An intelligent lance clogging monitoring system

Abstract

The application discloses an intelligentized spray gun blockage monitoring system, which comprises a multi-dimensional sensing and collecting unit, a signal conditioning and isolation transmission unit, an edge intelligent computing unit, a hierarchical early warning unit, a linkage control unit and a historical data storage and self-learning unit. The application realizes real-time collection, accurate analysis, quantitative determination, early warning, hierarchical linkage and continuous optimization of the spray gun blockage state, and upgrades passive post-repair to active predictive maintenance, thereby guaranteeing environmental protection emission standard and continuous safe operation of the unit.

Description

Technical Field

[0001] This invention belongs to the field of intelligent monitoring and control technology for industrial environmental protection equipment, and specifically relates to an intelligent spray gun clogging monitoring system. Background Technology

[0002] In the flue gas denitrification and desulfurization processes of coal-fired power plants, the spray gun is the core equipment for atomizing the reducing agent (ammonia water, urea solution) or desulfurizing agent slurry. Its operating status directly determines the pollutant removal efficiency and system safety. During long-term continuous operation, the spray gun is prone to partial or complete nozzle blockage due to impurities in the medium, crystal precipitation, particulate matter deposition, and scale adhesion. This leads to a decrease in spray flow, excessive atomized particle size, spray angle deviation, and uneven flow field distribution, ultimately causing a series of chain failures such as excessive NOx / SO2 emissions, overspray of reducing agent, air preheater blockage, increased differential pressure in the dust collector, corrosion of boiler heating surfaces, and tube rupture.

[0003] Currently, the industry commonly uses the following methods for spray gun clogging monitoring, all of which have obvious technical defects:

[0004] 1) Manual inspection and experience judgment: Relying on inspectors to observe the spray pattern, touch the pipeline pressure, and periodically disassemble and inspect, it is impossible to achieve 24-hour online continuous monitoring, blockage detection is seriously delayed, and the missed detection rate is high;

[0005] 2) Single pressure / flow monitoring: Using only a pressure sensor or flow sensor for single-point monitoring is susceptible to interference from pump pulsation, medium temperature, pipeline pressure drop, and valve opening fluctuations. Fixed thresholds can easily lead to a large number of false alarms and missed alarms, resulting in poor reliability.

[0006] 3) Only supports full blockage alarms: It cannot identify minor, gradual, or initial blockages, missing the best maintenance window and only allowing passive shutdown for handling;

[0007] 4) Lack of intelligent analysis and self-learning capabilities: It lacks the ability to model historical data, fit trends, and adapt to different operating conditions, and cannot adapt to different spray gun models, media concentrations, and changes in operating conditions.

[0008] 5) No system-level linkage control: The alarm is only a local audible and visual prompt and cannot be linked with the DCS system, the backup spray gun switching mechanism, the bypass valve, or the protection shutdown logic, which can easily lead to unplanned shutdowns, environmental penalties, and equipment damage.

[0009] The aforementioned technical deficiencies have become key bottlenecks restricting power plants from meeting environmental standards and ensuring the continuous and stable operation of generating units. Therefore, there is an urgent need in this field for an intelligent spray gun blockage monitoring system capable of multi-sensor fusion, intelligent algorithm discrimination, early prediction of progressive blockage, graded early warning, system linkage, and self-learning optimization. Summary of the Invention

[0010] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an intelligent spray gun blockage monitoring system that enables real-time acquisition, accurate analysis, quantitative judgment, early warning, hierarchical linkage, and continuous optimization of spray gun blockage status. This upgrades passive post-event repairs to proactive predictive maintenance, ensuring compliance with environmental emission standards and continuous safe operation of the unit.

[0011] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0012] An intelligent spray gun clogging monitoring system includes:

[0013] 1) Multi-dimensional sensing and acquisition unit: composed of flow sensor, static pressure sensor and vibration sensor; the flow sensor is installed in the straight pipe section 50-100mm in front of the liquid inlet of the spray gun, the static pressure sensor is installed at the pressure tapping point 30-80mm in front of the nozzle, and the vibration sensor is magnetically fixed to the central axis position of the outer wall of the nozzle.

