An oscillating balance dust concentration adaptive integration and mass classification method based on allan variance

Abstract

This invention discloses an adaptive integration and quality classification method for dust concentration based on an oscillating balance using Allan variance, belonging to the field of atmospheric particulate matter monitoring technology. The method includes: collecting oscillation frequency, temperature and humidity, flow rate, amplitude, and phase-locked state; establishing a frequency buffer and performing preprocessing, removing linear trends from the frequency sequences entering the Allan variance calculation; calculating the Allan variance of each candidate integration time, and calculating a comprehensive score based on factors such as temperature, humidity, and flow rate, adaptively selecting the optimal integration time accordingly; calculating the dust concentration after determining the integration time; and determining anomalies based on temperature, humidity, flow rate, frequency jumps, mechanical vibration, amplitude, and phase-locked state, marking the concentration data as Grade A (excellent), Grade B (usable), Grade C (reference), or Grade D (invalid), and simultaneously outputting the concentration value, integration time, quality level, and anomaly cause code.

Description

Technical Field

[0001] This invention relates to the field of atmospheric particulate matter concentration monitoring technology, and more specifically, to an adaptive integral and quality classification method for dust concentration based on an oscillating balance using Allan variance. Background Technology

[0002] Tapered Element Oscillating Microbalance (TEOM) dust concentration sensors are widely used instruments in the field of ambient air particulate matter monitoring. Their basic principle is as follows: a hollow conical tube, fixed at one end and fitted with a filter membrane at the other, serves as the oscillating element. When the sampling airflow passes through the filter membrane, airborne particles are deposited on it, increasing the membrane's mass and causing the conical tube's oscillation frequency to decrease. The controller measures the change in oscillation frequency, combining this with the sampling flow rate and sampling time to calculate the airborne dust concentration. These sensors offer advantages such as compact structure, high sensitivity, and continuous online monitoring capabilities, and are widely used in ambient air quality monitoring, industrial dust monitoring, and mine dust measurement.

[0003] In existing calculations of dust concentration using oscillating balances, the following calibration relationship is typically used between the oscillation frequency and the particulate matter mass:

[0004]

[0005] in, This represents the change in the mass of dust adhering to the filter membrane. For calibration coefficients, The average oscillation frequency of the current window. This is the no-load oscillation frequency. Then, based on the sampled flow rate... and points time Calculate the sampling volume Final dust concentration .

[0006] Most oscillating balance dust concentration products on the market currently calculate concentration by averaging the frequency over a fixed integration time, such as a fixed integration window of 30 or 60 seconds. This fixed integration method is simple to implement and requires little computation, but it also has several obvious shortcomings.

[0007] First, a fixed integration time makes it difficult to balance response speed and measurement stability. If the integration time is too short, frequency noise has a significant impact on the concentration calculation results, leading to noticeable fluctuations. If the integration time is too long, although noise is smoothed, the response to changes in dust concentration slows down, easily resulting in significant lag in scenarios with rapidly changing concentrations. This problem is particularly prominent in field monitoring—users desire faster response times in stable environments, while requiring longer integration times to ensure data stability in disturbed environments; a fixed integration time cannot simultaneously meet both of these needs.

[0008] Second, interference from ambient humidity is difficult to effectively identify and eliminate. When ambient humidity changes abruptly, the filter membrane or sampling component may adsorb water vapor, causing a change in the oscillation frequency. This frequency change is physically no different from the frequency change caused by particulate matter accumulation, and therefore can be misinterpreted as an increase in particulate matter mass, resulting in an artificially high concentration. Conversely, during the recovery period after humidity decreases, water vapor desorption may lead to an artificially low concentration. Traditional algorithms lack a mechanism for judging humidity conditions and cannot distinguish these situations from the official concentration output.

[0009] Third, short-term anomalies in the frequency signal may occur due to mechanical vibration, frequency jumps, phase-locked loop (PLL) malfunctions, or amplitude anomalies. Oscillating balances rely on the continuous oscillation of a conical tube to operate; any factor affecting the oscillation state—whether external mechanical shock, internal PLL loss of lock, or drive malfunction—will be directly reflected in the frequency measurement. Traditional algorithms lack real-time monitoring and response mechanisms for these anomalies, continuing to output official concentration values ​​even when the signal is abnormal, potentially leading to serious measurement errors.

[0010] Fourth, when the flow rate deviates from the set value, the calculation of the sampling volume will produce errors. The concentration calculation of the oscillating balance depends on the accurate measurement of the sampling volume. If the actual flow rate deviates from the rated flow rate but the algorithm still calculates based on the rated flow rate, the concentration result will inevitably be inaccurate. Traditional algorithms typically do not monitor and respond to the flow status in real time.

[0011] Fifth, traditional output information is limited. Currently, most oscillating balance dust concentration meters primarily output the concentration value itself, lacking synchronized data quality levels and anomaly cause codes. The host computer, display terminal, or data acquisition system cannot determine the reliability of the received concentration data or its suitability for alarm, statistical, or regulatory purposes. In some critical application scenarios, such as data reporting from automatic ambient air quality monitoring stations, data reliability assessment is crucial, but traditional products lack effective technical means for this.

[0012] To address the aforementioned issues, several improvement solutions have been proposed. For example, one solution adds a light scattering measurement module to help determine if the oscillating balance data is abnormal; another solution employs a dual weighing chamber structure, using the measurement results from the balancing weighing chamber to calibrate the real-time weighing chamber. However, these solutions all require additional hardware—a light scattering module or a second weighing chamber—which not only increases instrument cost and structural complexity but is also difficult to implement on existing deployed equipment.

