A method and system for real-time monitoring and control of parameters in a turbine spraying process

Abstract

The present application relates to the field of spraying control technology, and discloses a method and system for real-time monitoring and control of parameters in a turbine spraying process. The method comprises the following steps: collecting multimodal data in the spraying process and constructing a four-dimensional data structure; performing tensor decomposition to extract core feature tensors and factor matrices; constructing a reinforcement game strategy model and outputting a process parameter control strategy; introducing Lyapunov optimization to generate a control input; combining the control strategy with the control input to generate a spraying execution action; and performing feedforward compensation based on a disturbance observer to form a final control instruction. The present invention adopts multimodal process data fusion and tensor decomposition technology to construct a four-dimensional data tensor and extract the core features of the spraying process. Through this method, the system can more comprehensively capture the dynamic changes in the spraying process and provide an accurate process parameter control strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of spraying control, and in particular to a method and system for real-time monitoring and control of parameters in a turbine spraying process. Background Art

[0002] In modern manufacturing, the spray coating process for hydraulic turbines is crucial to the performance and service life of the equipment. The quality of the spray coating directly impacts the equipment's corrosion and wear resistance. With the continuous advancement of spray coating technology, precise control of the spray coating process has become increasingly important. However, existing technologies often face challenges in adapting to multiple changes in complex spray environments and are unable to achieve efficient and stable spray control in dynamic environments.

[0003] Current spray control systems are mostly driven by single-sensor data, using pressure sensors, temperature sensors, or flow meters to monitor key parameters in the spray process. These systems are capable of adjusting individual factors in the spray process, such as spray speed and coating quality. However, they typically rely on traditional closed-loop feedback control mechanisms and focus on optimizing a single objective, such as improving spray quality or reducing energy consumption. Through feedback regulation and simple algorithms, they can improve spray accuracy or efficiency to a certain extent, but their overall control capabilities and adaptability are relatively limited.

[0004] While existing technologies offer certain advantages, they still face numerous limitations in terms of multimodal data integration and dynamic disturbance compensation. First, traditional systems lack data acquisition capabilities and often fail to fully reflect the complex interactions of multiple factors during the spraying process, resulting in inaccurate control decisions. Second, existing technologies often focus solely on optimizing a single objective, failing to balance multiple process goals, such as spray quality and energy consumption. This results in low spraying efficiency and high energy consumption. Furthermore, traditional feedback control mechanisms react slowly to external environmental disturbances (such as temperature changes and humidity fluctuations), lacking real-time adjustment capabilities and easily leading to fluctuations in spray quality. Therefore, those skilled in the art have proposed a method and system for real-time monitoring and control of turbine spraying process parameters to address these issues. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a real-time monitoring and control method and system for turbine spraying process parameters, which solves the problems in the existing technology of insufficient spraying process control accuracy, inability to respond to multi-variable complex disturbances in real time, and inability to take into account multi-objective optimization.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for real-time monitoring and control of parameters during a turbine spraying process, comprising the following steps:

[0007] S1. Collecting multimodal process data during a spraying process, and constructing a four-dimensional data structure based on the multimodal process data;

[0008] S2. Based on the constructed four-dimensional data structure, a tensor decomposition operation is performed to extract a core feature tensor and a factor matrix of each dimension used to reflect the dynamic state of the spraying process;

[0009] S3. Using the core feature tensor and factor matrix as state input, constructing a reinforcement game strategy model, and outputting a corresponding spraying process parameter control strategy based on the reinforcement game strategy model;

[0010] S4. Based on the output results of the control strategy, a Lyapunov optimization method is introduced to conduct deviation state evaluation by constructing a virtual queue model for describing the stability of the spraying process, and a control input for regulation is generated according to the evaluation results;

[0011] S5. Based on the control strategy and the control input, the generation of the spraying execution action is completed, specifically including the robot arm path planning and the generation of the pulse control signal of the spraying valve;

[0012] S6. Based on the environmental disturbance parameters in the multimodal process data, a disturbance observer is used to generate a disturbance estimate, and the disturbance estimate is synthesized with the above-mentioned control strategy result through feedforward compensation to form a control instruction for driving the spray execution end.

