A control method of a device for deep removal of hexogen based on a magnesium nitrate solution

Abstract

The application provides a control method of a RDX deep removal device based on a magnesium nitrate solution, comprising the following steps: acquiring running data of the RDX deep removal device in real time, including physical parameters of an inner wall of the device, the physical parameters including temperature, pressure and flow rate physical parameters, and simultaneously establishing a physical parameter benchmark value under a scaling state; judging whether scaling occurs in the device by comparing the physical parameters with the benchmark value; if scaling occurs in the device, obtaining micro-morphological characteristics of the scaling by means of a scanning electron microscope, identifying and positioning scaling distribution, and acquiring component information of a scaling sample to obtain component, morphology and distribution information of the scaling; comparing the real-time monitored physical parameters of the device with removal effects under different scaling conditions obtained by prediction, and adjusting temperature, pressure and flow rate running parameters of the device by means of a preset control system if it is found that the device running parameters deviate from a target range.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a control method for a device for deep removal of RDX from magnesium nitrate solution. Background Technology

[0002] Scaling on the inner walls of RDX equipment has long been a problem hindering the industry's development. Scaling not only reduces heat transfer efficiency and increases energy consumption, but also leads to decreased production capacity and even safety accidents. Traditional methods rely mainly on periodic shutdowns for maintenance and manual judgment, which is inherently delayed, failing to detect early-stage scaling and often missing the optimal treatment window. Relying on experience lacks scientific basis and makes it difficult to accurately assess the severity and impact of scaling. Furthermore, even when scaling is detected, traditional cleaning methods typically employ strong acids and alkalis or physical methods like high-pressure water jets. These methods can not only corrode or damage the equipment but also generate secondary pollution. Therefore, achieving early warning of scaling in RDX production equipment has become an urgent problem to solve. Moreover, accurately determining the degree, location, and type of scaling without shutting down the equipment presents a significant challenge. In addition, developing precise, efficient, and non-destructive cleaning solutions based on the specific scaling conditions is a long-standing goal pursued by the industry. Finally, how to suppress further scaling and extend equipment lifespan by optimizing operating parameters after scaling has occurred is also an important research direction. Summary of the Invention

[0003] This invention provides a control method for a deep removal device of RDX in magnesium nitrate solution, mainly comprising:

[0004] Real-time acquisition of operating data from the RDX deep removal equipment, including physical parameters of the equipment's inner wall, such as temperature, pressure, and flow rate. Simultaneously, establishing benchmark values ​​for physical parameters under scaling conditions, and determining whether scaling has occurred in the equipment by comparing the physical parameters with the benchmark values.

[0005] If scaling occurs in the equipment, the microscopic morphological characteristics of the scaling are obtained by scanning electron microscopy, the distribution of the scaling is identified and located, and the composition information of the scaling sample is obtained to obtain the composition, morphology and distribution information of the scaling.

[0006] A multi-scale numerical model is established based on scaling information to simulate the interaction between RDX and the scaling layer under fluid action, outputting equipment status and scaling layer changes, establishing a correlation model between scaling characteristics and removal efficiency, and predicting the removal effect under different scaling conditions.

[0007] The real-time monitored physical parameters of the equipment are compared with the predicted removal effect under different scaling conditions. If the equipment operating parameters are found to deviate from the target range, the temperature, pressure, and flow rate operating parameters of the equipment are adjusted through the preset control system.

[0008] Continuously monitor the changes in physical parameters of the inner wall of the equipment to determine whether scaling has been inhibited. If the monitored physical parameters tend to stabilize or return to the baseline value of the preset scale-free state, and the scaling thickness or scaling area stops increasing or begins to decrease, it is determined that the scaling has been inhibited. The removal efficiency of RDX is then evaluated, and the evaluation results are fed back to the correlation model for optimization and update.

[0009] If, after adjusting the operating parameters, the monitored physical parameters of the equipment's inner wall continue to deviate from the baseline value under scale-free conditions, or if the scale thickness or scale area continues to increase, or even accelerates, then the scale problem has not been alleviated after adjusting the operating parameters. In this case, the preset equipment cleaning program will be started to clean the inner wall of the equipment and remove the scale layer.

[0010] Furthermore, the real-time acquisition of operating data from the RDX deep removal equipment includes physical parameters of the equipment's inner wall, such as temperature, pressure, and flow rate. Simultaneously, a baseline value for these physical parameters under scaling conditions is established. By comparing the physical parameters with the baseline value, it is determined whether scaling has occurred on the equipment, including:

[0011] Physical parameter values ​​of the RDX deep removal equipment under current operating conditions are obtained from temperature, pressure, and flow rate sensors. The data acquisition module periodically samples and records these physical parameters at preset sampling intervals. An exponentially weighted moving average method is used to smooth the collected temperature, pressure, and flow rate physical parameters. The statistical mean of the smoothed physical parameters is calculated and used as the baseline value for the physical parameters under scaling conditions. The difference between the smoothed physical parameters and the baseline value is calculated, and the standard deviation of the difference sequence is calculated using the least squares method to obtain the equipment's internal wall condition parameters. A preset scaling condition judgment value is compared with the equipment's internal wall condition parameters. If the internal wall condition parameters are greater than the scaling condition judgment value, scaling is determined to have occurred in the RDX deep removal equipment.

[0012] Furthermore, if scaling occurs in the equipment, the microscopic morphological characteristics of the scaling are obtained using a scanning electron microscope to identify and locate the scaling distribution, and the compositional information of the scaling sample is obtained to obtain information on the composition, morphology, and distribution of the scaling, including:

[0013] The scaled surface was scanned and imaged using a scanning electron microscope (SEM). Microscopic morphology images of the scale were acquired based on preset scanning area size and pixel resolution, resulting in high-resolution SEM images of the scaled surface. For these SEM images, an image segmentation algorithm was used to identify scaled regions, calculate the area ratio and thickness distribution of these regions, and generate a scale distribution density map. Based on this map, the region with the highest scale density was selected as a sampling point, and elemental analysis was performed using an energy dispersive spectroscopy (EDS) instrument. Peak identification and quantitative analysis of the acquired EDS data were conducted to obtain the mass fraction of each element in the scale sample. A 3D reconstruction algorithm was used to model the scaled region, and the scale thickness was calculated using vertical height information to generate a scale thickness distribution map. Based on the scale distribution density map, elemental mass fractions, and scale thickness distribution map, a scale feature dataset was established. A convolutional neural network was used to identify the phase structure characteristics of the scale, obtaining correlation feature data between scale composition and distribution. Based on this correlation feature data, the scaled regions were classified and labeled, generating a scale feature map containing information on scale composition, morphology, and distribution.

