Hydraulic hoist remote diagnosis control system and method thereof

Abstract

This invention relates to the field of intelligent control technology for water conservancy and hydropower engineering facilities and fluid transmission, specifically a remote diagnostic control system and method for hydraulic gate hoists. The system includes: constructing a deep reinforcement learning model based on physical information; initializing the system's safe operation boundary parameters and task urgency parameters; and triggering an adaptive diagnostic control process in response to a received remote flood discharge command, including: constructing a standardized temporal state tensor; extracting internal damage feature vectors representing the state of hydraulic components; determining the current system's functional remaining life entropy through microscopic fatigue damage accumulation calculation; generating a survival-priority control trajectory; and outputting the drive control signal of the hydraulic actuator through multimodal actuator adjustment. This invention avoids diagnostic model input distortion caused by data loss or communication interruption, ensuring the continuity and reliability of the diagnostic system under extreme environments.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for water conservancy and hydropower engineering facilities and fluid transmission, specifically to a remote diagnostic control system and method for hydraulic gate hoists. Background Technology

[0002] In the current operation and maintenance environment of large-scale water conservancy facilities, hydraulic gate hoists, as key execution units for flood discharge scheduling and flow regulation, often face the dual challenges of equipment aging and extreme working conditions, and their operating status needs to be monitored remotely in real time through multi-dimensional sensors.

[0003] To ensure system safety, existing solutions typically employ threshold-based state monitoring or manual experience-based scheduling and control. This involves collecting signals such as pressure and vibration to determine if limits are exceeded, and then having operators directly execute the start-up operation upon receiving a flood discharge command. While these solutions can maintain basic operation under normal conditions, they suffer from significant drawbacks in the face of unstable remote communication and complex hydraulic environments. Remote data transmission is prone to packet loss and outliers, and traditional linear interpolation repair methods struggle to preserve the nonlinear dynamic characteristics of the signals, leading to distorted diagnostic input. Strong external flow-induced vibration noise often masks weak internal damage signals such as cavitation and leakage, making it difficult to effectively decouple and identify early fault characteristics. More critically, existing control strategies lack a quantitative balance mechanism between the remaining functional lifespan of equipment and the urgency of the task. When dealing with urgent tasks on aging equipment, the pursuit of response speed can easily lead to sudden equipment failure, or over-protection can delay flood discharge. Therefore, achieving high-fidelity state awareness in high-noise and data-deficient environments, and generating adaptive diagnostic control strategies that balance equipment survivability and task execution efficiency, has become a pressing technical problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a remote diagnostic control system and method for hydraulic gate hoists, aiming to solve the problems of poor data communication quality and difficulty in identifying early faults due to strong background noise under remote extreme working conditions in existing technologies. It repairs nonlinear dynamic characteristics through time-series reconstruction and decouples weak internal damage feature vectors from strong flow-induced vibration using a nonlinear hydraulic fluid dynamics model. Furthermore, it can adaptively generate survival-priority control trajectories based on a quantitative assessment of functional remaining life entropy and task urgency, thereby achieving a dynamic balance between flood discharge task execution efficiency and equipment safety boundaries while ensuring that aging equipment does not suffer catastrophic failure. Specifically, the technical solution of this invention is as follows:

[0005] A remote diagnostic control method for hydraulic gate hoists includes:

[0006] Multi-dimensional sensor arrays are deployed at key nodes of the hydraulic gate hoist system, and a deep reinforcement learning model based on physical information is constructed.

[0007] Initialize the system's safe operating boundary parameters and task urgency parameters;

[0008] In response to receiving a remote flood discharge task command, an adaptive diagnostic control process is triggered, including:

[0009] Step 1: Collect real-time monitoring data from the multi-dimensional sensor array, perform temporal reconstruction and missing value imputation on the real-time monitoring data, and construct a standardized temporal state tensor;

[0010] Step 2: Input the time-series state tensor into the nonlinear hydraulic fluid dynamics model, and extract the internal damage feature vector characterizing the state of the hydraulic components through signal feature decoupling analysis;

[0011] Step 3: Input the internal damage feature vector into the pre-trained lifetime prediction network, and determine the functional remaining lifetime entropy of the current system through the accumulation calculation of micro fatigue damage.

[0012] Step 4: Input the functional remaining lifetime entropy, task urgency parameters, and safe operation boundary parameters into the physical information-based deep reinforcement learning model to generate a survival-priority control trajectory;

[0013] Step 5: Based on the survival-first control trajectory and real-time monitoring data, adjust the multi-modal actuator to output the drive control signal of the hydraulic actuator;

[0014] The timing interactions between the above steps and system modules are transmitted unidirectionally according to preset logical instructions and trigger verification.

[0015] Preferably, the multidimensional sensor array includes:

[0016] A pressure pulsation sensor is installed between the pump outlet and the main control valve in the hydraulic pipeline;

[0017] Vibration acceleration sensors installed at the gate connection points;

[0018] Oil contamination and temperature monitoring sensors are installed inside the hydraulic oil tank.

[0019] Preferably, the real-time monitoring data is processed through time-series reconstruction and missing value imputation, including:

[0020] Identify communication packet loss segments and abnormal outliers in real-time monitoring data;

[0021] Based on the spatiotemporal correlation of historical operational data, generative adversarial networks are used to interpolate data for lost communication packets.

[0022] The interpolated data is denoised and standardized to generate a time-series state tensor that includes pressure, vibration, and oil state.

[0023] Preferably, the internal damage feature vector of the hydraulic component is extracted through signal feature decoupling analysis, including:

[0024] Using a nonlinear hydraulic fluid dynamics model, an external disturbance reference signal under a preset flow-induced vibration is generated;

[0025] The temporal state tensor is mapped to the frequency domain space, and the components that overlap with the spectrum of the external interference reference signal are filtered out by an adaptive filtering algorithm.

