Pipeline flow monitoring device for perfluorotripropylamine production

Abstract

The present application provides a pipeline flow monitoring device for the production of perfluorotripropylamine. The pipeline flow monitoring device obtains the scaling image of the inner wall of the pipeline, extracts features and calculates the scaling thickness, and analyzes the correlation between the scaling thickness and the feed flow fluctuation in combination with the prediction model trained by historical data. According to the predicted flow fluctuation, it is judged whether the pipeline needs to be cleaned. At the same time, the feed pressure and flow are monitored in real time, the change trend is analyzed, and the feed pump speed is adjusted in time. In addition, the flow fluctuation data is subjected to spectrum analysis to determine whether there is a scaling risk.

Description

Technical Field

[0001] The invention relates to the field of mechanical equipment, and in particular to a pipeline flow monitoring device used for the production of perfluorotripropylamine. Background Art

[0002] The fluorination reaction control system in the continuous production of perfluorotripropylamine faces a key technical problem: in the fluorinating agent feed pipeline, the feed flow fluctuates due to the different thickness of the scale on the pipe wall. This feed flow fluctuation directly affects the conversion rate of the fluorination reaction, and then affects the product quality and yield. Specifically, as the production time goes by, the fluorinating agent contains impurities, so it will gradually deposit on the inner wall of the feed pipeline to form a layer of scale. The thickness distribution of the scale is not uniform, which causes the effective cross-sectional area in the pipeline to change. When the thicker scale parts and the thinner scale parts appear alternately, the feed flow will fluctuate periodically. The flow rate fluctuates, making the amount of fluorinating agent entering the reactor more and less, and the reaction conditions are difficult to maintain stability. Unstable feed amount will disrupt the material ratio and residence time distribution in the reactor, resulting in fluctuations in the conversion rate of the fluorination reaction. The fluctuation of the conversion rate will directly affect the quality indicators of downstream products, such as purity and yield. In addition, due to the instability of the reaction conversion rate, it may also cause safety hazards. The core of this technical problem lies in how to achieve uniform and stable feed of fluorinating agent in the presence of scaling and reduce the impact of feed fluctuation on reaction conversion rate. Summary of the invention

[0003] The present invention provides a pipeline flow monitoring device for perfluorotripropylamine production, comprising:

[0004] The pipeline flow monitoring device for perfluorotripropylamine production comprises a fluorinating agent feeding unit, an electrolytic cell and a control unit; the fluorinating agent feeding unit comprises a fluorinating agent feeding pipeline connected to the top of the electrolytic cell, a detection port arranged on the fluorinating agent feeding pipeline close to one end of the electrolytic cell, an endoscope capable of entering the interior of the fluorinating agent feeding pipeline through the detection port, and a pressure sensor, a flow sensor and a temperature sensor arranged on the fluorinating agent feeding pipeline; the electrolytic cell, the endoscope, the pressure sensor, the flow sensor and the temperature sensor are respectively connected to the control unit:

[0005] The control unit is used to obtain scaling image data of the inner wall of the fluorinating agent feeding pipeline, pre-process the image data, extract image features of the scaling area of ​​the pipeline wall, and calculate the scaling thickness value of the current pipeline inner wall according to a preset scaling thickness evaluation index;

[0006] The control unit is further used to analyze the correlation between the scaling thickness and the feed flow fluctuation through historical data, and predict the feed flow fluctuation within a preset time period under the current scaling thickness; determine whether it is necessary to clean and maintain the fluorinating agent feed pipeline according to the feed flow fluctuation within the preset time period under the current scaling thickness; if the predicted feed flow fluctuation exceeds a preset fluctuation safety threshold, the pipeline cleaning process is triggered to clean the inner wall of the pipeline to remove scaling;

[0007] The control unit is further used to collect the pressure and flow of the feed pipeline in real time during the fluorinating agent feeding process, and perform curve fitting processing on the pressure and flow of the feed pipeline to obtain the change trend of the pressure and flow, and obtain the change amplitude and frequency of the pressure and flow based on the change trend of the pressure and flow, and judge whether there is a sudden increase or decrease in pressure during the feeding process. If so, adjust the speed of the feed pump in time to control the fluctuation amplitude within a preset range;

[0008] The control unit is further used to obtain feed flow fluctuation data, use a spectrum analysis method to obtain the frequency characteristics of the fluctuation signal, and compare it with a preset frequency threshold to determine whether the frequency of the fluctuation is within a preset range of the fluctuation frequency. If it exceeds the preset range of the fluctuation frequency, an early warning message is generated to indicate the risk of pipeline scaling.

[0009] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0010] The present invention discloses a pipeline flow monitoring device for the production of perfluorotripropylamine. The pipeline flow monitoring device obtains a scaling image of the inner wall of the pipeline, extracts features and calculates the scaling thickness, and analyzes the correlation between the scaling thickness and the feed flow fluctuation in combination with a prediction model trained with historical data. According to the predicted flow fluctuation, it is determined whether the pipeline needs to be cleaned. At the same time, the feed pressure and flow are monitored in real time, the change trend thereof is analyzed, and the feed pump speed is adjusted in time. In addition, the flow fluctuation data is subjected to spectrum analysis to determine whether there is a scaling risk. Finally, the pipeline flow monitoring device can also establish a prediction model of the operating condition and the conversion rate based on the temperature and pressure of the reactor, and correlate the feed fluctuation with the conversion rate prediction result to establish a correction model. The present invention realizes comprehensive monitoring and optimization of the fluorination reaction process, and improves production efficiency and product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 The present invention is a schematic structural diagram of a pipeline flow monitoring device for the production of perfluorotripropylamine. DETAILED DESCRIPTION

[0012] The following will describe the technical solutions in the embodiments of the present invention in detail in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention.

[0013] like Figure 1 As shown, this embodiment is a pipeline flow monitoring device for the production of perfluorotripropylamine, and the pipeline flow monitoring device for the production of perfluorotripropylamine includes a fluorinating agent feeding unit 1, an electrolytic cell 2 and a control unit (not shown in the figure). The fluorinating agent feeding unit includes a fluorinating agent feeding pipeline 10 connected to the top of the electrolytic cell 2, a detection port 11 provided on the fluorinating agent feeding pipeline 10 near the end of the electrolytic cell 2, an endoscope 12 that can enter the interior of the fluorinating agent feeding pipeline 10 through the detection port 11, and a pressure sensor 13, a flow sensor 14 and a temperature sensor 15 provided on the fluorinating agent feeding pipeline 10. The electrolytic cell 2, the endoscope 12, the pressure sensor 13, the flow sensor 14 and the temperature sensor 15 are respectively connected to the control unit. The endoscope 12 can use a depth image acquisition probe, etc., which is not limited here.

[0014] The endoscope 12 is used to periodically extend into the interior of the fluorinating agent feeding pipeline 10 and obtain scaling image data of the inner wall of the fluorinating agent feeding pipeline 10 .