[0014] 2) Signal conditioning and isolation transmission unit: The input end is connected to the sensor acquisition unit, and the output end is connected to the edge intelligent computing unit. It is used for signal amplification, filtering, AD conversion, opto-isolation and long-distance transmission.

[0015] 3) Edge intelligent computing unit: It has a built-in processor, memory and congestion prediction algorithm, which is used for data preprocessing, dynamic benchmark update, anomaly calculation, congestion fusion, trend prediction, level determination and early warning output;

[0016] 4) Hierarchical early warning unit: connected to the edge intelligent computing unit, used for local audible and visual alarms and remote signal push;

[0017] 5) Linkage control unit: Connected to the edge intelligent computing unit, used to drive the switching of backup spray guns, opening of bypass valves, and system protection shutdown;

[0018] 6) Historical data storage and self-learning unit: bidirectionally connected to the edge intelligent computing unit, used for historical data storage, curve playback, model iteration and parameter optimization.

[0019] After power-on, this system enters the reference self-learning phase, collecting stable operating data for 10 consecutive minutes, automatically generating reference flow rate Q0, reference pressure P0, and reference vibration frequency V0. Based on the medium temperature T, medium concentration C, and system main pipe pressure S, a working condition correction coefficient k is introduced to obtain dynamic reference values.

[0020] Q0'=Q0×k(T,C,S)

[0021] P0'=P0×k(T,C,S)

[0022] V0'=V0×k(T,C,S)

[0023] During real-time operation, the sensor collects Q(t), P(t), and V(t) signals, which are then processed by Kalman filtering and a 10-second sliding window averaging. The signals are then sent to the edge intelligent computing unit to execute the congestion prediction algorithm, calculate the comprehensive congestion degree B(t), determine the congestion level, output a graded early warning signal, and drive the linkage control unit to perform corresponding actions.

[0024] Compared with the prior art, the present invention has the following significant advantages:

[0025] 1) Multi-sensor fusion discrimination: flow rate + pressure + vibration triple features, false alarm rate ≤2%, false alarm rate ≤1%;

[0026] 2) Adaptive dynamic threshold: The baseline is automatically corrected according to the operating conditions and is not affected by fluctuations in temperature, concentration, or pressure;

[0027] 3) Early detection of progressive blockage: Early warning 3-30 minutes in advance to avoid unplanned downtime;

[0028] 4) Quantitative classification of congestion: 0-100% continuous output, five-level judgment, suitable for refined operation and maintenance;

[0029] 5) Self-learning optimization: The model is iterated daily, and the accuracy continues to improve over the long term;

[0030] 6) System linkage control: Automatically switches to backup spray guns to ensure continuous compliance with emission standards;

[0031] 7) Significant cost reduction and efficiency improvement: Reduces reducing agent consumption by 5%-15% and reduces maintenance costs by 30%-60%. Attached Figure Description

[0032] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0033] As the intelligent computing core of this invention, the edge intelligent computing unit is equipped with a congestion prediction algorithm;

[0034] This algorithm is a multi-feature fusion adaptive threshold-time series trend fitting congestion prediction algorithm (MFAT-TS).

[0035] Algorithm parameter definition

[0036] Q(t): Real-time flow rate at time t, in L / min;

[0037] P(t): Static pressure in front of the nozzle at time t, in MPa;

[0038] V(t): The dominant frequency of vibration at time t, in Hz;

[0039] Q0 / P0 / V0: 10-minute stable operating baseline values;

[0040] Q0' / P0' / V0': Dynamic baseline values ​​after operating condition correction;

[0041] T: Medium temperature, in °C;

[0042] C: Medium mass concentration, in %

[0043] S: System main pressure, in MPa;

[0044] k(T,C,S): Working condition correction factor;

[0045] Dq: Traffic anomaly level, %

[0046] Dp: Pressure anomaly level, %

[0047] Dv: Vibration anomaly degree, %

[0048] B(t): Overall blockage rate, 0-100%;

[0049] ε: Trend correction term, -5% to +10%;

[0050] α: First-order exponential smoothing coefficient, 0.70;

[0051] Δt: Prediction time, 3-5 min.