[0013] Preferably, in step S1, the multimodal process data includes:

[0014] Spraying trajectory data is obtained through lidar and visual encoder;

[0015] Material rheological property data, including pressure, viscosity, and shear rate, are obtained through pressure sensors and flow meters;

[0016] Thermophysical state data, including spray surface temperature and heat flux, are obtained by infrared thermal imaging camera;

[0017] Environmental disturbance data, including air humidity, wind speed, and workshop temperature, is obtained through environmental sensors;

[0018] Constructing a four-dimensional data tensor from multimodal process data based on time series ;in: is the dimension of the spray trajectory data, is the dimension of the material rheological data, is the dimension of thermophysical data, is the dimension of the environmental perturbation data.

[0019] Preferably, in step S2, the tensor decomposition operation includes:

[0020] Perform normalization and denoising preprocessing on the constructed four-dimensional data structure;

[0021] Extract the core feature tensor and the corresponding dimensional factor matrix based on the tensor decomposition algorithm Tucker;

[0022] Use the sliding time window mechanism to incrementally update newly collected data to improve dynamic response capabilities;

[0023] The decomposition results are used to represent the main dynamic modes of the spraying process status and serve as input for subsequent strategies.

[0024] Preferably, the tensor decomposition operation satisfies the following mathematical expression:

[0025] ;

[0026] in: is the original four-dimensional data tensor, representing the multimodal spraying process data; is the decomposed core tensor, which represents the main dynamic characteristics of the spraying process; is the factor matrix of the spray trajectory data dimension; is the factor matrix of the material rheological property data dimension; is the factor matrix of the thermophysical state data dimension; is the factor matrix of the environmental disturbance data dimension; For tensors and matrices in The multiplication operation in the mode, that is, along the Dimensions are linearly mapped.

[0027] Preferably, in step S3, constructing the enhanced game strategy model includes:

[0028] Design a dual-agent game framework with process quality and energy consumption as optimization objectives respectively;

[0029] Use multi-round game processes to generate strategic game trajectories and realize the search for equilibrium solutions between different control strategies;

[0030] The extracted feature tensor and factor matrix are used as the policy state space;

[0031] The action space includes the process parameters of spraying speed, nozzle temperature and particle size distribution adjustment;

[0032] The reward function construction follows the comprehensive weighting scheme of indicators.

[0033] Preferably, in step S4, constructing a virtual queue model for describing the stability of the spraying process includes:

[0034] Define the control parameter drift as the basis for virtual queue update;

[0035] Set the threshold term and control penalty function, and combine the Lyapunov function to determine the deviation stability boundary;

[0036] Establish system stability indicators based on spray pressure changes and material deposition rate;

[0037] According to the queue status feedback control input generation rules, the output rate is dynamically adjusted.

[0038] Preferably, the Lyapunov stability control satisfies the following optimization function:

[0039] ;

[0040] in: For the current moment Control inputs include spray speed, path correction, and nozzle temperature; is the drift of the virtual queue, which measures the change of queue status between consecutive moments; is a positive scalar, called the drift-penalty balance factor, which is used to adjust the weight relationship between control performance and queue stability; is the expected value of the loss function, which represents the core tensor The stability or error of the current state is measured.

[0041] Preferably, in step S5, the generation of the spray execution action is completed including:

[0042] According to the control strategy optimization output, a set of spraying path points of the robot arm is generated;

[0043] Use spline interpolation for trajectory smoothing and acceleration limit control;

[0044] Convert the generated path into a servo control instruction and combine it with a pulse modulation signal to drive the spraying actuator;

[0045] The spray valve control signal adopts a multi-level modulation method, and the frequency and switch duty cycle are dynamically adjusted according to the pressure demand.