[0013] In summary, existing technologies lack a comprehensive solution that can simultaneously address the three challenges of adaptive optimization of integration time, identification of environmental and operational state interference, and quantitative evaluation of output data reliability without adding extra hardware. How to systematically solve these problems by utilizing existing sensor signals from the oscillating balance through algorithmic innovation is a key technical challenge that urgently needs to be overcome in this field. Summary of the Invention

[0014] 1. Technical problem to be solved:

[0015] This invention addresses the technical problems in existing oscillating balance dust concentration measurement, such as the difficulty in balancing response speed and stability with fixed integration time, the inability to effectively identify environmental humidity interference, the output of incorrect concentrations when there are abnormal signals such as mechanical vibration and phase-locking anomalies, the impact of flow deviation on calculation accuracy, and the lack of data quality evaluation due to the single output information. It provides an adaptive integration and quality classification method for oscillating balance dust concentration based on Allan variance.

[0016] 2. Technical Solution:

[0017] To solve the above problems, the present invention adopts the following technical solution.

[0018] An adaptive integral and quality classification method for dust concentration based on an oscillating balance using Allan variance, executed by an embedded controller built into the oscillating balance, includes the following steps:

[0019] In the signal acquisition step, the controller acquires various operating parameters of the oscillating balance, including the oscillation frequency, at a set acquisition period. ,temperature relative humidity Sampling flow rate ,amplitude The system collects data on oscillation frequency, sampling flow rate, amplitude, and phase-locked state at 1-second intervals, while temperature and relative humidity are collected at intervals ranging from 1 to 5 seconds. Of these signals, the oscillation frequency is used to calculate the Allan variance and dust mass increment; temperature and relative humidity are used to identify temperature and humidity abrupt changes; the sampling flow rate is used to calculate the sampling volume and determine if the flow rate is abnormal; and the amplitude and phase-locked state are used to determine if the oscillation system is operating stably.

[0020] Frequency buffering and preprocessing steps. The controller establishes a frequency ring buffer of a certain length (the buffer length is at least twice the longest candidate integration time to ensure sufficient statistical degrees of freedom for Allan variance calculation. This scheme uses a length of 300 seconds) to store the oscillation frequency sequence within the most recent period.

[0021] For ease of description, the data segment in the cache corresponding to a certain candidate integration time τ is defined as an integration window; the length of the integration window is the integration time τ. The frequency data within each integration window is calculated to obtain the score at that integration time.

[0022] Each time new frequency data is acquired, a data validity check is performed first—checking for obvious frequency jumps, phase-locking anomalies, or amplitude anomalies. If the current data meets the unlocking or frequency jump conditions (e.g., the absolute value of the oscillation frequency difference between two adjacent 1-second samples exceeds 0.25Hz), the concentration data at that moment is not used as the official output, and the controller enters the anomaly recovery protection process. For frequency sequences that pass the validity check and are to be included in the Allan variance calculation, linear trend removal processing is first performed. The purpose is to eliminate the slow frequency decline trend caused by normal particulate matter accumulation—this trend is a normal measurement signal rather than noise. If it is not removed, it will mask short-term noise characteristics and affect the accurate evaluation of frequency stability by Allan variance.

[0023] Allan variance calculation steps. For each candidate integral time... Calculate the Allan variance separately. The candidate integration time can be set according to actual needs, such as 10 seconds, 30 seconds, 60 seconds, 120 seconds, etc. The formula for calculating the Allan variance is:

[0024]

[0025] in, For the first indivual Average frequency over a period of time The partitionable parts in the current cache Number of time periods.

[0026] Allan variance was originally a time-domain characterization method used to analyze the frequency stability of oscillators. This invention applies it to the selection of integration time in the measurement of dust concentration using an oscillating balance. The smaller the Allan variance, the better the stability of the frequency sequence at that integration time. If the Allan variance no longer decreases or even increases over a longer integration time, it indicates that the signal has entered the drift-dominated region (the region dominated by systematic drift), and further extending the integration time will not help improve measurement accuracy.

[0027] The window scoring and integration time selection steps involve calculating a comprehensive score for each candidate integration time:

[0028] ,

[0029] in, This is the Allan variance normalized noise term, reflecting the frequency noise level; This represents the drift term between adjacent windows, reflecting the frequency drift trend; These are the temperature and humidity disturbance terms, reflecting changes in ambient temperature and humidity. This is the flow fluctuation term, reflecting the stability of the sampled flow. The weights of each term can be adjusted according to the specific application environment or sensor characteristics. A preferred weight configuration is as follows: , , , The optimal integration time is selected from the candidate integration times based on the comprehensive score. .

[0030] To avoid output instability caused by frequent integration time switching, this invention incorporates an integration time switching hysteresis mechanism: switching to a new integration time is only allowed when the overall score of the new candidate integration time improves by more than 15% relative to the overall score of the current integration time (or the absolute score difference exceeds 10% of the overall score of the current integration time), and the current integration time has been maintained for at least 30 seconds. Furthermore, if a sudden change in humidity, temperature, or flow rate is detected, the algorithm can force the selection of an integration time of at least 30 seconds to obtain more stable measurement results during environmental disturbances.