[0014] Furthermore, the process involves establishing a multi-scale numerical model based on scaling information to simulate the interaction between RDX and the scaling layer under fluid action, outputting equipment status and scaling layer changes, establishing a correlation model between scaling characteristics and removal efficiency, and predicting the removal effect under different scaling conditions, including:

[0015] Based on the composition, morphology, and spatial distribution data of the scale layer, a multi-scale numerical model of the scale layer was constructed using the finite element method. Mesh generation was performed on the porosity and surface roughness of the scale layer to obtain discretized geometric structure data. Based on this discretized geometric structure data, a set of fluid dynamics equations was established, including continuity and momentum equations. A two-phase flow solver was used to calculate the flow field distribution data at the fluid-scale layer interface. Based on the flow field distribution data and the mass transfer equation, the RDX concentration field was calculated, and a correlation equation between the concentration gradient and the mass transfer coefficient at the scale layer interface was established to obtain interfacial mass transfer flux data. The local dissolution rate of the scale layer was calculated using dissolution kinetics equations, and heat transfer data was obtained by solving the temperature field equation, resulting in dynamic evolution data of the scale layer's geometry. Based on this dynamic evolution data of the scale layer's geometry, a dissolution process dataset was established, including scale layer morphological characteristics, interfacial mass transfer flux, and local dissolution rate. A deep neural network was used to construct a correlation model between scale characteristics and removal efficiency. Based on the correlation model and preset operating parameters, the gradient boosting tree algorithm is used to calculate the amount of scale dissolved and the dissolution time under different scale distribution conditions, so as to obtain the predicted data of scale removal efficiency.

[0016] Furthermore, the process of comparing the real-time monitored physical parameters of the equipment with the predicted removal effects under different scaling conditions, and if the equipment operating parameters are found to deviate from the target range, adjusting the equipment's temperature, pressure, and flow rate operating parameters through a preset control system, includes:

[0017] The system collects current equipment operating parameters from temperature, pressure, and flow rate sensors. The collected parameter data is then processed using a moving average method according to a preset sampling period, set to one-quarter of the operating parameter response time, to obtain real-time equipment operating status data. This real-time operating status data is matched with standard operating condition data in a removal effect prediction database. The Manhattan distance is used to calculate the deviation values ​​of the operating parameters, yielding temperature, pressure, and flow rate deviation data. Parameter adjustment sequences are established based on these deviations. A long short-term memory network is used to predict the changing trends of each parameter, with input features including historical deviation values ​​and adjustment records, to obtain the adjustment direction of the operating parameters. Target values ​​are set for the temperature, pressure, and flow rate controllers based on the adjustment direction of the operating parameters. Control command sequences are generated based on the adjustment dead zone and adjustment step size of each controller. A proportional-integral-derivative (PID) arithmetic logic unit (PIA) is used to optimize the control command sequences. The proportional coefficient is determined by the slope of the parameter response curve, the integral time by the parameter adjustment period, and the derivative time by the parameter fluctuation period, resulting in the corrected control commands.

[0018] Furthermore, the continuous monitoring of changes in the physical parameters of the inner wall of the equipment determines whether scaling has been suppressed. If the monitored physical parameters tend to stabilize or return to the baseline value of the preset scale-free state, and the scaling thickness or scaling area stops increasing or begins to decrease, then it is determined that scaling has been suppressed. The removal efficiency of RDX is then evaluated, and the evaluation results are fed back to the correlation model for optimization and updating, including:

[0019] The equipment's internal wall temperature, pressure, and flow rate are collected according to a preset monitoring cycle. The short-term rate of change of these physical parameters is calculated using an exponentially weighted moving average method, and the fluctuation period is obtained through autocorrelation analysis. Trend decomposition is performed on the continuously collected physical parameter sequence, and the root mean square error (RMSE) of the parameter sequence is calculated based on the decomposition results to determine whether the physical parameter sequence has entered a stable state. Image segmentation algorithms are used to calculate the scale thickness and area, and the rate of change of scale characteristic quantities is obtained through differential operations. The degree of physical parameter recovery is calculated by comparing this rate of change with a preset scale-free state baseline. An inhibition threshold is set based on the degree of physical parameter recovery. If the fluctuation value of the physical parameters within the continuous monitoring cycle is less than the preset threshold, and the rate of change of the scale characteristic quantities is negative, then scaling is determined to be suppressed.

[0020] Furthermore, if, after adjusting the operating parameters, the monitored physical parameters of the equipment's inner wall continue to deviate from the baseline value under scale-free conditions, or if the scale thickness or area continues to increase, or even accelerates, then the scaling problem has not been alleviated after adjusting the operating parameters. In this case, a preset equipment cleaning program is initiated to clean the equipment's inner wall and remove the scale layer, including:

[0021] Physical parameters of the equipment's inner wall are collected using temperature, pressure, and flow rate sensors. The collected data is processed using a moving average according to a preset sampling period to calculate a sequence of differences between the physical parameters and the baseline values ​​for a scale-free state. Temperature deviation rate, pressure deviation rate, and flow rate deviation rate are calculated based on this sequence, and a weighted summation method is used to obtain a comprehensive deviation index to determine the deviation status of the physical parameters. Images of the scaled area are acquired using a scanning electron microscope. The scale thickness and area are calculated using an image segmentation algorithm, and the rate of change of scale characteristic quantities is obtained using a difference method. A convolutional neural network is used to extract scale morphology features, and a scale state assessment index is constructed based on the rate of change of scale characteristic quantities to determine the growth trend of the scale layer. A bivariate correlation matrix is ​​established based on the comprehensive deviation index and the scale state assessment index. A Markov chain is used to predict the development trend of the scale problem, obtaining a scale problem mitigation assessment value. For operating conditions where the scale problem mitigation assessment value exceeds a preset threshold, a chemical cleaning agent mixing command and a mechanical cleaning head position command are generated to activate the cleaning actuator. An acoustic sensor is used to monitor the scaling layer peeling status during the cleaning process, and an online turbidity meter is used to detect the scaling content in the cleaning solution, thereby achieving closed-loop control of the cleaning process.

