Electrostatic dust concentration measurement method and system based on spectrum analysis

Abstract

The application discloses a kind of electrostatic dust concentration measurement method and system based on spectrum analysis, belong to dust concentration measurement technical field, the present application is based on traditional electrostatic induction measurement method, by deep mining spectrum information, electrode pollution spectrum diagnosis and self-healing are realized, quantitative evaluation pollution index and realize signal reconstruction, non-steady-state condition adaptive measurement, dynamic adjustment analysis window and directly output mass flow, life cycle management, measurement precision is significantly improved.

Description

Technical Field

[0001] This invention relates to the field of dust concentration measurement technology, and more specifically, to a method and system for measuring electrostatic dust concentration based on spectrum analysis, which is particularly suitable for online measurement of dust concentration in industrial emission monitoring, occupational health protection, and process control. Background Technology

[0002] Currently, electrostatic dust concentration meters are widely used in industrial flues, mine ventilation, cleanrooms, and environmental monitoring. These instruments reflect dust concentration by detecting the induced current generated by charged particles in an electric field, offering advantages such as fast response, simple structure, and long-term online operation. However, in complex environments, factors such as dust particle size distribution, flow field disturbances, electrode contamination, and power supply noise can cause significant fluctuations in the induced signal, resulting in unstable measurement results or deviations from the actual concentration.

[0003] Existing methods for measuring electrostatic dust concentration mostly employ time averaging or signal amplitude analysis, relying solely on the overall current output by the sensor as a concentration indicator, failing to fully utilize the spectral characteristics of the signal. Since particles of different sizes, velocities, and charge states correspond to different frequency characteristics during the sensing process, ignoring the spectral structure leads to decreased recognition capability, especially at low concentrations or in complex flow conditions, easily resulting in zero-point drift, insufficient sensitivity, and poor anti-interference ability.

[0004] On the other hand, some improvement solutions attempt to extract some spectral energy by adding filter circuits or fixed-band amplifiers. However, this method has a fixed frequency band, poor adaptability to environmental changes, and cannot achieve automatic compensation and long-term calibration. Especially in industrial settings, fluctuations in temperature and humidity and changes in flow rate can cause shifts in the signal's characteristic frequency band, leading to the rapid failure of static calibration models.

[0005] Therefore, there is an urgent need for a method for measuring electrostatic dust concentration based on spectrum analysis, which can achieve stable, accurate and traceable measurement of dust concentration by dividing the frequency band and extracting features, combined with dynamic compensation of environmental parameters. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for measuring electrostatic dust concentration based on spectrum analysis. Through systematic extraction of spectrum features and fusion of multi-dimensional information, it achieves high-precision, high-reliability, and low-maintenance online monitoring of dust concentration.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] In a first aspect, the present invention provides a method for measuring electrostatic dust concentration based on spectrum analysis, comprising the following steps:

[0009] S1: Signal Acquisition and Front-End Protection

[0010] Raw electrical signals are collected from electrostatic dust sensors, and temperature, relative humidity, airflow, and equipment status parameters are recorded simultaneously. Overload and saturation protection circuits are set at the front end, and obviously abnormal spike signals and power-down segments are automatically marked to ensure the effectiveness of subsequent analysis.

[0011] In this step, the signal acquisition frequency should not be lower than 5kHz to ensure complete acquisition of high-frequency disturbance signals. The overload protection circuit uses a bidirectional transient voltage suppression diode or a gas discharge tube, and the saturation protection circuit uses automatic gain control or a limiting amplifier. The criteria for judging abnormal spike signals include: signal amplitude exceeding the normal range by more than 5 times, duration less than 1ms, and waveform shape significantly different from normal signals. The criteria for judging power-down segments include: signal amplitude consistently below the noise floor, and the device status register reading indicating a power-down state.

[0012] S2: Preprocessing and Steady-State Section Identification

[0013] The original electrical signal is subjected to baseline stabilization, trend removal, and power frequency notch filtering to obtain a clean time-domain sequence. Steady-state periods are identified based on sliding statistical criteria, and data segments with drastic flow fluctuations, alarm states, or maintenance states are filtered out to output an effective time window that can be used for spectrum analysis.

[0014] Baseline stabilization uses a moving average or median filtering method to subtract the DC bias and ultra-low frequency drift of the signal. Trend removal uses a high-pass filter to filter out the sensor's own thermal drift and slow changes. Power frequency notch filtering targets 50Hz or 60Hz power frequency interference and its harmonics, using notch filters or adaptive noise cancellation techniques.

[0015] The specific method for identifying steady-state segments is as follows: A sliding window is used to calculate the statistical characteristics of the time-domain sequence, including the variance, mean, and rate of change of the signal within the window. Based on a comparison of the statistical characteristics with preset thresholds, it is determined whether the current window is a steady-state period. The criteria for determining a steady-state period include: the variance of the signal within the window is lower than a first threshold, and the rate of change of the signal is lower than a second threshold. Data segments in non-steady-state states are filtered out. Non-steady-state states include instantaneous rate of change of gas flow exceeding the flow threshold, the system being in an alarm state, or the system being in a manual or automatic maintenance state. Multiple consecutive windows determined to be steady-state are spliced ​​together to form the minimum effective window whose length meets the requirements of spectral analysis. When the effective window length is insufficient, data from adjacent windows are used to supplement the current period or the current period is marked as an invalid analysis period.

[0016] S3: Spectrum Transformation and Background Noise Spectrum Estimation

[0017] The time-domain signal of the effective time window is converted to the frequency domain to obtain the spectral amplitude distribution; an adaptive strategy is used to estimate the background noise spectrum and the instrument's background noise, and corresponding reference curves are generated for subsequent spectral denoising and threshold setting.

[0018] The spectrum transformation employs either Fast Fourier Transform or Wavelet Transform to obtain a spectrum distribution that includes amplitude and phase information.

[0019] Adaptive strategy estimation of background noise spectrum includes at least one of the following methods:

[0020] Based on the spectrum corresponding to the period when the concentration is below the concentration threshold in historical data, its statistical average value is calculated as the background noise spectrum.

[0021] The minimum value tracking method of multi-frame spectrum is adopted to extract the minimum value of the spectrum of multiple consecutive time windows on a frequency-by-frequency basis to obtain the background noise spectrum estimate;

[0022] Based on the instrument's background noise spectrum at the time of manufacture, and combined with the sensor's cumulative operating time and pollution index, the background noise spectrum is dynamically updated through an aging model.