[0031] The dust concentration calculation steps, after determining the optimal integration time... Then, the dust concentration is calculated based on the frequency change and sampling volume within the integration window:

[0032] ;

[0033] in, This represents the change in the mass of dust adhering to the filter membrane. For calibration coefficients, The average oscillation frequency of the current window. This is the no-load oscillation frequency; This represents the sampling flow rate. It should be noted that in the formula... The sampling volume is represented by the sampling volume. However, in actual calculations, as a preferred implementation method, the sampling volume obtained by accumulating the actual flow within the integration window is used. This can avoid the sampling volume calculation error caused by short-term fluctuations in flow.

[0034] Anomaly detection steps. Based on one or more of the following: temperature, relative humidity, sampling flow rate, oscillation frequency, amplitude, and phase-locked state, determine whether an anomaly exists. In this embodiment, the criteria for detecting each anomaly are as follows:

[0035] Humidity abrupt change: A humidity abrupt change is defined as the absolute value of the difference between the current relative humidity and the relative humidity 60 seconds ago exceeding 5 percentage points. A relative humidity higher than 60% is defined as a high humidity state.

[0036] Temperature abrupt change: When the absolute value of the difference between the current temperature and the temperature 60 seconds ago exceeds 1°C, it is judged as a temperature abrupt change.

[0037] Traffic anomaly: When the deviation of the current traffic or the average traffic in the integral window from the rated traffic exceeds 3% but does not exceed 5%, it is judged as a minor traffic disturbance; when the deviation exceeds 5%, it is judged as a serious traffic anomaly.

[0038] Frequency jump: When the absolute value of the difference between the oscillation frequencies of two adjacent 1-second samples exceeds 0.25Hz, it is determined to be a frequency jump.

[0039] Mechanical vibration: When the standard deviation of the oscillation frequency exceeds 0.35 Hz in the last 10 seconds, or the standard deviation of the normalized amplitude exceeds 0.06 in the last 10 seconds, it is judged as mechanical vibration.

[0040] Amplitude anomaly: Normalize the amplitude to 1 with the nominal amplitude in a stable phase-locked state. If the normalized amplitude is below 0.75 or above 1.25, it is judged as amplitude anomaly.

[0041] Phase-locked failure: When the phase-locked status signal indicates a loss of lock, or when the controller cannot maintain the target phase difference within a predetermined time, it is determined to be a phase-locked failure.

[0042] The above thresholds can be calibrated and adjusted according to the natural frequency of the oscillating element, the normal noise baseline, the rated amplitude, and the application environment, and are not limited to the specific values ​​mentioned above.

[0043] Quality grading and output steps. Based on the anomaly determination results, each dust concentration value is marked with its corresponding data quality level, and the dust concentration, optimal integration time, data quality level, and anomaly cause code are output simultaneously.

[0044] Data quality is divided into four levels:

[0045] Grade A – Excellent data, indicating stable frequency, stable temperature and humidity, stable flow rate, and normal phase-locked loop, which can be displayed, recorded, and uploaded as official data.

[0046] Grade B – Available data, indicating that a relatively long points accumulation time is required but no obvious abnormalities occur, and the data can be uploaded normally and is eligible for statistics.

[0047] Level C – Reference data, indicating situations such as sudden changes in humidity, high humidity recovery period, or slight flow disturbances. The data is displayed as a reference only and is not included in formal alarm statistics.

[0048] Level D – Invalid data, indicating the presence of mechanical vibration, phase-locking abnormalities, amplitude abnormalities, frequency jumps, or severe flow abnormalities. This data will not be included in formal statistics or alarms.

[0049] When multiple abnormal states occur simultaneously, the lowest quality level is selected, and multiple corresponding abnormal reason codes are allowed to be output at the same time.

[0050] To prevent frequent quality level jumps, this invention incorporates a quality level anti-jitter mechanism: when recovering from a lower to a higher quality level, the higher quality level conditions must be met continuously for 5 seconds before switching is allowed; in the event of mechanical vibration, phase-locked loop anomaly, amplitude anomaly, frequency jump, or severe flow anomaly, the quality level immediately drops to level D. Furthermore, different recovery observation periods are set for different types of anomalies—a 25-second level D recovery observation period is maintained after severe anomalies such as mechanical vibration, phase-locked loop anomaly, or frequency jump disappear; a 60-second level C environmental recovery period is maintained after sudden humidity changes, high humidity, or temperature changes. These mechanisms ensure the stability and rationality of the output level.

[0051] 3. Beneficial effects:

[0052] Compared with the prior art, the technical solution provided by this invention has the following advantages:

[0053] (1) The algorithm can automatically select the optimal integration time based on the Allan variance of the oscillation frequency sequence. In a stable signal environment, the algorithm automatically selects a shorter integration time to improve the system's response speed to concentration changes; in an environment with significant signal disturbance or noise, the algorithm automatically selects a longer integration time or reduces the data quality level to avoid concentration misjudgment caused by short-term anomalies. Through simulation verification, compared with a fixed 60-second integration, the step response time of this invention is shortened from approximately 53 seconds to approximately 30 seconds, and the concentration error RMSE is reduced from approximately 0.1733 mg / m³. 3 Reduced to approximately 0.0388 mg / m³ 3 This fully demonstrates the significant advantages of the adaptive integration mechanism in balancing response speed and measurement accuracy.

[0054] (2) It can effectively distinguish abnormal situations such as sudden humidity changes, high humidity recovery period, flow disturbance, mechanical vibration, frequency jump, phase lock abnormality and amplitude abnormality from the formal concentration output. In these cases, the algorithm does not output formal A or B level data, but outputs C level reference data or D level invalid data, thereby greatly reducing false alarms and erroneous statistics.