[0022] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0023] This invention discloses a control method for a deep RDX removal device based on magnesium nitrate solution. The method establishes a scaling state baseline by real-time monitoring of the device's internal physical parameters and uses scanning electron microscopy to obtain the microscopic characteristics of the scaling. Based on the scaling information, a multi-scale numerical model is constructed to simulate the interaction between RDX and the scaling layer, predicting the removal effect under different scaling conditions. This invention compares the real-time monitoring data with the prediction results and automatically adjusts the device's operating parameters to inhibit scaling growth. If parameter adjustment is ineffective, a preset cleaning program is initiated. Through continuous monitoring and evaluation, this invention continuously optimizes the correlation model, improving RDX removal efficiency. This method achieves intelligent operation of the deep RDX removal device, effectively solving the scaling problem and improving device performance and lifespan. Attached Figure Description

[0024] Figure 1 This is a flowchart of a control method for a deep removal device for RDX in magnesium nitrate solution according to the present invention. Detailed Implementation

[0025] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments in this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0026] like Figure 1 The control method for a deep removal device of RDX in magnesium nitrate solution in this embodiment may specifically include:

[0027] S101, if the control function of the RDX deep removal equipment is triggered, the physical parameters of the equipment during operation, including temperature, pressure and flow rate, are collected in real time, and a baseline value under the scaling state is established through data processing. The comparison result between the real-time parameters and the baseline value is used to determine whether scaling has occurred on the inner wall of the equipment.

[0028] In this embodiment of the invention, the RDX deep removal device has a real-time monitoring function. This function can be activated by the device's built-in control system or remotely via external commands, with the specific triggering method determined according to the actual application scenario. The collected physical parameters originate from sensors on the inner wall of the device, including temperature sensors, pressure sensors, and flow rate sensors. The temperature sensor uses a high-precision platinum resistance thermometer, with a measurement range covering 0 to 200 degrees Celsius and an accuracy of 0.1 degrees Celsius; the pressure sensor adopts a piezoresistive design, with a range of 0 to 10 MPa and an accuracy of 0.01 MPa; the flow rate sensor is an electromagnetic flowmeter, with a measurement range of 0 to 15 meters per second and an accuracy of 0.1 meters per second. These sensors periodically record data at 10-second sampling intervals through a data acquisition module, forming time-series data for subsequent analysis.

[0029] S1011, physical parameters are collected by sensors and the data is smoothed to generate stable benchmark values. In this embodiment of the invention, the data acquisition module stores the temperature, pressure, and flow rate data collected by the sensors as a sampling sequence at preset time intervals. Since random interference may occur during equipment operation, such as local eddies or external vibrations, the raw data may contain noise. To ensure data reliability, an exponentially weighted moving average method is used to smooth the sampling sequence, with a smoothing coefficient set to 0.3 to preserve the parameter change trend and filter out instantaneous fluctuations. The smoothed data is used to calculate the statistical mean based on an 8-hour operating cycle, yielding a temperature benchmark of 85 degrees Celsius, a pressure benchmark of 3.5 MPa, and a flow rate benchmark of 8 meters per second, serving as reference standards under scaling conditions.

[0030] S1012 performs difference analysis between the smoothed data and the baseline value to calculate the internal wall condition parameters of the equipment and determine the scaling status.

[0031] After acquiring smoothed data, this embodiment of the invention analyzes the changes in the internal wall condition of the equipment by calculating the difference between real-time physical parameters and baseline values. The standard deviation of the difference sequence is calculated using the least squares method to generate the internal wall condition parameters. When scaling occurs in the equipment, the deposits on the internal wall cause a temperature increase of 2 to 5 degrees Celsius, a pressure increase of 0.5 to 1 MPa, and a flow rate decrease of 1 to 2 meters per second. Through statistical analysis of a large amount of operational data, a scaling threshold of 0.8 is set for the condition parameters. If the calculated condition parameters exceed this threshold, scaling is determined to have occurred on the internal wall of the equipment. At this time, the system records the deviation of the current parameters to provide a basis for subsequent processing.

[0032] In this embodiment of the invention, if scaling is detected, abnormal fluctuations in sensor data are further analyzed. Warning thresholds for temperature, pressure, and flow rate are set at 95 degrees Celsius, 5 MPa, and 12 m / s, respectively. Data points exceeding these thresholds are marked as abnormal. The marked data are corrected using a convolutional neural network. This model contains three convolutional layers and two fully connected layers. The input is a sequence of abnormal data at 128 time points and their temporal characteristics; the output is the corrected physical parameter values. The corrected data is used to dynamically update baseline values ​​to adapt to long-term changes in equipment operating conditions while maintaining sensitivity to scaling deviations.

[0033] In this embodiment of the invention, the above steps achieve accurate monitoring of the scaling status of the RDX deep removal equipment. Real-time acquisition and data processing can promptly detect scaling problems on the inner wall, preventing a decrease in heat transfer efficiency or flow channel blockage due to scaling deterioration. The high-precision design of the sensors and data smoothing processing ensure the reliability of the monitoring results, while the dynamic update mechanism enhances the method's adaptability to changes in equipment operation. By comparing with preset thresholds, not only can the occurrence of scaling be determined, but data support can also be provided for subsequent control strategies, thereby improving the operational stability of the equipment and the treatment effect of RDX.

[0034] It is understood that the embodiments of the present invention do not impose too many restrictions on the specific implementation of the sensor type and sampling interval, which can be adjusted by technicians according to actual needs to adapt to different production environments and equipment characteristics.

[0035] S102. If it is determined that scaling has occurred in the RDX deep removal equipment, a high-resolution imaging of the microscopic morphology of the scaling surface on the inner wall of the equipment is performed using a scanning electron microscope and features are extracted. At the same time, energy dispersive spectroscopy and three-dimensional reconstruction technology are combined to comprehensively obtain information on the composition, morphological characteristics and spatial distribution of the scaling, and establish a detailed scaling feature map to support the formulation of subsequent control strategies.

