value and the distribution of the actual slope change rate of the last 30 cycles is compared. When the median of the actual value deviates from the reference value by more than 20%, the reference value is recalibrated: the normal condition screening range is expanded to the previous 200 cycles, and the moving average is recalculated. The calibration process lasts for a maximum of 3 rounds, and if it still does not converge, an adsorption characteristic drift warning is generated.

[0129] The storage of the symmetry deviation degree sequence uses incremental encoding. Only the difference value with the previous cycle is stored in each regeneration cycle, and the absolute value is stored in the initial cycle. When reading, the complete sequence is reconstructed by accumulating the difference value, saving 70% of the storage space. The storage value is associated with the temperature compensation coefficient: when the ambient temperature exceeds 25 degrees Celsius, the symmetry deviation degree is corrected by 0.5% per degree Celsius, eliminating the influence of temperature on the adsorption equilibrium.

[0130] The fluctuation anomaly degree calculation adds a dynamic smoothing process. Take the weighted average of the last three calculation results: the current cycle weight is 50%, the previous week is 30%, and the previous two weeks are 20%. When a single result is detected to be more than 30% different from the previous one, data review is started: the original symmetry deviation degree sequence is re-extracted, and the review path is recorded throughout. The final result output includes a confidence index: according to the calculation of the number and quality of the data period used, when the confidence is less than 60%, the supplementary data acquisition instruction is triggered.

[0131] The range adaptive adjustment mechanism of the slope change rate offset is: statistics the distribution range of historical offset, update the display range every quarter. For example, when the maximum historical offset is 150%, the display range is set to 0%~200%; if a value exceeding 200% appears, the range is automatically expanded to 300%. This mechanism ensures the visibility of data at different aging stages.

[0132] When obtaining the current values of the slope change rate offset and the fluctuation anomaly degree, a data validity verification mechanism is established. The slope change rate offset is extracted from the output cache of step S6, and the fluctuation anomaly degree is obtained from the symmetry deviation degree analysis module. Before extraction, the data timestamp is verified: ensure that both correspond to the same regeneration cycle end time, and the time deviation does not exceed 5% of the average length of the regeneration cycle. When detecting that the value is out of bounds (offset > 1000% or anomaly > 100%), automatically trigger the original data review process: re-execute step S6 calculation and record the review log.

[0133] When reading the preset first threshold and second threshold, access the dynamic configuration table in the system parameter database. The initial value of the first threshold is 25%, which is set according to the adsorbent performance degradation model: through laboratory accelerated aging test, when the adsorbent capacity decreases by 15%, the typical offset is 25±3%. The initial value of the second threshold is 30%, based on the statistical distribution of the fluctuation anomaly degree of historical normal working conditions: take the 90% quantile of the maximum fluctuation anomaly degree of the previous 100 cycles. The threshold storage format is floating point, which supports runtime modification: after modification, automatically record the operator information and modification time.

[0134] When judging dual-threshold conditions, atomic transactions are used to ensure logical integrity. First, the current slope change rate offset value is locked, and the first threshold comparison is performed. If the offset is greater than 25%, condition A is marked as true. Then, the current fluctuation anomaly value is locked, and the second threshold comparison is performed. If the anomaly is greater than 30%, condition B is marked as true. The dual-condition judgment is completed within 10 milliseconds, avoiding inconsistent states caused by data updates. When both conditions A and B are met, a warning trigger event is generated. Event information includes the trigger time, the actual offset value, and the actual anomaly value.

[0135] The continuous cycle verification mechanism is initiated after the early warning judgment condition is triggered. A cycle counter is created with an initial value of 1. At the end of each subsequent regeneration cycle, the dual threshold judgment is repeated: if the condition is met again, the counter is incremented; if the condition is not met, the counter is reset to zero. A state retention function is set during the continuous verification period: if a cycle cannot be judged due to invalid data (such as pressure pulse disturbance), the counting state of the previous cycle is retained. When the counter reaches 3, an adsorbent replacement warning signal is generated and the counter is reset.

[0136] A multi-level warning system is defined when generating an adsorbent replacement warning signal. The basic signal contains three fields: warning level (enumeration type: mild / moderate / severe), trigger time (the end of the regeneration cycle), and diagnostic summary (text type). Level determination rules: When the offset is between 25% and 35% and the abnormality is between 30% and 40%, it is mild; when the offset is between 35% and 50% or the abnormality is between 40% and 60%, it is moderate; and the rest are severe. The diagnostic summary is automatically generated by splicing key parameter values ​​and trigger condition descriptions, for example, "Offset 28.6% > 25%, abnormality 33.2% > 30%, and the standards have been met for three consecutive cycles."