[0055] (3) Rich output information. While outputting dust concentration, the integral time, data quality level and anomaly cause code are output simultaneously, enabling the display terminal, 485 communication terminal and host computer system to accurately judge the credibility of each concentration data, providing a basis for the graded use of data and intelligent decision-making.

[0056] (4) No additional hardware is required. This invention is based entirely on the existing frequency acquisition, temperature and humidity acquisition, flow acquisition and phase-locked state acquisition capabilities of the oscillating balance controller. It can be deployed on existing equipment through firmware upgrades, and has good compatibility and economy.

[0057] (5) By employing mechanisms such as integral time switching hysteresis, quality level anti-jitter, and anomaly recovery protection, frequent jumps in algorithm output are avoided. These engineered processing mechanisms ensure the stability and reliability of the algorithm during long-term operation in actual products, and are more in line with the actual usage requirements of field monitoring equipment.

[0058] It should be noted that the structures not described in this invention are not related to the design points and improvement directions of this invention, and are the same as or can be implemented using existing technologies, so they will not be elaborated here. Attached Figure Description

[0059] Figure 1 This is a flowchart illustrating the adaptive integral and quality classification method for dust concentration based on Allan variance of the present invention.

[0060] Figure 2 This diagram illustrates the variation of Allan variance under different candidate integration times and the final selection of the optimal integration time.

[0061] Figure 3 This diagram illustrates the correspondence between data quality levels and abnormal states.

[0062] Figure 4 This is a schematic diagram of the test scenario used in the simulation verification of this invention.

[0063] Figure 5 This is a comparison chart of concentration output under the same test scenario between a fixed 60-second integral and the adaptive integral of this invention. Detailed Implementation

[0064] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to illustrate the present invention and should not be regarded as limiting the scope of the present invention.

[0065] System Composition and Signal Acquisition

[0066] This embodiment uses a typical oscillating balance dust concentration monitor as the hardware platform. The instrument mainly consists of an oscillating conical tube (including a filter membrane), a frequency detection circuit, a temperature and humidity sensor, a mass flow controller, an amplitude detection and phase-locked loop circuit, and an embedded controller. The controller uses an ARM Cortex-M series microcontroller, which has sufficient computing power and storage resources to execute the algorithm of this invention.

[0067] Signal acquisition is performed as follows: the oscillation frequency f(t) is measured by the frequency detection circuit with a period of 1 second and sent to the controller; the sampled flow rate Q is fed back to the controller by the mass flow controller with a period of 1 second; the amplitude A and phase-locked loop status are provided by the amplitude detection and phase-locked loop circuit with a period of 1 second; the temperature T and relative humidity RH are acquired by the temperature and humidity sensor with a period of 1 to 5 seconds. In this embodiment, a 5-second period is used to reduce power consumption. All signals are timestamped in the controller and stored in the corresponding data buffer.

[0068] The purpose of each input signal is shown in the table below:

[0069]

[0070] Frequency caching and preprocessing

[0071] The controller establishes a frequency circular buffer in memory with a length of 300 seconds. Every second, as new frequency data is collected, the new data is written to the buffer, while the oldest data is overwritten.

[0072] Perform a data validity check immediately after each new data is written:

[0073] Frequency jump check. Calculate the absolute value of the difference between the current frequency and the frequency of the previous second. If this difference exceeds 0.25Hz, it is considered a frequency jump. This threshold is set based on the frequency change rate of a typical oscillating balance under normal operating conditions—under normal conditions, due to the slow accumulation of particulate matter, the frequency change per second is usually on the order of 0.01Hz. A jump of 0.25Hz clearly exceeds the normal range and is considered an abnormal event.

[0074] Phase-locked loop (PLL) anomaly check. Read the PLL status signal. If the PLL status indicates a loss of lock, or if the controller fails to maintain the target phase difference after a predetermined number of consecutive attempts, a PLL anomaly is determined.

[0075] Amplitude anomaly check. Read the current amplitude value and normalize it to 1 (the nominal amplitude in a stable phase-locked state). If the normalized amplitude is lower than 0.75 or higher than 1.25, it is considered an amplitude anomaly.

[0076] If any of the above checks fails, the concentration data at that moment is marked as Class D invalid data, and no further integration time selection or concentration calculation will be performed. The controller then enters the anomaly recovery protection process. The specific mechanism of the anomaly recovery protection will be detailed later.

[0077] For frequency data that passes the validity check, linear trend removal is required before calculating the Allan variance. Specifically, the frequency sequence in the buffer is subjected to least-squares linear fitting to obtain a trend line of frequency change over time. This trend line is then subtracted from the original frequency sequence to obtain the residual sequence after trend removal. The purpose is that the normal accumulation of particulate matter on the filter membrane causes a continuous and slow decrease in oscillation frequency, which is a "signal" rather than "noise." If this is retained in the sequence for Allan variance calculation, the Allan variance will primarily reflect this low-frequency trend rather than high-frequency noise, thus failing to accurately evaluate the noise level at different integration times.

[0078] Allan variance calculation

[0079] In this embodiment, the candidate integration time is set to four levels: 10 seconds, 30 seconds, 60 seconds, and 120 seconds. For each candidate integration time... The Allan variance is calculated from the preprocessed frequency residual sequence.