[0036] In this embodiment of the invention, when the inner wall of the device is determined to be in a state of scaling, a scanning electron microscope (SEM) is immediately activated to perform a detailed analysis of the scaled area. The SEM acquires data using a partitioned scanning method, with the scanning area set at 500 μm × 500 μm and a pixel resolution of 2048 × 2048. Each pixel corresponds to an actual size of approximately 0.24 μm, ensuring the capture of minute structural details on the scaled surface. The operating voltage is set to 20 kV, and the working distance is maintained at 10 mm. High-contrast surface morphology images are generated using backscattered electron imaging technology, providing reliable basic data for subsequent feature extraction. This high-resolution imaging method can clearly present the microscopic texture and distribution characteristics of the scale, laying the foundation for accurate identification of scaled areas.

[0037] S1021 performs image segmentation and density distribution analysis based on scanning electron microscope images to generate a scale distribution density map.

[0038] In this embodiment of the invention, an image segmentation algorithm is used to automatically identify scaled areas in the acquired scanning electron microscope (SEM) images. The segmentation process first distinguishes between scaled and non-scaled areas using a grayscale thresholding method. Since scaled areas typically exhibit higher grayscale values ​​in backscattered electron images, a dynamic threshold can be set to adapt to image characteristics under different lighting conditions. After segmentation, the area ratio of the scaled region is calculated; for example, in one analysis, the scaled region occupies 35% of the field of view. A scale distribution density map is then generated. The map uses a pseudo-color visualization method, where red marks areas with a scale density exceeding 50%, and blue indicates areas with a density below 20%, visually reflecting the spatial distribution pattern of the scale. Analysis shows that scale distribution often exhibits a clustered characteristic, with local densities reaching up to 65%, which is closely related to fluid scouring and deposition conditions, providing a basis for subsequent sampling point selection.

[0039] S1022 uses energy dispersive spectroscopy to detect the scale components and combines them with three-dimensional reconstruction technology to extract the elemental composition and thickness distribution characteristics of the scale.

[0040] After determining the scale distribution density map, this embodiment of the invention selects the area with the highest density as the focus of analysis and uses an energy dispersive spectroscopy (EDS) instrument to detect elemental composition. The detector of the EDS instrument is held at a 40-degree angle to the sample surface to optimize the acquisition efficiency of X-ray signals. Five representative sampling points are selected in the high-density area for analysis, and the mass fraction of each element is obtained through peak identification and quantitative calculation. For example, the detection results show that the scale is mainly composed of calcium, sulfur, and oxygen, with mass fractions of 32.5%, 15.8%, and 48.2%, respectively, and trace elements account for about 3.5%, indicating that the scale may be mainly composed of calcium sulfate. Subsequently, image pairs are acquired using a scanning electron microscope at 0-degree and 45-degree tilt angles, and the three-dimensional structure of the scale is calculated using a three-dimensional reconstruction algorithm. The algorithm recovers height information through the principle of stereo vision and generates thickness distribution data. The results show that the scale thickness varies between 15 micrometers and 85 micrometers, with an average of 45 micrometers, and locally forms an accumulation band with a width of about 200 micrometers. This non-uniformity of thickness distribution reflects the influence of fluid dynamics on scale deposition and provides key input for subsequent model analysis.

[0041] In this embodiment of the invention, a scale feature dataset is constructed by integrating scale distribution density maps, elemental composition data, and thickness distribution information. To further analyze the phase structure of the scale, a convolutional neural network is used for feature recognition. The network design includes five convolutional layers, which extract and fuse morphological, compositional, and thickness features, respectively. The training input includes density distribution, elemental mass fraction, and thickness data. After training, the network identifies the scale as mainly composed of calcium sulfate crystals, with particle sizes between 10 and 30 micrometers, and a morphology that is predominantly a mixture of needle-like and plate-like structures. This phase characteristic is consistent with the banded distribution pattern in high-flow-rate regions, indicating that fluid shear force may be the main driving factor for scale morphology formation.

[0042] Finally, this embodiment of the invention generates a feature map containing information on the composition, morphology, and distribution of scale through the above analysis. The map uses color coding to mark different phase regions; for example, calcium sulfate scale is marked in red. Thickness contour lines are also overlaid to visually display the thickness variation trend. The map reveals the banded distribution characteristics of scale extending along the flow direction. This not only provides evidence for verifying fluid dynamics hypotheses but also provides precise data support for adjusting operating parameters and formulating cleaning strategies. It is understood that the scanning area size, resolution, and algorithm parameters can be flexibly adjusted by technicians according to equipment characteristics and scale complexity to ensure the applicability of the analysis results. Through the implementation of this step, the multidimensional characteristics of scale can be comprehensively grasped, laying a solid foundation for the dynamic control of RDX deep removal equipment.

[0043] S103. Based on the scaling information, a multi-scale numerical model is established to simulate the interaction between RDX and the scaling layer under fluid action. The model outputs the equipment status and scaling layer changes, establishes a correlation model between scaling characteristics and removal efficiency, and predicts the removal effect under different scaling conditions. The scaling information includes scaling composition, morphology and distribution information; the equipment status includes flow field, pressure, temperature and RDX concentration distribution.

[0044] In this embodiment of the invention, based on previously acquired scaling characteristic data, including elemental composition, microstructure, and spatial distribution information, numerical simulation is used to analyze the evolution process of the scaling layer under fluid action. The model construction first discretizes the geometric structure of the scaling layer. Considering the non-uniformity of its surface roughness and porosity, the finite element method is used to generate a mesh. In the high-roughness protruding regions, the mesh size is set to 5 micrometers to capture detailed features, while a 20-micrometer mesh is used in flat regions, ultimately generating a computational domain of approximately 500,000 hexahedral elements. This adaptive mesh generation method balances computational accuracy and efficiency, ensuring that the simulation results accurately reflect the complex geometric characteristics of the scaling layer.

[0045] S1031, calculates the flow field distribution at the scale layer interface and analyzes the fluid behavior using fluid dynamics equations.