[0046] Preferably, in step S6, the feedforward compensation synthesis of the disturbance estimation amount and the above-mentioned control strategy result includes:

[0047] Construct a disturbance observer to model the disturbance sources of temperature fluctuations and sudden changes in air velocity from the external environment;

[0048] Use a combination of exponential filtering and Kalman filtering to improve the accuracy of disturbance estimation;

[0049] Use disturbance estimators for feedforward compensation, including gain adjustment, delay correction, and error fitting;

[0050] The final compensation control instruction is used to drive the execution end in real time to maintain process stability.

[0051] A real-time monitoring and control system for parameters of a turbine spraying process, comprising:

[0052] Data acquisition module, used to collect multimodal data during the spraying process, including trajectory parameters, rheological parameters, thermal physical state and environmental disturbance parameters;

[0053] Tensor construction module, used to construct four-dimensional data structures and perform tensor decomposition to extract core feature tensors and factor matrices;

[0054] Strategy decision module, used to build a control strategy model based on game mechanism and output process control strategy;

[0055] Stability analysis module, which is used to construct a virtual platoon and generate stable control inputs based on the Lyapunov optimization method;

[0056] Execution control module, used to plan the robot arm path and generate spray valve pulse signals based on the strategy results;

[0057] Disturbance compensation module, used to estimate environmental disturbances and perform control feedforward compensation;

[0058] The central control unit is used to coordinate the operation of the above modules and output control instructions in real time to drive the spraying device to complete the spraying operation.

[0059] The present invention provides a method and system for real-time monitoring and control of parameters during the spraying process of a turbine. This method has the following beneficial effects:

[0060] 1. The present invention adopts multimodal process data fusion and tensor decomposition technology to construct a four-dimensional data tensor and extract the core features of the spraying process. Through this method, the system can more comprehensively capture the dynamic changes in the spraying process and provide accurate process parameter control strategies. Compared with the control system driven by single sensor data in the existing technology, the present invention solves the problems of incomplete data collection and missing information in traditional technology, and improves the stability and accuracy of the spraying process.

[0061] 2. The present invention introduces a multi-agent strategy optimization scheme based on reinforcement game, combines process quality and energy consumption control goals, and finds the optimal solution through game algorithm. This innovative scheme ensures that the spraying process can achieve high-quality coatings while minimizing resource consumption. Compared with the existing control methods that usually focus on single-target optimization, the dual optimization strategy of the present invention solves the defect that traditional methods can only focus on one dimension, greatly improving the economy and efficiency of the process.

[0062] 3. Under the Lyapunov optimization framework, the present invention constructs a virtual queue model to adjust the stability of the spraying process. By optimizing the control input, the system can dynamically respond to disturbances and deviations in the spraying process and maintain the spraying quality and stability. Compared with the existing solutions that are difficult to handle complex disturbances, the present invention effectively compensates for the shortcomings of traditional technology in dealing with external disturbances and system instability through the introduction of the virtual queue model, thereby improving the robustness of the system.

[0063] 4. The present invention combines disturbance estimation with feedforward compensation mechanism, generates disturbance estimation in real time through disturbance observer, and performs feedforward compensation in combination with control strategy. This method ensures that even if there are changes in external environment or equipment errors during the spraying process, the control system can make rapid adjustments to maintain spraying accuracy and quality. Compared with traditional solutions in the prior art that often rely only on feedback control, the present invention effectively reduces the impact of disturbances on the spraying process through feedforward compensation technology, thereby improving the system's adaptability and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 Schematic diagram of the method flow of the present invention;

[0065] Figure 2 Schematic diagram of the system architecture of the present invention. DETAILED DESCRIPTION

[0066] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0067] Please see the attached Figure 1 The embodiment of the present invention provides a method and system for real-time monitoring and control of parameters of a turbine spraying process, comprising the following steps:

[0068] S1. Collect multimodal process data during the spraying process and construct a four-dimensional data structure based on the multimodal process data;

[0069] Specifically, in this embodiment, a comprehensive process data model is first constructed by collecting multimodal process data from the spraying process. This step forms the foundation of the entire spraying process parameter monitoring and control system and provides data support for subsequent processing and decision-making. By monitoring various key parameters in real time during the spraying process, rich input information can be provided for subsequent tensor decomposition, process parameter control, and dynamic adjustment, ensuring the accuracy and stability of the spraying process.

