Full process monitoring and fault diagnosis method and system for pneumatic conveying system

Abstract

The present application provides a full-process monitoring and fault diagnosis method and system for a pneumatic conveying system. The method includes: collecting multi-source monitoring data of a pneumatic conveying system; processing the multi-source monitoring data to obtain standardized monitoring data; constructing a training set based on historical operation data, training a pre-constructed deep learning model, and obtaining a system state evaluation model; based on the system state evaluation model and the multi-source monitoring data, using a system state vector algorithm to generate a differential flatness index; based on the differential flatness index, designing a k-induction control algorithm to generate control parameters for the induced draft fan speed and the opening of the material distribution valve of the pneumatic conveying system; constructing a disturbance observer with an adaptive safety boundary, and controlling the pneumatic conveying system based on the control parameters. The method realizes intelligent monitoring and safety control of the pneumatic conveying system by combining deep learning and intelligent control, and improves the system operation efficiency and reliability.

Description

Technical Field

[0001] The present application relates to the field of industrial automation and artificial intelligence technology, and in particular to a full-process monitoring and fault diagnosis method and system for a pneumatic conveying system. Background Art

[0002] As an important means of material transportation in industrial production, the safe and stable operation of pneumatic conveying system is crucial to the production process. The existing monitoring and diagnosis of pneumatic conveying system mainly adopts the following methods:

[0003] The threshold monitoring method based on a single sensor sets an alarm by setting thresholds for key parameters (such as pressure and flow rate), but this method is too simple and cannot cope with complex fault scenarios.

[0004] Traditional fault diagnosis methods, such as rule-based expert systems, can handle some equipment failures, but they are difficult to adapt to the dynamic changes in equipment status and operating conditions, and their diagnostic accuracy is not high.

[0005] Although the simple PID control method can realize basic control functions, it is difficult to achieve precise control and energy consumption optimization due to the lack of in-depth understanding of the dynamic characteristics of the system.

[0006] The above-mentioned existing technologies are difficult to achieve full-process intelligent monitoring of the pneumatic conveying system, resulting in low system operation efficiency. At the same time, there is a lack of accurate modeling and control of the dynamic characteristics of the system, which affects the conveying stability and energy utilization efficiency. Summary of the invention

[0007] In view of this, the present application provides a full-process monitoring and fault diagnosis method and system for a pneumatic conveying system, which solves the problem in the prior art that it is difficult to achieve full-process intelligent monitoring and precise control of the system.

[0008] A full-process monitoring and fault diagnosis method for a pneumatic conveying system, the method comprising:

[0009] Collect multi-source monitoring data of pneumatic conveying systems;

[0010] Processing the multi-source monitoring data to obtain standardized monitoring data;

[0011] Building a training set based on the historical operation data of the pneumatic conveying system, training a pre-built deep learning model, and obtaining a system state assessment model, wherein the deep learning model includes an input layer corresponding to the transportation parameters of the pneumatic conveying system, a plurality of hidden layers for extracting features, and an output layer for outputting an assessment result;

[0012] Based on the system state assessment model and the multi-source monitoring data, a system state vector algorithm is used to generate a differential flatness index, where the differential flatness index is used to characterize system state change characteristics;

[0013] According to the differential flatness index, a k-summary control algorithm is designed to generate control parameters for the induced draft fan speed and the opening of the material distribution valve of the pneumatic conveying system;

[0014] A disturbance observer with an adaptive safety margin is constructed to control the pneumatic conveying system based on the control parameters.

[0015] Optionally, the collecting of multi-source monitoring data of the pneumatic conveying system includes:

[0016] A distributed sensor network is used to collect multi-source monitoring data of the pneumatic conveying system, wherein the multi-source monitoring data includes: pressure data, flow rate data, temperature data and equipment vibration data;

[0017] Based on threshold parameters adapted to the types of the multi-source monitoring data, filtering and time synchronization operations are performed on the multi-source monitoring data.

[0018] Optionally, the multi-source monitoring data is processed to obtain standardized monitoring data, including:

[0019] The multi-source monitoring data is subjected to noise reduction processing by a sliding window median filtering technique to obtain noise-reduced multi-source monitoring data;

[0020] Using a statistical analysis method to identify abnormal data in the noise-reduced multi-source monitoring data, and performing an operation of eliminating the abnormal data to obtain target multi-source monitoring data;

[0021] Interpolating the missing values ​​of the target multi-source monitoring data to obtain a complete data sequence;

[0022] Normalization processing is performed based on the multi-source monitoring data in the complete data sequence to obtain the standardized monitoring data.

[0023] Optionally, the step of constructing a training set based on the historical operation data of the pneumatic conveying system, training a pre-constructed deep learning model, and obtaining a system state assessment model comprises:

[0024] Based on the transport parameters of the pneumatic conveying system, a multi-layer neural network structure is designed, and a nonlinear activation function is used in the hidden layer;

[0025] Adding a normalization layer and a random dropout layer to the multi-layer neural network structure;

[0026] Building a training set based on historical operation data of the pneumatic conveying system, wherein the historical operation data includes normal operation data and fault state data;

[0027] The deep learning model is optimized by using a gradient descent method and an early stopping strategy to obtain the system state evaluation model.

[0028] Optionally, the input layer includes 14 feature nodes, wherein the 14 feature nodes correspond to the feed inlet pressure of the pneumatic conveying system, the conveying speed of 4 measuring points, 3 material level data, 2 induced draft fan vibration values, 2 distribution valve vibration values, and 2 dust concentration values;

[0029] The hidden layer includes a first hidden layer, a second hidden layer, a third hidden layer, and a fourth hidden layer, the number of nodes in the first hidden layer is 128, the number of nodes in the second hidden layer is 256, the number of nodes in the third hidden layer is 128, and the number of nodes in the fourth hidden layer is 64, and the first hidden layer, the second hidden layer, the third hidden layer, and the fourth hidden layer all use a ReLU activation function, wherein a BatchNormalization layer is added between the second hidden layer and the third hidden layer, and a Dropout layer with a dropout rate of 0.3 is added after the third hidden layer and the fourth hidden layer respectively;

[0030] The output layer includes 8 feature nodes, which respectively correspond to: system blockage risk assessment value, pipeline wear warning value, induced draft fan failure warning value, distribution valve failure warning value, system leakage risk value, and predicted pressure values ​​of three key parts.

[0031] Optionally, the generating a differential flatness index based on the system state assessment model and the multi-source monitoring data using a system state vector algorithm includes:

[0032] Outputting the prediction results of the multi-source monitoring data through the system state assessment model, wherein the prediction results include a multi-dimensional state vector of the pneumatic conveying system;

[0033] Based on the system state vector algorithm, calculating the first-order difference value and the second-order difference value of the multidimensional state vector;

[0034] constructing a flatness matrix based on the first-order difference value and the second-order difference value;

[0035] Performing singular value decomposition on the flatness matrix to obtain the differential flatness index.