[0023] The background noise spectrum is compared with the current spectrum to generate a reference curve for spectral denoising and a threshold curve for signal detection. The threshold curve is used to distinguish between effective signal components and noise components, and is usually set to a multiple (such as 3 or 5 times) of the background noise spectrum.

[0024] S4: Frequency Band Division and Feature Extraction

[0025] Based on the sensor's structural characteristics and typical particle size-velocity response relationship, the spectral amplitude distribution is divided into multiple functional frequency bands, which include at least two of the following: low-frequency operating condition band, mid-frequency collision band, and high-frequency charge disturbance band. Spectral features are extracted from each functional frequency band to form a feature vector, which is then spliced ​​with the environmental features of temperature, relative humidity, and flow rate within the same time window.

[0026] The functional frequency bands are divided according to the following criteria:

[0027] Low-frequency operating band: frequency range 0.1-20Hz, used to characterize the collision induction between dust particles and electrodes, and is related to dust concentration and particle size distribution;

[0028] Mid-frequency collision band: frequency range 20-60Hz, used for 50Hz power supply interference, pipeline vibration and mechanical operating condition changes;

[0029] High-frequency charge disturbance band: frequency range 60-1000Hz, used to characterize magnetic field interference, transient electromagnetic pulses, etc.

[0030] The spectral features extracted within each functional frequency band specifically include:

[0031] Total energy: the sum of squares or the area of ​​the integral of the amplitudes at all frequencies within this frequency band;

[0032] Main peak location: The frequency value corresponding to the maximum amplitude within this frequency band;

[0033] Peak-valley contrast: the ratio of the amplitude of the main peak to the amplitude of the adjacent valley, or the ratio of the amplitude of the main peak to the average amplitude of the frequency band;

[0034] In-band energy ratio: The ratio of the total energy in this frequency band to the total energy of the entire frequency band;

[0035] Inter-band energy ratio: The ratio between the total energy of different functional frequency bands, including the energy ratio between the mid-frequency band and the low-frequency band, and the energy ratio between the high-frequency band and the mid-frequency band;

[0036] Spectral flatness: The ratio of the geometric mean to the arithmetic mean of the spectral amplitudes within a frequency band;

[0037] Spectral slope: The rate of change of the spectral amplitude with frequency within this frequency band, obtained by linear fitting slope in logarithmic coordinate system.

[0038] The above spectral features are normalized and then concatenated with the environmental features of temperature, relative humidity, and flow rate within the same time window to form a complete feature vector for model input.

[0039] S5: Establishment and Online Calibration of Spectrum-Concentration Mapping Model

[0040] Based on standard dust samples or historical calibration data, a mapping model from the feature vector to dust concentration is established, and a built-in model validity detection mechanism and offset correction mechanism are implemented. When low dust or clean reference conditions are met, the background noise spectrum and zero-point offset parameters are automatically updated. When environmental parameters or flow distribution change, the adaptive adjustment of model parameters is triggered.

[0041] The mapping model is built using at least one of the following algorithms: multiple linear regression, support vector regression, random forest, gradient boosting tree, neural network, or deep learning model. During model training, the feature vector is used as input, and the concentration value measured by the standard reference method is used as output. The prediction error is minimized through an optimization algorithm.

[0042] The model effectiveness detection mechanism includes: periodically calculating the deviation between the model's predicted values ​​and the standard reference values, calculating the time series stability index of the predicted values, and calculating the matching degree between the model's input features and the feature distribution of the modeling dataset. When the deviation exceeds the deviation threshold, the stability falls below the stability threshold, or the matching degree falls below the matching degree threshold, the model calibration process is triggered.

[0043] Low dust or clean reference conditions include: the concentration is consistently below the concentration threshold (e.g., 1 mg / m³) for more than a first time threshold (e.g., 10 minutes), the gas flow rate is within a stable range, and no abnormal operating conditions occur. When the above conditions are met, the current spectrum is automatically used as the new background noise spectrum, and the current model output is compared with the theoretical zero point to update the zero-point offset parameters.

[0044] Model parameter adaptation includes: when an environmental parameter or flow distribution change exceeding an environmental change threshold is detected, fine-tuning some weight parameters of the model using an online learning algorithm, including stochastic gradient descent or incremental learning. During the adaptation process, the deviation between the model output and the auxiliary reference measurement is monitored; when the deviation exceeds a safety threshold, the adaptation is terminated and the model parameters are restored to their original state.

[0045] S6: Combined environmental and fluid regime compensation

[0046] Joint compensation is implemented for spectral drift caused by temperature, relative humidity, flow rate, power-on duration, and recent cleaning and maintenance status; when the flow rate change exceeds the preset threshold, the characteristic frequency band is dynamically weighted and adjusted according to the preset response law to reduce the estimation deviation caused by flow state changes; when electrode contamination or signs of heavy dust accumulation are detected, the noise reduction intensity is automatically increased and maintenance prompts are generated.

[0047] Environmental compensation employs a multivariate function model: C_compensated = C_raw × f(T, RH, Q, t_on, clean_status), where T is temperature, RH is relative humidity, Q is flow rate, t_on is power-on duration, and clean_status is the most recent cleaning and maintenance status. The compensation function f is obtained through experimental calibration and can be implemented using multinomial regression or a lookup table method.

[0048] Dynamic weighted adjustment includes: when the flow rate change exceeds the flow rate change threshold, dynamically adjusting the weight coefficients of each functional frequency band in the feature vector according to the pre-calibrated flow-frequency band response relationship. The flow-frequency band response relationship is as follows: when the flow rate increases, the weight of the low-frequency operating band decreases, while the weights of the mid-frequency collision band and the high-frequency charge disturbance band increase. The weight adjustment uses linear interpolation or a sigmoid function transition to avoid abrupt changes.

[0049] Detection of electrode contamination or increasing dust accumulation includes: monitoring the attenuation rate of high-frequency charge disturbance energy, monitoring changes in the noise floor of a specific frequency band, and monitoring trends in spectral flatness. When contamination indicators exceed the contamination threshold, the noise reduction intensity of the digital filter is automatically increased, and maintenance prompts are generated, including recommended cleaning times and methods.

[0050] S7: Output Integration, Quality Control and Lifecycle Management

[0051] The system performs weighted fusion of concentration estimates from multiple consecutive time windows, outputs the final concentration value and corresponding confidence level, and simultaneously outputs quality indicators. Key parameters, alarm records, and maintenance records are written to non-volatile storage media to form a traceable lifecycle archive, and automatically prompts for review when abnormal trends are detected.