[0080] The specific calculation process is as follows: Assume that there are N data points in the current cache that are sampled at a frequency of 1 second (the actual cache length is 300 seconds, so N=300). For a given... Divide the N data points into as many lengths as possible. The time intervals (allowing overlap) are used to average the frequency values ​​within each time interval, resulting in M ​​average values ​​y(1), y(2), …, y(M). Then, the Allan variance is calculated using the following formula:

[0081] ,

[0082] Allan variance is used to characterize the stability of a frequency sequence at different time scales. If the Allan variance for a short integration time (e.g., 10 seconds) is small, it indicates that the frequency is sufficiently stable at a short time scale, and a shorter integration time can be chosen to obtain a faster response. If the Allan variance for a short integration time is large while the Allan variance for a long integration time (e.g., 60 seconds or 120 seconds) decreases significantly, it indicates the presence of strong short-term noise in the frequency, requiring a longer integration time for smoothing. If the Allan variance for a long integration time no longer decreases or even increases, it indicates that the signal has entered the drift-dominated region—at this point, frequency changes are mainly driven by systematic factors such as particulate matter accumulation rather than random noise, and further extending the integration time will not only fail to improve the signal-to-noise ratio but will also reduce the response speed.

[0083] Figure 2 The graph shows the variation of Allan variance under different candidate integration times and the final selected integration time. It can be seen from the graph that during periods of stable environment, the algorithm tends to choose shorter integration times (e.g., 10 seconds or 30 seconds); during periods of increased noise or disturbance, the algorithm tends to choose longer integration times (e.g., 60 seconds or 120 seconds).

[0084] Window rating and points time selection

[0085] Using only Allan variance to select the integration time is insufficient, as Allan variance only reflects the characteristics of frequency noise and does not consider other factors affecting measurement quality, such as changes in ambient temperature and humidity, and flow fluctuations. Therefore, this invention further introduces a comprehensive scoring mechanism.

[0086] For each candidate integration time Calculate the overall score:

[0087]

[0088] The meanings and calculation methods for each item are as follows:

[0089] —Allan variance normalized noise term. The Allan variances corresponding to the four candidate integration times are normalized (e.g., divided by the maximum value) to obtain normalized values ​​between 0 and 1. The smaller the Allan variance, the better. The larger the value, the lower the noise level and the better the integration time.

[0090] —Drift term between adjacent integration windows. Calculate the difference in the mean frequency between the current integration window and the previous integration window, normalize it, and use it as the drift term. The smaller the drift, the lower the drift. The larger.

[0091] —Temperature and humidity disturbance terms. Calculated based on the changes in current temperature and relative humidity relative to historical averages. The more stable the temperature and humidity, the better. The larger; when temperature and humidity change abruptly, This decreases the overall score.

[0092] —Flow fluctuation item. Calculated based on the absolute value of the deviation of the current flow rate from the rated flow rate. The more stable the flow rate, the better. The larger the value, the greater the flow deviation. The smaller.

[0093] One preferred configuration of the weights is as follows: , , , This weighting configuration is primarily based on frequency noise, while also taking into account drift, temperature, humidity, and flow rate. In practical applications, the weights can be adjusted according to the noise characteristics of the specific sensor and the application scenario.

[0094] After calculating the comprehensive score of each of the four candidate integration times, the one with the highest score is selected as the optimal integration time. .

[0095] To prevent frequent switching of the points accumulation period, a switching hysteresis mechanism is implemented. Specifically: assuming the currently used points accumulation period is... The newly calculated optimal integration time is Only when both of the following conditions are met is it permitted to [transfer from / begin]. Switch to (1) Compare (2) More than 15% higher (or the score difference exceeds a preset absolute threshold); It has been used continuously for at least 30 seconds. If the condition is not met, continue using it. .

[0096] Furthermore, when a sudden change in humidity, temperature, or flow rate is detected, the algorithm can force an integration time of no less than 30 seconds. This is done to obtain more stable measurement results during environmental disturbances and to avoid amplifying the effects of disturbances due to an excessively short integration time.

[0097] Dust concentration calculation

[0098] After confirming Then, the dust concentration was calculated according to the standard formula of the oscillating balance.

[0099] First, calculate The average value of all frequency samples within the time window is used to obtain the average oscillation frequency of the current window. Then, based on the basic calibration relationship of the oscillating balance, the change in the mass of dust on the filter membrane is calculated:

[0100]

[0101] in, These are calibration coefficients, determined by factory calibration or on-site calibration. The no-load oscillation frequency (i.e., the oscillation frequency when there is no dust on the filter membrane) is also determined during the calibration process.

[0102] In obtaining Then, calculate the dust concentration using the following formula:

[0103]

[0104] For sampling flow rate, This is the currently selected optimal integration time. This is the sampling volume within the integration window.

[0105] It should be noted that the sampling flow rate used in the above calculation method In practice, there are several ways to obtain the value. One of the simplest methods is to directly use the instrument's rated flow rate (e.g., the target flow rate set by the mass flow controller). This method is simple to implement, but it may introduce some volume calculation error when the actual flow rate fluctuates. A better implementation method is to accumulate the actual flow rate values ​​collected every second within the integration window to obtain the actual sampling volume. Then use Calculate the dust concentration.

[0106] The method of accumulating actual flow rates can effectively avoid the impact of short-term flow fluctuations on concentration calculation results, and is particularly suitable for situations where flow control accuracy is limited or where there are periodic flow fluctuations.

[0107] In this embodiment, since the actual flow accumulation method is equally simple to implement in engineering (it only requires accumulating the flow value per second within the integration window) and can improve the accuracy of volume measurement, the actual flow accumulation method is preferred for calculating the sampling volume.