[0015] The control unit is used to pre-process the image data, extract the image features of the scaling area of ​​the pipe wall, and calculate the scaling thickness value of the current pipe inner wall according to the preset scaling thickness evaluation index.

[0016] The inner wall of the fluorinating agent feed pipeline 10 is circumferentially scanned by a depth image acquisition probe to obtain an original depth image, and a denoised image is obtained based on the original depth image by using a contrast enhancement operator and a Gaussian filter; the denoised image is binarized by using a maximum inter-class variance method based on the pipe wall reference position and the gray value distribution of the pipe cross section, and a scaling area image is obtained by a morphological corrosion and dilation operation; depth data is extracted from the scaling area image to construct a three-dimensional point cloud, a random sample consistency algorithm is used to fit the pipe wall reference plane to obtain a scaling depth map, and the depth data at the intersection of the scanning direction and the contour line is used to obtain the actual measured value of the scaling thickness.

[0017] Exemplarily, the inner wall of the fluorinating agent feed pipe 10 is scanned circumferentially by a depth image acquisition probe to obtain an original depth image, a contrast enhancement operator is used to highlight the scaling boundary features of the original depth image to obtain an enhanced image, and then a Gaussian filter is used to eliminate noise points in the enhanced image to obtain a denoised image, and the pipe wall reference position is identified from the denoised image based on the grayscale gradient change. According to the pipe wall reference position and the grayscale value distribution of the pipe cross section, the maximum inter-class variance method is used to calculate the optimal segmentation threshold, the denoised image is binarized to obtain a binary image, the edge fragments are eliminated from the binary image by morphological corrosion and expansion operations to obtain a scaling area image, and the edge contour line is extracted from the scaling area image based on the gradient operator to obtain a scaling contour image. Depth data is extracted from the scaling contour image to construct a three-dimensional point cloud, and a random sample consistency algorithm is applied to the three-dimensional point cloud to fit the pipe wall reference plane, and the distance from the scaling area point cloud to the reference plane is calculated to construct a scaling depth map, and the depth data at the intersection of the scanning direction and the contour line is used to obtain the actual measured value of the scaling thickness. According to the preset circumferential scanning resolution and depth measurement accuracy, the fan-shaped sampling units are divided according to the scaling area. The depth mean value in the sampling unit is obtained from the scaling depth map as the scaling thickness of the unit, and the complete scaling thickness distribution curve of the inner wall of the pipeline is obtained by interpolation calculation. The scaling detection of the inner wall of the pipeline first obtains image data by scanning with a depth image acquisition probe. During the circumferential scanning, the probe rotates along the pipe wall at a fixed interval, for example, the rotation angle interval is 0.5 degrees, and a depth image is collected at each position. The image resolution is 1024×768 pixels and the depth accuracy is 0.1 mm. In this way, complete pipe wall surface depth data can be obtained. For the original depth image, histogram equalization is used to enhance the contrast and enhance the grayscale difference between the scaling and the pipe wall. Then, a Gaussian filter is used for noise reduction. The filter kernel size is 5×5 and the standard deviation is 1. The pipe wall reference position recognition is to determine the pipe wall boundary at the maximum gradient by calculating the grayscale gradient in the vertical direction of the image, so as to establish a coordinate reference. When segmenting the denoised image, the maximum inter-class variance method is used to adaptively calculate the segmentation threshold. For example, for an 8-bit grayscale image, the grayscale value when the inter-class variance is the largest is 128, which can be used as the binarization threshold. After the scaling area is extracted, morphological processing is used to remove the connected domains with an area less than 100 pixels to eliminate speckle noise. In the three-dimensional reconstruction stage, depth data is obtained from the scaling contour image, and each pixel corresponds to the three-dimensional coordinates in the actual space. The random sample consistency algorithm is used to randomly select 3 points from the point cloud, fit the plane equation, iterate 500 times, and select the plane with the most points as the pipe wall reference plane. When measuring the scaling thickness, a depth measurement point is taken every 1 degree along the scanning direction, and the vertical distance from the point to the reference plane is calculated as the scaling thickness. The construction of the scaling thickness distribution curve is to divide the 360-degree circumferential range of the pipeline into sector units, each sector unit has a central angle of 5 degrees, collect 10 evenly distributed thickness measurement points in each unit, and calculate the average value as the scaling thickness value of the unit.Cubic spline interpolation is used between adjacent units to achieve smooth transition, and finally a continuous scaling thickness distribution curve is obtained. In practical applications, for a pipe with a diameter of 200 mm, if the scaling thickness measurement value is in the range of 2 to 8 mm, the measurement accuracy can reach 0.2 mm.

[0018] The control unit is further used to analyze the correlation between scaling thickness and feed flow fluctuation through historical data, and predict the feed flow fluctuation within a preset time period under the current scaling thickness.

[0019] The scaling thickness value and flow value in the historical data of pipeline operation are obtained, and multiple time period data are obtained according to the change rate of the scaling thickness value; a physical model is established for the time period data, the flow resistance coefficient is calculated by the physical model, and the corresponding relationship between the pressure loss and the flow is obtained by polynomial fitting; a sliding window is set according to the corresponding relationship, and a random forest algorithm is used to extract features of the data in the sliding window to obtain the scaling thickness gradient value and the flow fluctuation amplitude; a predictor is constructed by using a long short-term memory network, and the scaling thickness gradient value and the flow fluctuation amplitude are input into the predictor to generate flow fluctuation prediction data.