[0052] Weighted fusion employs at least one of the following methods:

[0053] Time decay weighting: The concentration estimates of the most recent N time windows are assigned weights that decay over time, with more recent windows having higher weights;

[0054] Confidence weighting: A weighting coefficient is calculated based on the quality label of each window, with higher quality windows having higher weights;

[0055] Adaptive Kalman filtering: The concentration estimates of a continuous window are used as the observation sequence, and the optimal estimate is obtained through the Kalman filtering algorithm.

[0056] Quality labels include:

[0057] Percentage of valid data: The proportion of time windows deemed valid within the current fusion cycle out of the total number of windows;

[0058] Background estimation error: the degree of deviation between the current background noise spectrum and the historical background noise spectrum, or the degree of deviation between the current background noise spectrum and the factory background noise spectrum;

[0059] Model consistency score: The degree of matching between the current input feature vector and the feature distribution of the modeling dataset, calculated using Mahalanobis distance or local outlier factor.

[0060] The lifecycle profile includes: full lifecycle operational data of the sensor, calibration records, maintenance records, alarm event records, and key parameter evolution trajectories. Abnormal trends include at least one of the following: accelerated zero-point drift, continuously increasing background noise, accelerated response decay in a specific frequency band, and decreased recovery effectiveness after cleaning. When an abnormal trend is detected, an early warning message is generated, prompting manual review or planned maintenance.

[0061] S8: Adaptive Measurement for Unsteady Conditions

[0062] This step is optional and is used to extend the applicability of the method to unsteady conditions.

[0063] The instantaneous rate of change of the signal is monitored in real time, and the operating condition index is calculated; when the operating condition index exceeds the steady-state threshold, it is determined that the system has entered an unsteady-state operating condition.

[0064] Under non-steady-state conditions, the length of the spectrum analysis window is dynamically adjusted according to the magnitude of the operating condition index. The window length is negatively correlated with the operating condition index, and the shortest window length is not less than the minimum window length required to ensure the lowest frequency resolution (e.g., 0.2 seconds, corresponding to a 5Hz frequency resolution).

[0065] Within the dynamically adjusted window, spectral transformation is performed to extract transient spectral features, and the spectral evolution trajectory of multiple consecutive windows is recorded. The evolution trajectory includes the temporal changes of energy in each frequency band, the offset trajectory of the main peak position, and the temporal evolution parameters of the spectral shape.

[0066] An end-to-end mapping model from transient spectral evolution features to mass flow rate is established, directly outputting the mass flow rate measurement in kg / s. The model can employ a long short-term memory network or a temporal convolutional network, with the input being a sequence of feature vectors from multiple consecutive transient windows, and the output being the mass flow rate at the current moment.

[0067] When the operating condition index falls below the steady-state threshold and remains below it for a preset time (e.g., 3 seconds), the system is determined to return to steady-state operating conditions, gradually restores the standard window length, and switches the measurement mode from mass flow rate back to concentration measurement.

[0068] Secondly, the present invention provides an electrostatic dust concentration measurement system based on spectrum analysis, comprising:

[0069] Sensor module: Includes electrostatic dust sensor, temperature sensor, humidity sensor, and flow sensor, used to collect raw electrical signals and environmental and operating parameters. The electrostatic dust sensor adopts one of the following structures: ring electrode, rod electrode, or flat electrode, and the sensor material is stainless steel, titanium alloy, or metal with a surface coating of wear-resistant material.

[0070] Signal processing module: Connected to the sensor module, it includes overload protection circuit, saturation protection circuit, filtering circuit, and analog-to-digital conversion circuit, used for front-end protection and preprocessing of the raw electrical signal. The signal processing module also includes:

[0071] Programmable gain amplifier, used to automatically adjust the amplification factor according to the signal strength;

[0072] Anti-aliasing filters are used to filter out frequency components above the Nyquist frequency before analog-to-digital conversion;

[0073] Isolation circuits are used to achieve electrical isolation between signal ground and digital ground, and to suppress common-mode interference.

[0074] Spectrum analysis and feature extraction module: Connected to the signal processing module, it is used to convert the time-domain signal to the frequency domain, divide the functional frequency band, extract spectral features, and form a feature vector.

[0075] Storage module: Used to store historical data, calibration data, model parameters, background noise spectrum and lifecycle archives.

[0076] Central Processing Module: Connected to the signal processing module, spectrum analysis and feature extraction module, and storage module respectively, the central processing module is configured to execute the steps of the method described in the first aspect of the present invention. The central processing module includes at least one of a digital signal processor, a field-programmable gate array, or an embedded microprocessor, and has a built-in fast Fourier transform hardware acceleration unit for real-time spectrum calculation.

[0077] Output and Interaction Module: Connected to the central processing module, this module outputs dust concentration measurements, confidence levels, quality indicators, maintenance prompts, and early warning information. The output and interaction module includes a display screen, indicator lights, an audible and visual alarm, a wired communication interface, and a wireless communication interface. The wired communication interface includes RS485, Ethernet, or CAN bus, while the wireless communication interface includes 4G / 5G, Wi-Fi, or LoRa.

[0078] Cleaning execution module: Connected to the central processing module, it automatically performs backflushing or cleaning operations based on the contamination diagnosis results. The cleaning execution module includes a high-pressure gas backflushing device and / or a cleaning fluid spraying device, and is equipped with an intelligent control unit that automatically adjusts the cleaning intensity and duration according to the contamination index.

[0079] Beneficial effects:

[0080] Compared with the prior art, the present invention has the following significant advantages:

[0081] (1) Measurement accuracy is significantly improved

[0082] This invention utilizes a systematic spectral feature extraction method to fully leverage multi-dimensional information in electrostatic induction signals, including energy distribution, peak position, and spectral morphology. Compared to traditional methods that only utilize signal amplitude, this method offers a stronger characterization capability for dust concentration. The frequency band division strategy enables the model to distinguish signal components generated by different physical mechanisms, effectively suppressing operational interference.

[0083] (2) The anti-interference ability is greatly enhanced.

[0084] This invention effectively suppresses interference from factors such as temperature drift, humidity, flow fluctuations, and electrode contamination through multiple mechanisms including adaptive background noise spectrum estimation, joint environmental and flow regime compensation, and dynamic frequency band weighting. It maintains stable measurement performance even in complex industrial environments. In particular, it exhibits excellent suppression capabilities against common industrial interferences such as power frequency interference and mechanical vibration interference.