[0026] Characteristic frequency components representing cavitation, leakage, and wear are extracted from the filtered residual signal to generate an internal damage feature vector.

[0027] Preferably, the remaining functional lifetime entropy of the current system is determined by calculating the cumulative micro-fatigue damage, including:

[0028] Using a pre-defined probability mapping model, the internal damage feature vector is mapped to the probability of hydraulic oil film rupture and the probability of valve core thermal jamming.

[0029] Calculate the information entropy value at the current moment based on the joint distribution of the probability of hydraulic oil film rupture and the probability of valve core thermal jamming;

[0030] A weighted integral operation is performed on the information entropy values ​​within the historical time window to generate the functional remaining lifetime entropy, which represents the degree of cumulative irreversible damage to the system.

[0031] Preferably, a survival-priority control trajectory is generated based on functional remaining lifetime entropy, task urgency parameters, and safe operation boundary parameters, including:

[0032] If the remaining lifespan entropy of the function is lower than the preset safety threshold, a continuous and rapid activation trajectory will be generated with the goal of maximizing response speed.

[0033] If the remaining functional lifetime entropy is higher than or equal to the preset safety threshold, then a pulse-type micro-motion trajectory or a pressure unloading recovery trajectory is generated with the goal of suppressing the rate of entropy growth.

[0034] Survival-first control trajectory represents the target pressure and displacement sequence of the hydraulic actuator within a future time window.

[0035] Preferably, the drive control signal of the hydraulic actuator is adjusted by a multi-modal actuator, including:

[0036] The survival-first control trajectory is decomposed into proportional valve opening commands and pump station speed commands;

[0037] Monitor the real-time pressure response during execution. If the real-time pressure response exceeds the safe operating boundary parameters, force an emergency unloading control signal to be output.

[0038] The hydraulic gate hoist is driven to operate according to the proportional valve opening command and the pump station speed command.

[0039] A remote diagnostic control system for hydraulic gate hoists includes:

[0040] The sensing module is configured to collect data from a multi-dimensional sensor array at key nodes of the hydraulic gate hoist system.

[0041] The initialization module is configured to set safe operation boundary parameters and task urgency parameters, and load a deep reinforcement learning model based on physical information.

[0042] The diagnostic control module includes:

[0043] The data preprocessing unit is configured to collect real-time monitoring data and perform temporal reconstruction and missing value imputation on the real-time monitoring data to construct a standardized temporal state tensor.

[0044] The feature decoupling unit is configured to combine the time-series state tensor with the nonlinear hydraulic fluid dynamics model, and extract the internal damage feature vector of the hydraulic component through signal feature decoupling analysis;

[0045] The entropy calculation unit is configured to input the internal damage feature vector into the life prediction network and determine the functional remaining life entropy of the current system through the accumulation calculation of micro fatigue damage.

[0046] The strategy generation unit is configured to input the functional remaining lifetime entropy, task urgency parameters and safe operation boundary parameters into the physical information-based deep reinforcement learning model to generate a survival-priority control trajectory.

[0047] The execution control unit is configured to adjust the output of the hydraulic actuator drive control signal based on the survival-priority control trajectory and real-time monitoring data through the multi-modal actuator.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1. This invention employs a generative adversarial network based on spatiotemporal correlation to interpolate data from lost communication packets, effectively solving the data integrity problem caused by unstable remote communication. Compared with traditional linear interpolation, this method can accurately identify outliers and preserve the nonlinear dynamic characteristics of the signal, avoiding distortion of the diagnostic model input due to data loss or communication interruption, and ensuring the continuity and reliability of the diagnostic system in extreme environments.

[0050] 2. This invention combines a nonlinear hydraulic fluid dynamics model with an adaptive filtering algorithm to achieve accurate removal of weak internal fault signals under high-intensity water flow background noise. By filtering out components that overlap with the spectrum of external interference reference signals, the system can effectively extract characteristic frequencies representing cavitation, leakage and wear from complex flow-induced vibrations, solving the problem of strong noise masking early damage characteristics and significantly improving the accuracy of fault identification in complex hydraulic environments.

[0051] 3. This invention introduces functional remaining lifetime entropy as an indicator to quantify the vulnerability of a system. Through the calculation of microscopic fatigue damage accumulation, it can keenly capture the trend of the system transitioning from a steady state to a chaotic state. This indicator breaks through the limitations of traditional fixed window summation and correctly characterizes the degree of cumulative irreversible damage that increases monotonically over time, providing an accurate quantitative decision boundary for the survival priority control of aging equipment under extreme working conditions.

[0052] 4. This invention constructs a deep reinforcement learning model based on physical information, realizing adaptive switching of control strategies between task urgency and equipment health; the pulse-type micro-motion or pressure unloading recovery trajectory generated when the system is on the verge of failure breaks through the limitations of human experience, can significantly extend the effective working time of faulty equipment under extreme conditions, avoid sudden equipment failure caused by blind forced operation, and effectively balance flood discharge efficiency and inherent safety. Attached Figure Description

[0053] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0054] Figure 1 This is a flowchart of the method of the present invention;

[0055] Figure 2 This is a structural diagram of the system of the present invention. Detailed Implementation

[0056] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0057] Example 1:

[0058] Please see Figure 1 A remote diagnostic control method for hydraulic gate hoists includes: deploying multi-dimensional sensor groups at key nodes of the hydraulic gate hoist system and constructing a deep reinforcement learning model based on physical information; initializing the system's safe operation boundary parameters and task urgency parameters; and triggering an adaptive diagnostic control process in response to receiving a remote flood discharge task command, including:

[0059] Step 1: Collect real-time monitoring data from the multi-dimensional sensor array, perform temporal reconstruction and missing value imputation on the real-time monitoring data, and construct a standardized temporal state tensor;

[0060] Step 2: Input the time-series state tensor into the nonlinear hydraulic fluid dynamics model, and extract the internal damage feature vector characterizing the state of the hydraulic components through signal feature decoupling analysis;

[0061] Step 3: Input the internal damage feature vector into the pre-trained lifetime prediction network, and determine the functional remaining lifetime entropy of the current system through the accumulation calculation of micro fatigue damage.