[0108] Anomaly Detection

[0109] Anomaly detection is a crucial step in this invention. Its purpose is to identify various abnormal states that may affect the accuracy of measurements and to provide a basis for subsequent data quality grading.

[0110] The criteria for determining each anomaly and the preferred thresholds are as follows:

[0111] Humidity abrupt change. Compare the current relative humidity with the relative humidity 60 seconds ago. If the absolute value of the difference exceeds 5 percentage points, it is considered a humidity abrupt change. During a humidity abrupt change, the filter membrane may experience rapid adsorption or desorption of water vapor, causing the frequency change to be misinterpreted as a change in dust quality. It should be noted that some oscillating balance products eliminate humidity interference through full-process heating, but this requires additional heating hardware and power consumption. This invention does not rely on heating hardware, but instead uses an algorithm to identify humidity abrupt changes and accordingly reduce the data quality level.

[0112] High humidity. A relative humidity level above 60% is considered a high humidity condition. In a high humidity environment, water vapor adsorption on the filter membrane may continue, causing a systematically higher concentration measurement.

[0113] Temperature abrupt change. Compare the current temperature with the temperature 60 seconds ago. If the absolute value of the difference exceeds 1°C, it is considered a temperature abrupt change. Temperature changes affect the elastic modulus and geometry of the oscillating element, thus causing frequency drift.

[0114] Flow anomaly. Calculate the absolute percentage deviation of the current flow rate (or the average flow rate within the integration window) from the rated flow rate. A deviation exceeding 3% but not exceeding 5% is considered a minor flow disturbance; a deviation exceeding 5% is considered a severe flow anomaly. Flow deviation directly affects the accuracy of sampling volume calculation.

[0115] Frequency jump. A frequency jump is defined as an oscillation frequency difference exceeding 0.25 Hz between two consecutive 1-second samples. As mentioned earlier, this threshold is much larger than the frequency change rate under normal operating conditions and is used to detect sudden frequency anomalies.

[0116] Mechanical vibration. An oscillation frequency with a standard deviation exceeding 0.35 Hz or a normalized amplitude with a standard deviation exceeding 0.06 Hz within the last 10 seconds is considered mechanical vibration. Mechanical vibration can affect the oscillation state of the tapered tube through mechanical transmission, causing additional fluctuations in the frequency measurement.

[0117] Amplitude anomaly. Normalization is performed with the nominal amplitude in a stable phase-locked state as 1. An amplitude anomaly is defined as the current normalized amplitude being below 0.75 or above 1.25. Amplitude anomalies may originate from drive circuit failures, changes in the mechanical characteristics of the tapered tube, or external impacts. The nominal amplitude is determined during instrument factory calibration or dynamically updated after each phase-locked state stabilization.

[0118] Phase-locked loop (PLL) malfunction. A PLL malfunction is diagnosed when the PLL status signal indicates a loss of lock, or when the controller fails to maintain the target phase difference within a predetermined time. The PLL is the key circuit for maintaining the continuous oscillation of the tapered tube; a PLL malfunction means the oscillation state is uncontrollable, and the frequency measurement value is unreliable.

[0119] The above threshold values ​​are all preferred values. In practical applications, they can be calibrated and adjusted according to the natural frequency of the oscillating element, the normal noise baseline, the rated amplitude, and the application environment.

[0120] When multiple anomalies occur simultaneously, they are recorded independently and do not affect each other.

[0121] Data quality grading and output

[0122] Based on the anomaly assessment results, each concentration value is classified into quality grades. The grading criteria are as follows:

[0123] Grade A – Excellent data. No abnormal conditions were triggered; frequency, temperature, humidity, flow rate, and phase-locked loop were all stable. Grade A data can be displayed, recorded, and uploaded as official data.

[0124] Grade B – Usable Data. Although no abnormal state was triggered, the algorithm selected a longer integration time (e.g., 60 seconds or 120 seconds) due to the high signal-to-noise level. Grade B data can still be uploaded and used in statistics, but users will know that it has undergone a longer smoothing process.

[0125] Level C – Reference Data. This data indicates that one or more of the following events have been triggered: sudden humidity changes, high humidity conditions, or minor flow disturbances. Level C data is displayed as a reference only and is not included in formal alarm statistics, but it can be uploaded for user reference.

[0126] Level D – Invalid data. This data triggered one or more of the following: mechanical vibration, phase-locked loop anomaly, amplitude anomaly, frequency jump, or severe flow anomaly. Level D data is not included in formal statistics and alarms, and it is generally not recommended to display it to end users.

[0127] When multiple anomalies occur simultaneously, the lowest quality level is used. For example, if both a sudden humidity change (Level C) and a slight flow disturbance (Level C) exist simultaneously, Level C is used; if both a sudden humidity change (Level C) and mechanical vibration (Level D) exist simultaneously, Level D is used. Multiple anomaly cause codes can be output simultaneously, for example, " This facilitates fine-grained processing by the host computer.

[0128] To prevent frequent changes in quality levels, the following protection mechanisms are implemented:

[0129] Quality level stabilization. When the quality level recovers from a lower level (such as C or D) to a higher level (such as A or B), the conditions of the higher level must be met continuously for 5 seconds before switching is allowed. This mechanism avoids quality level switching due to brief fluctuations in abnormal conditions.

[0130] Immediate downgrade. When a serious anomaly occurs (such as mechanical vibration, phase-locked loop anomaly, frequency jump, etc.), the quality level is immediately downgraded to level D, with no anti-jitter delay. This is to ensure that the system can mark the data as invalid immediately when a real anomaly occurs.