[0020] Exemplarily, the scaling thickness value, feed flow value, pipeline pressure value and temperature value are extracted from the historical data of pipeline operation to form an original data set (wherein the feed flow value, pipeline pressure value and temperature value can be obtained in real time through sensors), and the original data set is divided into multiple time periods according to the scaling thickness change rate. The time delay relationship between scaling thickness and flow change is calculated for each time period, and the noise value and abnormal value in the data are removed to obtain a cleaned data set. Based on the cleaned data set, a physical model between scaling thickness, fluid viscosity and pressure loss is constructed, and the flow resistance coefficient under different scaling thicknesses is calculated by pipeline fluid mechanics parameters. The corresponding relationship between pressure loss and flow is calculated by polynomial fitting, and the influence function of scaling thickness on flow is obtained from the corresponding relationship. Parameter calibration is performed for the influence function, the sliding window width is set to 12 hours, the sampling interval is set to 10 minutes, and the random forest algorithm is used to extract features from the data in the window. The three main eigenvalues ​​of scaling thickness gradient, flow fluctuation amplitude and fluctuation period are calculated, and the eigenvalues ​​are weighted according to the physical meaning to obtain a comprehensive characteristic index. Based on the comprehensive characteristic index, the data is denoised and smoothed, and the predictor is established using a long short-term memory network. The number of input layer units is set to 72, the number of hidden layer units is set to 128, and the number of output layer units is set to 24. The current scaling thickness value is input into the predictor to generate the flow fluctuation forecast data for the next 24 hours, and the flow change trend curve is obtained by superimposing the calculation. The pipeline scaling thickness prediction starts with data preprocessing. The original data contains four key parameters: scaling thickness, flow, pressure, and temperature. The data collection frequency is once per minute, and the data span is 30 days. The scaling growth rate is obtained by calculating the difference in scaling thickness between two adjacent measurements and dividing it by the time interval. When the rate mutation exceeds 0.5 mm / day, the data is divided into different time periods, and the length of each data segment is between 48 hours and 96 hours. For the abnormal data judgment criteria, data points with flow fluctuations exceeding 20% ​​of the rated value, pressure fluctuations exceeding 30%, and temperature fluctuations exceeding 10 degrees Celsius are marked as abnormal values. The fluid properties are taken into consideration in the construction of the physical model. At a temperature of 20 degrees Celsius, the fluid viscosity is 15 centipoise, the density is 1.2 grams per cubic centimeter, and the inner diameter of the pipe is 200 mm. When the scale thickness increases from 2 mm to 8 mm, the flow resistance coefficient increases from 0.025 to 0.045, and the pressure loss increases with the increase in scale thickness in a quadratic function relationship. At a rated flow of 100 cubic meters per hour, the pressure loss increases by 0.15 MPa for every 1 mm increase in scale thickness. In the feature extraction stage, a 12-hour width sliding window is used, and the window is translated for 10 minutes each time. The 720 data points in the window are processed, and the scale thickness gradient eigenvalue range is 0 to 0.1 mm per hour, the flow fluctuation amplitude eigenvalue range is -10% to +10%, and the fluctuation period eigenvalue range is 2 hours to 8 hours. The weights of the three eigenvalues ​​are 0.5, 0.3, and 0.2 respectively.When training the prediction model, the historical data is divided into a training set and a validation set in a ratio of 8:2. The input layer receives 72 historical moment data, and features are learned through 128 hidden layer units to output the flow values ​​at 24 prediction moments. The prediction time span is 24 hours. The prediction results show that when the scaling thickness is 4 mm, the flow fluctuation amplitude is 7%, the main fluctuation period is 4 hours, and the flow rate shows an overall downward trend, with a 5% drop in 24 hours. When the scaling thickness of the pipeline reaches 6 mm, the pressure loss increases to 0.9 MPa, the fluid flow resistance causes the flow fluctuation to intensify, the fluctuation amplitude increases to 12%, and the fluctuation period is shortened to 2.5 hours. The prediction shows that the flow rate will drop by 8% within 24 hours, and the pipeline cleaning and maintenance needs to be carried out in time.

[0021] The control unit is further used to determine whether it is necessary to clean and maintain the fluorinating agent feed pipeline 10 according to the fluctuation of the feed flow rate within a preset time period under the current scaling thickness. If the predicted feed flow rate fluctuation exceeds the preset fluctuation safety threshold, the pipeline cleaning process is triggered to clean the inner wall of the pipeline to remove scaling. The cleaning and maintenance of the fluorinating agent feed pipeline 10 can be achieved by a cleaning pipeline 16 provided in the fluorinating agent feed pipeline 10. The cleaning pipeline 16 is connected to a cleaning agent supply unit (not shown in the figure), and the fluorinating agent feed pipeline 10 is cleaned during the cleaning and maintenance process.

[0022] The real-time feed flow is obtained and the deviation value between the real-time feed flow and the standard flow is calculated, and the deviation value is smoothed by using a moving average filter to obtain a flow fluctuation curve; the fluctuation amplitude and the fluctuation period are extracted by using a peak and valley detection operator for the flow fluctuation curve, and a fluctuation over-limit state judgment result is obtained according to the fluctuation amplitude and the fluctuation period; the fluctuation over-limit state judgment result is analyzed by time series statistics, and if the fluctuation amplitude continuously exceeds a preset threshold, an exponentially weighted average calculation is used to obtain a cleaning trigger signal; according to the cleaning trigger signal, historical cleaning records are read from a database, and a Kalman filter is used to recursively estimate the historical cleaning records, and a cleaning execution instruction is generated according to the recursive estimation result.

[0023] Exemplarily, the standard flow deviation value is calculated according to the real-time data of the feed flow, the deviation value is smoothed by moving average filtering to obtain the flow fluctuation curve, the peak and valley detection operator is used to extract the fluctuation amplitude and fluctuation period of the fluctuation curve, and the fluctuation over-limit state judgment result is calculated from the fluctuation characteristic parameters. Time series statistics are performed on the fluctuation over-limit state judgment result. If the fluctuation amplitude exceeds the set threshold or the fluctuation period is less than the set duration within 5 consecutive monitoring periods, the exponential weighted average is used to calculate the degree of fluctuation aggravation to obtain the cleaning trigger signal, and the cleaning level identification is generated in combination with the current scaling thickness and the running time. The historical cleaning records are read from the database, the flow recovery degree and maintenance time after the last cleaning are extracted, the cleaning effect data is recursively estimated by the Kalman filter, the corresponding relationship between the scaling removal rate and the cleaning parameters is established, and the cleaning execution instruction containing the three parameters of the cleaning agent concentration, the cleaning pressure, and the cleaning duration is generated. The proportional integral regulator is used to realize the closed-loop control of pressure and flow during the cleaning process, the conductivity value of the cleaning liquid is monitored by the online conductivity detector, the cleaning parameter value is adjusted according to the conductivity change trend, and the scaling removal degree is judged to determine the cleaning termination time point. The flow fluctuation curve is calculated with a monitoring cycle of 10 minutes. The real-time flow data is sampled in each cycle. The difference between the current flow value and the standard flow value of 100 cubic meters per hour is the flow deviation. The deviation sequence is smoothed using a moving average window consisting of 19 sampling points to obtain a continuous fluctuation curve. The peak point on the fluctuation curve corresponds to the maximum flow deviation, the valley point corresponds to the minimum deviation, the difference between adjacent peak and valley values ​​is the fluctuation amplitude, and the time interval between adjacent peak points is the fluctuation cycle. The fluctuation over-limit judgment adopts a double threshold criterion. The fluctuation amplitude exceeds 15% of the standard flow or the fluctuation cycle is less than 4 hours, which is considered an abnormal state. When the abnormal state is maintained for 5 consecutive monitoring cycles, the degree of fluctuation aggravation is calculated. Under the condition of a scaling thickness of 5 mm and a running time of more than 720 hours, the degree of fluctuation aggravation is 0.8, triggering secondary cleaning. The historical cleaning records show that after the last cleaning, the flow rate recovered to 95% of the standard value, the maintenance time was 360 hours, the cleaning agent concentration was 4%, the cleaning pressure was 0.6 MPa, the cleaning time was 4 hours, and the scaling removal rate reached 90%. The Kalman filter predicts the cleaning parameters under the current working conditions based on historical data, and recommends that the cleaning agent concentration be adjusted to 4.5%, the cleaning pressure be increased to 0.7 MPa, and the cleaning time be extended to 4.5 hours. The proportional integral control of pressure and flow during the cleaning process uses a 0.5-second sampling period. When the cleaning pressure fluctuation exceeds ±5% of the set value, the controller outputs a compensation signal to adjust the pump speed. The baseline value of the cleaning fluid conductivity is 2000 micro-Siemens per centimeter. As the scale dissolves, the conductivity gradually increases. When the conductivity increment is less than 50 micro-Siemens per centimeter and lasts for 10 minutes, the cleaning effect is as expected and the cleaning is terminated.In actual operation, when the pipe scale thickness was 6 mm, the flow fluctuation curve showed obvious abnormalities, with a fluctuation amplitude of 18%, a fluctuation period shortened to 2.5 hours, and four consecutive monitoring periods exceeded the limit, with the fluctuation intensity reaching 0.9, triggering pipeline cleaning. During the cleaning process, the conductivity increased from 2000 to 3500 microSiemens per centimeter and then stabilized. The cleaning time was 4.2 hours, and the flow rate recovered to 93% of the standard value after cleaning.