[0062] Step 4: Input the functional remaining lifetime entropy, task urgency parameters, and safe operation boundary parameters into the physical information-based deep reinforcement learning model to generate a survival-priority control trajectory;

[0063] Step 5: Based on the survival-first control trajectory and real-time monitoring data, adjust the multi-modal actuator to output the drive control signal of the hydraulic actuator.

[0064] This embodiment details the execution logic of a remote diagnostic control method for hydraulic gate hoists. This method aims to address the balance between survival and efficiency of aging equipment under extreme flood discharge conditions. Specifically, the system constructs a perception and decision-making architecture at the physical level. This is achieved by deploying multi-dimensional sensor arrays at key nodes of the hydraulic system to acquire physical state data and building a deep reinforcement learning model based on physical information. - This model is derived from the fusion of constraints from nonlinear hydraulic dynamics equations and the exploratory capabilities of deep reinforcement learning. - Architecture; specifically, The network consists of an input layer and two layers, each with a specific function. and Hidden layers of 1 neuron, using Activation function, and Output layer; The networks have the same hidden layer structure, and the output layer is a linear output of a single neuron;

[0065] Deep reinforcement learning model based on physical information - By introducing physical constraint terms into the loss function The hydraulic dynamics boundary is embedded into the network training process, where... The penalty coefficient is... For the output pressure predicted by the network, To determine the upper limit of the rated safe pressure of the hydraulic system, the model employs a deep deterministic strategy gradient algorithm for network training, by statically or dynamically adding this physical constraint term to the total loss function. This forces the network to prioritize satisfying the hydrodynamic safety boundary when updating parameters via gradient descent, thereby ensuring the unique executability and universality of the specific method of embedding physical information constraints on the generated action sequence; whereby, the state space... Defined as a five-dimensional feature vector ,in, For real-time pressure, The amplitude of the vibration. The remaining lifetime entropy of the function. Due to the urgency of the task, The current valve opening and operating space. Defined as a continuous action vector ,in, For target pressure, The target displacement; to address the inconsistency in dimensions caused by directly adding different physical quantities, the reward function... The design incorporates a normalization factor as follows:

[0066]

[0067] in, The dimensionless weighting coefficients are determined through hyperparameter search in this embodiment. The parameter values ​​given here are preferred exemplary values ​​for the specific working conditions of this embodiment. In practical applications, those skilled in the art can make adaptive adjustments according to the specific hydraulic hoist model and operating environment. Target pressure, unit: , Actual pressure, unit: , Maximum pressure at the safety boundary, unit: , The entropy growth rate is expressed in units of 1000 m / s. , In this embodiment, the reference entropy growth rate is set to [value]. The value is taken from the average entropy increase during the steady-state operation of the system. As an indicator function, when the state Exceeding safe operating boundaries The value is 1 when the condition is met and 0 otherwise. The physical meaning of the indicator function is to impose a penalty term constraint on dangerous states of the system that exceed the physical safety boundary. To ensure the computability and dimensional consistency of the formula, the entropy growth rate is given here. Calculation method:

[0068]

[0069] in, The entropy value at the current moment, in units of , The entropy value at the previous time step, in units of... , The sampling period is set to [value] in this embodiment. , The entropy growth rate is expressed in units of 1000 m / s. The selection of these experimental parameters is based on Shannon's sampling theorem. The general adjustment rule is that the sampling period should be less than half the reciprocal of the system's highest dynamic response frequency, and the reference entropy growth rate can be dynamically and adaptively calibrated by extracting the average value of the historical steady-state fault-free operating range of different equipment models. Its physical meaning is to search for atypical control strategies in the solution space under the premise of satisfying physical conservation laws. On this basis, the system initialization sets the safe operating boundary parameters. With task urgency parameter ;

[0070] in, The method of obtaining the data is as follows: a destructive physical limit test is performed on the same model of hydraulic cylinder, and the maximum pressure before plastic deformation occurs is recorded. and maximum vibration The specific test standard for this destructive test is as follows: Under normal temperature and pressure test conditions, a step-increase hydraulic loading method is used. The criterion for determining failure is that the sensor detects that the amount of irreversible plastic deformation of the component exceeds 0.2% of the original size. In this embodiment, for the QPP-2000 hydraulic press, the actual test... ,set up The coefficient The selection criteria are derived from Joint analysis of criteria and sensor measurement error: Considering the overall measurement error of the pressure sensor is... And the variance of the material's yield strength, in order to ensure that... Within the confidence interval, the system's operating state remains within the material's elastic deformation region, with reserved... Safety margins are necessary; their physical meaning is the set of physical limits under which a system will not experience catastrophic failure.