[0137] Signal storage utilizes a spatiotemporal correlation architecture. The time dimension converts the regeneration cycle end time into a UNIX timestamp (accurate to the millisecond) as the primary key. The spatial dimension associates the adsorption tower number with the adsorbent batch number. Storage implements a dual-write operation: real-time data is written to an in-memory database for access by the monitoring system and asynchronously persisted to a disk database. Records are retained in the in-memory database for seven days, while the disk database stores them permanently. Environmental context is also included during storage, including auxiliary parameters such as the average temperature during the regeneration phase and peak metal dust concentration.

[0138] The dynamic threshold calibration mechanism is automatically performed quarterly. Calibration data sources are historical warning records and adsorbent test reports. The minimum offset / anomaly value during actual adsorbent replacement cycles is calculated and used as the new threshold baseline. The first threshold update rule: 80% of the baseline value is used as the new threshold (no less than 20%). The second threshold update rule: 90% of the baseline value is used as the new threshold (no less than 25%). Changes exceeding 10% of the original value require manual confirmation.

[0139] Continuous period verification setting exception interrupt processing. If valid data cannot be obtained for two consecutive periods (such as device maintenance downtime), the verification is suspended and an interrupt log is generated. After the system resumes operation, the counting is restarted from the most recent valid period. When the cumulative interrupts exceed 5 periods, the verification process is forcibly reset.

[0140] After the early warning signal is generated, the linkage action is triggered. When a severe early warning is generated, the following linkage actions are automatically executed: freezing the adsorption tower operating parameters to prevent misadjustment, activating the standby adsorption tower to run in parallel, and sending a short message and an email alarm to the maintenance team. When a moderate early warning is generated, the weekly report is triggered: the characteristic trend chart of the last 10 periods is summarized. When a mild early warning is generated, only a log is recorded, which is used for subsequent trend analysis.

[0141] The signal storage database implements encryption and integrity protection. The early warning content is encrypted using the AES-256 algorithm, and the key is changed every 72 hours. When data is written, a SHA-256 hash value is generated and stored in a separate verification area, and when data is read, the data integrity is automatically verified. Database access is implemented with three-level permission control: operators can only query, engineers can modify thresholds, and administrators can correct historical records.

[0142] The calculations involved in the embodiments are all dimensionless numerical calculations, and the preset parameters and threshold values in the calculations are set by those skilled in the art according to the actual situation.

[0143] The above embodiments can be realized wholly or partially by software, hardware, firmware, or any combination thereof. When realized by software, the above embodiments can be realized in the form of a computer program product, wholly or partially.

[0144] Those skilled in the art can realize that the modules and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application of the technical solution and the constraints of the invention. Those skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0145] In addition, the functional modules in each embodiment of the present application can be integrated in one processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0146] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the modules is only a logical function division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0147] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