[0085] (3) Intelligent maintenance reduces operation and maintenance costs

[0086] This invention monitors electrode contamination status in real time through spectrum analysis, quantifies and assesses the contamination index, and dynamically optimizes cleaning strategies based on the degree of contamination, achieving predictive maintenance. Cleaning is automatically triggered when the contamination index exceeds a threshold, avoiding the blindness of regular manual maintenance. In cases of mild contamination where immediate maintenance is not possible, a signal reconstruction algorithm extends the effective measurement cycle, significantly reducing the frequency and cost of on-site maintenance.

[0087] (4) Reliable measurement throughout the entire life cycle

[0088] This invention achieves online self-calibration through a built-in model validity detection and offset correction mechanism, ensuring the measurement accuracy of the sensor during long-term operation. A digital lifecycle archive records the sensor's entire lifecycle evolution trajectory, providing a basis for data validity verification. The introduction of confidence levels and quality labels enables users of monitoring data to accurately assess the reliability of the data, providing reliable support for environmental enforcement and process control.

[0089] (5) Breakthrough in applicability to unsteady operating conditions

[0090] This invention overcomes the limitation of traditional methods, which are only applicable to steady-state conditions, by employing an adaptive measurement procedure for unsteady-state operating conditions. In unsteady-state processes such as material blockage, chute fluctuations, and unstable pneumatic conveying, high-precision measurement is achieved by dynamically adjusting the analysis window and establishing a direct mass flow mapping model, filling a technological gap in this field.

[0091] (6) Strong standardization and compatibility

[0092] The system architecture of this invention adopts a modular design with clear interfaces between modules, facilitating integration into existing monitoring systems. It supports multiple communication protocols, enabling easy interface with upper-level platforms such as environmental data platforms, distributed control systems, and manufacturing execution systems. The sensor structure is compatible with the installation dimensions of existing electrostatic sensors, allowing for in-situ upgrades of older equipment. Attached Figure Description

[0093] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below.

[0094] Figure 1 The overall flowchart of the electrostatic dust concentration measurement method based on spectrum analysis provided in the embodiments of the present invention is shown.

[0095] Figure 2 This is a detailed flowchart of the preprocessing and steady-state segment identification in an embodiment of the present invention.

[0096] Figure 3 This is a schematic diagram of spectrum transformation and background noise spectrum estimation in an embodiment of the present invention.

[0097] Figure 4 This is a schematic diagram of frequency band division and feature extraction in an embodiment of the present invention.

[0098] Figure 5 This is a flowchart illustrating the establishment and online calibration of the spectrum-concentration mapping model in an embodiment of the present invention.

[0099] Figure 6 This is a schematic diagram of the combined compensation of environment and flow regime in an embodiment of the present invention.

[0100] Figure 7 This is a flowchart of output fusion, quality control, and lifecycle management in an embodiment of the present invention.

[0101] Figure 8 This is a detailed flowchart of adaptive measurement under unsteady conditions in an embodiment of the present invention.

[0102] Figure 9 The structural block diagram of the electrostatic dust concentration measurement system based on spectrum analysis provided in the embodiments of the present invention is shown. Detailed Implementation

[0103] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are only for explaining the present invention and are not intended to limit the scope of protection of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0104] Example 1:

[0105] This embodiment provides an electrostatic dust concentration measurement system based on spectrum analysis, such as... Figure 9 As shown, it includes the following modules:

[0106] Sensor module: Includes an electrostatic dust sensor, temperature sensor, humidity sensor, flow sensor, and pressure sensor. The electrostatic dust sensor uses a ring electrode structure with titanium alloy electrodes coated with a ceramic layer. The temperature sensor uses a PT100 platinum resistance thermometer with a measurement range of -40-150℃. The humidity sensor uses a polymer capacitive type with a measurement range of 0-100%RH. The flow sensor uses a thermal mass flow meter with a measurement range of 0-30 m / s.

[0107] Signal processing module: Connects to the sensor module, including:

[0108] Overload protection circuit: Uses a bidirectional TVS diode and a PTC thermistor in series, with a clamping voltage of ±5V.

[0109] Saturation protection circuit: Employs a programmable gain amplifier with a gain range of 1-128 times and automatic adjustment.

[0110] Anti-aliasing filter: Fourth-order Butterworth low-pass filter, cutoff frequency 5kHz

[0111] Isolation circuit: Employs a magnetic coupler isolator with an isolation voltage of 2500Vrms.

[0112] Analog-to-digital converter: 24-bit Δ-Σ type ADC, sampling frequency 20kHz

[0113] Spectrum analysis and feature extraction module: Implemented on an FPGA, with a built-in FFT hardware acceleration core, capable of parallel processing of 1024-point FFTs in less than 0.1ms. This module receives the digital signal output from the signal processing module in real time, completes frequency band division and feature extraction, and outputs a set of feature vectors per second.

[0114] Storage modules include DDR3 SDRAM (1GB) for temporary data caching, eMMC flash memory (8GB) for storing historical data and model parameters, and ferroelectric memory (FRAM) for storing critical lifecycle parameters.

[0115] Central Processing Module: Employs an ARM Cortex-A9 dual-core processor with an 800MHz clock speed, running an embedded Linux operating system. The central processing module is configured to execute the method steps of this invention, including steady-state segment identification, background noise estimation, mapping model calculation, environmental compensation, quality control, and lifecycle management. The module incorporates a watchdog timer and a real-time clock to ensure long-term reliability.

[0116] Output and Interaction Modules:

[0117] Display screen: Industrial-grade touch screen, displaying real-time concentration, trend curves, and equipment status.

[0118] Indicator lights: Power indicator, Operation indicator, Alarm indicator (red, yellow, and green)

[0119] Audible and visual alarm: buzzer + rotating warning light

[0120] Wired communication: The RS485 interface supports the Modbus RTU protocol, and the Ethernet interface supports Modbus TCP and MQTT protocols.

[0121] Wireless communication: The 4G module supports all network networks and is used for remote data transmission and alarm push notifications.

[0122] Cleaning execution module:

[0123] High-pressure gas backflush device: includes solenoid valve, gas tank, and nozzle, with backflush pressure adjustable from 0.4 to 0.8 MPa.