[0071] The command levels originate from the flood control dispatch center, and the specific quantitative mapping relationship is as follows: Level 1 emergency command corresponds to... Level 2 emergency order Level 3 regular instructions correspond The physical meaning is a scalar that quantifies the time requirement of the current flood discharge task; in response to receiving a remote flood discharge task instruction, the system triggers an adaptive diagnostic control process;

[0072] Real-time monitoring data from a multi-dimensional sensor array is collected, and generative adversarial networks are used to perform temporal reconstruction and missing value imputation on the data, constructing a standardized temporal state tensor. Temporal state tensor The dimension is defined as ,in, For batch size, For time step, The feature dimension includes pressure, vibration, temperature, contamination level, and flow rate. This tensor is input into a nonlinear hydraulic fluid dynamics model. External water flow interference is filtered out through signal feature decoupling analysis, and internal damage feature vectors characterizing the state of hydraulic components are extracted. Internal damage feature vector It is a 12-dimensional dimensionless normalized vector containing components such as the proportion of cavitation energy and the proportion of leakage frequency band energy; The input is fed into a pre-trained lifetime prediction network, and the functional remaining lifetime entropy of the current system is determined through the accumulation of microscopic fatigue damage. ;

[0073] This entropy value originates from the combination of information entropy theory and fatigue damage model, and its physical meaning is to quantify the degree of uncertainty of the system's distance from functional sudden death; , and enter - Model, generating survival-priority control trajectories This trajectory aims to plan the target pressure and displacement sequence of the hydraulic actuator within a future time window; according to Based on real-time monitoring data, the multi-modal actuator is adjusted to output drive control signals for the hydraulic actuator. .

[0074] Example 2:

[0075] A multi-dimensional sensor array, including:

[0076] A pressure pulsation sensor is installed between the pump outlet and the main control valve in the hydraulic pipeline;

[0077] Vibration acceleration sensors installed at the gate connection points; oil contamination and temperature monitoring sensors installed inside the hydraulic oil tank.

[0078] This embodiment further specifies the multi-dimensional sensor deployment strategy in Embodiment 1. This deployment strategy follows the principle of covering critical failure points, aiming to capture the holographic state of the hydraulic system. The pressure pulsation sensor is deployed on the high-pressure side of the hydraulic pipeline, i.e., between the pump outlet and the main control valve. This sensor is configured to collect high-frequency pressure waves. Its physical meaning is to monitor the pressure pulsation in the early stage of hydraulic shock and cavitation, because the high-pressure side is the area where the pressure shock is most intense and the pipe is most prone to bursting; at the same time, the vibration acceleration sensor is placed at the gate connection, that is, near the trunnion of the hydraulic cylinder piston rod.

[0079] The sensor is configured to collect vibration signals. Its physical meaning lies in sensing the mixed signals of external flow-induced vibration and internal mechanical jamming; this location is the coupling point for the transmission of hydraulic load to the mechanical system. Furthermore, oil contamination and temperature monitoring sensors are deployed in the return oil area inside the hydraulic tank. These sensors are configured to monitor the physicochemical properties of the oil, including temperature. It determines the oil viscosity, leakage rate, and contamination level. It reflects the degree of accumulated wear and tear within the system.

[0080] Example 3:

[0081] Real-time monitoring data will be processed through time-series reconstruction and missing value imputation, including:

[0082] Identify communication packet loss segments and abnormal outliers in real-time monitoring data;

[0083] Based on the spatiotemporal correlation of historical operational data, generative adversarial networks are used to interpolate data for lost communication packets.

[0084] The interpolated data is denoised and standardized to generate a time-series state tensor that includes pressure, vibration, and oil state.

[0085] This embodiment is a further specification of the data preprocessing step in Embodiment 1; this step employs a spatiotemporal correlation generative adversarial network. - This system aims to solve data integrity issues caused by unstable remote communication. It performs an anomaly detection step, using a sliding window statistical method to scan real-time monitoring data and identify lost packet segments, i.e., those with values... Or zero data segments, and outliers exceeding physical limits; utilizing - Data interpolation is performed on the lost packet segments; - The specific network structure configuration is as follows: Generator The U-Net architecture is adopted, and both its encoder and decoder consist of 3-layer bidirectional LSTM, with the hidden layer dimension set to 1. Used to capture long-term sequence dependencies; discriminator Employing a 5-layer one-dimensional convolutional neural network (1D-CNN), with a convolutional kernel size of [missing information]. Used to extract local time-frequency features; loss function Designed as a weighted sum of adversarial loss and reconstruction loss:

[0086]

[0087] in, This represents the mathematical expectation operation. For true and complete data, Input data for the mask. It is a binary mask matrix. To reconstruct the weights, in this embodiment they are set to... To balance the magnitude of the countermeasure loss and the reconstruction loss, The Hadamard product represents the element-wise multiplication of corresponding positions in a matrix. Its core meaning is to implement a physical operation for masking and filtering invalid or missing data. The training process employs an alternating optimization strategy: fixed... optimization To maximize the accuracy of discrimination, then fix optimization To minimize Until the generated data passes The discrimination threshold is set, and the convergence condition for network training is set as follows: the discriminator output probability stabilizes around 0.5 and the reconstruction loss tends to a stable lower bound. The root mean square error (RMSE) index is used to verify the model performance. When the RMSE of the validation set is less than the preset tolerance, the network training is considered complete. In response to the completion of data interpolation, the system performs wavelet threshold denoising on the complete data and uses Z-Score normalization. A time-series state tensor containing pressure, vibration, and oil state is generated. This serves as a unified input format for subsequent models;

[0088] This embodiment utilizes the spatiotemporal distribution patterns of data learned by generative adversarial networks to achieve data self-repair in the event of sensor failure or communication interruption. Compared with traditional linear interpolation, this method can preserve the nonlinear dynamic characteristics of the signal, avoid the distortion of diagnostic model input caused by data loss, and ensure the continuity and reliability of the diagnostic system under extreme environments.

[0089] Example 4:

[0090] Through signal feature decoupling analysis, the internal damage feature vector of hydraulic components is extracted, including:

[0091] Using a nonlinear hydraulic fluid dynamics model, an external disturbance reference signal under a preset flow-induced vibration is generated;

[0092] The temporal state tensor is mapped to the frequency domain space, and the components that overlap with the spectrum of the external interference reference signal are filtered out by an adaptive filtering algorithm.