[0036] Optionally, the k-summary control algorithm is designed according to the differential flatness index to generate control parameters for the induced draft fan speed and the opening of the material distribution valve of the pneumatic conveying system, including:

[0037] Based on the differential flatness index, a system control objective function in the form of a Lyapunov function is constructed, and the system control objective function includes:

[0038] V(t) = 1 / 2[(p(t)-p )² / p ² + (v(t)-v )² / v ²],

[0039] Among them, p is the preset pressure, v The preset conveying speed;

[0040] According to the control objective function, a control law including a proportional term and an integral term is designed. The control law takes the following form:

[0041] (t) = - (p(t)-p )- (p(t)-p )dt,

[0042] (t) = - (v(t)-v )- (v(t)-v )dt,

[0043] in, is the first control gain parameter, is the second control gain parameter, is the third control gain parameter, is the fourth control gain parameter;

[0044] The control parameter ranges of the rotation speed of the induced draft fan and the adjustment speed of the opening of the distribution valve are determined based on the Lyapunov stability theory.

[0045] Optionally, one or more of the following are met:

[0046] Based on the implementation of hierarchical control according to the differential flatness index, the method includes: when the differential flatness index is lower than a first threshold, maintaining the current control parameter; when the differential flatness index is between the first threshold and the second threshold, smoothly adjusting the control parameter; when the differential flatness index is higher than the second threshold, triggering an emergency control strategy to reduce the speed of the induced draft fan and the opening of the material distribution valve;

[0047] The design of the control law includes: constructing a dual-loop control structure including a pressure control loop and a speed control loop; in the pressure control loop, designing a pressure control law based on the pressure deviation; in the speed control loop, designing a speed control law based on the speed deviation; and designing a control gain adaptive adjustment mechanism according to the differential flatness index.

[0048] Optionally, the construction of a disturbance observer with an adaptive safety margin to control the pneumatic conveying system based on the control parameters comprises:

[0049] Establishing a nominal model including pipeline transport dynamics equations, wherein the nominal model is used to represent the relationship between pressure, flow rate and material concentration;

[0050] A disturbance estimator is constructed based on the nominal model. The disturbance estimator includes a nominal state observer and a disturbance reconstructor. The state equation of the state observer is:

[0051] (t+1) = A (t)+Bu(t)+L(y(t)- (t))+ (t)

[0052] Among them, among them, (t+1) is the estimated value of the state at time t+1; (t): the state estimate at the current moment; A is the system state transfer matrix; B is the control input matrix, u(t) is the control input vector, L is the observer gain matrix, which is used to correct the state estimate; y(t) is the actual measured output; (t) is the estimated output value; (t) is the estimated disturbance term used to compensate for the unmodeled dynamics and external disturbances in the system;

[0053] Using recursive least squares method to identify key parameters of the pneumatic conveying system online;

[0054] Based on the output of the disturbance estimator, dynamically adjust the pressure, speed and material level safety margins of the system;

[0055] Designing a hierarchical control quantity limiting strategy according to the safety boundary;

[0056] Setting a control amount buffer zone in the control amount limiting strategy;

[0057] Predicting the system state evolution trend based on the current system state, and adjusting the control strategy according to the system state evolution trend;

[0058] Adjust control parameters based on real-time system response.

[0059] The present application embodiment provides a full-process monitoring and fault diagnosis system for a pneumatic conveying system, comprising:

[0060] A data acquisition unit, used to collect multi-source monitoring data of the pneumatic conveying system;

[0061] A data processing unit, used to process the multi-source monitoring data to obtain standardized monitoring data;

[0062] A model training unit, used to construct and train a deep learning model based on the standardized monitoring data to obtain a system status assessment model;

[0063] An indicator generating unit, configured to construct a training set based on the historical operation data of the pneumatic conveying system, train a pre-constructed deep learning model, and obtain a system state evaluation model, wherein the deep learning model comprises an input layer corresponding to the transportation parameters of the pneumatic conveying system, a plurality of hidden layers for extracting features, and an output layer for outputting an evaluation result;

[0064] A control parameter generating unit, for generating a differential flatness index based on the system state evaluation model and the multi-source monitoring data by using a system state vector algorithm, wherein the differential flatness index is used to characterize the system state change characteristics, and for designing a k-summary control algorithm according to the differential flatness index to generate control parameters for the induced draft fan speed and the opening of the material distribution valve of the pneumatic conveying system;

[0065] A safety control unit is used to construct a disturbance observer with an adaptive safety margin and control the pneumatic conveying system based on the control parameters.

[0066] This application has the following technical effects:

[0067] By combining deep learning with intelligent control, intelligent monitoring and fault diagnosis of the pneumatic conveying system are realized, thus improving the reliability of system operation.

[0068] The method combining differential flatness and k-induction control algorithm is adopted to improve the system control accuracy and effectively ensure the stability of the conveying process;

[0069] A disturbance observer with adaptive safety margin is designed to enhance the system's anti-disturbance capability and improve operational safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following is a brief introduction to the drawings required for use in the embodiments. The drawings herein are incorporated into the specification and constitute a part of the specification. These drawings illustrate embodiments consistent with the present disclosure and are used together with the specification to illustrate the technical solutions of the present disclosure. It should be understood that the following drawings only illustrate certain embodiments of the present disclosure and should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can also be obtained based on these drawings without creative work.

[0071] Figure 1 It is a flow chart of a method for full-process monitoring and fault diagnosis of a pneumatic conveying system provided in an embodiment of the present application;

[0072] Figure 2 It is a structural schematic diagram of a full-process monitoring and fault diagnosis system for a pneumatic conveying system provided in an embodiment of the present application;

[0073] Figure 3 is a schematic diagram of the network structure of the deep learning model provided in the embodiment of the present application;

[0074] Figure 4 It is a schematic diagram of the differential flatness calculation process provided in an embodiment of the present application. DETAILED DESCRIPTION

[0075] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0076] The embodiments of the present application are further described in detail below with reference to the accompanying drawings.

[0077] First, see Figure 1 The full-process monitoring and fault diagnosis method for a pneumatic conveying system provided in this application comprises the following steps:

[0078] S1: Collect multi-source monitoring data of pneumatic conveying system.

[0079] S1.1: A distributed sensor network is used to collect multi-source monitoring data of the pneumatic conveying system, wherein the multi-source monitoring data includes: pressure data, flow rate data, temperature data and equipment vibration data.

[0080] In S1.1, a distributed sensor network is used to collect the operating data of the pneumatic conveying system.

[0081] In one embodiment, in the lithium iron phosphate powder conveying system, the specific deployment is as follows:

[0082] An explosion-proof pressure sensor is installed at the inlet of the pneumatic conveying pipeline with a measuring range of 0-0.6MPa and an accuracy of 0.075% to monitor the system feed inlet pressure.

[0083] Flow velocity sensors with a measuring range of 0-30m / s are arranged at key positions of the horizontal and vertical sections of the conveying pipeline to monitor the powder conveying speed.

[0084] Material level sensors are installed at the end of the conveying pipeline and in the intermediate buffer tank, with a measuring range of 0-5m, to monitor the material accumulation height.

[0085] Vibration sensors with a measuring range of 0.5-10kHz are installed on key equipment such as induced draft fans and dosing valves to monitor the vibration status of the equipment.

[0086] In addition, explosion-proof dust concentration detectors are installed at key points of the system to monitor dust leakage. An explosion-proof programmable controller is used as the field control unit to collect data through the EtherNet / IP network with a sampling period of 50ms, and transmit the data to the upper monitoring system through the OPC-UA protocol. Finally, a 15-minute real-time data cache is established using the TimescaleDB time series database, and a double buffer mechanism is used for storage to ensure data integrity in the event of network fluctuations.