[0124] Example 2: cc

[0125] This embodiment provides a method for measuring electrostatic dust concentration based on spectrum analysis, applicable to online dust concentration monitoring. For example... Figures 1-8 As shown, the method includes the following steps:

[0126] Step S1: Signal Acquisition and Front-End Protection

[0127] An electrostatic dust sensor with a ring electrode structure is installed on the flue. The electrode material is 316L stainless steel, and the insulator is made of polytetrafluoroethylene. A temperature sensor (PT100 platinum resistance thermometer), a humidity sensor (polymer capacitive type), and a flow sensor (differential pressure type) are installed simultaneously to collect environmental and operating parameters.

[0128] The signal acquisition circuit employs a 24-bit high-precision analog-to-digital converter (ADC) with a sampling frequency of 10kHz. An overload protection circuit is included at the front end, using bidirectional transient voltage suppression diodes with a clamping voltage of ±5V. The saturation protection circuit utilizes automatic gain control, automatically reducing the amplification factor when the signal amplitude exceeds 90% of the ADC's full-scale range.

[0129] The raw signal is monitored in real time. When the signal amplitude exceeds the normal range by more than 10 times and the duration is less than 0.5ms, it is marked as a spike signal. When the device status register reading is abnormal and the signal amplitude remains below the noise floor for more than 1 second, it is marked as a power-down segment. The marked data segments are not included in subsequent analysis.

[0130] Step S2: Preprocessing and Steady-State Section Identification

[0131] The original signal was preprocessed as follows: baseline stabilization was performed using the moving average method with a window length of 100ms; trend removal was performed using a first-order high-pass filter with a cutoff frequency of 0.1Hz; and power frequency interference was removed using a 50Hz notch filter and its third harmonic notch filter.

[0132] Steady-state segment identification employs a sliding window method, with a window length of 2 seconds and a sliding step of 1 second. The variance and rate of change of the signal within each window are calculated. A variance threshold of V_th = 0.01 (normalized value) and a rate of change threshold of R_th = 0.05 / s are set. A window is considered a steady-state window when its variance is below V_th and its rate of change is below R_th.

[0133] Simultaneously monitor gas flow rate. When the instantaneous change rate of flow rate exceeds 10% / s, mark the period as unsteady. When the system is in alarm or maintenance state, the corresponding period is also marked as unsteady.

[0134] Window segments continuously determined to be in a steady state are stitched together. When the stitch length reaches 4 seconds, it is output as a valid time window. If the stitch length is less than 4 seconds, the system continues to wait for subsequent windows. If a valid 4-second window cannot be obtained after waiting for more than 30 seconds, the current time segment is marked as an invalid analysis segment, and a maintenance prompt is output.

[0135] Step S3: Spectrum Transformation and Background Noise Spectrum Estimation

[0136] A fast Fourier transform is performed on the time-domain signal within the effective time window, and a Hanning window is used to suppress spectral leakage to obtain the spectral amplitude distribution X(f).

[0137] Background noise spectrum estimation employs a multi-frame minimum value tracking method: The spectra of the most recent 100 valid windows are stored, and the minimum value of these 100 frames is taken for each frequency point to obtain the initial background noise spectrum N0(f). Based on this, a first-order recursive smoothing update is used.

[0138] N_new(f) = α × N_old(f) + (1-α) × min(X(f), N_old(f))

[0139] The smoothing coefficient α is set to 0.95. A threshold curve T(f) = N(f) × 3 is generated to distinguish between valid signals and noise.

[0140] Step S4: Frequency band division and feature extraction

[0141] Based on the sensor's structural characteristics and the properties of the dust, the frequency spectrum is divided into three functional bands:

[0142] Low-frequency operating range: 0.1-50Hz

[0143] Mid-frequency collision band: 50-500Hz

[0144] High-frequency charge perturbation band: 500-2000Hz

[0145] The following features are extracted within each frequency band:

[0146] Total energy: E = Σ|X(f)|², f∈frequency band

[0147] Main peak location: f_peak = argmax|X(f)|, f∈frequency band range

[0148] Peak-valley contrast: C = |X(f_peak)| / mean(|X(f)|), f∈frequency band.

[0149] In-band energy percentage: R_band = E_band / E_total

[0150] Inter-band energy ratio: R_mid_low = E_mid / E_low, R_high_mid = E_high / E_mid

[0151] Spectral flatness: F = exp(mean(ln|X(f)|)) / mean(|X(f)|)

[0152] Spectral slope: The slope obtained by linearly fitting log(f) - log(|X(f)|) is the spectral slope.

[0153] Temperature T, relative humidity RH, and flow rate Q are extracted from the same window. All features are normalized to form a 31-dimensional feature vector (3 frequency bands × 7 features + 3 environmental features + 1 intercept term).

[0154] Step S5: Establishment and online calibration of the spectrum-concentration mapping model

[0155] Support vector regression was used to establish a mapping model. Training data came from laboratory calibration: cement dust was used as the test dust, and data were collected at 5 concentration points (10, 30, 50, 100, 200 mg / m³) and 3 flow rates (10, 15, 20 m / s). Ten sets of valid data were collected for each operating condition, for a total of 150 samples. Radial basis function kernels were used, and the penalty parameter C and kernel parameter γ were optimized through cross-validation.

[0156] The model has a built-in validity detection mechanism: at 2:00 AM every day (during low emission periods), it automatically calculates the deviation between the model's predicted value and the reference value (measured synchronously using the isokinetic sampling-weighing method). If the deviation exceeds 10% for three consecutive days, the model calibration process is triggered.

[0157] Low dust reference conditions: When the concentration remains below 2 mg / m³ for more than 15 minutes and the flow rate is stable within ±5%, the current spectrum will be automatically updated to the background noise spectrum, and the model output will be compared with the theoretical zero point to update the zero point offset parameters.

[0158] When the temperature change exceeds 10℃ or the flow rate change exceeds 20%, model adaptation is triggered: the model weights are fine-tuned using stochastic gradient descent with a learning rate of 0.01 and an update step count not exceeding 100 steps. During the adaptation process, if the instantaneous deviation between the model output and the reference value exceeds 20%, the adaptation is immediately stopped and the original model is restored.

[0159] Step S6: Joint compensation of environment and flow regime

[0160] An environmental compensation model was established. The compensation function f(T,RH,Q,t_on,clean_status) was obtained through experimental calibration. Using 20℃, 50%RH, 15m / s, 30 minutes after power-on, and 1 day after cleaning as baseline conditions, the spectral changes under different conditions were measured, and the multivariate compensation function was obtained by fitting the model.

[0161] Dynamic weighting of traffic flow: Dynamic weighting is activated when the traffic flow rate changes by more than 5% / s. The preset traffic flow-frequency band response relationship is as follows: for every 1 m / s increase in traffic flow, the weight of the low-frequency band decreases by 5%, the weight of the mid-frequency band increases by 3%, and the weight of the high-frequency band increases by 2%. Linear interpolation is used to achieve a smooth transition of weights.