[0093] Characteristic frequency components representing cavitation, leakage, and wear are extracted from the filtered residual signal to generate an internal damage feature vector.

[0094] This embodiment is a further specification of the feature extraction step in Embodiment 1; this step aims to extract weak internal fault signals from high-intensity water flow background noise using a nonlinear hydraulic fluid dynamics model. Generate external interference reference signal This model is based on the fluid continuity equation, and its core dynamic equations are as follows:

[0095]

[0096] in, For effective volume, For bulk modulus, The rate of change of pressure, For input traffic, Leakage coefficient, The disturbance angular frequency; to achieve input capability of the model, external control commands need to be established. With model input flow The mapping relationship is shown in the following formula:

[0097]

[0098] in, This is the flow gain coefficient. For control commands, For oil supply pressure, For symbolic functions, The formula uses absolute values; by introducing a sign function and absolute value operations, it corrects the original model in... To address the domain defects where values ​​within the square root are negative under extreme conditions such as load impact or cavitation reverse flow, the numerical stability of simulation calculations under extreme conditions is ensured; among them, the fluid disturbance parameters The method of obtaining it is:

[0099]

[0100] in, The coefficients are experimental fitting coefficients, which are calibrated based on wind tunnel experiments in this embodiment. The specific calibration process is as follows: A scale of [missing information] is constructed. A scaled-down model of a transparent acrylic valve body was tested in a recirculation wind tunnel; the adjustable wind speed range covered... from to Within a given range, the pulsating velocity of fluid particles at different Reynolds numbers was recorded using PIV (Particle Image Velocimetry) technology, and the amplitude of the equivalent flow pulsation induced by fluid disturbance was calculated. The least squares method was used to analyze the experimental data points. Perform nonlinear regression fitting and assess goodness of fit. To ensure dimensional consistency, the solution is set to... , Let be the Reynolds number; for the computability of the formula, the Reynolds number is... The definition is as follows:

[0101]

[0102] in, The average flow velocity of the fluid. The hydraulic diameter is taken as the inner diameter of the main oil inlet pipe in this embodiment. , Let the kinematic viscosity be taken as... Viscosity value of No. 46 hydraulic oil To solve the problem of average fluid velocity To address the issue of calculating the source of the flow velocity, this embodiment introduces a flow velocity calculation formula based on Bernoulli's principle, establishing its relationship with the upstream and downstream water level difference. Physical constraints:

[0103]

[0104] in, The velocity coefficient is taken as in this embodiment. , Take the acceleration due to gravity , The difference in water level between upstream and downstream is expressed in units of 1. This data is acquired in real time by water level sensors deployed before and after the gate, or directly sent from the flood control dispatch center system interface; based on the above equation, the current control command is input. Water level difference between upstream and downstream Calculated ;

[0105] Temporal state tensor The data is mapped to the frequency domain and an adaptive filter is used; specifically, the Least Mean Square (LMS) algorithm is employed, and its weight update formula is as follows:

[0106]

[0107] in, For weight vectors, The step size factor is set to 0.005. For a moment The filtering error signal, i.e., the difference between the observed signal and the estimated signal, For a moment The input time-series state tensor vector; so-called spectral overlap is mathematically defined as satisfying the condition The set of frequencies, that is, for any frequency point Only when the power spectral density of the external interference reference signal Power spectral density of real-time signal All are greater than the energy threshold When this occurs, it is determined that aliasing interference exists at that frequency point; among which, For power spectral density, Take the energy threshold ;by Using the spectrum as a reference, from Frequency components that overlap with external water flow interference are filtered out, and the frequency domain residual signal is calculated. ;from Extracting characteristic frequency components representing specific faults, including cavitation characteristic energy in the high-frequency band of 2kHz-5kHz. Leakage characteristic energy in the flow noise frequency band and wear characteristic energy at specific meshing frequencies Combine the above components to generate an internal damage feature vector. .

[0108] Example 5:

[0109] The remaining functional lifetime entropy of the current system is determined by calculating the cumulative effect of micro-fatigue damage, including:

[0110] Using a pre-defined probability mapping model, the internal damage feature vector is mapped to the probability of hydraulic oil film rupture and the probability of valve core thermal jamming.

[0111] Calculate the information entropy value at the current moment based on the joint distribution of the probability of hydraulic oil film rupture and the probability of valve core thermal jamming;

[0112] A weighted integral operation is performed on the information entropy values ​​within the historical time window to generate the functional remaining lifetime entropy, which represents the degree of cumulative irreversible damage to the system.

[0113] This embodiment is a further specification of the lifetime prediction step in Embodiment 1; this step introduces entropy theory to quantify the vulnerability of the system; and utilizes a preset probability mapping model, i.e., a pre-trained lifetime prediction network, to convert the internal damage feature vector... Mapped to failure probability; the specific network structure is configured as follows: the first layer LSTM contains The second LSTM layer contains 1 neuron. The third LSTM layer contains [number] neurons. One neuron; to ensure the completeness and constraints of the probability space. The output layer of the network is designed to contain three nodes. The layers correspond to the probability of hydraulic oil film rupture. Probability of valve core thermal seizure and normal operating probability Strictly meet ;

[0114] The network was trained using the Adam optimizer, with an initial learning rate set to... loss function Select mean square error:

[0115]

[0116] in, For batch size, For real labels, This is a predicted value;

[0117] Regarding the preparation of training samples: the data comes from a hydraulic component accelerated life test bench, collecting data throughout the entire life cycle. Sequence; sample labels are defined as normalized remaining lifetimes. ,in, The time of failure The dataset is divided into training, validation, and test sets in a 7:2:1 ratio, representing the current time.