[0070] In one possible implementation, multimodal process data includes, but is not limited to, spray trajectory data, material rheological properties data, thermophysical state data, and environmental disturbance data. Each type of data is collected in real time by different sensor devices to ensure data accuracy and reliability.

[0071] Specifically, the spray trajectory data can be collected through lidar and visual encoders. LiDAR can accurately measure the real-time position and motion trajectory of the spraying equipment in space, while the visual encoder can provide more detailed information about the spraying path. Material rheological property data, including pressure, viscosity, shear rate and other information, are usually collected through pressure sensors and flow meters. Infrared thermal imagers are used to collect thermal physical state data, which can monitor the temperature distribution and heat flux information of the spray surface in real time. Environmental disturbance data, such as air humidity, wind speed, workshop temperature, etc., are obtained in real time through environmental sensors.

[0072] Based on the above multimodal process data, we further use the time series as the basis to construct a four-dimensional data tensor In this tensor, the four dimensions represent the following four main parameters:

[0073] : The dimension of the spraying trajectory data, which represents the position information of the spraying equipment in space.

[0074] : The dimension of material rheological data, which represents the physical properties of the material during the spraying process.

[0075] : The dimension of thermophysical data, representing the change of heat distribution during the spraying process.

[0076] : The dimension of environmental disturbance data, which represents the impact of the external environment on the spraying process.

[0077] In this embodiment, these data are collected at high frequency and synchronously processed to form a complete, high-dimensional, multimodal dataset, facilitating subsequent analysis and processing. The construction of the four-dimensional data tensor is based on a time series data fusion strategy, ensuring that temporal changes accurately reflect the various process data, further providing strong support for subsequent tensor decomposition operations.

[0078] Generally, to improve data processing accuracy, data is normalized and denoised to reduce interference caused by sensor errors or external environmental disturbances. For example, when collecting spray trajectory data, the output data of the LiDAR and visual encoder may be affected by other objects or external interference. Appropriate data preprocessing can significantly improve the accuracy of subsequent analysis.

[0079] As an option, the collection of various data during the spraying process can be set to real-time or periodic collection. Specifically, real-time collection is suitable for scenarios where instantaneous changes in the spraying process need to be accurately captured, while periodic collection is suitable for scenarios where parameters change more slowly during the spraying process and the data update frequency is infrequent.

[0080] In one possible implementation, using time series data to construct a four-dimensional data tensor ensures that the spraying process at each point in time can be accurately captured and that changes in various parameters can be processed synchronously. This data will provide strong support for subsequent tensor decomposition and the construction of enhanced game strategies.

[0081] S2. Based on the constructed four-dimensional data structure, perform tensor decomposition operations to extract the core feature tensor and factor matrices of each dimension used to reflect the dynamic state of the spraying process;

[0082] Specifically, in step S1, multimodal data acquisition and four-dimensional data structure construction yield a multidimensional dataset describing the spraying process. Building on this foundation, the core task of step S2 is to perform tensor decomposition on this data to extract the core features of the spraying process. This step not only facilitates dimensionality reduction and noise removal, but also maps the high-dimensional data into a simpler, lower-dimensional representation that is more amenable to subsequent analysis.

[0083] In this embodiment, first, the four-dimensional data tensor constructed in step S1 is To ensure the accuracy and validity of the subsequent decomposition results, the data tensor is first normalized. This process helps eliminate dimensional differences between modal data, allowing all data to be compared and analyzed at the same scale.

[0084] After data preprocessing, the data tensor is decomposed by Tucker tensor decomposition algorithm. Tucker decomposition is a common high-dimensional data decomposition method that can decompose a high-dimensional tensor into the product of multiple low-dimensional factor matrices and can effectively represent the multimodal characteristics of the data. In this embodiment, Tucker decomposition is used to extract the core feature tensor from the high-dimensional four-dimensional data tensor. And the factor matrix of each dimension , , , these factor matrices represent the main dynamic patterns of the data.