[0024] The control unit is further used to collect the pressure and flow of the fluorinating agent feed pipeline 10 in real time during the fluorinating agent feeding process, and perform curve fitting processing on the pressure and flow of the fluorinating agent feed pipeline 10 to obtain the changing trend of the pressure and flow, and obtain the changing amplitude and frequency of the pressure and flow based on the changing trend of the pressure and flow, and judge whether a sudden increase or decrease in pressure occurs during the feeding process. If so, the rotation speed of the feed pump is adjusted in time to control the fluctuation amplitude within a preset range.

[0025] A pressure transmitter and a flow meter are used to obtain pressure data and flow data of a fluorinating agent feed pipeline 10, and the pressure data and flow data are processed by median filtering to obtain filtered data, and the filtered data are eliminated by removing abnormal values ​​to obtain valid data; according to the valid data, an exponential weighted average is used to calculate the pressure change rate and the flow change rate, and the mean of the pressure change rate and the flow change rate is calculated through a moving window to determine whether a sudden change state is entered; for the sudden change state, standard state parameters are read from a database, and a proportional integral differential controller is used to calculate the feed pump speed adjustment amount according to the deviation value between the standard state parameter and the real-time parameter; an adaptive filter is used to process feedback data after the feed pump speed is adjusted, and the pressure fluctuation amplitude and the flow fluctuation amplitude are calculated through a sliding time window, and the control parameters of the proportional integral differential controller are adjusted online.

[0026] Exemplarily, the pressure data and flow data of the fluorination agent feed pipeline 10 are collected by a pressure transmitter and a flow meter, the sampling period is set to 0.1 seconds, the sampled data is subjected to median filtering to eliminate periodic interference, the data points that deviate from the mean by 5 times the standard deviation are eliminated, and the pressure change rate and the flow change rate are calculated by exponential weighted average to obtain a fluctuation trend curve. For the real-time monitoring of the pressure change rate and the flow change rate in the fluctuation trend curve, the moving window length of 1 second is used to calculate the change rate mean. When the pressure change rate exceeds 0.5 MPa per second or the flow change rate exceeds 5 cubic meters per hour per second, it is determined to enter a mutation state, and the standard state parameters corresponding to the current working condition are read from the database. According to the standard state parameter and the real-time parameter deviation value, the proportional integral differential controller is used to adjust the feed pump speed, the control period is synchronized with the sampling period, the integral time is 2 seconds, the differential time is 0.5 seconds, the speed output limit is plus or minus 30% of the rated speed, and the pump speed is adjusted in real time. An adaptive filter is used to process the feedback data after the pump speed is adjusted. The sliding time window length is set to 5 seconds. The pressure fluctuation amplitude and flow fluctuation amplitude are calculated in real time. The controller parameters are corrected through online tuning to keep the pressure fluctuation amplitude within 0.2 MPa and the flow fluctuation amplitude within 2 cubic meters per hour. During the operation data collection process of the fluorination agent feed pipeline 10, the pressure transmitter and the flow meter are arranged at the outlet of the feed pump and the end of the pipeline respectively. The sampling period is set to 0.1 seconds to achieve a rapid response to pressure fluctuations. Under standard working conditions, the pressure value is 0.8 MPa and the flow value is 100 cubic meters per hour. There is periodic interference caused by equipment vibration in the measured data, which is manifested as pressure values ​​fluctuating between 0.75 and 0.85 MPa and flow values ​​fluctuating between 95 and 105 cubic meters per hour. After the original data is processed by median filtering, the abnormal points that deviate from the mean by 5 times the standard deviation are eliminated. When the pipeline is suddenly blocked, the pressure value rises from 0.8 MPa to 1.2 MPa within 0.2 seconds. The pressure change rate is calculated to be 2 MPa per second, which is far more than the threshold of 0.5 MPa per second, triggering the sudden change state judgment. At this time, the flow rate drops sharply to 60 cubic meters per hour, and the change rate is 40 cubic meters per hour per second. In response to this sudden change condition, the controller immediately starts to respond, the proportional coefficient is 0.8, the integral time is 2 seconds, the differential time is 0.5 seconds, the rated speed of the feed pump is 1500 rpm, and the pump speed adjustment is calculated to be -300 rpm. The pump speed is reduced to 1200 rpm within 0.3 seconds, and the pressure begins to drop. During the adjustment process, a 5-second time window is used to monitor the pressure and flow. The maximum pressure in the window is 1.2 MPa, the minimum is 0.75 MPa, and the fluctuation amplitude is 0.45 MPa. Through adaptive parameter setting, the proportional coefficient is adjusted to 0.6, and the integration time is extended to 2.5 seconds. The pressure gradually stabilizes at 0.85 MPa, and the fluctuation amplitude is reduced to within 0.15 MPa.In actual applications, when the scaling degree of the fluorination agent feed pipeline 10 increases, the frequency of pressure mutations gradually increases, the duration of a single fluctuation is extended from the initial 3 seconds to 8 seconds, and the number of controller responses increases from 5 times to 12 times within 4 hours. Through the analysis of fluctuation characteristics, it is found that the pressure mean gradually increases. When the pressure baseline value exceeds 1 MPa, the pipeline cleaning process is automatically triggered. After the cleaning is completed, the pressure returns to 0.8 MPa, and the flow rate is maintained in a stable range of 98 to 102 cubic meters per hour.

[0027] The control unit is further used to obtain feed flow fluctuation data, use a spectrum analysis method to obtain the frequency characteristics of the fluctuation signal, and compare it with a preset frequency threshold to determine whether the frequency of the fluctuation is within a preset range of the fluctuation frequency. If it exceeds the preset range of the fluctuation frequency, an early warning message is generated to indicate the risk of pipeline scaling.