[0118] Output using this model and Based on the joint probability distribution, calculate the instantaneous information entropy of the system at the current moment. The calculation formula is as follows:

[0119]

[0120] in, Represents the natural logarithm. Derived from real-time computing, its physical meaning is the uncertainty of the system state at the current moment, and its unit is nanoseconds; This represents the probability of hydraulic oil film rupture. The probability of valve core thermal seizure. This represents the probability of the system operating normally.

[0121] Considering the cumulative effect of damage, it is necessary to perform cumulative calculations on the information entropy value over the entire lifespan to correct the numerical fluctuations caused by the original fixed-window summation; specifically, a recursive accumulation formula is used to calculate the remaining lifetime entropy. The calculation formula is as follows:

[0122]

[0123] in, The accumulated entropy at the current moment, in units of , The initial value is the accumulated entropy from the previous time step. , The instantaneous information entropy at the current moment, in units of , The sampling interval is expressed in units of 1 / 2. , The damage conversion coefficient is set to a value of [value to be filled in]. The physical meaning is the entropy-induced damage conversion rate per unit time, used as the dimension in the equilibrium formula; the fixed window in the original scheme has been removed here. By limiting and directly calculating the cumulative entropy value over the entire time domain, the degree of cumulative irreversible damage that monotonically increases over time can be accurately characterized; and the functional remaining lifetime entropy can be generated. The physical meaning of this parameter is to characterize the degree of irreversible damage accumulated in the system;

[0124] This embodiment discloses in detail the network topology, hyperparameter setting, and data construction method of the lifetime prediction network, ensuring that those skilled in the art can reproduce this key model; the entropy value calculated by this model can keenly capture the trend of the system transitioning from a steady state to a chaotic state, providing a quantitative decision boundary for subsequent survival-priority control.

[0125] Example 6:

[0126] Based on functional remaining lifetime entropy, task urgency parameters, and safe operation boundary parameters, a survival-priority control trajectory is generated, including:

[0127] If the remaining lifespan entropy of the function is lower than the preset safety threshold, a continuous and rapid activation trajectory will be generated with the goal of maximizing response speed.

[0128] If the remaining functional lifetime entropy is higher than or equal to the preset safety threshold, then a pulse-type micro-motion trajectory or a pressure unloading recovery trajectory is generated with the goal of suppressing the rate of entropy growth.

[0129] Survival-first control trajectory represents the target pressure and displacement sequence of the hydraulic actuator within a future time window.

[0130] This embodiment is a further specification of the decision generation step in Embodiment 1; this step aims to dynamically switch control strategies based on system health, and the system will calculate the remaining functional lifetime entropy in real time. With preset safety threshold Comparison; here The configuration logic is based on historical fault data. Distribution analysis to determine the preset safety threshold The maximum likelihood estimation method is used to calculate... Shape parameters of the distribution With scale parameters The calculation formula is as follows:

[0131]

[0132] in, This is a sample of historical failure times. Given the total number of samples; construct the failure rate function. The calculation formula is as follows:

[0133]

[0134] Calculate the rate of change of failure rate And define the inflection point from the accidental failure period to the wear-out failure period. for First time exceeding the preset acceleration threshold The formula for calculating the time is as follows:

[0135]

[0136] In this embodiment, based on fatigue test data from a large number of devices of the same model, the following settings are made: ,in, Representing hours, the logic for determining this threshold is as follows: Second derivative analysis was performed on the full life cycle data of hydraulic cylinders of the same model to obtain the average curvature abrupt change point in the bathtub curve from the steady period to the wear period. This value is the statistical average value of the acceleration of the failure rate change; this threshold corresponds to the critical point where the curvature of the failure rate curve undergoes a significant abrupt change.

[0137] Query historical database The system entropy value corresponding to a given time is set as a preset safety threshold. The physical basis for this is that when the system entropy exceeds a certain point, even a small disturbance will trigger a nonlinear chain of damage. The quantitative judgment standard and general setting method for this safety threshold are as follows: For any model of hydraulic gate hoist, the corresponding historical full life cycle failure data is extracted and fitted with a Weber distribution. The curvature abrupt change point of its failure rate curve is then calculated as a benchmark for quantitative setting. Below This indicates that the system's health is acceptable, and the system enters task mode, aiming to maximize response speed by generating continuous and rapid activation trajectories to meet the urgent needs of flood discharge; conversely, responding to... Higher than or equal to This indicates that the system is on the verge of failure and switches to survival mode; in this mode, the system aims to suppress the rate of entropy increase. To achieve this, generate pulsed micro-motion trajectories or pressure unloading recovery trajectories; specifically, the entropy growth rate... The calculation formula is as follows:

[0138]

[0139] in, The entropy value at the current moment. The sampling period pulse-type micro-motion trajectory aims to control the valve core to perform high-frequency, small-amplitude inching to disrupt static friction and prevent the valve core from jamming; its trajectory generation function... The calculation formula is as follows:

[0140]

[0141] in, To maintain the displacement via micro-motion, the steady-state position command value of the valve core at the current moment is taken. The pulse amplitude is set to 0.5 mm. The pulse frequency is set to 5Hz; the above parameters and The method is determined based on experiments on the static friction characteristics of the valve core. Specifically, it involves setting a fixed frequency on an offline test bench. Perform from arrive Frequency sweep test, and at the same time The increment value of the step size Record the minimum energy point when the valve core starts; experimental results show that, and At the same time, it can minimize fluid resonance energy while destroying static friction; to ensure that static friction is effectively destroyed without causing excessive system vibration; the pressure unloading recovery trajectory aims to actively reduce system pressure in non-critical operating ranges and rebuild the oil film during the interval; output survival-priority control trajectory. , representing the target pressure and displacement sequence of the hydraulic actuator within a future time window;

[0142] This embodiment achieves adaptive switching of control strategies, prioritizing speed when the system is healthy and protecting equipment when the system is in danger. In particular, the generation of pulse micro-motion and unloading recovery strategies breaks through the experience limitations of human operators, which can significantly extend the effective working time of faulty equipment under extreme conditions and avoid sudden equipment failure caused by blind forced operation.