[0085] Specifically, the mathematical expression of Tucker decomposition is:

[0086] ;

[0087] in: is the original four-dimensional data tensor, representing the multimodal spraying process data; is the decomposed core tensor, which represents the main dynamic characteristics of the spraying process; is the factor matrix of the spray trajectory data dimension; is the factor matrix of the material rheological property data dimension; is the factor matrix of the thermophysical state data dimension; is the factor matrix of the environmental disturbance data dimension; For tensors and matrices in The multiplication operation in the mode, that is, along the Dimensions are linearly mapped.

[0088] During this tensor decomposition process, They represent the dimensionality reduction ranks of the spray trajectory data dimension, material rheological property data dimension, thermophysical state data dimension, and environmental disturbance data dimension in the tensor decomposition process, respectively. , in order to achieve effective data compression and feature extraction; The original data dimensions corresponding to each dimension of the multimodal process data are: Represents the original dimension of the spray trajectory data , corresponds to the location information dimension of the spraying equipment in space; The original dimension representing the material rheological property data , which corresponds to the parameter dimension of the physical properties of the material during the spraying process; Represents the original dimension of the thermophysical state data ,parameter dimension corresponding to the change of heat distribution during spraying; Represents the original dimension of environmental perturbation data , corresponding to the parameter dimensions of external environmental influencing factors.

[0089] Through this decomposition, the original four-dimensional data tensor can be effectively mapped to a more compact representation, the core tensor and the factor matrix The key dynamic patterns and correlations contained in different modal data are revealed respectively.

[0090] Generally speaking, the Tucker decomposition method is suitable for processing data with high-dimensional structures and complex associations. It can extract the main dynamic characteristics of the spraying process and map them into a low-dimensional space to facilitate subsequent analysis and modeling. In this way, the various change patterns in the spraying process can be effectively captured, avoiding the redundancy of the original data and providing more representative features for subsequent control strategies.

[0091] Alternatively, in some embodiments, tensor decomposition can employ other types of tensor decomposition algorithms, such as CP decomposition, depending on specific application requirements. However, Tucker decomposition is selected as the primary decomposition method in the present invention because it can reduce computational complexity while preserving the core features of the data.

[0092] In another implementation, to further improve data processing efficiency, the present invention introduces a sliding window mechanism during tensor decomposition to incrementally update newly acquired data. This approach ensures that the system can reflect dynamic changes in the spraying process in real time, further improving the system's response speed and sensitivity.

[0093] S3. Using the core feature tensor and factor matrix as state input, a reinforcement game strategy model is constructed, and the corresponding spraying process parameter control strategy is output based on the reinforcement game strategy model;

[0094] Specifically, in step S2, the core feature tensor of the spraying process is obtained through the Tucker tensor decomposition operation And the factor matrix of each mode , ,These data provide key input information for building a reinforcement game strategy model for the spraying process.,Based on this, the goal of step S3 is to use the extracted core features to build a reinforcement game strategy model,and output a control strategy suitable for the spraying process.

[0095] In this example, the reinforcement game strategy model is a two-agent game model, in which the two agents represent the optimization objectives of spraying process quality and energy consumption. These two objectives are in conflict during the spraying process, so using a game theory model can effectively find a balance that achieves the optimal compromise between spraying process quality and energy consumption.

[0096] Specifically, the construction process of the reinforcement game strategy model is as follows:

[0097] First, the extracted core feature tensor and the factor matrix , The input state space of the game model reflects the multimodal dynamic characteristics of the spraying process and can provide real-time updated data support for strategic decision-making.

[0098] The action space of each intelligent agent includes key control parameters in the spraying process, such as spraying speed, nozzle temperature, and particle size distribution adjustment. These parameters are key to process quality and energy consumption control.

[0099] Each agent continuously adjusts its strategy through the learning game process to achieve the optimal process control goal. During the game, the agents search for equilibrium solutions between different control strategies through multiple rounds of game interaction, ensuring that the spraying process can operate stably under different control strategies.