[0162] Contamination Detection: Monitors the energy attenuation rate of high-frequency charge disturbance bands. Based on a clean state, an energy attenuation exceeding 30% is considered light contamination; exceeding 50% is considered moderate contamination; and exceeding 70% is considered heavy contamination. Simultaneously, it monitors the noise floor in the 300-400Hz frequency band. If the noise floor rises more than three times, it indicates potential oil film contamination. When moderate or higher levels of contamination are detected, the median filter window length is automatically increased, and a maintenance prompt is generated.

[0163] Step S7: Output Integration, Quality Control, and Lifecycle Management

[0164] A confidence-weighted fusion method is adopted. The confidence level of each time window is determined by three parts: window variance (the smaller the variance, the higher the confidence level), deviation from background noise (the smaller the variance, the higher the confidence level), and the matching degree between the model output and historical trends (the better the matching degree, the higher the confidence level). The concentration estimates of the most recent 5 windows are weighted and averaged, with the weights proportional to the confidence level.

[0165] Output quality identifier:

[0166] Percentage of valid data: Number of valid windows in the most recent hour / Total number of windows;

[0167] Background estimation error: the normalized root mean square error between the current background noise spectrum and the background noise spectrum 24 hours ago;

[0168] Model consistency score: The Mahalanobis distance is used to calculate the matching degree between the current feature vector and the modeling dataset;

[0169] Write the following data to non-volatile storage: hourly average, maximum, and minimum concentrations; calibration records (time, type, parameter changes); maintenance records (time, type, operator); alarm event records (time, type, duration). Build a sensor lifecycle profile.

[0170] Abnormal trend detection: Linear regression analysis is used to analyze the zero-point drift trend. If the zero-point drift slope exceeds 0.1 mg / m³ / month for 30 consecutive days, it indicates that the sensor may be aging. If the high-frequency energy recovery rate is less than 80% after cleaning and occurs 3 times in a row, it indicates that the cleaning device may be malfunctioning. If the background noise continues to increase and exceeds twice the factory value, it indicates that the sensor may need to be replaced.

[0171] Example 3: S8: Adaptive Measurement Example for Unsteady-State Operating Conditions

[0172] This embodiment, based on Embodiment 2, adds an adaptive measurement function for unsteady-state operating conditions and is applied to the dust removal pipeline of blast furnace gas in a steel plant. This operating condition involves frequent purging and ash removal processes, resulting in drastic fluctuations in the flow pattern, leading to significant errors with traditional measurement methods.

[0173] Step S81: Transient condition detection

[0174] The instantaneous rate of change of the signal is calculated in real time. A sliding window of 0.2 seconds is used to calculate the variance rate of change dVar / dt and the mean rate of change dMean / dt within the window. The operating condition index S is defined as (dVar / dt) × (dMean / dt) / (Var_base × Mean_base), where Var_base and Mean_base are the baseline values ​​for steady-state conditions. When S exceeds a threshold of 10, the system is considered to have entered a non-steady-state operating condition.

[0175] Step S82: Adaptive Window Adjustment

[0176] Under unsteady conditions, the length L of the spectrum analysis window is dynamically adjusted according to the magnitude of S:

[0177] L = 2 / (1 + 0.1 × S), in seconds

[0178] When S=10, L=1 second; when S=50, L=0.33 seconds. The minimum window length is set to 0.2 seconds to ensure that the frequency resolution is not lower than 5Hz.

[0179] Step S83: Transient Spectral Feature Extraction

[0180] FFT transformation is performed within an adaptive window to extract transient spectral features. Simultaneously, the spectral evolution trajectory of 10 consecutive windows is recorded, including: the temporal change sequence of energy in each frequency band, the offset trajectory of the main peak position, the change sequence of the spectral centroid (centroid frequency), and the change sequence of spectral broadening (second moment).

[0181] Step S84: Direct Modeling of Mass Flow Rate

[0182] An end-to-end mapping model from transient spectral evolution features to mass flow rate is established. A Long Short-Term Memory (LSTM) network is adopted, and the network structure is as follows: input layer (10 time steps × 20-dimensional features) - LSTM layer (128 units) - fully connected layer (64 units) - output layer (1 unit, mass flow rate).

[0183] Training data was obtained through synchronous isokinetic sampling: during a non-steady-state process, 30 seconds of spectral evolution data and synchronous isokinetic sampling weighing data were continuously collected, resulting in a total of 200 training samples. The model was trained using the Adam optimizer with a learning rate of 0.001 and 100 training epochs.

[0184] Step S85: State Smoothing and Switching

[0185] When the operating condition index S falls below 10 and remains below 10 for 3 seconds, the system is determined to return to steady-state operating conditions. A linear transition method is used, gradually restoring the window length from the current value to 2 seconds within 5 seconds, and switching the measurement mode from mass flow rate back to concentration measurement. During the switching process, the output value is a weighted average of the results from the two modes, with the weights changing linearly with time.

[0186] Cleaning fluid spray device: including diaphragm pump, liquid storage tank and nozzle, used for cleaning oily contaminants.

[0187] Intelligent control unit: Automatically selects cleaning methods and parameters based on the pollution index sent by the central processing module.