[0028] A flow sensor 14 is used to obtain fluctuation sampling data, and the fluctuation sampling data is filtered through a Butterworth low-pass filter to obtain a filtered fluctuation sequence; a fast Fourier transform and a power spectrum density calculation are performed on the filtered fluctuation sequence, a peak detection threshold is set according to the mean value of the power spectrum density, and a main frequency component is extracted from the fluctuation sequence; if the main frequency component exceeds a preset frequency threshold range, a support vector machine is used to classify the main frequency component to obtain a fluctuation state determination result; according to the fluctuation state determination result, the number of frequency exceeding limit times is counted, and when the number of frequency exceeding limit times exceeds a preset number threshold within a preset time, the scaling thickness value under the corresponding frequency characteristic is obtained from the database to generate an early warning information.

[0029] Exemplarily, the flow fluctuation data is collected by the feed flow sensor 14 at a sampling frequency of 100 Hz, and the fluctuation data is filtered by a Butterworth low-pass filter with a cutoff frequency of 20 Hz to eliminate the interference of the power frequency 50 Hz and high-frequency noise, so as to obtain a filtered fluctuation sequence, and the fluctuation sequence is processed in segments with a period of 10 minutes. The spectrum is calculated by fast Fourier transform for the segmented fluctuation sequence, the power spectrum density is calculated by the autocorrelation function, the peak detection threshold is set to 3 times the power spectrum mean, the main frequency components in the spectrum are extracted, and the main frequency components are sorted according to the amplitude to generate a frequency characteristic sequence. The fluctuation frequency range is determined according to the frequency characteristic sequence. If the main frequency component exceeds the preset threshold range of 0.1 Hz to 0.5 Hz, or the amplitude ratio of the frequency component exceeds the preset threshold, the frequency characteristic sequence is classified by a support vector machine to obtain the fluctuation state determination result. Based on the result of the fluctuation state judgment, the frequency exceeding the limit is counted according to the 1-minute update frequency. When the number of exceeding the limit exceeds 5 times within 10 minutes, the scaling thickness value under the corresponding frequency feature in the scaling record is extracted from the database to generate warning information. The warning information includes the exceeding frequency value, exceeding time, scaling thickness estimation value, and warning level number. The collection of flow fluctuation data adopts a sampling frequency of 100 Hz, and 100 data points are obtained per second to accurately capture rapid fluctuations. There is 50 Hz power frequency interference in the industrial site, mainly from motor vibration and electromagnetic interference. The Butterworth low-pass filter with a cutoff frequency of 20 Hz can effectively eliminate these interferences. During normal operation, the flow value fluctuates around 100 cubic meters per hour. The filtered fluctuation sequence shows that the fluctuation amplitude is within the range of plus or minus 2 cubic meters per hour. In the spectrum analysis stage, the fluctuation data is segmented into 10 minutes, each segment of data contains 60,000 sampling points, and the spectrum diagram is obtained by fast Fourier transform. The spectrum diagram during normal operation shows that the main frequency components are concentrated between 0.2 Hz and 0.3 Hz, with a corresponding period of 3 to 5 seconds, which is determined by the normal working characteristics of the feed pump. When the pipeline begins to scale, a high-frequency component of 0.8 Hz appears in the spectrum diagram, and the amplitude gradually increases, exceeding 3 times the mean of the power spectrum. A double threshold is set for the fluctuation state judgment, and the frequency range threshold is 0.1 Hz to 0.5 Hz, with a corresponding period of 2 seconds to 10 seconds. This is the normal fluctuation range obtained based on a large amount of operation data statistics. When a frequency component exceeds this range and its amplitude ratio exceeds 0.3, the support vector machine determines it as an abnormal fluctuation state. In actual operation, the initial scaling is manifested as a high-frequency fluctuation of 0.7 Hz to 0.9 Hz, with an amplitude ratio of 0.4 to 0.5. The early warning judgment is based on the frequency over-limit statistics, and the statistical value is updated every minute. When the number of over-limits exceeds 5 times within 10 minutes, the early warning is triggered.Historical data shows that when there is a continuous frequency fluctuation of 0.8 Hz, the scaling thickness of the pipeline has reached 4 mm to 6 mm. The warning information generated at this time includes an over-limit frequency of 0.8 Hz, an over-limit duration of 8 minutes, a predicted scaling thickness of 5 mm, and a warning level of 2. After the warning information is pushed, the operation and maintenance personnel can reasonably arrange the cleaning time according to the degree of scaling. In the process of gradually increasing scaling in the pipeline, the spectrum characteristics show an obvious change pattern. The fluctuation frequency gradually increases from the initial 0.25 Hz to 0.8 Hz, and the amplitude ratio increases from 0.1 to 0.5, reflecting the degree of impact of scaling on fluid flow. When the early warning system determines that the scaling thickness exceeds 8 mm, the pipeline must be cleaned immediately to avoid drastic fluctuations in flow caused by scaling.

[0030] In other embodiments, the control unit can be further used to obtain the temperature and pressure of the reactor during the fluorination reaction (which can be obtained by a temperature sensor and a pressure sensor arranged in the electrolytic cell 2), and predict the reaction conversion rate level under the current operating conditions in real time through the temperature and pressure in the reactor, and evaluate the completion and efficiency of the reaction process through conversion rate indicators, which include product yield, single-pass conversion rate and total conversion rate.

[0031] The temperature data and pressure data of the reactor are collected, and an operating condition deviation vector is obtained through median filtering, wherein the operating condition deviation vector is composed of a temperature deviation value and a pressure deviation value; historical data that meets the reaction time and purity requirements are screened out from a database according to the operating condition deviation vector, and a gradient boosting tree algorithm is used to train the historical data to obtain a conversion rate prediction function; the instantaneous consumption rate of raw materials and the product generation rate in the reactor are calculated through the conversion rate prediction function, and after obtaining the consumption rate and the generation rate, the material balance equation is solved to obtain a single-pass conversion rate and a total conversion rate; a sliding time window is used to update the single-pass conversion rate and the total conversion rate, and a conversion rate prediction value is obtained through the operating condition deviation vector calculation within the window, and process data is recorded when the window slides to a preset period.

[0032] Exemplarily, the reactor data is collected at a sampling interval of 1 second through a temperature sensor and a pressure transmitter, and a 5-point median filter is used to eliminate instantaneous fluctuations. The temperature data is normalized by subtracting the reaction standard temperature and the pressure data is normalized by subtracting the standard pressure to obtain the operating condition deviation vector. The operating condition deviation value and material conversion rate records in a single batch reaction process are read from the database, and historical data with reaction time within the set range and product purity that meets the standard are screened out. The selected historical data is trained using a gradient boosting tree algorithm, with the input variables including the operating condition deviation vector and the reaction time variable, and the output variables including the raw material consumption rate and the product generation rate. The number and depth parameters of the tree are determined through 5-fold cross validation, and a conversion rate prediction function is generated. The instantaneous consumption rate of the reactants and the product generation rate under the current operating conditions are calculated from the prediction function.

[0033]

[0034] , K represents the number of cross-validation folds, x i represents the i-th input variable, represents the average value of the i-th input variable in the k-th compromise, and n represents the total number of input variables. According to the instantaneous consumption rate and generation rate, the three material balance values ​​of raw material input, residual amount, and conversion amount in the reactor are cumulatively calculated, and the single-pass conversion rate and product yield are calculated from the material balance values. The total amount balance equation is constructed based on the material flow relationship between the single-pass reaction and the cyclic feeding, and the total conversion rate value is obtained by solving the equation. The product yield refers to the ratio of the actual amount of product obtained to the maximum amount of product that can be obtained from the raw material in theory.