[0100] The game process adopts the classic reinforcement learning method, in which the goal of each agent is to optimize its control strategy through the reward function. Specifically, the reward function and They correspond to the spraying quality and energy consumption control targets respectively, and their expressions are as follows:

[0101] ;

[0102] ;

[0103] in: It is an indicator of spraying quality, usually indicating the uniformity and adhesion of the coating; It is a resource consumption indicator, reflecting the energy consumption during the spraying process; is the spraying pressure gradient, which indicates the unevenness of pressure distribution during the spraying process; is the trace of the core tensor, which represents the overall state of the spraying process; , , , is the corresponding weight coefficient, which is used to balance the weight relationship between quality and energy consumption.

[0104] In this game model, each agent selects a corresponding control action based on its current state and strategy, and updates its strategy based on feedback from the reward function. Over multiple iterations, the game gradually converges to an equilibrium state, where the spraying process achieves an optimal balance between quality and energy consumption.

[0105] In general, the reinforcement game model can optimize key control parameters in the spraying process through the interaction of multiple game processes, achieving dual optimization of process quality and energy consumption. This game-based control strategy can not only effectively improve the quality of the spraying process, but also reduce energy consumption, and has high practical application value.

[0106] Alternatively, in some embodiments, the enhanced game model can also incorporate the influence of external disturbances, such as device aging or environmental changes, to enhance the model's robustness and adaptability. This extension allows the system to maintain good control performance in the face of uncertainty and external disturbances.

[0107] In another possible implementation, in order to further improve the accuracy of the strategy, the state space and action space in the game process can be refined and hierarchically designed, adding more process parameters and adjustment variables to more accurately adjust the spraying process.

[0108] S4. Based on the output of the control strategy, the Lyapunov optimization method is introduced to construct a virtual queue model to describe the stability of the spraying process, perform deviation state evaluation, and generate control inputs for regulation based on the evaluation results;

[0109] Specifically, in step S3, the control strategies for the spraying process were output through the reinforcement game strategy model. These control strategies provide a preliminary optimization solution for the spraying process. However, in actual application, the system requires further adjustments to address potential stability issues. The main task of step S4 is to introduce a virtual queue model using the Lyapunov optimization method to analyze and control the stability of the system. Specifically, this step ensures that the spraying process can maintain stability and precision in a dynamic environment by evaluating the deviation state and adjusting the control input.

[0110] In this example, a virtual queue model is first constructed to describe the stability of the spraying process. The core concept of this virtual queue model is to treat the deviation of the spraying system as an accumulated amount in the queue and dynamically adjust it using the Lyapunov optimization method. In this way, unstable factors in the system can be effectively suppressed while maintaining process quality, ensuring the smooth progress of the spraying process.

[0111] Specifically, the virtual queue model is constructed based on the update of control parameter drift. For a spraying system, control parameter drift can cause instability in the spraying process. Therefore, drift is introduced to simulate dynamic changes in the control system. This drift serves as a key variable in the virtual queue model, helping the system to provide feedback and adjust to unstable factors.

[0112] The update rules for the virtual queue are as follows:

[0113] ;

[0114] in: is the current state of the virtual queue, indicating the accumulated deviation of the system; For the core tensor in the control strategy relevant adjustment factors; is the spraying trajectory factor matrix Related regulatory factors; Represents the current time step.

[0115] Through this formula, the system can monitor and adjust the status of the virtual queue in real time to avoid excessive accumulation of deviations, thereby achieving stable control of the spraying process.

[0116] Furthermore, to ensure the spraying process remains stable in a dynamic environment, the Lyapunov optimization method was incorporated into the virtual queue model. The Lyapunov optimization method is a mathematical optimization technique based on system stability and is widely used in the regulation of dynamic systems. In this embodiment, the Lyapunov optimization method uses a drift-penalty balance factor to adjust the relationship between control performance and queue stability.