[0046] In this embodiment of the invention, based on the discretized geometry, a set of fluid dynamics equations, including continuity and momentum equations, is established to describe the interaction between the fluid and the scale layer. The inlet boundary of the computational domain is set as a velocity inlet condition with a fixed flow velocity of 2 m / s, while the outlet boundary uses a pressure outlet condition to simulate actual working conditions. The flow field distribution is calculated using a VOF two-phase flow solver. The results show that the fluid velocity on the scale layer surface exhibits significant spatial differences, with the highest velocity reaching 3.5 m / s in the raised region and dropping to as low as 0.5 m / s in the recessed region. This non-uniformity in velocity distribution stems from the geometrical blocking effect of the scale surface, directly affecting subsequent mass transfer and dissolution processes. During the calculation, the k-ε model is used as the turbulence model to further optimize the simulation of shear force in the high-velocity region, ensuring the reliability of the flow field data. S1032 calculates the mass transfer flux and dissolution rate of the scale layer using mass transfer and heat transfer equations.

[0047] After acquiring the flow field distribution data, this embodiment of the invention further constructs a mass transfer equation to analyze the changes in the RDX concentration field. The inlet RDX concentration is set to 500 mg / L, and the concentration gradient and interfacial mass transfer coefficient are calculated using the convection-diffusion equation. The results show that the mass transfer coefficient in the raised region reaches 0.002 m / s due to the higher flow velocity, while it is only 0.0005 m / s in the recessed region, reflecting the significant influence of the flow field on mass transfer behavior. Simultaneously, the local dissolution rate of the scale layer is calculated by combining the dissolution kinetics equation and the temperature field equation. The temperature field solution shows that at an operating temperature of 85°C, the surface temperature of the scale layer is distributed between 80 and 90°C, and the dissolution rate in the locally high-temperature and high-mass-transfer-coefficient regions can reach up to 0.05 mm / h. This multi-physics coupled calculation method can comprehensively reveal the synergistic effect of fluid, concentration, and temperature on scale dissolution, providing multi-dimensional data support for predicting removal efficiency.

[0048] In this embodiment of the invention, based on the above calculation results, a dynamic evolution dataset containing scale morphology characteristics, interfacial mass transfer flux, and local dissolution rate is generated. The dataset covers 2000 sets of data under different operating conditions, each set including 48 feature parameters, such as porosity, surface roughness, flow rate, and temperature. A deep neural network is used to construct a correlation model between scale characteristics and removal efficiency. The network is designed with a 5-layer structure, with the number of neurons being 48, 96, 64, 32, and 16 respectively. The ReLU activation function is used, and the model is trained after 50,000 iterations. The trained model is combined with a gradient boosting tree algorithm to predict the scale removal effect under different operating conditions. For example, tests were conducted within the range of inlet flow rate of 1 to 4 m / s, temperature of 70 to 100 degrees Celsius, and RDX concentration of 300 to 700 mg / L. The prediction results show that under optimal operating conditions, the scale layer can be completely dissolved within 4 hours, while under unfavorable operating conditions, the dissolution time is extended to more than 12 hours.

[0049] Through multi-scale numerical simulations in this invention, not only can the flow field, pressure, temperature, and RDX concentration distribution within the equipment be output, but the geometric changes of the scale layer can also be dynamically tracked, providing a scientific basis for process parameter optimization. It is understood that the mesh generation strategy and solver selection can be adjusted by technicians according to equipment characteristics to adapt to the simulation needs of different scale types and operating conditions. This method, by coupling physical field calculations with machine learning predictions, significantly improves the accuracy of removal efficiency predictions, laying the foundation for intelligent control of RDX deep removal equipment.

[0050] S104 compares the physical parameters monitored in real time by the RDX deep removal equipment with the predicted scaling removal effect. If the operating parameters are found to deviate from the expected target, the temperature, pressure and flow rate are dynamically adjusted using the preset control system to optimize equipment performance and improve scaling treatment efficiency.

[0051] In this embodiment of the invention, the monitoring of equipment operating parameters relies on a high-precision sensor array, including a temperature sensor employing a platinum resistance thermometer with a measurement range of 0 to 200 degrees Celsius and an accuracy of 0.1 degrees Celsius; a piezoresistive pressure sensor with a range of 0 to 10 MPa and an accuracy of 0.01 MPa; and a flow velocity sensor composed of an electromagnetic flowmeter with a measurement range of 0 to 15 m / s and an accuracy of 0.1 m / s. The collected parameter data are sampled according to their respective dynamic response characteristics, with a temperature response time of 60 seconds and a sampling period of 15 seconds; a pressure response time of 10 seconds and a sampling period of 2.5 seconds; and a flow velocity response time of 5 seconds and a sampling period of 1.25 seconds. The collected data is smoothed using a moving average method, with the window size set to 5 times the sampling period to filter out instantaneous noise and generate stable real-time operating status data, providing a reliable basis for subsequent deviation analysis.

[0052] S1041 calculates parameter deviations based on Manhattan distance and establishes adjustment sequences to predict adjustment directions.

[0053] After acquiring real-time operating status data, this embodiment of the invention matches it with the optimal parameter combination in a standard operating condition database. The standard operating condition database stores reference data for different degrees of scaling; for example, for light scaling, the temperature is 85 degrees Celsius, the pressure is 3.5 MPa, and the flow rate is 2 m / s, while for heavy scaling, the temperature needs to be increased to 95 degrees Celsius, the pressure to 4.5 MPa, and the flow rate to 3 m / s. The Manhattan distance algorithm is used to calculate the deviation between the real-time data and the standard data; the formula is the sum of the absolute differences of each parameter, and the results generate temperature deviation, pressure deviation, and flow rate deviation, respectively. If the deviation exceeds a preset threshold, i.e., a temperature deviation greater than 5 degrees Celsius, a pressure deviation greater than 0.5 MPa, or a flow rate deviation greater than 0.3 m / s, an adjustment process is triggered. Subsequently, the deviation data is organized into a time series and input into a Long Short-Term Memory (LSTM) network for trend prediction. This network adopts a two-layer structure, containing 128 memory units. Input features include deviation values ​​from the past 4 hours and historical adjustment records, with 12 data points collected per hour. By analyzing historical trends, the network outputs the direction of parameter adjustment, such as whether the temperature needs to be increased or the flow rate needs to be increased. The prediction accuracy can reach 92% after training, which significantly improves the scientific nature of regulation decisions.