[0035]

[0036] , P(t) represents the actual product quantity at time t, Y max It represents the maximum amount of product that can be theoretically obtained from the raw material, and F0 represents the initial raw material amount.

[0037]

[0038] , C(t) represents the number of conversions at time t, F0 represents the initial sample size, and this formula is used to calculate the one-way conversion rate, which is the percentage of the number of conversions at time t to the initial sample size.

[0039]

[0040] , F0 represents the initial feed amount, R represents the circulating feed amount, C(t) represents the consumption at time t, F(t) represents the outflow at time t, and P(t) represents the product amount at time t.

[0041]

[0042] , TCR represents the total conversion rate, F0 represents the initial feed amount, R represents the circulating feed amount, C(t) and P(t) represent the consumption and product amount at time t, respectively. A 60-second sliding time window is used to update the conversion rate value. The operating condition deviation vector is input to the prediction function at 1-second intervals in the window, and the conversion rate prediction value is output. When the window sliding distance reaches the preset update cycle, the conversion rate value at the end of the window is stored to form a process record including timestamp, operating condition parameters, and three conversion rates. The operating condition monitoring of the fluorination reaction process adopts a 1-second sampling frequency to achieve a rapid response to the temperature and pressure of the reactor. The standard reaction temperature is set at 120 degrees Celsius and the pressure is set at 0.6 MPa. In actual operation, the temperature fluctuates between 115 and 125 degrees Celsius and the pressure fluctuates between 0.55 and 0.65 MPa. The use of 5-point median filtering can eliminate high-frequency fluctuations caused by the agitator. The operating condition deviation vector reflects the degree of deviation between the actual operating parameters and the standard operating conditions. During normal operation, the temperature deviation is within plus or minus 5 degrees Celsius, and the pressure deviation is within plus or minus 0.05 MPa. The historical data screening focuses on the reaction time and product purity. The single batch reaction time is in the range of 4 hours to 6 hours, and the data with a product purity of more than 98% are used as training samples. When training the gradient boosting tree algorithm, the number of trees is set to 100 and the maximum depth is 6. The optimal parameter combination is determined by 5-fold cross validation, and the prediction error on the validation set is controlled within 3%. The prediction results show that under standard conditions, the raw material consumption rate is 200 kg per hour and the product generation rate is 180 kg per hour. During the material balance process, the raw material input is 1,000 kg. When the reaction is carried out for 2 hours, the measured remaining raw material amount is 600 kg, and the amount of product generated is 360 kg. The calculated single-pass conversion rate is 40% and the product yield is 90%. Considering that the unreacted raw materials are recycled after separation, 200 kg of fresh raw materials are added in each batch. After three cycles, the total conversion rate reaches 85%, realizing efficient utilization of raw materials. The online monitoring adopts a 60-second sliding window, which contains 60 sampling points. A new set of operating data is input every second and the predicted value is output. When the temperature rises to 130 degrees Celsius, the prediction shows that the raw material consumption rate increases to 250 kg per hour, but the product generation rate only increases to 200 kg per hour, indicating that the side reaction is aggravated. At this time, the single-pass conversion rate rises to 50%, but the product yield drops to 85%, and the reaction temperature needs to be adjusted in time. In actual application, it is observed that the reaction temperature and conversion rate show a significant correlation. For every 5 degrees Celsius increase in temperature, the raw material consumption rate increases by about 15%, while the product yield decreases by 2 to 3 percentage points. The effect of pressure on the conversion rate is relatively small. In the range of 0.55 to 0.65 MPa, the conversion rate changes by no more than 1 percentage point. Through real-time prediction and online monitoring, the impact of operating condition fluctuations on the conversion rate is timely discovered, and precise control of the reaction process is achieved.

[0043] In other embodiments, the control unit can be further used to perform correlation analysis between the feed flow fluctuation data and the reaction conversion rate prediction results, explore the influence rules between the feed fluctuation frequency, amplitude and conversion rate, perform prediction correction of the conversion rate under the influence of feed fluctuation, and use the percentage change of the fluctuation frequency and amplitude to correct the conversion rate prediction value to obtain the corrected conversion rate.

[0044] The feed flow data collected by the flow meter is obtained, and a fast Fourier transform is performed on the feed flow data to obtain spectrum data, and a main frequency component is extracted from the spectrum data; the main frequency component is segmented according to a preset frequency interval, and the correlation coefficient between each frequency band of the main frequency component and the reaction conversion rate is calculated, and a frequency band influence function is constructed using a long short-term memory network to obtain a frequency correction coefficient; a real-time frequency deviation value is calculated according to the frequency correction coefficient, and the frequency change percentage and the amplitude change percentage are calculated for the frequency deviation value, and a correction function is used to obtain a conversion rate correction amount; the conversion rate correction amount is updated using a sliding time window, and it is determined whether the error value between the corrected conversion rate and the measured conversion rate exceeds a preset threshold value, and if the error value exceeds the preset threshold value, parameter training of the correction function is triggered.

[0045] Exemplarily, the feed flow data is collected by the flow meter at a sampling frequency of 100 Hz, the flow data is fast Fourier transformed to obtain spectrum data, the frequency components in the range of 0.1 Hz to 10 Hz are extracted from the spectrum data, the main frequency components are screened according to the standard that the amplitude is greater than 3 times the mean, and the proportion of the main frequency components in the total fluctuation energy is calculated to obtain the frequency weight value. The reaction conversion rate data within a 10-minute time period is extracted from the database, the main frequency components are segmented at 0.1 Hz intervals, the correlation coefficient between the fluctuation amplitude and the conversion rate in each frequency band is calculated, and the frequency band influence function is constructed using a long short-term memory network. The number of input layer units is set to 24, the number of hidden layer units is set to 48, and the number of output layer units is set to 1, and the correction coefficient of the frequency change to the conversion rate is obtained. The real-time frequency deviation is calculated based on the correction coefficient of the frequency change, the frequency change percentage is calculated based on the standard frequency value, and the amplitude change percentage is calculated based on the standard amplitude value. The frequency change percentage and the amplitude change percentage are substituted into the correction function, and the conversion rate correction amount under the current working condition is calculated to correct the original predicted conversion rate.