[0052] Q_filter(t): The steady-state flow rate at time t after Kalman filtering and 10-second sliding window averaging.

[0053] P_filter(t): The steady-state pressure at time t after Kalman filtering and 10-second sliding window averaging.

[0054] Q_pre(t+Δt): Predicted flow rate at time Δt in the future.

[0055] P_pre(t+Δt): Predicted pressure at time Δt in the future.

[0056] Trend_Q: Traffic change trend value

[0057] Trend_P: Pressure change trend value

[0058] Th_q_pre: Early warning threshold for traffic, set to 65% of the dynamic baseline traffic Q0'.

[0059] Th_p_pre: Early warning threshold for pressure, set to 135% of the dynamic reference pressure P0'.

[0060] B_label: The actual congestion level label, manually or calibrated by operations and maintenance personnel.

[0061] ω1, ω2, ω3: weighting coefficients for flow rate, pressure, and vibration, respectively.

[0062] N: Total number of valid historical data samples within 24 hours

[0063] η: Learning rate, with a value of 0.05

[0064] λ: Regularization coefficient, with a value of 0.01.

[0065] Loss: The value of the loss function, representing the deviation between the predicted congestion level and the calibrated congestion level.

[0066] Sum_ω: The sum of the updated weights

[0067] Algorithm Steps

[0068] 1) Data preprocessing

[0069] Kalman filtering is performed on Q(t), P(t), and V(t), followed by 10-second sliding window mean processing to remove noise and pump pulsation interference.

[0070] 2) Dynamic benchmark update

[0071] The baselines Q0, P0, and V0 are updated using stable data for 10 consecutive minutes.

[0072] Calculate the working condition correction factor:

[0073] k(T,C,S)=0.0008×T+0.0012×C+0.0005×S+0.96

[0074] Dynamic benchmark:

[0075] Q0'=Q0×k

[0076] P0'=P0×k

[0077] V0'=V0×k

[0078] 3) Calculation of single feature anomaly degree

[0079] Dq=|(Q0'−Q(t)) / Q0'|×100%

[0080] Dp = |(P(t)−P0') / P0'|×100%

[0081] Dv=|(V(t)−V0') / V0'|×100%

[0082] 4) Overall congestion calculation

[0083] B(t)=0.42×Dq+0.38×Dp+0.20×Dv+ε

[0084] 5) Time series trend forecasting

[0085] The system performs short-term time series predictions on the filtered stable flow rate and stable pressure to identify the progressive clogging trend of the spray gun and provide early warning before the clogging reaches the threshold.

[0086] The forecast employs a first-order exponential smoothing method, with an exponential smoothing coefficient α of 0.70 and a forecast time Δt of 3 or 5 minutes. The forecasted flow rate and forecasted pressure are calculated using the following formula:

[0087] Q_pre(t+Δt)=α×Q_filter(t)+(1−α)×Q_filter(t−1)

[0088] P_pre(t+Δt)=α×P_filter(t)+(1−α)×P_filter(t−1)

[0089] Simultaneously calculate the flow trend value and the pressure trend value:

[0090] Trend_Q=Q_filter(t)−Q_filter(t−1)

[0091] Trend_P=P_filter(t)−P_filter(t−1)

[0092] When Trend_Q < 0 and Trend_P > 0, the spray gun's operating state is determined to be progressing towards worsening blockage. If the predicted flow rate is lower than the flow rate early warning threshold, or the predicted pressure is higher than the pressure early warning threshold, the system triggers a progressive blockage early warning. The flow rate early warning threshold is set to 65% of the dynamic baseline flow rate, and the pressure early warning threshold is set to 135% of the dynamic baseline pressure.