[0143] Example 7:

[0144] The drive control signal of the hydraulic actuator is adjusted through a multimodal actuator, including:

[0145] The survival-first control trajectory is decomposed into proportional valve opening commands and pump station speed commands;

[0146] Monitor the real-time pressure response during execution. If the real-time pressure response exceeds the safe operating boundary parameters, force an emergency unloading control signal to be output.

[0147] The hydraulic gate hoist is driven to operate according to the proportional valve opening command and the pump station speed command.

[0148] This embodiment is a further specification of the execution control steps in Embodiment 1; this step adopts a two-layer architecture of AI decision-making and rule-based fallback; and implements a survival-first control trajectory. , i.e., target pressure and displacement This is decomposed into specific actuator instructions; this decomposition process employs a decoupled PID control strategy, and the calculation formula is as follows:

[0149]

[0150]

[0151] in, This is the control voltage for the proportional valve, in units of... , This is a pump station speed command, in units of... , These are real-time displacement monitoring values, in units of... , This is a real-time pressure monitoring value, in units of... , To control the gain; in this embodiment, the control gain parameter is tuned using the Ziegler-Nichols method and determined through on-site debugging. The specific on-site debugging method is the critical proportional gain method, gradually increasing the proportional gain until the system exhibits constant-amplitude oscillations to record the critical gain and period. The specific value is then fine-tuned using overshoot less than 5% and settling time minimized as debugging indicators. The specific values ​​are: The control voltage of the proportional valve controlling the flow rate and direction is obtained through the above calculations. And the pump speed command for controlling the source pressure of the control system. Among them, the proportional valve controls the voltage. It has a linear mapping relationship with the proportional valve opening command and is used to drive the proportional valve spool to achieve the target opening degree; during the driving execution process, the system continuously monitors the real-time pressure response. ; Response to determination Exceeding safe operating boundary parameters For example, if a pressure surge is detected, the system will ignore the output of the upper-level model and force an emergency unloading control signal; this signal directly controls the overflow valve to fully open, allowing the high-pressure oil to flow back to the tank rapidly; if the emergency control is not triggered, according to... and Coordinated drive of hydraulic gate hoist operation;

[0152] This embodiment achieves pump-valve collaborative optimization through instruction decomposition, thereby improving the control's energy efficiency ratio. At the same time, the underlying emergency unloading rule mechanism ensures the absolute safety baseline of the physical system, preventing catastrophic consequences that might result from exploratory strategies by deep learning models, and realizing an organic combination of intelligent control and intrinsic safety.

[0153] Example 8:

[0154] Please see Figure 2 A remote diagnostic control system for hydraulic gate hoists, comprising:

[0155] The sensing module is configured to collect data from a multi-dimensional sensor group at key nodes of the hydraulic gate hoist system; the initialization module is configured to set safe operation boundary parameters and task urgency parameters, and load a deep reinforcement learning model based on physical information.

[0156] The diagnostic control module is equipped with a high-speed communication bus interface to realize data interaction and serial processing timing between the following units, including: a data preprocessing unit, configured to collect real-time monitoring data and perform timing reconstruction and missing value imputation on the real-time monitoring data to construct a standardized timing state tensor;

[0157] The feature decoupling unit is configured to combine the time-series state tensor with the nonlinear hydraulic fluid dynamics model, and extract the internal damage feature vector of the hydraulic component through signal feature decoupling analysis;

[0158] The entropy calculation unit is configured to input the internal damage feature vector into the life prediction network and determine the functional remaining life entropy of the current system through the accumulation calculation of micro fatigue damage.

[0159] The strategy generation unit is configured to input the functional remaining lifetime entropy, task urgency parameters and safe operation boundary parameters into the physical information-based deep reinforcement learning model to generate a survival-priority control trajectory.

[0160] The execution control unit is configured to adjust the output of the hydraulic actuator drive control signal based on the survival-priority control trajectory and real-time monitoring data through the multi-modal actuator.

[0161] This embodiment provides a specific hardware and software architecture for a remote diagnostic control system for a hydraulic gate hoist. This system serves as the physical carrier of the aforementioned method. The sensing module consists of a pressure pulsation sensor, a vibration acceleration sensor, and an oil monitoring sensor, configured to collect multi-dimensional physical data. The initialization module is configured to store the design parameters of the hydraulic system and load pre-trained... - Model weights; the diagnostic control module, as the core computing unit, contains multiple dedicated processing units; among them, the data preprocessing unit is built-in. - Algorithm chips perform data cleaning and reconstruction;

[0162] The feature decoupling unit runs a nonlinear hydraulic dynamics simulation program to separate internal and external signals in real time; the entropy calculation unit refreshes the system's functional remaining life entropy in real time based on the cumulative damage model; the strategy generation unit runs a deep reinforcement learning inference engine to dynamically plan the control trajectory based on the current entropy value and task urgency; the execution control unit includes a PLC controller and a drive amplifier, which is responsible for converting the digital trajectory into analog voltage signals to drive the valve group and pump station, and has a hard-wired emergency unloading circuit.

[0163] This embodiment integrates complex physical sensing, life prediction and intelligent decision-making algorithms through modular design; the close collaboration between the modules realizes the full life cycle health management of large hydraulic gate hoists. Especially when facing extreme flood discharge tasks, the system can provide highly fault-tolerant control capabilities, ensuring the safe operation of water conservancy facilities.