[0046]

[0047] , Δη represents the conversion rate correction, f represents the correction function, P f Indicates the percentage of frequency change, P aIndicates the percentage change in amplitude. A 60-second sliding window is used to continuously update the conversion rate correction. The window slides once per second, and the latest frequency component and amplitude data are input in real time to calculate the new correction and update the conversion rate value. When the error between the corrected conversion rate and the measured conversion rate exceeds the preset threshold, the correction function parameter retraining is triggered. The feed flow fluctuation analysis uses a 100 Hz sampling frequency to obtain 100 data points per second to accurately capture rapid fluctuations. Under standard feed conditions, the flow value is maintained at 100 cubic meters per hour, and the fluctuation amplitude is within the range of plus or minus 2 cubic meters per hour. Spectral analysis shows that the main frequency components during normal operation are concentrated in the range of 0.2 to 0.3 Hz, corresponding to a fluctuation period of 3 to 5 seconds. When pipeline scaling or feed pressure fluctuations occur, high-frequency components of 0.8 to 1 Hz will appear in the spectrum, and the amplitude exceeds the mean by 3 times. The correlation analysis is based on data within a 10-minute time window, including 600 conversion rate sampling points. Taking the standard frequency of 0.25 Hz as the benchmark, the measured frequency is divided into multiple frequency bands at intervals of 0.1 Hz. Under normal working conditions, the fluctuation amplitude of the 0.2 to 0.3 Hz frequency band is negatively correlated with the conversion rate, with a correlation coefficient of -0.75, indicating that the increased fluctuation in this frequency band will lead to a decrease in the conversion rate. The correlation coefficient between the fluctuation amplitude of the 0.8 to 1 Hz frequency band and the conversion rate reaches -0.92, indicating that the impact of high-frequency fluctuations on the conversion rate is more significant. In the calculation of the correction coefficient, the long short-term memory network receives frequency data through 24 input units and outputs the correction value after processing by 48 hidden units. When the frequency shifts from 0.25 Hz to 0.8 Hz, the frequency change percentage is 220%, resulting in a 6 percentage point decrease in the conversion rate. At the same time, the fluctuation amplitude increases from 2 cubic meters per hour to 6 cubic meters per hour, and the amplitude change percentage is 200%, causing the conversion rate to further decrease by 4 percentage points. The correction function comprehensively considers these two factors, corrects the original predicted conversion rate of 95%, and obtains a correction value of 85%. The real-time update process uses a 60-second sliding window, which contains 6,000 sampling points. When the feed pump outlet pressure fluctuation is detected to be intensified, the amplitude of the 0.8 Hz component in the spectrum increases by 150% within 10 seconds, and the correction function responds immediately and outputs a new correction value. If the deviation between the corrected conversion rate and the measured value exceeds 3 percentage points, the correction function parameter update is triggered. In one operation, when the flow fluctuation intensified due to scaling in the fluorination agent feed pipeline 10, the frequency shifted to 0.9 Hz and the fluctuation amplitude increased to 8 cubic meters per hour. The correction function reduced the predicted conversion rate from 92% to 80%, which is in good agreement with the measured value of 79.5%.

[0048] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A pipeline flow monitoring device for the production of perfluorotripropylamine, characterized in that: The pipeline flow monitoring device for perfluorotripropylamine production comprises a fluorinating agent feeding unit, an electrolytic cell and a control unit; the fluorinating agent feeding unit comprises a fluorinating agent feeding pipeline connected to the top of the electrolytic cell, a detection port arranged on the fluorinating agent feeding pipeline close to one end of the electrolytic cell, an endoscope capable of entering the interior of the fluorinating agent feeding pipeline through the detection port, and a pressure sensor, a flow sensor and a temperature sensor arranged on the fluorinating agent feeding pipeline; the electrolytic cell, the endoscope, the pressure sensor, the flow sensor and the temperature sensor are respectively connected to the control unit: the endoscope is used to obtain scaling image data of the inner wall of the fluorinating agent feeding pipeline; The control unit is used to pre-process the image data according to the scaling image data of the inner wall of the fluorinating agent feeding pipeline, extract the image features of the scaling area of ​​the pipe wall, and calculate the scaling thickness value of the current inner wall of the pipeline according to the preset scaling thickness evaluation index; The control unit is also used to analyze the correlation between the scaling thickness and the feed flow fluctuation through historical data, and predict the feed flow fluctuation within a preset time period under the current scaling thickness; determine whether it is necessary to clean and maintain the fluorinating agent feed pipeline according to the feed flow fluctuation within the preset time period under the current scaling thickness; if the predicted feed flow fluctuation exceeds a preset fluctuation safety threshold, the pipeline cleaning process is triggered to clean the inner wall of the pipeline to remove scaling; The control unit is also used to collect the pressure and flow of the fluorinating agent feeding pipeline in real time during the fluorinating agent feeding process, and perform curve fitting processing on the pressure and flow of the fluorinating agent feeding pipeline to obtain the change trend of the pressure and flow, and obtain the change amplitude and frequency of the pressure and flow based on the change trend of the pressure and flow, and judge whether the pressure rises or drops suddenly during the feeding process. If so, adjust the speed of the feed pump in time to control the fluctuation amplitude within a preset range; The control unit is also used to obtain feed flow fluctuation data, use a spectrum analysis method to obtain the frequency characteristics of the fluctuation signal, and compare it with a preset frequency threshold to determine whether the fluctuation frequency is within a preset range of the fluctuation frequency. If it exceeds the preset range of the fluctuation frequency, an early warning message is generated to indicate the risk of pipeline scaling. According to the scaling image data of the inner wall of the fluorinating agent feeding pipeline, the image data is preprocessed to extract the image features of the scaling area of ​​the pipe wall, and the scaling thickness value of the current inner wall of the pipeline is calculated according to the preset scaling thickness evaluation index, including: performing circumferential scanning on the inner wall of the fluorinating agent feeding pipeline by a depth image acquisition probe to obtain an original depth image, and obtaining a denoised image according to the original depth image by using a contrast enhancement operator and a Gaussian filter; binarizing the denoised image by using the maximum inter-class variance method based on the pipe wall reference position and the gray value distribution of the pipe cross section, and obtaining a scaling area image by a morphological corrosion and expansion operation; extracting depth data from the scaling area image to construct a three-dimensional point cloud, using a random sample consistency algorithm to fit the pipe wall reference plane to obtain a scaling depth map, and using the depth data at the intersection of the scanning direction and the contour line to obtain the actual measured value of the scaling thickness; The method analyzes the correlation between scaling thickness and feed flow fluctuation through historical data to predict the feed flow fluctuation within a preset time period under the current scaling thickness, including: obtaining scaling thickness value and flow value in the pipeline operation history data, and obtaining multiple time period data according to the change rate of the scaling thickness value; establishing a physical model for the time period data, calculating the flow resistance coefficient through the physical model, and obtaining the corresponding relationship between pressure loss and flow by polynomial fitting; setting a sliding window according to the corresponding relationship, and extracting features from the data in the sliding window by using a random forest algorithm to obtain a scaling thickness gradient value and a flow fluctuation amplitude; and constructing a predictor by using a long short-term memory network, inputting the scaling thickness gradient value and the flow fluctuation amplitude into the predictor, and generating flow fluctuation prediction data.