[0093] 6. Congestion Level Determination

[0094] Based on the comprehensive congestion degree calculation result B(t), the system divides the congestion state into five levels according to the preset range and outputs the corresponding warning or alarm signal.

[0095] The specific judgment rules are as follows:

[0096] If the overall congestion rate is between 0% and 20%, it is considered a normal state and no warning will be triggered.

[0097] If the overall congestion level is between 21% and 40%, it is considered a minor congestion, and a green warning signal is issued.

[0098] If the overall congestion rate is between 41% and 70%, it is judged as moderate congestion, and a yellow warning signal is issued.

[0099] If the overall blockage rate is between 71% and 95%, it is considered a severe blockage, and an orange alarm signal is output.

[0100] If the overall blockage rate is between 96% and 100%, it is considered a complete blockage, and a red alarm signal is output.

[0101] Each level of signal is used to prompt observation, suggest maintenance, prepare for switching, and take emergency protection actions.

[0102] 7) Self-learning optimization

[0103] The system uses valid historical operating data from the previous 24 hours to self-learn and update the flow weight, pressure weight, and vibration weight daily, in order to adapt to spray gun wear, changes in media characteristics, and deviations in on-site operating conditions, thereby improving the accuracy of long-term operation judgment.

[0104] The self-learning mechanism aims to minimize the error between the predicted congestion level and the actual calibrated congestion level, and constructs a loss function with regularization constraints:

[0105] Loss=(1 / N)×Σ[(ω1Dq+ω2Dp+ω3Dv)−B_label]²+λ(ω1²+ω2²+ω3²)

[0106] The gradient descent method is used to calculate the gradient of each weight, and the original weights are iteratively updated according to the learning rate η:

[0107] ω1_new=ω1_old−η×∂Loss / ∂ω1

[0108] ω2_new=ω2_old−η×∂Loss / ∂ω2

[0109] ω3_new=ω3_old−η×∂Loss / ∂ω3

[0110] To ensure that the congestion level remains stable within the 0–100% range, the updated weights are normalized so that the sum of the three is 1:

[0111] Sum_ω=ω1_new+ω2_new+ω3_new

[0112] ω1=ω1_new / Sum_ω

[0113] ω2 = ω2_new / Sum_ω

[0114] ω3=ω3_new / Sum_ω

[0115] At the same time, the weights are constrained within a reasonable range: ω1∈[0.35,0.50], ω2∈[0.35,0.45], ω3∈[0.10,0.25], to avoid abnormal weight shifts and ensure stable and reliable system judgment.

[0116] Example Hardware Configuration

[0117] Flow sensor: range 0-15L / min, accuracy ±0.3%FS, response ≤50ms;

[0118] Static pressure sensor: measuring range 0-2.5MPa, accuracy ±0.2%FS, temperature compensation -20℃~85℃;

[0119] Vibration sensor: piezoelectric, frequency 10Hz-10kHz, amplitude resolution 0.1mg;

[0120] Edge controller: ARM Cortex-A72, 400MHz clock speed, 8GB Flash;

[0121] Signal conditioning: amplification, filtering, opto-isolation, 4-20mA / RS485 output;

[0122] Early warning module: Three-color audible and visual alarm, ≥90dB;

[0123] Linkage module: 8-channel relay, AC220V / 5A.

[0124] Installation and wiring

[0125] 1) The flow sensor is installed in the 80mm straight pipe section at the front end of the spray gun inlet to ensure full pipe measurement;

[0126] 2) The static pressure sensor is installed at a pressure tapping point 50mm in front of the nozzle, with the pressure tapping port flush with the pipe wall;

[0127] 3) The vibration sensor is magnetically fixed to the central axis of the nozzle outer wall to ensure good coupling;

[0128] 4) The sensor signal line is connected to the signal conditioning unit, and after isolation, it is connected to the edge controller;

[0129] 5) The early warning and linkage module is connected to the DCS system and the backup spray gun switching valve via relays.