[0087] S1.2: Based on threshold parameters adapted to the types of the multi-source monitoring data, filtering and time synchronization operations are performed on the multi-source monitoring data.

[0088] In S1.2, the collected data are initially filtered and time synchronized.

[0089] In one embodiment, in a lithium iron phosphate delivery system:

[0090] First, the validity of the pressure sensor data is checked and the pressure values ​​exceeding 0.5MPa are marked as abnormal (to avoid powder agglomeration caused by exceeding the safety pressure).

[0091] Next, the flow rate sensor data is checked, and data below 8 m / s (below the minimum fluidization velocity of the powder) or above 25 m / s (exceeding the speed at which the powder is easily worn) are marked as suspicious values.

[0092] Secondly, the material level data is screened and data exceeding 4.5m (close to the safe material level of the storage tank) will trigger an alarm.

[0093] Furthermore, the dust concentration data is monitored and an alarm is immediately triggered when the detection value exceeds 30mg / m³.

[0094] In some embodiments, due to the differences in sampling periods of different sensors (for example, the sampling period of pressure sensors and flow rate sensors is 50ms, the sampling period of material level sensors is 200ms, and the sampling period of vibration is 1s), a timestamp alignment algorithm is used to perform cubic spline interpolation on the material level and vibration data based on the minimum sampling period of 50ms to achieve data synchronization. Finally, the synchronized data is stored in TimescaleDB in time series, and each record contains the measurement value, timestamp, sensor number, device location information, and alarm level identification.

[0095] It should be noted that, first, step S1.1 completes the deployment of sensors and the establishment of a data acquisition network suitable for the lithium battery positive electrode material delivery system, selects explosion-proof sensors and clarifies the measurement range and installation location of various sensors, providing high-precision multi-source monitoring data for step S1.2. Second, based on these multi-source monitoring data, step S1.2 performs preliminary filtering by setting the thresholds of key parameters in the lithium iron phosphate powder delivery process, focusing on the characteristics of the powder (such as fluidization velocity, agglomeration risk, etc.), and solving the problem of inconsistent sampling cycles of different sensors, and finally outputs a time-aligned valid data stream, providing a basis for subsequent data analysis and safety monitoring.

[0096] S2: Processing the multi-source monitoring data to obtain standardized monitoring data.

[0097] S2.1: performing noise reduction processing on the multi-source monitoring data by using a sliding window median filtering technique to obtain noise-reduced multi-source monitoring data;

[0098] In S2.1, data denoising is performed using a sliding window median filtering technique.

[0099] The specific implementation is, first, using different window sizes for different types of sensor data to balance the filtering effect and real-time performance.

[0100] In one embodiment, for pressure data, a window size of 15 sampling points (wherein the time span is 750 ms) is used to effectively suppress transient fluctuations caused by material transportation.

[0101] In one embodiment, for flow rate data, a window size of 11 sampling points (with a time span of 550 ms) is used to retain the rapid change characteristics of the delivery speed.

[0102] In one embodiment, for the material level data, a window size of 7 sampling points (wherein the time span is 350 ms) is adopted to timely reflect the material level change trend.

[0103] In one embodiment, for vibration data, a window size of 21 sampling points (wherein the time span is 1050 ms) is used to effectively remove high-frequency noise while retaining characteristic vibration of the equipment.

[0104] Secondly, when processing the data in each window, the data in the window is first sorted by numerical value, and then the median is taken as the output value of the window to effectively suppress abnormal peaks.

[0105] S2.2: using a statistical analysis method to identify abnormal data in the noise reduction multi-source monitoring data, and performing an operation of removing the abnormal data to obtain target multi-source monitoring data;

[0106] In S2.2, statistical analysis methods are used to identify abnormal data.

[0107] First, the statistical characteristics of each data in a 20-minute time window are calculated, including the mean μ and standard deviation σ.

[0108] In one embodiment, for pressure data, the 3σ criterion is used to mark data that deviates from the mean by more than ±3σ as abnormal.

[0109] In one embodiment, for flow rate data, combined with the physical limitations of the equipment, data that deviates from the mean by more than ±2.5σ and is lower than 8 m / s or higher than 25 m / s is determined to be abnormal.

[0110] In one embodiment, for the material level data, data points with a rate of change exceeding 0.5 m / s are marked as abnormal.

[0111] In one embodiment, for vibration data, an interquartile range (IQR) method is used to determine data exceeding Q3+1.5IQR or below Q1-1.5IQR as abnormal, where Q1 is the first quartile and Q3 is the third quartile.

[0112] Secondly, for data points marked as abnormal, the system will record the time when the abnormality occurred, the sensor location and the degree of abnormality for subsequent analysis.

[0113] S2.3: interpolating the missing values ​​of the target multi-source monitoring data to obtain a complete data sequence;

[0114] In S2.3, interpolation is performed for missing data. Different interpolation strategies are used depending on the length of time the data is missing:

[0115] In one embodiment, for a single point missing for a short time (no more than 5 sampling periods), a linear interpolation method is used.

[0116] In one embodiment, for medium-length (5-20 sampling periods) missing data, cubic spline interpolation is used to maintain data continuity and smoothness.

[0117] In one embodiment, for missing data of longer duration (greater than 20 sampling periods), a local polynomial regression method is used, which takes into account the historical trend and periodicity characteristics of the data.

[0118] In particular, for the missing vibration data, an interpolation method based on spectrum analysis is used to maintain the frequency domain characteristics of the signal.

[0119] S2.4: Perform normalization processing based on the multi-source monitoring data in the complete data sequence to obtain the standardized monitoring data.

[0120] In S2.4, the complete data sequence is normalized.

[0121] First, for each type of sensor data, a suitable normalization method is selected according to its physical meaning.

[0122] In one embodiment, for the pressure data, a maximum and minimum value normalization method is adopted, and the mapping range is [0, 1].

[0123] In one embodiment, the flow velocity data is normalized using a design condition-based normalization method and scaled based on a rated flow velocity of 15 m / s.

[0124] In one embodiment, a piecewise linear normalization method is used for the material level data, and different mapping relationships are used for different material level intervals.

[0125] In one embodiment, for the vibration data, amplitude normalization is first performed, and then a Z-score normalization method is used.

[0126] In addition, in order to adapt to changes in system operating conditions, the normalization parameters are updated every 12 hours to ensure the validity of the normalization results. For new incoming data points, the system will perform normalization processing in real time and output standardized monitoring data streams.

[0127] It should be noted that in the entire data processing process, the system adopts a streaming processing architecture, and the output of each processing step will be passed to the next step in a timely manner, ensuring the real-time nature of data processing. At the same time, both the original data and the processed data will be saved for later analysis and model optimization. The entire data processing process fully considers the characteristics of the pneumatic conveying system and the needs of real-time monitoring, providing high-quality input data for subsequent deep learning models.

[0128] S3: construct a training set based on the historical operation data of the pneumatic conveying system, train the pre-constructed deep learning model, and obtain a system status assessment model, wherein the deep learning model is as follows: Figure 3 As shown, it includes an input layer corresponding to the transportation parameters of the pneumatic conveying system, a plurality of hidden layers for extracting features, and an output layer for outputting evaluation results.