[0148] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. An intelligent monitoring method for an argon recovery system, characterized in that: The steps include: S1. During the regeneration stage of the adsorption tower, real-time collection of concentration time series data of the regenerated tail gas; S2. Extracting the slope change rate of the desorption rate rising segment and the area integral value of the desorption peak decay segment from the concentration time series data; S3, continuously extracting the area integral values ​​of the current regeneration cycle and the previous preset number of regeneration cycles, and constructing an area integral value sequence, including: Get the end time of the current regeneration cycle and the historical end time of the previous preset number of regeneration cycles; Retrieving the area integral value corresponding to the end time of each regeneration cycle from the stored concentration time series database; Verify the integrity of the area integral values, and arrange the qualified area integral values ​​in chronological order according to the end time of the regeneration cycle to form an area integral value sequence; S4. Calculating the symmetric deviation between the area integral value of the current regeneration cycle and the mean of the area integral value sequence, including: Calculate the arithmetic mean of the area integral value sequence as the mean of the area integral value sequence; Determine the absolute difference between the area integral value of the current regeneration cycle and the mean of the area integral value sequence; Obtain the standard deviation of the area integral value sequence as the dispersion benchmark; Divide the absolute difference by the dispersion benchmark to obtain the preliminary deviation; The time decay factor of the cumulative running time of the adsorbent is introduced to correct the initial deviation and obtain a symmetrical deviation. S5. Identify the pressure pulse disturbance period in the regeneration phase, remove the slope change rate and area integral value within the pressure pulse disturbance period, and obtain effective period data, including: Obtaining the time series of pressure sensor data during the regeneration phase; Calculate the pressure change rate time series of pressure sensor data; identifying consecutive periods in the pressure change rate time series where the pressure change rate threshold is exceeded and the pressure sensor data exceeds the pressure absolute value threshold; When the duration of a continuous period is greater than the minimum duration threshold, it is marked as a pressure pulse disturbance period; In the time interval of the desorption rate rising section and the desorption peak decay section, the slope change rate and area integral value of the period overlapping with the pressure pulse disturbance period are eliminated; The uneliminated slope change rate and the uneliminated area integral value are combined into valid period data; S6. Calculate the slope change rate offset and the fluctuation anomaly of the symmetric deviation based on the valid period data, including: Extract the slope change rate of the current regeneration cycle from the valid period data; Obtain the moving average of the slope change rate under historical normal operating conditions as a reference value for historical normal operating conditions; Calculate the absolute difference between the slope change rate of the current regeneration cycle and the historical normal operating condition reference value, and divide it by the historical normal operating condition reference value to obtain the slope change rate offset; Continuously extracting the symmetry deviation of the current regeneration cycle and the previous preset number of regeneration cycles; Calculate the standard deviation of the symmetric deviation sequence for a preset number of consecutive regeneration cycles; Divide the standard deviation by the arithmetic mean of the symmetric deviation sequence of a preset number of consecutive regeneration cycles to obtain the fluctuation anomaly of the symmetric deviation; S7. When the slope change rate offset exceeds a first threshold and the fluctuation abnormality exceeds a second threshold, an adsorbent replacement warning signal is generated.

2. The intelligent monitoring method for an argon recovery system according to claim 1, characterized in that: During the regeneration stage of the adsorption tower, the concentration time series data of the regenerated tail gas is collected in real time, including: A mid-infrared laser gas analyzer is used to obtain the original signal of the regenerated exhaust gas concentration at a sampling frequency of not less than once per second; Synchronously receive the metal dust concentration detection results and pressure sensor data of the regeneration pipeline; Perform sliding mean filtering on the original signal of regenerated exhaust gas concentration to eliminate instantaneous noise interference; The filtered concentration signal, metal dust concentration detection results and pressure sensor data are associated with timestamps and stored as concentration time series data.

3. The intelligent monitoring method for an argon recovery system according to claim 2, characterized in that: The slope change rate of the desorption rate rising section and the area integral value of the desorption peak decay section are extracted from the concentration time series data, including: Identify the starting and ending points of the desorption rate increase segment in the concentration time series data; The desorption rate rising section is divided into multiple sub-intervals at fixed time intervals, and the linear regression slope of the concentration change in each sub-interval is calculated. The standard deviation of the slopes of all sub-intervals is used as the slope change rate; Determine the start and end time of the desorption peak decay segment; The concentration values ​​and time intervals of all adjacent sampling points in the desorption peak attenuation section are accumulated in a trapezoidal manner, and the accumulated results are used as the area integral value.

4. The intelligent monitoring method for an argon recovery system according to claim 3, characterized in that: The starting point is the moment when the concentration value exceeds the average concentration value for the first time, and the end point is the moment when the concentration value reaches the peak concentration for the first time; the starting time is the moment corresponding to the peak concentration, and the ending time is the moment when the concentration value falls back to the average concentration value.

5. The intelligent monitoring method for an argon recovery system according to claim 1, characterized in that: The integrity of the area integral value is verified as follows: if the area integral value of the current regeneration cycle does not contain the pressure pulse disturbance period data, and the area integral value of the historical regeneration cycle is continuous and complete, it is judged as qualified data.

6. The intelligent monitoring method for an argon recovery system according to claim 3, characterized in that: When the slope change rate offset exceeds a first threshold and the fluctuation abnormality exceeds a second threshold, an adsorbent replacement warning signal is generated, including: Get the current values ​​of the slope change rate offset and fluctuation anomaly; Reading a preset first threshold and a second threshold; Determine whether the slope change rate offset is greater than a first threshold, and simultaneously determine whether the fluctuation abnormality is greater than a second threshold; When the slope change rate offset is greater than the first threshold and the fluctuation abnormality is greater than the second threshold, the warning judgment condition is triggered; Repeated detection of warning determination conditions within three consecutive regeneration cycles; If the warning judgment conditions are met for three consecutive regeneration cycles, an adsorbent replacement warning signal is generated; The adsorbent replacement warning signal is associated with the end time of the current regeneration cycle and stored.

Images