2. The pipeline flow monitoring device for perfluorotripropylamine production according to claim 1, characterized in that: The method determines whether it is necessary to clean and maintain the fluorinating agent feed pipeline according to the feed flow fluctuation within a preset time period under the current scaling thickness. If the predicted feed flow fluctuation exceeds a preset fluctuation safety threshold, the pipeline cleaning process is triggered to clean the inner wall of the pipeline to remove scaling, including: Acquire the real-time feed flow rate and calculate the deviation between the real-time feed flow rate and the standard flow rate, and use moving average filtering to smooth the deviation value to obtain a flow fluctuation curve; A peak-valley detection operator is used to extract the fluctuation amplitude and fluctuation period of the flow fluctuation curve, and a fluctuation over-limit state judgment result is obtained according to the fluctuation amplitude and the fluctuation period; By analyzing the fluctuation over-limit state judgment result through time series statistics, if the fluctuation amplitude continuously exceeds the preset threshold, an exponentially weighted average calculation is adopted to obtain a cleaning trigger signal; According to the cleaning trigger signal, historical cleaning records are read from a database, a Kalman filter is used to perform recursive estimation on the historical cleaning records, and a cleaning execution instruction is generated according to the recursive estimation result.

3. The pipeline flow monitoring device for perfluorotripropylamine production according to claim 1, characterized in that: The real-time collection of the pressure and flow rate of the fluorinating agent feeding pipeline during the fluorinating agent feeding process, and the curve fitting processing of the pressure and flow rate of the fluorinating agent feeding pipeline to obtain the change trend of the pressure and flow rate, the change amplitude and frequency of the pressure and flow rate based on the change trend of the pressure and flow rate, and the judgment of whether the pressure rises or drops suddenly during the feeding process, if so, timely adjusting the speed of the feed pump to control the fluctuation amplitude within a preset range, including: A pressure transmitter and a flow meter are used to obtain pressure data and flow data of a fluorinating agent feed pipeline, the pressure data and flow data are processed by median filtering to obtain filtered data, and the filtered data are eliminated by removing abnormal values ​​to obtain valid data; Calculate the pressure change rate and the flow change rate by exponential weighted average according to the valid data, calculate the mean of the pressure change rate and the flow change rate by moving the window, and judge whether it enters the mutation state; Reading standard state parameters from a database for the sudden change state, and using a proportional-integral-differential controller to calculate a feed pump speed adjustment amount according to a deviation between the standard state parameters and the real-time parameters; An adaptive filter is used to process the feedback data after the speed adjustment of the feed pump. The pressure fluctuation amplitude and flow fluctuation amplitude are calculated through a sliding time window, and the control parameters of the proportional-integral-differential controller are adjusted online.

4. The pipeline flow monitoring device for perfluorotripropylamine production according to claim 1, characterized in that: The feed flow fluctuation data is obtained, and a spectrum analysis method is used to obtain the frequency characteristics of the fluctuation signal. By comparing with a preset frequency threshold, it is determined whether the fluctuation frequency is within a preset range of the fluctuation frequency. If it exceeds the preset range of the fluctuation frequency, an early warning message is generated to indicate the risk of pipeline scaling, including: A flow sensor is used to obtain fluctuation sampling data, and the fluctuation sampling data is filtered by a Butterworth low-pass filter to obtain a filtered fluctuation sequence; Performing fast Fourier transform and power spectrum density calculation on the filtered fluctuation sequence, setting a peak detection threshold according to the mean value of the power spectrum density, and extracting a main frequency component from the fluctuation sequence; If the main frequency component exceeds the preset frequency threshold range, a support vector machine is used to classify the main frequency component to obtain a fluctuation state determination result; The frequency exceeding limit times are counted according to the fluctuation state determination result. When the frequency exceeding limit times exceeds a preset threshold within a preset time, the scaling thickness value under the corresponding frequency characteristic is obtained from the database to generate warning information.

5. The pipeline flow monitoring device for perfluorotripropylamine production according to claim 1, characterized in that: The control unit is also used to obtain the temperature and pressure of the reactor during the fluorination reaction, and to predict the reaction conversion rate level under the current working conditions in real time through the temperature and pressure in the reactor, and to evaluate the completion and efficiency of the reaction process through the conversion rate index, wherein the conversion rate index includes product yield, single-pass conversion rate and total conversion rate; The control unit is also used to perform correlation analysis on the feed flow fluctuation data and the reaction conversion rate prediction results, explore the influence rules between the feed fluctuation frequency, amplitude and conversion rate, perform prediction correction of the conversion rate under the influence of feed fluctuation, and use the percentage change of the fluctuation frequency and amplitude to correct the conversion rate prediction value to obtain the corrected conversion rate.

6. The pipeline flow monitoring device for the production of perfluorotripropylamine according to claim 5, characterized in that: The temperature and pressure of the reactor during the fluorination reaction are obtained, the reaction conversion rate level under the current working conditions is predicted in real time by the temperature and pressure in the reactor, and the completion and efficiency of the reaction process are evaluated by the conversion rate index, wherein the conversion rate index includes product yield, single-pass conversion rate and total conversion rate, including: The temperature data and pressure data of the reactor are collected, and a working condition deviation vector is obtained by median value filtering, wherein the working condition deviation vector is composed of a temperature deviation value and a pressure deviation value; According to the working condition deviation vector, historical data that meets the reaction time and purity requirements are screened from the database, and the historical data are trained using a gradient boosting tree algorithm to obtain a conversion rate prediction function; The instantaneous consumption rate of the raw materials and the product generation rate in the reactor are calculated by the conversion rate prediction function. After obtaining the consumption rate and the generation rate, the material balance equation is solved to obtain the single-pass conversion rate and the total conversion rate. The single-pass conversion rate and the total conversion rate are updated using a sliding time window, and the conversion rate prediction value is obtained by calculating the operating condition deviation vector within the window. Process data is recorded when the window slides to a preset period.

7. The pipeline flow monitoring device for perfluorotripropylamine production according to claim 6, characterized in that: The feed flow fluctuation data is correlated with the reaction conversion rate prediction result, the influence law between the feed fluctuation frequency, amplitude and conversion rate is explored, the prediction correction of the conversion rate under the influence of feed fluctuation is performed, and the conversion rate prediction value is corrected by the change percentage of the fluctuation frequency and amplitude to obtain the corrected conversion rate, including: Obtaining feed flow data collected by a flow meter, performing fast Fourier transform on the feed flow data to obtain spectrum data, and extracting a main frequency component from the spectrum data; According to the main frequency component, the frequency is divided into segments according to preset frequency intervals, and the correlation coefficient with the reaction conversion rate is calculated for each frequency band of the main frequency component, and a frequency band influence function is constructed using a long short-term memory network to obtain a frequency correction coefficient; Calculate the real-time frequency deviation value according to the frequency correction coefficient, calculate the frequency change percentage and the amplitude change percentage for the frequency deviation value, and use the correction function to obtain the conversion rate correction amount; The conversion rate correction amount is updated using a sliding time window to determine whether the error value between the corrected conversion rate and the measured conversion rate exceeds a preset threshold. If the error value exceeds the preset threshold, parameter training of the correction function is triggered.

Images