[0129] S3.1: Based on the transport parameters of the pneumatic conveying system, a multi-layer neural network structure is designed, and a nonlinear activation function is used in the hidden layer;

[0130] The input layer includes 14 characteristic nodes, wherein the 14 characteristic nodes correspond to the inlet pressure of the pneumatic conveying system, the conveying speed of 4 measuring points, 3 material level data, 2 induced draft fan vibration values, 2 distribution valve vibration values, and 2 dust concentration values;

[0131] The hidden layer includes a first hidden layer, a second hidden layer, a third hidden layer, and a fourth hidden layer, the number of nodes in the first hidden layer is 128, the number of nodes in the second hidden layer is 256, the number of nodes in the third hidden layer is 128, and the number of nodes in the fourth hidden layer is 64, and the first hidden layer, the second hidden layer, the third hidden layer, and the fourth hidden layer all use a ReLU activation function, wherein a BatchNormalization layer is added between the second hidden layer and the third hidden layer, and a Dropout layer with a dropout rate of 0.3 is added after the third hidden layer and the fourth hidden layer respectively;

[0132] The output layer contains 8 feature nodes, which correspond to: system blockage risk assessment value (0-1), pipeline wear warning value (0-1), induced draft fan failure warning value (0-1), distribution valve failure warning value (0-1), system leakage risk value (0-1), and predicted pressure values ​​of 3 key parts.

[0133] The following is a detailed description of the 8 feature nodes in the output layer of the deep learning model and their specific meanings.

[0134] Among the five risk assessment values ​​in the output layer, the system blocking risk assessment value reflects the possibility of blockage in the pipeline, and the value range is 0-1. When the value is greater than 0.8, it means that the system is very likely to be blocked; when it is between 0.5-0.8, it means that there is a risk of blocking and needs to be paid special attention; when it is lower than 0.5, it means that the system is operating normally.

[0135] Among them, the following factors are mainly considered during the assessment: abnormal pipeline pressure gradient, sudden drop in conveying speed, sudden change in material level, etc. For example, when a sudden increase in horizontal section pressure and a drop in conveying speed of more than 20% are detected, the blockage risk assessment value will quickly rise to above 0.7.

[0136] The pipeline wear warning value represents the degree of wear of the transmission pipeline, and the value range is 0-1. When the value is greater than 0.9, it means that the pipeline is severely worn and needs to be replaced in time; when it is between 0.6-0.9, it means that the pipeline is obviously worn and needs to be monitored more closely; when it is lower than 0.6, it means that the pipeline is in good condition.

[0137] The system mainly evaluates the pipe wear condition by analyzing the pressure loss characteristics, flow velocity distribution and vibration characteristics. For example, when the pressure loss of the system gradually decreases at the same flow velocity and the local vibration intensifies, the wear warning value will gradually increase.

[0138] The fault warning value of the induced draft fan and the fault warning value of the dosing valve reflect the health status of these two key devices, and the value range is 0-1. For the induced draft fan, the system mainly monitors its vibration characteristics, current fluctuations and efficiency changes; when the vibration amplitude exceeds 1.5 times the normal value or the efficiency drops by more than 10%, the fault warning value will exceed 0.6.

[0139] Among them, for the dosing valve, the system focuses on its action response time, opening control accuracy and sealing performance; when the response time is extended to more than 200ms or a small leak occurs, the fault warning value will rise to above 0.7.

[0140] The system leakage risk value is used to evaluate the sealing performance of the system, and the value range is 0-1. This value is mainly based on the data of the dust concentration detector and the system pressure balance analysis. When the dust concentration detected at a certain place exceeds 25mg / m³ and is on an upward trend, the leakage risk value will rise rapidly; when the value exceeds 0.8, the system will trigger the emergency treatment plan.

[0141] Regarding the predicted pressure values ​​of 3 key parts, the key parts here specifically refer to:

[0142] The first key part is the feed port, which is located at the starting end of the conveying system. Its pressure value directly affects the material conveying stability. The system will predict the pressure change in the next 5 seconds, and the accuracy of the predicted value is required to be ±0.02MPa.

[0143] The second key part is the transition between the horizontal section and the vertical section, where material accumulation and blockage are prone to occur. The prediction of the pressure here is crucial for timely detection and prevention of blockage accidents. The prediction time window is 3 seconds and the accuracy requirement is ±0.015MPa.

[0144] The third key part is the inlet of the storage silo at the end of the system. The pressure there directly reflects the operating status of the entire conveying system. The system will predict the pressure change trend in the next 10 seconds in real time, with a prediction accuracy requirement of ±0.01MPa, to guide the adjustment of the overall control strategy of the system.

[0145] The selection of these key locations is based on the following considerations: first, these locations are the parts of the system most prone to failure; second, the pressure changes at these locations can better reflect the overall operating status of the system; finally, the pressure data at these locations has important guiding significance for system control. By accurately predicting the pressure of these three key locations, the system can promptly detect potential problems and take preventive measures, significantly improving the safety and stability of operation.

[0146] In S3.1, a multi-layer neural network structure is designed based on the transport parameters of the pneumatic conveying system.

[0147] For example, in the lithium iron phosphate powder conveying system, a multi-layer deep learning model is constructed as follows:

[0148] First, the design input layer contains 14 feature nodes, corresponding to the key parameters of the system: feed inlet pressure, conveying speed at 4 measuring points (2 at the horizontal section and 2 at the vertical section), 3 material level data (2 at the intermediate buffer tank and 1 at the terminal silo), 2 induced draft fan vibration values, 2 distribution valve vibration values, and 2 dust concentration values.

[0149] Secondly, 4 hidden layers are designed, with the number of nodes being 128, 256, 128, and 64 respectively. ReLU activation function is used in each layer to improve the model's ability to express the nonlinear characteristics of the conveying system. BatchNormalization layer is added between the second and third hidden layers to accelerate training convergence and improve model stability.

[0150] Then, a Dropout layer with a dropout rate of 0.3 is added after the third and fourth hidden layers to prevent the model from overfitting.

[0151] Finally, the output layer is designed to be 8 nodes, corresponding to: system blockage risk assessment value (0-1), pipeline wear warning value (0-1), induced draft fan failure warning value (0-1), distribution valve failure warning value (0-1), system leakage risk value (0-1), and predicted pressure values ​​of 3 key parts.

[0152] S3.2: Add a normalization layer and a random dropout layer to the multi-layer neural network structure;

[0153] In S3.2, optimize the training stability of the neural network.

[0154] First, add a BatchNormalization layer between the second hidden layer and the third hidden layer, use a momentum parameter of 0.99 for exponential moving average calculation, and set epsilon to 1e-5 to avoid numerical instability. At the same time, add gamma and beta learnable parameters after each BatchNormalization layer to restore the expressiveness of the data.

[0155] Secondly, a Dropout layer is added after the third and fourth hidden layers, and the dropout rate is set to 0.3. 30% of the neurons are randomly inactivated during the training phase to effectively prevent the model from overfitting. In addition, the He initialization method is used to initialize the network weights so that the variance of each layer output remains consistent, which is conducive to the convergence of the network.

[0156] S3.3: constructing a training set based on historical operation data of the pneumatic conveying system, wherein the historical operation data includes normal operation data and fault state data;

[0157] In S3.3, a training dataset is constructed based on 2 years of operation data of the lithium iron phosphate delivery system (approximately 5 million records).