[0054] In this embodiment of the invention, target values ​​for the temperature, pressure, and flow rate controllers are set according to the predicted adjustment direction, and an initial control command sequence is generated. The dead zone of each controller is set to twice the measurement accuracy, i.e., the temperature dead zone is 0.2 degrees Celsius, the pressure dead zone is 0.02 MPa, and the flow rate dead zone is 0.2 m / s; the adjustment step size is 2 degrees Celsius, 0.2 MPa, and 0.1 m / s, respectively, based on the response characteristics. To further optimize the commands, a proportional-integral-derivative (PID) arithmetic unit is used to finely adjust the sequence. The proportional coefficient is determined based on the slope of the parameter response curve, for example, 2 for temperature control, 1.5 for pressure, and 1.2 for flow rate; the integral time is linked to the adjustment period, and is 180 seconds, 30 seconds, and 15 seconds, respectively; the derivative time is related to the fluctuation period, and is 45 seconds, 7.5 seconds, and 3.75 seconds, respectively. The optimized control commands are output through the actuator, and the objective function, i.e., the root mean square error between the operating parameters and the target value, is calculated every 60 seconds using the gradient descent method, dynamically optimizing to ensure that the parameters converge quickly to the target range.

[0055] Through the above control strategy, this embodiment of the invention achieves coordinated adjustment of multiple parameters. In actual operation, temperature fluctuations are controlled within ±1 degree Celsius, pressure fluctuations do not exceed ±0.1 MPa, and flow rate fluctuations are limited to ±0.05 m / s. Compared with traditional single-parameter adjustment methods, this significantly improves the stability of equipment operation and scale removal efficiency. It is foreseeable that the sampling period, adjustment step size, and network parameters can be further adjusted according to the equipment model and operating conditions to adapt to more complex operating environments, thereby providing more flexible technical support for the intelligent management of RDX deep removal equipment.

[0056] S105 continuously monitors the changes in physical parameters of the inner wall of the RDX deep removal equipment, analyzes the scaling inhibition effect and evaluates the removal efficiency. If the parameters tend to stabilize and approach the benchmark value of the scale-free state and the scaling characteristic quantity no longer increases, the correlation model is optimized to improve the prediction accuracy and control effect.

[0057] In this embodiment of the invention, the temperature, pressure, and flow rate parameters of the inner wall of the equipment are continuously collected via high-frequency sampling, with sampling periods set to 15 seconds, 5 seconds, and 2 seconds, respectively, to capture dynamic changes in the parameters. The collected data is smoothed using an exponentially weighted moving average method, with a weighting coefficient set to 0.3, which preserves short-term trends while reducing noise interference. The smoothed data is used to calculate the short-term rate of change of the physical parameters, and the fluctuation period is determined through autocorrelation analysis. For example, the temperature fluctuation period is approximately 30 minutes, while the pressure and flow rate fluctuation periods are 10 minutes and 5 minutes, respectively. These periods reflect the dynamic characteristics of equipment operation and scaling treatment, providing a timescale basis for subsequent trend analysis.

[0058] S1051, the physical parameter sequence is decomposed into trends and stability indices are calculated to determine the operating status. In this embodiment of the invention, the continuously collected physical parameter sequence is decomposed into three components: a long-term trend term, a periodic term, and a random term, with a 24-hour cycle. The long-term trend term reflects the direction of parameter change with scaling treatment, the periodic term captures periodic fluctuations, and the random term quantifies the noise level. The stability of the parameters is assessed by calculating the root mean square error of the random term. If the error value is less than 0.5% and the amplitude of the periodic term is less than 1% of the mean, the parameter sequence is considered to have entered a stable state. For example, in actual monitoring, after the temperature dropped from 95 degrees Celsius to 85 degrees Celsius, the error value dropped to 0.3%, indicating that the treatment process has become stable. This decomposition method improves the accuracy of parameter stability judgment through multidimensional analysis, laying the foundation for evaluating the scaling inhibition effect.

[0059] In this embodiment of the invention, the monitoring of scaling characteristics is achieved through an image segmentation algorithm. An adaptive thresholding method is used to identify scaling regions and calculate their thickness and area. During the process, the scaling area gradually decreases from an initial 350 square centimeters to 150 square centimeters, and the thickness decreases from a maximum of 85 micrometers to 35 micrometers. The rate of change of the characteristics is calculated through differential calculation. The results show that the area reduction rate is 8 square centimeters per hour, and the thickness reduction rate is 2 micrometers per hour, both negative and continuously decreasing. The suppression judgment threshold is set according to the degree of parameter recovery: the temperature fluctuation threshold is ±1 degree Celsius, the pressure threshold is ±0.1 MPa, and the flow rate threshold is ±0.05 meters per second. When continuous monitoring for 8 hours shows that the temperature is stable between 84 and 86 degrees Celsius, the pressure is between 3.45 and 3.55 MPa, the flow rate is between 1.95 and 2.05 meters per second, and the scaling characteristics no longer increase, it is determined that scaling has been effectively suppressed.

[0060] S1052, construct removal efficiency evaluation indicators and optimize the correlation model through machine learning.

[0061] After determining scaling inhibition, this embodiment of the invention constructs a RDX removal efficiency evaluation index, including physical parameter recovery time, scaling characteristic quantity reduction magnitude, and parameter stability. Recovery time is defined as the time required for parameters to reach stability, typically 4 to 6 hours; reduction magnitude is the percentage difference between the initial and final values, for example, an area reduction of approximately 57% and a thickness reduction of approximately 59%; stability is characterized by the standard deviation of fluctuation, which must be less than 5% of a threshold. These indices are modeled using a support vector machine with a radial basis function kernel function. The kernel parameter is optimized to 0.8 through cross-validation, generating an evaluation dataset. Subsequently, a recurrent neural network is used to predict removal efficiency. The network design has 128 hidden layer units, inputting the parameter change sequence from the previous 4 hours, with training data from 1000 historical records. During optimization, the momentum method is used to update the weights, with a learning rate of 0.01 and a momentum factor of 0.9. The model is updated every 50 new data sets to ensure a prediction accuracy exceeding 90%, thereby achieving dynamic optimization of the correlation model and improving the adaptability of subsequent control.