[0164] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A hydraulic hoist remote diagnosis control method, characterized by, include: Multi-dimensional sensor arrays are deployed at key nodes of the hydraulic gate hoist system, and a deep reinforcement learning model based on physical information is constructed. Initialize the system's safe operating boundary parameters and task urgency parameters; In response to receiving a remote flood discharge task command, an adaptive diagnostic control process is triggered, including: Step 1: Collect real-time monitoring data from the multi-dimensional sensor array, perform temporal reconstruction and missing value imputation on the real-time monitoring data, and construct a standardized temporal state tensor; Step 2: Input the time-series state tensor into the nonlinear hydraulic fluid dynamics model, and extract the internal damage feature vector characterizing the state of the hydraulic components through signal feature decoupling analysis; Step 3: Input the internal damage feature vector into the pre-trained lifetime prediction network, and determine the functional remaining lifetime entropy of the current system through the accumulation calculation of micro fatigue damage. Step 4: Input the functional remaining lifetime entropy, task urgency parameters, and safe operation boundary parameters into the physical information-based deep reinforcement learning model to generate a survival-priority control trajectory; Step 5: Based on the survival-first control trajectory and real-time monitoring data, adjust the multi-modal actuator to output the drive control signal of the hydraulic actuator.

2. The hydraulic hoist remote diagnosis control method of claim 1, wherein A multi-dimensional sensor array, including: A pressure pulsation sensor is installed between the pump outlet and the main control valve in the hydraulic pipeline; Vibration acceleration sensors installed at the gate connection points; Sensors for monitoring oil contamination and temperature are installed inside the hydraulic oil tank.

3. The hydraulic hoist remote diagnosis control method of claim 2, wherein Real-time monitoring data will be processed through time-series reconstruction and missing value imputation, including: Identify communication packet loss segments and abnormal outliers in real-time monitoring data; Based on the spatiotemporal correlation of historical operational data, generative adversarial networks are used to interpolate data for lost communication packets. The interpolated data is denoised and standardized to generate a time-series state tensor that includes pressure, vibration, and oil state.

4. The hydraulic hoist remote diagnosis control method of claim 3, wherein, Through signal feature decoupling analysis, the internal damage feature vector of hydraulic components is extracted, including: Using a nonlinear hydraulic fluid dynamics model, an external disturbance reference signal under a preset flow-induced vibration is generated; The temporal state tensor is mapped to the frequency domain space, and the components that overlap with the spectrum of the external interference reference signal are filtered out by an adaptive filtering algorithm. Characteristic frequency components representing cavitation, leakage, and wear are extracted from the filtered residual signal to generate an internal damage feature vector.

5. The hydraulic hoist remote diagnostic control method of claim 4, wherein, The remaining functional lifetime entropy of the current system is determined by calculating the cumulative micro-fatigue damage, including: Using a pre-defined probability mapping model, the internal damage feature vector is mapped to the probability of hydraulic oil film rupture and the probability of valve core thermal jamming. Calculate the information entropy value at the current moment based on the joint distribution of the probability of hydraulic oil film rupture and the probability of valve core thermal jamming; A weighted integral operation is performed on the information entropy values ​​within the historical time window to generate the functional remaining lifetime entropy, which represents the degree of cumulative irreversible damage to the system.

6. The hydraulic hoist remote diagnostic control method of claim 5, wherein, Based on functional remaining lifetime entropy, task urgency parameters, and safe operation boundary parameters, a survival-priority control trajectory is generated, including: If the remaining lifespan entropy of the function is lower than the preset safety threshold, a continuous and rapid activation trajectory will be generated with the goal of maximizing response speed. If the remaining functional lifetime entropy is higher than or equal to the preset safety threshold, then a pulse-type micro-motion trajectory or a pressure unloading recovery trajectory is generated with the goal of suppressing the rate of entropy growth. Survival-first control trajectory represents the target pressure and displacement sequence of the hydraulic actuator within a future time window.

7. The hydraulic hoist remote diagnostic control method of claim 6, wherein, The drive control signal of the hydraulic actuator is adjusted through a multimodal actuator, including: The survival-first control trajectory is decomposed into proportional valve opening commands and pump station speed commands; Monitor the real-time pressure response during execution. If the real-time pressure response exceeds the safe operating boundary parameters, force an emergency unloading control signal to be output. The hydraulic gate hoist is driven to operate according to the proportional valve opening command and the pump station speed command.

8. A hydraulic hoist remote diagnosis control system applied to the hydraulic hoist remote diagnosis control method of any one of claims 1-7, characterized in that, include: The sensing module is configured to collect data from a multi-dimensional sensor array at key nodes of the hydraulic gate hoist system. The initialization module is configured to set safe operation boundary parameters and task urgency parameters, and load a deep reinforcement learning model based on physical information. The diagnostic control module includes: The data preprocessing unit is configured to collect real-time monitoring data and perform temporal reconstruction and missing value imputation on the real-time monitoring data to construct a standardized temporal state tensor. The feature decoupling unit is configured to combine the time-series state tensor with the nonlinear hydraulic fluid dynamics model, and extract the internal damage feature vector of the hydraulic component through signal feature decoupling analysis; The entropy calculation unit is configured to input the internal damage feature vector into the life prediction network and determine the functional remaining life entropy of the current system through the accumulation calculation of micro fatigue damage. The strategy generation unit is configured to input the functional remaining lifetime entropy, task urgency parameters and safe operation boundary parameters into the physical information-based deep reinforcement learning model to generate a survival-priority control trajectory. The execution control unit is configured to adjust the output of the hydraulic actuator drive control signal based on the survival-priority control trajectory and real-time monitoring data through the multi-modal actuator.

Images