[0158] First, classify according to data type: normal operation data (about 80%), minor blockage data (about 8%), pipeline wear data (about 5%), equipment failure data (about 5%), and material leakage data (about 2%).

[0159] Secondly, a stratified sampling strategy was used to construct the training set (70%), validation set (15%), and test set (15%) to ensure that the proportions of different types of data in each dataset were consistent.

[0160] Then, each type of fault data is enhanced by adding Gaussian noise, time shifting and other methods to expand the number of samples and improve the model's ability to recognize abnormal conditions.

[0161] Finally, a data pre-reading queue is established and multi-threading technology is used to accelerate the data loading process.

[0162] S3.4: Use the gradient descent method and early stopping strategy to optimize the parameters of the deep learning model to obtain the system state evaluation model.

[0163] In S3.4, the gradient descent method and early stopping strategy are used for model training.

[0164] In the lithium iron phosphate delivery system, the small batch stochastic gradient descent method is first used for model training. The batch size is set to 256, the initial learning rate is 0.001, and the Adam optimizer is used with β1=0.9 and β2=0.999. The learning rate adopts the cosine annealing strategy, and the learning rate is decayed every 50 epochs.

[0165] Secondly, an early stopping strategy was used to stop training when the validation set loss did not improve for 10 consecutive epochs. In order to evaluate the model performance, a 5-fold cross validation was used, and the following indicators were achieved on the test set: the accuracy of material blockage warning was 95.2%, the accuracy of pipeline wear warning was 93.8%, the accuracy of equipment failure warning was 94.5%, the accuracy of leakage warning was 96.3%, and the average relative error of pressure prediction was less than 2.5%. In addition, the model integration technology was used to train 5 models with different random initializations, and the stability of prediction was improved through the voting mechanism.

[0166] It should be noted that the design of this deep learning model fully considers the characteristics of the pneumatic conveying system, extracts features through multi-layer nonlinear transformation, and improves the generalization ability of the model through technologies such as BatchNormalization and Dropout. At the same time, strategies such as stratified sampling and data enhancement are used to solve the data imbalance problem, and the reliability of prediction is improved through cross-validation and model integration. A variety of optimization strategies are used in the model training process to ensure that the model can accurately identify various abnormal conditions and provide strong support for the safe and stable operation of the system.

[0167] S4: Based on the system state assessment model and the multi-source monitoring data, a system state vector algorithm is used to generate a differential flatness index, where the differential flatness index is used to characterize system state change characteristics.

[0168] like Figure 4 , S4 includes:

[0169] S4.1: Outputting prediction results of the multi-source monitoring data through the system state assessment model, wherein the prediction results include a multi-dimensional state vector of the pneumatic conveying system.

[0170] In S4.1, the prediction results of multi-source monitoring data are output through the system state assessment model.

[0171] First, construct the system state vector x(t), which contains the following key parameters: (t) is the feed port pressure value, unit: MPa; (t) (t) is the pressure value at two points of the horizontal pipe section, unit: MPa; (t) is the vertical pipe section pressure value, unit: MPa; (t) (t) is the conveying speed at two points in the horizontal section, in m / s; (t) is the vertical conveying speed, in m / s; (t) (t) is the height of the intermediate buffer tank, in meters. The state vector is expressed as: x(t) = [ (t), (t), (t), (t), (t), (t), (t), (t), (t) The deep learning model predicts the input state vector every 50ms, and the output includes various risk assessment values ​​of the system and pressure prediction values ​​of key parts.

[0172] S4.2: Based on the system state vector algorithm, calculate the first-order difference value and the second-order difference value of the multidimensional state vector.

[0173] In S4.2, the first-order difference and the second-order difference values ​​of the state vector are calculated.

[0174] First, a 50ms sampling period is used to calculate the state difference at adjacent moments and obtain the difference vector: Δx(t) = x(t) - x(t-1). In order to better capture the dynamic characteristics of the system, the second-order difference is calculated at the same time: Δ²x(t) = Δx(t) -Δx(t-1). Among them, the first-order difference reflects the rate of change of the system state, and the second-order difference reflects the acceleration characteristics of the system state. These two differential values ​​are of great significance for identifying the sudden change characteristics and trend changes of the system. Especially when detecting faults such as blockage or leakage in the powder conveying system, the acceleration characteristics of the state variables are often important early warning indicators.

[0175] S4.3: constructing a flatness matrix based on the first-order difference value and the second-order difference value;

[0176] In S4.3, the flatness matrix M(t) is constructed based on the first-order and second-order differences: M(t) = [Δx(t), Δ²x(t)]. The dimension of this matrix is ​​9×2, and each row corresponds to the first-order and second-order differences of a state variable.

[0177] The construction of the flatness matrix fully considers the coupling relationship between various parameters in the pneumatic conveying system, such as the mutual influence between pressure and flow rate, and between material level and pressure. By considering both the first-order difference and the second-order difference, the dynamic characteristics of the system can be more comprehensively described, providing a basis for the subsequent singular value decomposition.

[0178] S4.4: Perform singular value decomposition on the flatness matrix to obtain the differential flatness index.

[0179] In S4.4, the flatness matrix M(t) is subjected to singular value decomposition (SVD): M(t) = U(t)Σ(t)V(t . The diagonal elements of Σ(t) are ≥ ≥ 0 is a singular value.

[0180] The differential flatness index of the system is defined as: μ(t) = / .

[0181] It should be noted that the closer μ(t) is to 0, the closer the system state change is to the plane, and the easier it is to control; the closer μ(t) is to 1, the farther the system state change is from the plane, and stronger control is required.

[0182] In practical applications, based on a large amount of experimental data, the following control strategy is set: when μ(t) < 0.2, the system state changes smoothly and the current control parameters are maintained; when 0.2 ≤ μ(t) < 0.5, the system state fluctuates and the control parameters need to be adjusted appropriately; when μ(t) ≥ 0.5, the system state changes drastically and there may be a risk of blockage or leakage, and the emergency control strategy needs to be activated.

[0183] This differential flatness calculation method is particularly suitable for lithium iron phosphate powder conveying systems because: first, it takes into account the data of multiple pressure measurement points, which can timely detect the risk of local blockage; second, it introduces speed and material level data, which can fully reflect the conveying status; third, it captures the acceleration characteristics of the system through the second-order difference, which is conducive to predicting the stability of the system; finally, the hierarchical control strategy takes into account the characteristics of powder conveying and takes timely intervention in working conditions with large fluctuations. The entire calculation process provides a quantitative basis for the dynamic characteristics of the system for the subsequent k-induction control algorithm.

[0184] S5: Designing a k-summary control algorithm according to the differential flatness index to generate control parameters for the induced draft fan speed and the opening of the material distribution valve of the pneumatic conveying system.

[0185] S5.1: Based on the differential flatness index, construct a system control objective function in the form of a Lyapunov function, wherein the system control objective function includes:

[0186] V(t) = 1 / 2[(p(t)-p )² / p ² + (v(t)-v )² / v ²],

[0187] Among them, p is the preset pressure, v The preset conveying speed.

[0188] In S5.1, the system control objective function in the form of a Lyapunov function is constructed based on the differential flatness index.