[0130] System operation process

[0131] 1) Power-on self-test: Initialization of sensor, communication, and algorithm parameters;

[0132] 2) Benchmark learning: Collect stable data continuously for 10 minutes and generate Q0, P0, and V0;

[0133] 3) Real-time data acquisition: Q(t), P(t), and V(t) are acquired at 1-second intervals;

[0134] 4) Preprocessing: Kalman filter + 10s sliding window mean;

[0135] 5) Algorithm calculation: Dynamic baseline → Anomaly degree → Congestion degree → Trend prediction → Level determination;

[0136] 6) Early warning output: Trigger audible and visual alarms and remote push notifications according to the level;

[0137] 7) Linkage control: minor blockage alert, moderate blockage suggests maintenance, severe / complete blockage switches to backup spray gun;

[0138] 8) Data storage: Historical data is stored every second for ≥365 days;

[0139] 9) Self-learning: Iterate the algorithm weights every morning at midnight to improve long-term accuracy.

[0140] Test Results

[0141] After applying this system to the SCR denitrification spray guns of a 300MW unit in a power plant:

[0142] Congestion identification accuracy ≥98%;

[0143] False alarm rate ≤ 1.5%;

[0144] The average early warning time for gradual congestion is 12 minutes.

[0145] Unplanned downtime decreased by 100%;

[0146] The consumption of reducing agent decreased by 8%.

Claims

1. An intelligent spray gun clogging monitoring system, characterized in that, It includes a multi-dimensional sensing and acquisition unit, a signal conditioning and isolation transmission unit, an edge intelligent computing unit, a hierarchical early warning unit, a linkage control unit, and a historical data storage and self-learning unit; The multi-dimensional sensing and acquisition unit is used to acquire the real-time flow rate Q(t), nozzle inlet pressure P(t), and vibration main frequency V(t) of the spray gun; The edge intelligent computing unit is configured to perform the following steps: (1) Perform Kalman filtering and 10s sliding window mean preprocessing on Q(t), P(t), and V(t); (2) Update the baseline values ​​Q0, P0, and V0 based on the stable operation data for 10 minutes, and obtain the dynamic baselines Q0', P0', and V0' through the working condition correction coefficient k(T,C,S)=0.0008×T+0.0012×C+0.0005×S+0.96; (3) Calculate the flow anomaly degree Dq=|(Q0'−Q(t)) / Q0'|×100%, the pressure anomaly degree Dp=|(P(t)−P0') / P0'|×100%, and the vibration anomaly degree Dv=|(V(t)−V0') / V0'|×100%; (4) Calculate the overall congestion degree B(t) according to the formula B(t)=0.42×Dq+0.38×Dp+0.20×Dv+ε, where ε is a trend correction term of -5% to +10%; (5) First-order exponential smoothing Q(t+Δt)=0.70×Q(t)+0.30×Q(t-1) and P(t+Δt)=0.70×P(t)+0.30×P(t-1) are used to predict the operating trend in the next 3-5 minutes and realize early warning of gradual congestion; (6) Classify the congestion level into five levels: 0-20%, 21-40%, 41-70%, 71-95%, and 96-100%, and output graded early warnings; (7) Use the historical data from the previous 24 hours each day to perform self-learning and optimize the algorithm weights; The system outputs graded early warning signals based on the level of blockage and simultaneously executes actions such as switching to backup spray guns, opening bypass valves, or shutting down the system for protection.

2. The intelligent spray gun clogging monitoring system according to claim 1, characterized in that, The multi-dimensional sensing and acquisition unit includes: a flow sensor installed on a straight pipe section 50-100mm from the front of the spray gun inlet, a static pressure sensor installed at a pressure tapping point 30-80mm from the front of the nozzle, and a vibration sensor magnetically fixed to the outer wall of the nozzle.

3. The intelligent spray gun clogging monitoring system according to claim 1, characterized in that, The tiered early warning system includes: green alert, yellow alert, orange alert, and red alert, and supports integration with DCS, SIS, MES, and CEMS systems.

4. The intelligent spray gun clogging monitoring system according to claim 1, characterized in that, Historical data storage period is 1 second / data entry, and data retention time is ≥365 days.

Images