[0062] In this embodiment of the invention, continuous monitoring and data analysis not only verified the scaling inhibition effect, but also enhanced the predictive ability of the model through a feedback mechanism. It is foreseeable that the sampling frequency and threshold settings can be adjusted according to the equipment operating environment to adapt to different scaling types and treatment needs, ensuring the high efficiency and stability of the RDX deep removal equipment during long-term operation.

[0063] S106 If, after adjusting the operating parameters of the RDX deep removal equipment, the monitored internal wall physical parameters continue to deviate from the scale-free state benchmark or the scale characteristics show an increasing trend, then the scaling problem is not alleviated through multi-dimensional data analysis, and a cleaning program is initiated to efficiently remove the scale layer.

[0064] In this embodiment of the invention, real-time monitoring of the physical parameters of the inner wall of the equipment is achieved using high-precision sensors. The temperature sensor covers a range of 0 to 200 degrees Celsius with an accuracy of 0.1 degrees Celsius and a sampling period of 15 seconds. The pressure sensor has a range of 0 to 10 MPa with an accuracy of 0.01 MPa and a sampling period of 5 seconds. The flow rate sensor has a measurement range of 0 to 15 meters per second with an accuracy of 0.1 meters per second and a sampling period of 2 seconds. The collected data is averaged over a 60-second sliding window. After filtering out random fluctuations, the data is compared with a scale-free baseline value to generate a difference sequence. The difference sequence is weighted and summed to calculate a comprehensive deviation index, with weights of 0.4 for temperature, 0.3 for pressure, and 0.3 for flow rate. If the index exceeds 0.8, it indicates a severe deviation. For example, when the temperature deviation rate reaches 15%, the pressure deviation rate exceeds 20%, and the flow rate deviation rate decreases by 25%, the comprehensive deviation significantly exceeds the warning range, indicating an aggravated scaling problem.

[0065] S1061 utilizes scanning electron microscopy and image analysis techniques to monitor the dynamic changes in scale characteristics and assess growth trends.

[0066] While monitoring deviations in physical parameters, this embodiment of the invention uses a scanning electron microscope (SEM) to perform high-resolution imaging of the scaled area. The scanning area is set to 500 μm × 500 μm, achieving a resolution of 2048 × 2048 pixels, which clearly captures the microscopic morphology of the scale. After adaptive threshold segmentation, the scale thickness and area are calculated; for example, the thickness increases from 25 μm to 85 μm, and the area expands from 150 square centimeters to 350 square centimeters. The rate of change of characteristic quantities is calculated using the difference method, showing that the thickness growth rate reaches 2 μm per hour, and the area growth rate is approximately 8 square centimeters per hour, exhibiting an accelerating trend. To further analyze the scale morphology, an 8-layer convolutional neural network is used to extract features. The input is a high-resolution image, and the output is a scale state vector. The network is trained based on 1000 sets of samples, and an evaluation index is constructed by combining the rate of change, confirming that the scale problem is worsening and providing data support for subsequent cleaning decisions.

[0067] In this embodiment of the invention, a bivariate correlation matrix is ​​formed by combining the comprehensive deviation index and scaling characteristics, and a Markov chain is used to predict the scaling development trend. Matrix analysis reveals a high correlation between parameter deviation and scaling growth. The Markov chain has 5 state nodes, ranging from no scaling to severe scaling. The transition probability is trained based on 2000 sets of historical data, predicting an 80% probability of scaling deterioration in the next 24 hours. If the mitigation assessment value exceeds the threshold, a cleaning program is initiated, generating instructions for chemical cleaning agent mixing ratio and mechanical cleaning head position. The pH value of the cleaning agent is controlled between 2.5 and 3.0, the pressure of the mechanical cleaning head is 2.5 MPa, and the rotation speed is 300 rpm, ensuring a balance between cleaning power and safety.

[0068] S1062 achieves closed-loop control of the cleaning process and verifies the removal effect through acoustic wave and turbidity monitoring.

[0069] During the cleaning process, this embodiment of the invention uses a 1 MHz acoustic sensor to monitor the scaling layer peeling status. The echo signal intensity weakens as the scaling layer decreases, reflecting the cleaning progress in real time. Simultaneously, an online turbidimeter detects the scaling content in the cleaning solution, which initially rises from 50 mg / L to 350 mg / L and then gradually decreases, indicating that the scaling layer is being gradually removed. The cleaning process lasts for 4 hours until the turbidity value stabilizes, verifying the recovery of the equipment's heat transfer efficiency. This closed-loop control method optimizes cleaning parameters through dynamic feedback, avoiding potential damage to the equipment from over-cleaning while ensuring complete scaling removal.

[0070] In this embodiment of the invention, the aforementioned monitoring and cleaning process effectively addresses scenarios where parameter adjustments fail, promptly restoring equipment performance. It is foreseeable that cleaning parameters and sensor configurations can be flexibly adjusted according to the type of scaling to adapt to different operating conditions and improve the long-term operating efficiency of the RDX deep removal equipment.