[0189] Taking the lithium iron phosphate powder conveying system as an example, the control goal of the system is to maintain the conveying pressure p(t) at the target pressure p =0.3MPa, while keeping the conveying speed v(t) at the ideal speed v = around 15m / s. Based on these two core control quantities, the objective function is constructed: V(t) = 1 / 2[(p(t)-p )² / p ² + (v(t)-v )² / v ²]. This objective function actually measures the degree of deviation between the current state of the system and the ideal operating point.

[0190] For example, when the actual pressure is 0.35 MPa and the actual speed is 13 m / s, V(t)=0.0354 can be calculated, indicating that the system has a certain deviation but is still within the controllable range.

[0191] S5.2: According to the control objective function, a control law including a proportional term and an integral term is designed, and the control law is in the following form:

[0192] (t) = - (p(t)-p )- (p(t)-p )dt

[0193] (t) = - (v(t)-v )- (v(t)-v )dt

[0194] in, is the first control gain parameter, is the second control gain parameter, is the third control gain parameter, is the fourth control gain parameter.

[0195] In S5.2, a control law containing proportional and integral terms is designed.

[0196] Control input for induced draft fan speed (t), design control law: (t) = - (p(t)-p ) - (p(t)-p )dt.

[0197] Taking the actual working condition as an example, when the system pressure is detected to rise from 0.3MPa to 0.35MPa, assuming =2.5, =0.8, the calculated instantaneous correction is -0.125, the integral correction is about -0.04, and the final control input (t) causes the speed of the induced draft fan to decrease by about 16.5%.

[0198] Control input for dosing valve opening (t), design control law: (t) = - (v(t)-v ) - (v(t)-v )dt. For example, when the conveying speed decreases from 15m / s to 13m / s, set =1.8, = 0.6, the calculated control input (t) Increase the opening of the dosing valve by about 12%, thereby increasing the conveying speed.

[0199] Among them, the hierarchical control is implemented based on the differential flatness index, including: when the differential flatness index is lower than a first threshold, maintaining the current control parameters; when the differential flatness index is between the first threshold and the second threshold, smoothly adjusting the control parameters; when the differential flatness index is higher than the second threshold, triggering an emergency control strategy to reduce the speed of the induced draft fan and the opening of the dosing valve.

[0200] The design of the control law includes: constructing a dual-loop control structure including a pressure control loop and a speed control loop; in the pressure control loop, designing a pressure control law based on the pressure deviation; in the speed control loop, designing a speed control law based on the speed deviation; and designing a control gain adaptive adjustment mechanism according to the differential flatness index.

[0201] S5.3: Determine the control parameter range of the rotation speed of the induced draft fan and the adjustment speed of the opening of the distribution valve based on the Lyapunov stability theory.

[0202] In S5.3, the control parameter range is determined based on the Lyapunov stability theory.

[0203] Taking the lithium iron phosphate powder conveying system as an example, the adjustment range of the induced draft fan speed is first determined. Considering the fluidization characteristics of lithium iron phosphate powder, the change rate of the induced draft fan speed is limited to no more than 5% of the rated speed per second to prevent drastic fluctuations in the conveying pressure.

[0204] For example, when the differential flatness index μ(t)=0.6, it indicates that the system state changes dramatically. At this time, the maximum adjustment rate of the induced draft fan speed is limited to 3% / s, and the speed adjustment range is limited to 60%-90% of the rated speed.

[0205] For the opening of the dosing valve, the adjustment speed is limited to no more than 3% of the full range per second, and the adjustment range is limited to between 25% and 85%. These parameter limits ensure the smoothness of the control action and avoid powder agglomeration or system oscillation that may be caused by drastic adjustments.

[0206] At the same time, the system will dynamically adjust the control parameters according to different working conditions. For example, when the moisture content of the conveyed material increases, the system automatically reduces the control gain ( From 2.5 to 1.8, From 1.8 to 1.2, slow down the control response speed to prevent excessive adjustment from causing material blockage. When the local resistance of the pipeline is detected to increase, the system will appropriately increase the weight of the integral term ( From 0.8 to 1.2), the system has enhanced its ability to correct continuous deviations. This adaptive parameter adjustment strategy ensures that the system can maintain good control effects under different working conditions.

[0207] S6: constructing a disturbance observer with an adaptive safety margin, and controlling the pneumatic conveying system based on the control parameters.

[0208] S6.1: Establish a nominal model including pipeline transport dynamics equations, wherein the nominal model is used to represent the relationship between pressure, flow rate and material concentration;

[0209] In S6.1, a nominal model including pipeline transport dynamics equations is established.

[0210] In the lithium iron phosphate powder conveying system, the horizontal section conveying dynamics model is first established. Considering that the pressure drop in the pipeline mainly comes from the friction loss of gas-solid two-phase flow, the model contains the following key parameters: pipeline equivalent diameter D(t), powder volume density ρ(t), gas dynamic viscosity μ(t) and pipeline friction coefficient λ(t). The pressure drop equation for the horizontal section is: Δp = λ(t)·L·ρ(t)·v²(t) / (2D(t)), where L is the length of the pipe section. For the vertical section, the influence of the gravity term needs to be considered additionally, and the pressure drop equation is corrected to: Δp = λ(t)·L·ρ(t)·v²(t) / (2D(t)) + ρ(t)·g·H, where H is the vertical lifting height. These parameters are all changing in real time and will change with changes in operating conditions. For example, pipeline wear will cause the equivalent diameter to increase, and changes in the moisture content of the material will affect the volume density.

[0211] S6.2: Construct a disturbance estimator based on the nominal model, wherein the disturbance estimator includes a nominal state observer and a disturbance reconstructor, and the state equation of the state observer is:

[0212] (t+1) = A (t)+Bu(t)+L(y(t)- (t))+ (t)

[0213] Among them, among them, (t+1) is the estimated value of the state at time t+1; (t): the state estimate at the current moment; A is the system state transfer matrix; B is the control input matrix, u(t) is the control input vector, L is the observer gain matrix, which is used to correct the state estimate; y(t) is the actual measured output; (t) is the estimated output value; (t) is the estimated disturbance term used to compensate for the unmodeled dynamics and external disturbances in the system;

[0214] Determine the optimal parameters of the observer through simulation analysis: sliding mode gain =1.5, convergence gain =0.8, and the observer bandwidth is set to 50 Hz to balance the estimation accuracy and noise suppression capability.

[0215] In order to improve the robustness of the observer, an adaptive law can be introduced to dynamically adjust the gain parameters so that the observer can adapt to changes in system conditions.

[0216] S6.3: Use recursive least squares method to perform online identification of key parameters of the pneumatic conveying system.

[0217] In S6.3, online identification of key system parameters is achieved.

[0218] Using the recursive least squares method, the parameter identification algorithm is designed: (t) = (t-1) + P(t) (t)[y(t)- (t) (t-1)], where: is the parameter estimate, P is the covariance matrix, is the regression vector.

[0219] The forgetting factor of the identification algorithm is set to 0.95 to ensure sufficient tracking capability for changes in system parameters.

[0220] For example, when an abnormal increase in the pressure drop in a vertical pipe section is detected, parameter identification is used to determine whether it is caused by pipe scaling (equivalent diameter reduction) or changes in material viscosity.

[0221] The actual test shows that this method can detect the change of pipeline equivalent diameter by more than 5% within 2 seconds. At the same time, a trend database of parameter changes is established for predictive maintenance.

[0222] S6.4: Based on the output of the disturbance estimator, dynamically adjust the pressure, speed and material level safety margins of the system.