[0131] Recovery observation period. Different recovery observation periods are set for different types of anomalies. After severe anomalies such as mechanical vibration, phase-locked loop anomalies, or frequency jumps disappear, a 25-second Class D recovery observation period is maintained—even if no new anomalies are detected during these 25 seconds, the data is still marked as Class D to ensure that the oscillation system has fully recovered to a stable state. After sudden changes in humidity, high humidity, or temperature, a 60-second Class C environmental recovery period is maintained—during these 60 seconds, the data is marked as Class C reference data to ensure complete temperature and humidity balance.

[0132] Figure 3 It illustrates the correspondence between quality levels and abnormal states, as well as the roles of the recovery period and anti-shake mechanism.

[0133] Finally, the controller outputs the following information via a communication interface (such as a RS-485 bus, serial port, or Ethernet): dust concentration value, optimal integration time. The output includes data quality level (A / B / C / D) and anomaly reason codes (if any). This multi-dimensional output method enables the host computer system to make differentiated processing decisions based on the data quality level—for example, only A and B level data are included in the daily average statistics, C level data is only used for reference, and D level data is discarded directly.

[0134] Simulation verification

[0135] To verify the effectiveness of this invention, simulation scenarios were constructed that included various operating conditions such as concentration step changes, humidity abrupt changes, flow disturbances, mechanical vibrations, and phase-locked loop anomalies. Figure 4 As shown, the fixed 60-second integration algorithm and the adaptive integration algorithm of this invention were compared and tested under the same set of frequency noise and disturbance conditions.

[0136] Step response test. When the concentration changes stepwise, the response time of a fixed 60-second integration is approximately 53 seconds (from the occurrence of the step to the output value reaching 90% of the target value), while the response time of the adaptive integration of this invention to the final output is approximately 30 seconds. This is because during the signal stabilization phase, this invention automatically selects a shorter integration time (such as 10 seconds or 30 seconds), thus providing a faster response to concentration changes.

[0137] Accuracy testing. During the stabilization phase, the concentration error RMSE with a fixed 60-second integral was approximately 0.1733 mg / m³. 3 The RMSE of the officially output A / B level data of this invention is approximately 0.0388 mg / m³. 3The improved accuracy stems from two factors: firstly, the adaptive selection of the optimal integration time, and secondly, the marking of outlier data as C or D grade, preventing its inclusion in the formal output statistics.

[0138] Anomaly Handling Test. During periods of sudden humidity changes, flow disturbances, mechanical vibration, and phase-locked loop anomalies, this invention marks the data as Grade C reference or Grade D invalid, thereby preventing abnormal data from being included in formal statistics. However, a fixed 60-second integral still outputs formal concentration values ​​under the same conditions, leading to deviations in the statistical results.

[0139] Figure 5 The concentration output curves of the fixed 60-second integral and the adaptive integral of this invention were compared under the same test scenario. The graph clearly shows that the fixed integral responds slowly during concentration steps and still outputs outliers during disturbances; while the adaptive integral maintains a fast response in the stable phase and suppresses the normal output or marks the data as reference / invalid during disturbances, thus effectively improving data reliability while ensuring response speed.

[0140] Implementation of variant instructions

[0141] The thresholds described in the above embodiments (such as a sudden change in humidity of 5 percentage points, a sudden change in temperature of 1°C, a frequency jump of 0.25 Hz, and mechanical vibration of 0.35 Hz) are preferred values ​​determined experimentally by those skilled in the art on a typical oscillating balance platform. In practical applications, these thresholds can be adjusted according to the characteristics of the specific sensor.

[0142] For oscillating elements with a high inherent frequency, the threshold for frequency jumps can be appropriately increased;

[0143] For instruments operating in high humidity environments, the threshold for determining high humidity (60%) can be adjusted;

[0144] For applications with high requirements for response speed, shorter increments (such as 5 seconds) can be added to the candidate integration time.

[0145] For scenarios with high stability requirements, longer speed settings can be added (such as 180 seconds or 300 seconds).

[0146] In addition, the weighting of the overall score ( , , , It is also adjustable. For example, in application environments with drastic temperature and humidity changes, the weight of We can be appropriately increased; on equipment with insufficient flow control, the weight of Wq can be increased.

[0147] All of the above modifications do not depart from the basic concept and scope of protection of this invention.

[0148] Industrial applicability

[0149] This invention can be widely applied to various types of oscillating balance dust concentration monitoring equipment, including but not limited to:

[0150] Online PM10 and PM2.5 monitoring instruments at automatic ambient air quality monitoring stations;

[0151] Online dust concentration monitoring system for industrial production sites;

[0152] Coal mine underground dust concentration monitoring device;

[0153] Equipment for monitoring particulate matter emissions from stationary pollution sources.

[0154] Since this invention is entirely based on an algorithm and does not involve any hardware modifications, it can be deployed on existing equipment through firmware upgrades, thus exhibiting good industrial applicability and application prospects.