[0117] Specifically, the optimization objective function of Lyapunov stability control is as follows:

[0118] ;

[0119] in: For the current moment Control inputs include spraying speed, path correction, nozzle temperature, etc. is the drift of the virtual queue, which measures the change of queue status between consecutive moments; is a positive scalar, called the drift-penalty balance factor, which is used to adjust the weight relationship between control performance and queue stability; is the expected value of the loss function, which represents the core tensor The stability or error of the current state is measured.

[0120] By optimizing the objective function, the system can minimize queue drift and optimize the stability of the core tensor based on the control input. This optimization method effectively ensures that the spraying process can maintain a stable operation state in the face of external disturbances or dynamic changes.

[0121] Generally speaking, the Lyapunov optimization method can effectively avoid instabilities during the spraying process by adjusting the control inputs through real-time feedback when system deviations occur. In this way, the control parameters of the spraying process can be continuously optimized to achieve global stability of the system.

[0122] Alternatively, in some embodiments, the Lyapunov optimization method can be combined with other optimization algorithms (such as PID control and fuzzy control) to further enhance the system's adaptability and robustness. Specifically, by combining this with other control strategies, the system can achieve more efficient and stable regulation in complex environments.

[0123] In another possible implementation, the design of the virtual queue model can be customized according to different spraying processes to adapt to the actual needs of different production environments. This flexibility ensures the wide applicability of the present invention and can be adjusted according to specific circumstances.

[0124] S5. Based on the control strategy and control input, the generation of the spraying execution action is completed, specifically including the robot arm path planning and the generation of the pulse control signal of the spray valve;

[0125] Specifically, in step S4, the system uses the Lyapunov optimization method to regulate the stability of the spraying process, ensuring the adaptability and stability of the control input. Building on this, step S5 aims to generate the actual spraying execution action based on the control strategy and stability adjustment results. This involves planning the robot arm path and generating pulse control signals for the spray valve. The key to this step is converting virtual control signals into physical execution actions, ensuring the precise execution of the spraying process.

[0126] In this embodiment, the control strategy output obtained in the previous steps is used as input, and the resulting adjustments to the control input are combined to generate a path plan for the robotic arm. Spray path planning precisely designs the robotic arm's motion trajectory by considering the real-time demands and environmental constraints of the spraying process. During this process, the robotic arm must perform high-precision spraying operations along a predetermined path, tailored to the size and shape of the target area.

[0127] Specifically, the path planning uses B-spline interpolation for smoothing. This method can effectively avoid unnecessary vibration or acceleration peaks during the movement of the robot arm, thereby ensuring the stability and uniformity of the spraying process. B-spline interpolation is a commonly used smooth path planning technology that defines multiple control points. To construct a smooth spray path, the specific expression is:

[0128] ;

[0129] in: is the path curve, which represents the spraying trajectory of the robot arm; For the indivual Order B-spline basis function; The factor matrix The extracted control points specifically reflect the geometric shape of the spraying trajectory.

[0130] By using the B-spline interpolation method, path smoothing can effectively reduce the problem of uneven movement of the robot arm caused by path corners, thereby improving the spraying quality and reducing energy consumption during the spraying process.

[0131] In addition, the generation of the pulse control signal for the spray valve also depends on the output of the control strategy. To achieve precise spraying, the switching frequency and duty cycle of the spray valve need to be dynamically adjusted according to process parameters such as spray pressure, nozzle temperature, and paint flow rate. The spray control signal is usually generated using pulse width modulation (PWM). The specific model is:

[0132] ;

[0133] in: Control voltage for spray valve; and are the modulation amplitudes, respectively; The spraying frequency determines the switch cycle of the spray valve; is a pulse function, which represents the switch signal of the spray valve; is the pulse interval time; is the number of pulses.

[0134] This control method can accurately adjust the on / off state of the spray valve, thereby controlling the spraying amount of paint and ensuring the stability of the spraying process and the coating quality.

[0135] Generally speaking, the coordination of spray valve control signal generation and robot arm path planning is crucial, as they together determine coating uniformity and spray efficiency. Through this precise control, the system can achieve full control of the spray process, thereby improving process accuracy and production efficiency.