[0223] In S6.4, the system safety margin is dynamically adjusted based on the output of the disturbance observer.

[0224] First, set multiple safety boundaries: pressure safety boundary [0.1MPa, 0.4MPa], speed safety boundary [8m / s, 25m / s], and material level safety boundary [0.5m, 4.5m].

[0225] It should be pointed out that these boundaries are not fixed, but are dynamically adjusted according to the system state estimated by the observer. For example, when a scaling trend is observed in the pipeline (the equivalent diameter reduction trend exceeds 3% / day), the system automatically reduces the upper pressure limit from 0.4MPa to 0.35MPa, and at the same time reduces the upper speed limit from 25m / s to 20m / s to prevent the risk of blockage. For example, when a sudden increase in the moisture content of the material is detected (identified by the pressure drop characteristics), the system will further shrink the safety margin and adjust the speed range to [10m / s, 18m / s]. In addition, the system will dynamically adjust the safety margin according to the size of the disturbance. The larger the disturbance, the larger the reserved safety margin, ensuring that the system always operates within the safe range.

[0226] S6.5: Design a hierarchical control quantity limiting strategy based on the safety boundary.

[0227] In S6.5, a hierarchical control variable limiting strategy is designed.

[0228] Taking the lithium iron phosphate powder conveying system as an example, the strategy includes three layers of restrictions:

[0229] The first level is physical limitations, including the maximum speed of the induced draft fan (not exceeding 95% of the rated speed) and the opening range of the dosing valve (15%-90%).

[0230] The second level is safety restrictions based on current operating conditions. For example, when it is detected that the moisture content of the material increases, the upper limit of the induced draft fan speed is automatically reduced to 85% of the rated speed, and the upper limit of the dosing valve opening is reduced to 75%.

[0231] The third level is to optimize the restrictions based on energy consumption and equipment life. For example, when the material conveying volume is small, the energy consumption is reduced by limiting the maximum speed of the induced draft fan to below 80% of the rated speed. When the cumulative running time of the induced draft fan exceeds 5,000 hours, the maximum allowable speed is dynamically adjusted according to its vibration state. This hierarchical limiting strategy ensures that the system can achieve long-term stable operation of the equipment while ensuring safety and efficiency.

[0232] S6.6: Setting a control amount buffer zone in the control amount limiting strategy.

[0233] In S6.6, a control amount buffer is set in the control amount limiting strategy.

[0234] During implementation, a 10% buffer zone is set for each layer of restriction.

[0235] Taking the induced draft fan control as an example, if the upper limit of the safe speed under the current working conditions is 85% of the rated speed, the system will start to take preventive adjustments when it reaches 81%, and avoid reaching the limit through relatively gentle control actions. Similarly, for the control of the dosing valve, in the range of 8% before the opening limit, the controller will automatically reduce the adjustment rate from the normal 3% / s to 1% / s. Practice has shown that after setting the buffer zone, the system's control action is smoother, greatly reducing the frequency of limit triggering alarms, from an average of 8 times a day to less than 2 times.

[0236] S6.7: Predict the system state evolution trend based on the current system state, and adjust the control strategy according to the system state evolution trend.

[0237] In S6.7, the future state evolution trend is predicted based on the current system state.

[0238] A prediction model based on deep learning is used to use the operating data of the past 10 minutes to predict the system status in the next 3-5 seconds. For example, when the prediction model shows that the material concentration may exceed 25mg / m³ (the safety threshold is 30mg / m³) in 5 seconds, the control system will adjust the opening of the dosing valve in advance, gradually reducing it from the current 65% to 55%, rather than waiting until the actual exceeding of the standard occurs before responding. At the same time, the system will adjust the speed of the induced draft fan in advance based on the predicted pressure change trend to ensure that the pressure is always maintained within the ideal range. This predictive control strategy significantly improves the stability of the system and reduces the pressure fluctuation range from ±0.05MPa to ±0.02MPa.

[0239] S6.8: Adjust control parameters according to the real-time response of the system.

[0240] In S6.8, the control parameters are dynamically adjusted according to the real-time response of the system. First, a control performance evaluation index system is established, including: pressure control deviation root mean square (target less than 0.015MPa), speed fluctuation rate (target less than 5%), energy consumption index (no more than 75% of rated power), etc.

[0241] Based on these indicators, the system evaluates the control effect every 30 minutes. When it is found that the control performance has declined, for example, the pressure fluctuation exceeds the target value by more than 20%, the system will automatically start the parameter optimization program. The system model under the current working conditions is obtained through online identification, and then the controller parameters are re-optimized based on the genetic algorithm. During the optimization process, both control performance and robustness are considered to ensure that the new parameter settings can improve the control effect without making the system too sensitive to external interference. Practical data show that this adaptive parameter adjustment mechanism keeps the control performance of the system at a good level, the average pressure control deviation is reduced by 30%, and the system energy consumption is reduced by 15%.

[0242] In addition, see Figure 2 The present application also provides a monitoring and diagnostic system for a pneumatic conveying system, comprising:

[0243] A data acquisition unit, used to collect multi-source monitoring data of the pneumatic conveying system;

[0244] A data processing unit, used to process the multi-source monitoring data to obtain standardized monitoring data;

[0245] A model training unit, used to construct and train a deep learning model based on the standardized monitoring data to obtain a system status assessment model;

[0246] An indicator generating unit, configured to construct a training set based on the historical operation data of the pneumatic conveying system, train a pre-constructed deep learning model, and obtain a system state evaluation model, wherein the deep learning model comprises an input layer corresponding to the transportation parameters of the pneumatic conveying system, a plurality of hidden layers for extracting features, and an output layer for outputting an evaluation result;

[0247] A control parameter generating unit, for generating a differential flatness index based on the system state evaluation model and the multi-source monitoring data by using a system state vector algorithm, wherein the differential flatness index is used to characterize the system state change characteristics, and for designing a k-summary control algorithm according to the differential flatness index to generate control parameters for the induced draft fan speed and the opening of the material distribution valve of the pneumatic conveying system;

[0248] A safety control unit is used to construct a disturbance observer with an adaptive safety margin and control the pneumatic conveying system based on the control parameters.

Claims

1. A full-process monitoring and fault diagnosis method for a pneumatic conveying system, characterized in that: The method comprises: Collect multi-source monitoring data of pneumatic conveying systems; Processing the multi-source monitoring data to obtain standardized monitoring data; Building a training set based on the historical operation data of the pneumatic conveying system, training a pre-built deep learning model, and obtaining a system state assessment model, wherein the deep learning model includes an input layer corresponding to the transportation parameters of the pneumatic conveying system, a plurality of hidden layers for extracting features, and an output layer for outputting an assessment result; Based on the system state assessment model and the multi-source monitoring data, a system state vector algorithm is used to generate a differential flatness index, where the differential flatness index is used to characterize system state change characteristics; According to the differential flatness index, a k-summary control algorithm is designed to generate control parameters for the induced draft fan speed and the opening of the material distribution valve of the pneumatic conveying system; constructing a disturbance observer with an adaptive safety margin to control the pneumatic conveying system based on the control parameters; Wherein, the k-summary control algorithm is designed according to the differential flatness index to generate control parameters for the induced draft fan speed and the opening of the material distribution valve of the pneumatic conveying system, including: Based on the differential flatness index, a system control objective function in the form of a Lyapunov function is constructed, and the system control objective function includes: ， Wherein, p* is the preset pressure, v* is the preset delivery speed, v(t) represents the delivery speed, and p(t) represents the delivery pressure; According to the control objective function, a control law including a proportional term and an integral term is designed. The control law takes the following form: ， ， Among them, k1 is the first control gain parameter, k2 is the second control gain parameter, k3 is the third control gain parameter, k4 is the fourth control gain parameter, u1(t) is the control input of the induced draft fan speed, and u2(t) is the control input of the dosing valve opening; The control parameter ranges of the rotation speed of the induced draft fan and the adjustment speed of the opening of the distribution valve are determined based on the Lyapunov stability theory.