[0155] The above-described embodiments are merely illustrative of certain implementations of the present invention, and are described in a relatively specific and detailed manner. However, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements are all within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. An Allan variance based oscillating balance dust concentration adaptive integration and mass classification method, characterized in that, include: Signal acquisition step, with a 1 second cycle, of the oscillation frequency of the oscillating balance , temperature , relative humidity , sampling flow , amplitude and phase-locked state; The frequency buffering and preprocessing steps involve establishing a frequency ring buffer to judge the jump and phase-lock anomaly of each acquired oscillation frequency. If the current data meets the conditions for loss of lock or frequency jump, the concentration at that moment will not be used as the formal output, and the anomaly recovery protection process will be triggered. During the anomaly recovery, the formal concentration output will be suspended. After the anomaly recovery, the corresponding level data will be output according to the quality grading steps. The frequency sequence entering the Allan variance calculation is subjected to least squares linear fitting to remove the low-frequency trend caused by normal accumulation of particulate matter. Allan variance calculation steps, on candidate integration times Allan variances are computed separately: ， wherein, is the average frequency of the first number of time periods, is the average frequency of the second number of time periods, is the number of time periods that can be divided out of the current cache. The window scoring and integration time selection steps involve calculating a comprehensive score for each candidate integration time: ， wherein, is a Allan variance normalized noise term, is a neighboring window drift term, is a temperature and humidity disturbance term, is a flow fluctuation term; , , , is a corresponding weight; the optimal integration time is selected from the candidate integration times according to the comprehensive score ; Dust concentration calculation step, according to the optimal integration time Dust concentration is calculated from the amount of frequency change and the sampling volume within the corresponding integration window : ； wherein, is the change in the mass of dust adhering to the filter membrane, is a calibration factor, is the current window average oscillation frequency, is the empty load oscillation frequency, the sampling volume being determined from the sampling flow and the optimal integration time; The anomaly detection step determines whether an anomaly exists based on at least one of the following: temperature, relative humidity, sampling flow rate, oscillation frequency, amplitude, and phase-locked state. The quality grading and output steps mark each dust concentration value as the corresponding data quality level based on the anomaly judgment result, and simultaneously output the dust concentration, optimal integration time, data quality level and anomaly reason code.

2. The method according to claim 1, characterized in that, The candidate integration time is 10s, 30s, 60s, or 120s; the weight in the comprehensive score is... , , , The window scoring and integration time selection steps are configured with an integration time switching hysteresis: the integration time switching is allowed only after the comprehensive score of the new candidate integration time is improved by more than 15% relative to the comprehensive score of the current integration time, or the absolute score difference exceeds 10% of the comprehensive score of the current integration time, and the current integration time has been maintained for at least 30 seconds.

3. The method according to claim 1, characterized in that, In the anomaly determination step, a humidity change is determined when the absolute value of the difference between the current relative humidity and the relative humidity 60 seconds ago exceeds 5 percentage points; a high humidity state is determined when the relative humidity is higher than 60%; a temperature change is determined when the absolute value of the difference between the current temperature and the temperature 60 seconds ago exceeds 1°C; if a humidity change, temperature change, or flow abnormality is detected, an integration time of not less than 30 seconds is forcibly selected.

4. The method according to claim 1, characterized in that, In the anomaly determination step, the method for determining flow anomalies is as follows: when the absolute value of the deviation between the current flow or the average flow of the integral window and the rated flow exceeds 3% but does not exceed 5%, it is determined to be a slight flow disturbance; when the absolute value of the deviation exceeds 5%, it is determined to be a serious flow anomaly; when the absolute value of the difference between the oscillation frequencies of two adjacent samples exceeds 0.25Hz, it is determined to be a frequency jump.

5. The method according to claim 1, characterized in that, In the anomaly determination step, mechanical vibration is determined when the standard deviation of the oscillation frequency exceeds 0.35 Hz in the last 10 seconds, or when the standard deviation of the normalized amplitude exceeds 0.06 in the last 10 seconds; normalization is performed with the nominal amplitude in the stable phase-locked state as 1, and the current normalized amplitude is determined to be below 0.75 or above 1.25; phase-locked state signal indicates loss of lock or the controller cannot maintain the target phase difference within a predetermined time, which is determined to be a phase-locked anomaly.

6. The method according to claim 1, characterized in that, In the quality grading and output steps, the data quality levels include: Grade A: Excellent data, indicating stable frequency, stable temperature and humidity, stable flow rate, and normal phase lock. Grade B: Available data, indicating that a relatively long integration time is required but no obvious anomalies are found; Grade C: Reference data, indicating the presence of sudden humidity changes, high humidity recovery period, or slight flow disturbances; Grade D: Invalid data, indicating the presence of mechanical vibration, phase-locking abnormality, amplitude abnormality, frequency jump, or severe flow abnormality; The data quality levels decrease sequentially from A to D.

7. The method according to claim 6, characterized in that, In the quality grading and output steps, when multiple abnormal states occur simultaneously, the lowest data quality level is taken, and multiple corresponding abnormal reason codes are output simultaneously. The abnormal recovery protection process includes: a 25s-second Class D recovery observation period after the disappearance of mechanical vibration, phase-locking abnormality, amplitude abnormality, frequency jump, or severe flow abnormality; and a 60s-second Class C environmental recovery period after the end of humidity change, high humidity, or temperature change.

8. The method according to claim 6, characterized in that, In the quality grading and output steps, a grade anti-jitter mechanism is set: when the data quality grade recovers from a lower grade to a higher grade, the higher grade conditions must be met continuously for 5 seconds before switching is allowed; when mechanical vibration, phase-locking abnormality, amplitude abnormality, frequency jump or severe flow abnormality occurs, it will immediately drop to grade D.

9. The method according to claim 1, characterized in that, In the dust concentration calculation step, the length of the frequency ring buffer is 300s; in the signal acquisition step, the oscillation frequency, sampling flow rate, amplitude and phase-locked state are acquired in 1s cycles, and the temperature and relative humidity are acquired in 1s to 5s cycles.

Images