[0071] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A control method for a deep removal device of RDX in magnesium nitrate solution, characterized in that, The method includes: Real-time acquisition of operational data from the RDX deep removal equipment, including physical parameters of the equipment's inner wall such as temperature, pressure, and flow rate, is performed. Simultaneously, baseline values ​​for these physical parameters under scaling conditions are established. By comparing these physical parameters with the baseline values, it is determined whether scaling has occurred. If scaling has occurred, the microscopic morphological characteristics of the scaling are obtained using a scanning electron microscope (SEM) to identify and locate the scaling distribution. The compositional information of the scaling sample is also acquired, providing information on the scaling's composition, morphology, and distribution. This includes: performing partitioned scanning imaging of the scaling surface using an SEM, acquiring high-resolution SEM images of the scaling surface based on a preset scanning area size; and performing image segmentation based on the high-resolution SEM images. The algorithm identifies scaling areas and obtains a scaling density map by calculating the area ratio of the scaling areas. For the scaling density map, the region with the highest scaling density is selected, and elemental analysis using an energy dispersive spectroscopy (EDS) instrument is performed to obtain the mass fraction of each element in the scaling sample. Based on the mass fraction of each element in the scaling sample, a 3D reconstruction algorithm is used to create a three-dimensional model of the scaling area, and a scaling feature map is calculated using vertical height information. A multi-scale numerical model is established based on the scaling information to simulate the interaction between RDX and the scaling layer under fluid action, outputting equipment status and scaling layer changes. A correlation model between scaling characteristics and removal efficiency is established to predict the removal effect under different scaling conditions, including using the finite element method. The scale layer is meshed to obtain discretized geometric structure data. A set of fluid dynamics equations is established based on this data, and the flow field distribution data at the scale layer interface is calculated using these equations. A mass transfer equation is established based on the flow field distribution data, and the mass transfer flux data at the scale layer interface is calculated using these equations. The local dissolution rate of the scale layer is calculated based on the interface mass transfer flux data, and the predicted removal efficiency data is obtained from this local dissolution rate. The real-time monitored equipment physical parameters are compared with the predicted removal effects under different scaling conditions. If the equipment operating parameters deviate from the target range, a preset control system is activated. The system adjusts the temperature, pressure, and flow rate operating parameters of the equipment; it continuously monitors changes in the physical parameters of the equipment's inner wall to determine whether scaling has been suppressed. If the monitored physical parameters tend to stabilize or return to the baseline values ​​of the preset scale-free state, and the scale thickness or scale area stops increasing or begins to decrease, it is determined that scaling has been suppressed. The removal efficiency of RDX is then evaluated, and the evaluation results are fed back to the correlation model for optimization and updates. If, after adjusting the operating parameters, the monitored physical parameters of the equipment's inner wall continue to deviate from the baseline values ​​of the scale-free state, or the scale thickness or scale area continues to increase, or even accelerates, it is determined that the scaling problem has not been alleviated after adjusting the operating parameters. In this case, the preset equipment cleaning program is initiated to clean the inner wall of the equipment and remove the scale layer.

2. The method according to claim 1, characterized in that, The real-time acquisition of operating data from the RDX deep removal equipment includes physical parameters of the equipment's inner wall, such as temperature, pressure, and flow rate. Simultaneously, a baseline value for these physical parameters under scaling conditions is established. By comparing the physical parameters with the baseline value, it is determined whether scaling has occurred on the equipment. This includes: The physical parameters are obtained from temperature sensors, pressure sensors, and flow rate sensors. The data acquisition module periodically samples and records the physical parameters at preset sampling time intervals to obtain a sampling data sequence. The sampled data sequence is smoothed using an exponentially weighted moving average method, and the statistical mean is calculated based on the smoothed physical parameters to obtain the baseline value of the physical parameters. The difference between the smoothed physical parameters and the reference physical parameter values ​​is calculated, and the standard deviation of the difference sequence is calculated using the least squares method to obtain the internal wall state parameters of the equipment. The scale formation state is compared with the preset scale formation state judgment value and the internal wall state parameter of the equipment. If the internal wall state parameter of the equipment is greater than the scale formation state judgment value, it is determined that scale has occurred in the RDX deep removal equipment.

3. The method according to claim 1, characterized in that, The process involves comparing the real-time monitored physical parameters of the equipment with the predicted removal effects under different scaling conditions. If the equipment operating parameters are found to deviate from the target range, the temperature, pressure, and flow rate operating parameters of the equipment are adjusted through a preset control system, including: The real-time operating status data of the equipment is obtained by collecting operating parameter data from temperature sensors, pressure sensors and flow rate sensors and processing them through moving average. The Manhattan distance is used to calculate the matching degree between the real-time operating status data of the equipment and the standard operating condition database, and to obtain temperature deviation data, pressure deviation data and flow rate deviation data. A parameter adjustment sequence is established based on the temperature deviation data, the pressure deviation data, and the flow rate deviation data, and the direction of operating parameter adjustment is predicted by a long short-term memory network. The controller target value is set according to the adjustment direction of the operating parameters, and the control command sequence is optimized by a proportional-integral-differential arithmetic unit to obtain the corrected control command.

4. The method according to claim 1, characterized in that, The continuous monitoring device monitors changes in its internal physical parameters to determine whether scaling has been suppressed. If the monitored physical parameters stabilize or return to the baseline value of a preset scale-free state, and the scaling thickness or area stops increasing or begins to decrease, then scaling is considered suppressed. The removal efficiency of RDX is then evaluated, and the evaluation results are fed back to the correlation model for optimization and updates, including: The temperature, pressure, and flow rate parameters of the inner wall of the equipment are collected, and the rate of change of the physical parameters is obtained by using the exponential weighted moving average method based on the physical parameters. The physical parameter sequence is decomposed according to the rate of change of the physical parameters, and the root mean square error value is calculated from the trend decomposition result. The physical parameter sequence is processed using an image segmentation algorithm. Based on the image segmentation results, the scale thickness value and scale area value are obtained. The rate of change of scale characteristic quantities is obtained by performing a difference operation on the scale thickness value and scale area value. Based on the root mean square error value, a suppression determination threshold is set. If the rate of change of the scaling characteristic quantity is less than the suppression determination threshold, it is determined that scaling has been suppressed.

5. The method according to claim 1, characterized in that, If, after adjusting the operating parameters, the monitored physical parameters of the equipment's inner wall continue to deviate from the baseline value under scale-free conditions, or if the scale thickness or area continues to increase, or even accelerates, then the scaling problem has not been alleviated after adjusting the operating parameters. In this case, a preset equipment cleaning program will be initiated to clean the equipment's inner wall and remove the scale layer, including: The physical parameters collected by the temperature sensor, pressure sensor, and flow rate sensor are acquired, and a difference sequence is calculated based on the physical parameters and a preset reference value. The images of the scaled regions corresponding to the difference sequence are obtained by scanning electron microscopy, and the scaled region images are processed by an image segmentation algorithm to obtain scaled feature quantities. A bivariate correlation matrix is ​​established based on the difference sequence and the scaling characteristics, and the scaling problem mitigation assessment value is calculated through the bivariate correlation matrix. For areas where the scaling problem mitigation assessment value exceeds a preset threshold, an acoustic sensor is used to detect the scaling layer peeling status, and an online turbidity meter is used to measure the scaling content in the cleaning fluid to obtain cleaning process control commands.

Images