2. The method according to claim 1, characterized in that The multi-source monitoring data of the pneumatic conveying system is collected, including: A distributed sensor network is used to collect multi-source monitoring data of the pneumatic conveying system, wherein the multi-source monitoring data includes: pressure data, flow rate data, temperature data and equipment vibration data; Based on threshold parameters adapted to the types of the multi-source monitoring data, filtering and time synchronization operations are performed on the multi-source monitoring data.

3. The method according to claim 1, characterized in that The multi-source monitoring data is processed to obtain standardized monitoring data, including: The multi-source monitoring data is subjected to noise reduction processing by a sliding window median filtering technique to obtain noise-reduced multi-source monitoring data; Using a statistical analysis method to identify abnormal data in the noise-reduced multi-source monitoring data, and performing an operation of eliminating the abnormal data to obtain target multi-source monitoring data; Interpolating the missing values ​​of the target multi-source monitoring data to obtain a complete data sequence; Normalization processing is performed based on the multi-source monitoring data in the complete data sequence to obtain the standardized monitoring data.

4. The method according to claim 1, characterized in that: The method of constructing a training set based on the historical operation data of the pneumatic conveying system, training a pre-constructed deep learning model, and obtaining a system status assessment model includes: Based on the transport parameters of the pneumatic conveying system, a multi-layer neural network structure is designed, and a nonlinear activation function is used in the hidden layer; Adding a normalization layer and a random dropout layer to the multi-layer neural network structure; Building a training set based on historical operation data of the pneumatic conveying system, wherein the historical operation data includes normal operation data and fault state data; The deep learning model is optimized by using a gradient descent method and an early stopping strategy to obtain the system state evaluation model.

5. The method according to claim 1 or 4, characterized in that: The input layer includes 14 characteristic nodes, wherein the 14 characteristic nodes correspond to the feed port pressure of the pneumatic conveying system, the conveying speed of 4 measuring points, 3 material level data, 2 induced draft fan vibration values, 2 distribution valve vibration values, and 2 dust concentration values; The hidden layer includes a first hidden layer, a second hidden layer, a third hidden layer, and a fourth hidden layer, the number of nodes in the first hidden layer is 128, the number of nodes in the second hidden layer is 256, the number of nodes in the third hidden layer is 128, and the number of nodes in the fourth hidden layer is 64, and the first hidden layer, the second hidden layer, the third hidden layer, and the fourth hidden layer all use a ReLU activation function, wherein a BatchNormalization layer is added between the second hidden layer and the third hidden layer, and a Dropout layer with a dropout rate of 0.3 is added after the third hidden layer and the fourth hidden layer respectively; The output layer contains 8 feature nodes, which correspond to: system blockage risk assessment value, pipeline wear warning value, induced draft fan failure warning value (0-1), distribution valve failure warning value, system leakage risk value, and predicted pressure values ​​of three key parts.

6. The method according to claim 1, characterized in that The method of generating a differential flatness index based on the system state assessment model and the multi-source monitoring data by using a system state vector algorithm includes: Outputting the prediction results of the multi-source monitoring data through the system state assessment model, wherein the prediction results include a multi-dimensional state vector of the pneumatic conveying system; Based on the system state vector algorithm, calculating the first-order difference value and the second-order difference value of the multidimensional state vector; constructing a flatness matrix based on the first-order difference value and the second-order difference value; Performing singular value decomposition on the flatness matrix to obtain the differential flatness index.

7. The method according to claim 1, characterized in that Based on the implementation of hierarchical control according to the differential flatness index, the method includes: when the differential flatness index is lower than a first threshold, maintaining the current control parameter; when the differential flatness index is between the first threshold and the second threshold, smoothly adjusting the control parameter; when the differential flatness index is higher than the second threshold, triggering an emergency control strategy to reduce the speed of the induced draft fan and the opening of the material distribution valve; The design of the control law includes: constructing a dual-loop control structure including a pressure control loop and a speed control loop; in the pressure control loop, designing a pressure control law based on the pressure deviation; in the speed control loop, designing a speed control law based on the speed deviation; and designing a control gain adaptive adjustment mechanism according to the differential flatness index.

8. The method according to claim 1, characterized in that The construction of a disturbance observer with an adaptive safety margin to control the pneumatic conveying system based on the control parameters includes: Establishing a nominal model including pipeline transport dynamics equations, wherein the nominal model is used to represent the relationship between pressure, flow rate and material concentration; A disturbance estimator is constructed based on the nominal model. The disturbance estimator includes a nominal state observer and a disturbance reconstructor. The state equation of the state observer is: ， Where x̂(t+1) is the state estimate at time t+1; x̂(t): the state estimate at the current time; A is the system state transfer matrix; B is the control input matrix, u(t) is the control input vector, and L is the observer gain matrix used to correct the state estimate; y(t) is the actual measured output; ŷ(t) is the estimated output value; d̂(t) is the estimated disturbance term used to compensate for the unmodeled dynamics and external disturbances in the system; Using recursive least squares method to identify key parameters of the pneumatic conveying system online; Based on the output of the disturbance estimator, dynamically adjust the pressure, speed and material level safety margins of the system; Designing a hierarchical control quantity limiting strategy according to the safety boundary; Setting a control amount buffer zone in the control amount limiting strategy; Predicting the system state evolution trend based on the current system state, and adjusting the control strategy according to the system state evolution trend; Adjust control parameters based on real-time system response.

9. A full-process monitoring and fault diagnosis system for a pneumatic conveying system, characterized in that: include: A data acquisition unit, used to collect multi-source monitoring data of the pneumatic conveying system; A data processing unit, used to process the multi-source monitoring data to obtain standardized monitoring data; A model training unit, used to construct and train a deep learning model based on the standardized monitoring data to obtain a system status assessment model; An indicator generating unit, configured to construct a training set based on the historical operation data of the pneumatic conveying system, train a pre-constructed deep learning model, and obtain a system state evaluation model, wherein the deep learning model comprises an input layer corresponding to the transportation parameters of the pneumatic conveying system, a plurality of hidden layers for extracting features, and an output layer for outputting an evaluation result; A control parameter generating unit, for generating a differential flatness index based on the system state evaluation model and the multi-source monitoring data and using a system state vector algorithm, wherein the differential flatness index is used to characterize the system state change characteristics, and for designing a k-summary control algorithm according to the differential flatness index to generate control parameters for the induced draft fan speed and the opening of the material distribution valve of the pneumatic conveying system; A safety control unit is used to construct a disturbance observer with an adaptive safety margin and control the pneumatic conveying system based on the control parameters.

Images