[0188] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for measuring electrostatic dust concentration based on spectrum analysis, characterized in that, Includes the following steps: S1: Signal Acquisition and Front-End Protection: Raw electrical signals are acquired from electrostatic dust sensors, and temperature, relative humidity, airflow, and equipment status parameters are recorded simultaneously; overload protection and saturation protection circuits are set at the front end, and obvious abnormal spike signals and power-down segments are automatically marked to ensure the effectiveness of subsequent analysis; S2: Preprocessing and steady-state segment identification: The original electrical signal is subjected to baseline stabilization processing, trend removal processing, and power frequency notch filtering to obtain a clean time-domain sequence; steady-state periods are identified based on sliding statistical criteria, and data segments with drastic flow fluctuations, alarm states, or maintenance states are filtered out, and an effective time window that can be used for spectrum analysis is output. S3: Spectrum Transformation and Background Noise Spectrum Estimation: The time-domain signal of the effective time window is transformed to the frequency domain to obtain the spectrum amplitude distribution; an adaptive strategy is used to estimate the background noise spectrum and the instrument background noise, and corresponding reference curves are generated for subsequent spectral denoising and threshold setting; S4: Frequency band division and feature extraction: Based on the sensor structural characteristics and typical particle size-velocity response relationship, the spectral amplitude distribution is divided into multiple functional frequency bands. The functional frequency bands include at least two of the following: low-frequency operating condition band, mid-frequency collision band, and high-frequency charge disturbance band. Spectral features are extracted in each functional frequency band. The spectral features include at least one of the following: total energy, main peak position, peak-valley contrast, energy ratio within the band, energy ratio between bands, spectral flatness, and spectral slope. These features are then used to form a feature vector, which is concatenated with the environmental features of temperature, relative humidity, and flow rate within the same time window. S5: Spectrum-Concentration Mapping Model Establishment and Online Calibration: Based on standard dust samples or historical calibration data, a mapping model from the feature vector to dust concentration is established, and a built-in model validity detection mechanism and offset correction mechanism are implemented; when low dust or clean reference conditions are met, the background noise spectrum and zero-point offset parameters are automatically updated; when environmental parameters or flow distribution change, the adaptive adjustment of model parameters is triggered. S6: Environmental and Flow Condition Joint Compensation: Joint compensation is performed for spectral drift caused by temperature, relative humidity, flow rate, power-on duration, and recent cleaning and maintenance status; when the flow rate change exceeds the preset threshold, the characteristic frequency band is dynamically weighted and adjusted according to the preset response law to reduce the estimation deviation caused by flow condition changes; when electrode contamination or signs of heavy dust accumulation are detected, the noise reduction intensity is automatically increased and maintenance prompts are generated. S7: Output Fusion, Quality Control and Lifecycle Management: Weighted fusion of concentration estimates from multiple consecutive time windows, outputting the final concentration value and corresponding confidence level; synchronously outputting quality indicators, which include at least one of the following: effective data percentage, background estimation error and model consistency score; writing key parameters, alarm records and maintenance records into non-volatile storage media to form a traceable lifecycle archive, and automatically prompting for review when abnormal trends are detected.

2. The electrostatic dust concentration measurement method based on spectrum analysis according to claim 1, characterized in that, The preprocessing includes: calculating the statistical characteristics of the time-domain sequence using a sliding window, the statistical characteristics including the variance, mean, and rate of change of the signal within the window; determining whether the current window is a steady-state period based on a comparison of the statistical characteristics with a preset threshold; the criteria for determining a steady-state period include: the variance of the signal within the window is lower than a first threshold, and the rate of change of the signal is lower than a second threshold; and / or The steady-state segment identification includes: filtering data segments in a non-steady-state state, where the non-steady-state state includes the instantaneous rate of change of gas flow exceeding a flow threshold, the system being in an alarm state, or the system being in a manual or automatic maintenance state; and / or Multiple consecutive windows that are determined to be in a steady state are spliced ​​together to form the minimum effective window whose length meets the requirements of spectrum analysis. When the effective window length is insufficient, data from adjacent windows are used to supplement the window or the current time period is marked as an invalid analysis period.

3. The electrostatic dust concentration measurement method based on spectrum analysis according to claim 1, characterized in that, The spectral transformation includes: converting the time-domain signal to the frequency domain using a Fast Fourier Transform or Wavelet Transform to obtain a spectral distribution containing amplitude and phase information; and / or The adaptive strategy estimates the background noise spectrum, including: Based on the spectrum corresponding to periods when the concentration was below the concentration threshold in historical data, its statistical average was calculated as the background noise spectrum; and / or A minimum value tracking method using multi-frame spectra is employed to extract the minimum value of the spectrum at each frequency point across multiple consecutive time windows, thereby obtaining a background noise spectrum estimate; and / or Based on the instrument's baseline noise spectrum at the time of manufacture, combined with the sensor's cumulative operating time and pollution index, the baseline noise spectrum is dynamically updated through an aging model; and / or The background noise spectrum is compared with the current spectrum to generate a reference curve for spectral denoising and a threshold curve for signal detection; the threshold curve is used to distinguish between effective signal components and noise components.

4. The electrostatic dust concentration measurement method based on spectrum analysis according to claim 1, characterized in that, The criteria for dividing the functional frequency bands include: Low-frequency operating condition band: frequency range 0.1-50Hz, used to characterize airflow fluctuations, pipeline vibrations and changes in mechanical operating conditions; Mid-frequency collision band: frequency range 50-500Hz, used to characterize the collision induction between dust particles and electrodes, and is related to dust concentration and particle size distribution; High-frequency charge perturbation band: frequency range 500-2000Hz, used to characterize the charge perturbation carried by dust particles, and is related to dust material, charge characteristics and moisture content; The spectral features extracted within each functional frequency band specifically include: Total energy: the sum of squares or the area of ​​the integral of the amplitudes at all frequencies within this frequency band; Main peak location: The frequency value corresponding to the maximum amplitude within this frequency band; Peak-valley contrast: the ratio of the amplitude of the main peak to the amplitude of the adjacent valley, or the ratio of the amplitude of the main peak to the average amplitude of the frequency band; In-band energy ratio: The ratio of the total energy in this frequency band to the total energy of the entire frequency band; Inter-band energy ratio: The ratio between the total energy of different functional frequency bands, including the energy ratio between the mid-frequency band and the low-frequency band, and the energy ratio between the high-frequency band and the mid-frequency band; Spectral flatness: The ratio of the geometric mean to the arithmetic mean of the spectral amplitudes within a frequency band; Spectral slope: The rate of change of spectral amplitude with frequency within this frequency band, obtained by linear fitting slope in logarithmic coordinate system; The spectral features are normalized and then concatenated with the environmental features of temperature, relative humidity, and flow rate within the same time window to form a complete feature vector for model input.

5. The electrostatic dust concentration measurement method based on spectrum analysis according to claim 1, characterized in that, The mapping model is established using at least one of the following algorithms: multiple linear regression, support vector regression, random forest, gradient boosting tree, neural network, or deep learning model. and / or The model effectiveness detection mechanism includes: periodically calculating the deviation between the model's predicted values ​​and the standard reference values, calculating the time-series stability index of the predicted values, and calculating the matching degree between the model's input features and the feature distribution of the modeling dataset; triggering the model calibration process when the detected deviation exceeds a deviation threshold, stability falls below a stability threshold, or matching degree falls below a matching degree threshold; and / or The low-dust or clean reference conditions include: the concentration continuously being lower than the concentration threshold exceeding a first time threshold, the gas flow rate being within a stable range, and no abnormal operating conditions occurring; when the conditions are met, the current spectrum is automatically used as the new background noise spectrum, and the current model output is compared with the theoretical zero point to update the zero-point offset parameters; and / or The model parameter adaptation includes: when an environmental parameter or flow distribution change exceeding the environmental change threshold is detected, an online learning algorithm is used to fine-tune some weight parameters of the model, including stochastic gradient descent or incremental learning; during the adaptation process, the deviation between the model output and the auxiliary reference measurement is monitored, and when the deviation exceeds the safety threshold, the adaptation is stopped and the model parameters are restored to the original parameters.

6. The electrostatic dust concentration measurement method based on spectrum analysis according to claim 1, characterized in that, The joint compensation of environment and flow state includes establishing an environmental compensation model, which is a multivariate function: C_compensated = C_raw × f(T, RH, Q, t_on, clean_status), where T is temperature, RH is relative humidity, Q is flow rate, t_on is power-on duration, and clean_status is the most recent cleaning and maintenance status; and / or The dynamic weighted adjustment includes: when the flow rate change exceeds the flow rate change threshold, dynamically adjusting the weight coefficients of each functional frequency band in the feature vector according to a pre-calibrated flow-frequency band response relationship; the flow-frequency band response relationship is: when the flow rate increases, the weight of the low-frequency operating band is reduced, and the weights of the mid-frequency collision band and the high-frequency charge disturbance band are increased; and / or The detection of electrode contamination or heavy dust accumulation includes: monitoring the attenuation rate of high-frequency charge disturbance energy, monitoring changes in the noise floor of a specific frequency band, and monitoring the trend of spectral flatness. When contamination signs are detected exceeding the contamination threshold, the noise reduction intensity of the digital filter is automatically increased, and maintenance prompts are generated, including suggested cleaning times and suggested cleaning methods.

7. The electrostatic dust concentration measurement method based on spectrum analysis according to claim 1, characterized in that, The weighted fusion is performed in the following manner: Time-decay weighted: Concentration estimates for the most recent N time windows are assigned weights that decay over time, with more recent windows receiving higher weights; and / or Confidence weighting: A weighting coefficient is calculated based on the quality label of each window, with higher quality windows having higher weights; and / or Adaptive Kalman filtering: The concentration estimates of a continuous window are used as the observation sequence, and the optimal estimate is obtained through the Kalman filtering algorithm; and / or The quality identifier includes: Percentage of valid data: The proportion of time windows deemed valid within the current fusion cycle out of the total number of windows; and / or Background estimation error: the degree of deviation between the current background noise spectrum and the historical background noise spectrum, or the degree of deviation between the current background noise spectrum and the factory-set background noise spectrum; and / or Model consistency score: the degree of matching between the current input feature vector and the feature distribution of the modeling dataset, calculated using Mahalanobis distance or local outlier factor; and / or The lifecycle file includes: the sensor's full lifecycle operation data, each calibration record, each maintenance record, alarm event records, and the evolution trajectory of key parameters; the abnormal trends include at least one of the following: accelerated zero-point drift, continuous increase in background noise, accelerated attenuation of response in a specific frequency band, and decreased recovery effect after cleaning; when an abnormal trend is detected, an early warning message is generated and prompts for manual review or planned maintenance.

8. The electrostatic dust concentration measurement method based on spectrum analysis according to claim 1, characterized in that, It also includes step S8: adaptive measurement for unsteady conditions, which includes: The instantaneous rate of change of the signal is monitored in real time, and the operating condition index is calculated; when the operating condition index exceeds the steady-state threshold, it is determined that the system has entered an unsteady-state operating condition. Under non-steady-state conditions, the length of the spectrum analysis window is dynamically adjusted according to the magnitude of the operating condition index. The window length is negatively correlated with the operating condition index, and the shortest window length is not less than the minimum window length required to ensure the lowest frequency resolution. Spectrum transformation is performed within the dynamically adjusted window to extract transient spectrum features, and the spectrum evolution trajectory of multiple consecutive windows is recorded. The evolution trajectory includes the temporal changes of energy in each frequency band, the offset trajectory of the main peak position, and the temporal evolution parameters of the spectrum shape. Establish an end-to-end mapping model from transient spectrum evolution characteristics to mass flow rate, and directly output the mass flow rate measurement value in kg / s; When the operating condition index falls below the steady-state threshold and remains below it for a preset time, the system is determined to return to steady-state operating conditions, gradually restores the standard window length, and switches the measurement mode from mass flow rate back to concentration measurement.

9. A system for measuring electrostatic dust concentration based on spectrum analysis, characterized in that, include: Sensor module: includes electrostatic dust sensor, temperature sensor, humidity sensor and flow sensor, used to collect raw electrical signals and environmental and operating parameters; Signal processing module: connected to the sensor module, including overload protection circuit, saturation protection circuit, filtering circuit and analog-to-digital conversion circuit, used for front-end protection and preprocessing of the raw electrical signal; Spectrum analysis and feature extraction module: connected to the signal processing module, used to convert time-domain signals to frequency domain, divide functional frequency bands and extract spectral features to form feature vectors; Storage module: Used to store historical data, calibration data, model parameters, background noise spectrum, and lifecycle archives; Central processing module: connected to the signal processing module, the spectrum analysis and feature extraction module and the storage module respectively, the central processing module is configured to execute the method steps of any one of claims 1-8; Output and Interaction Module: Connected to the central processing module, used to output dust concentration measurement values, confidence levels, quality indicators, maintenance prompts, and early warning information; Cleaning execution module: Connected to the central processing module, used to automatically perform backflushing or cleaning operations based on the pollution diagnosis results.

10. The electrostatic dust concentration measurement system based on spectrum analysis according to claim 9, characterized in that: The signal processing module further includes: Programmable gain amplifier, used to automatically adjust the amplification factor according to the signal strength; Anti-aliasing filters are used to filter out frequency components above the Nyquist frequency before analog-to-digital conversion; Isolation circuits are used to achieve electrical isolation between signal ground and digital ground, and to suppress common-mode interference; The central processing module includes at least one of a digital signal processor, a field-programmable gate array, or an embedded microprocessor, and has a built-in fast Fourier transform hardware acceleration unit for real-time spectrum calculation. The output and interaction module includes a display screen, indicator lights, an audible and visual alarm, a wired communication interface, and a wireless communication interface. The wired communication interface includes RS485, Ethernet, or CAN bus, and the wireless communication interface includes 4G / 5G, Wi-Fi, or LoRa.

Images