[0136] As an option, in some embodiments, the control signal of the spray valve can be compensated in real time in combination with environmental parameters such as air humidity and workshop temperature to further optimize the stability and quality of the spray process.

[0137] In another implementation, to further improve spraying accuracy, the algorithms for spray path planning and pulse control signal generation can be adaptively optimized and adjusted based on real-time feedback during the spraying process. This approach can make the spraying process more intelligent and adaptable to the needs of different workpieces and changes in the external environment.

[0138] S6. Based on the environmental disturbance parameters in the multimodal process data, a disturbance observer is used to generate a disturbance estimate, and the disturbance estimate is synthesized with the above-mentioned control strategy results through feedforward compensation to form the final control instruction for driving the spray execution end.

[0139] Specifically, in step S5, the spray execution action has been generated, including robot arm path planning and pulse control signal generation for the spray valve. This provides preliminary control signals for the spraying process. However, the spraying process may be affected by external disturbances (such as ambient temperature changes, wind speed fluctuations, etc.) or internal equipment factors (such as mechanical errors, sensor deviations, etc.), which may lead to a decrease in spraying accuracy. To this end, the goal of step S6 is to use a disturbance observer to generate a disturbance estimate and perform feedforward compensation with the aforementioned control strategy, thereby optimizing the spray control instructions in real time to ensure the stability and efficiency of the spraying process.

[0140] In this embodiment, the disturbance source is first modeled using a disturbance observer based on the environmental disturbance parameters (such as air humidity, workshop temperature, etc.) in the multimodal process data obtained in the previous steps. The function of the disturbance observer is to monitor and estimate the disturbance in the spraying process in real time and generate the corresponding disturbance estimate. Specifically, the disturbance observer calculates the current disturbance to the system in real time through mathematical modeling and data fusion, combining environmental sensor input and feedback information during the spraying process.

[0141] The calculation method of the disturbance estimator usually adopts filtering-based technology, such as exponential filtering or Kalman filtering, which can effectively improve the accuracy of disturbance estimation. Specifically, the disturbance estimator The update formula can be shown as follows:

[0142] ;

[0143] in: For the moment The disturbance estimator of ; is the gain matrix, which is used to adjust the weight of the disturbance estimation; is the time decay coefficient, which is used to control the estimated time response; Represents a core tensor The partial derivative with respect to time reflects the rate of dynamic change of the system; is the tensor product operation between a tensor and a factor matrix, representing the combined impact on the system state; is the factor matrix of environmental disturbance data, representing the impact of the external environment on the system; is the integral variable, which represents the time variable.

[0144] In general, the accuracy of disturbance estimation is directly related to the stability and precision of the spraying process. By accurately modeling and estimating the disturbance source in real time, the system can quickly respond to external disturbances and effectively compensate for uncertainties in the spraying process.

[0145] Alternatively, the accuracy of the disturbance estimate can be optimized by adjusting the parameters of the filtering algorithm. For example, under different operating conditions, the gain matrix and the time decay coefficient Adaptive adjustments can be made to accommodate different disturbance characteristics and dynamic changes.

[0146] In obtaining the disturbance estimator Finally, the system uses the feedforward compensation mechanism to synthesize the disturbance estimate with the output of the aforementioned control strategy to generate the final control instruction. This feedforward compensation method can pre-compensate for disturbances during the spraying process, reduce the impact of disturbances on the control system, and ensure the accuracy and stability of the spraying process. The calculation formula is as follows:

[0147] ;

[0148] in: It is the final control instruction after compensation; The initial control instructions generated according to the optimization strategy; is the disturbance gain adjustment matrix, which controls the amplitude of disturbance compensation.

[0149] Specifically, the control instructions It includes optimized parameters such as spraying speed, path correction, nozzle temperature, etc. These parameters are adjusted based on disturbance estimation to ensure high-precision execution of the spraying process.

[0150] In some embodiments, feedforward compensation can be used not only to compensate for environmental disturbances but also for internal disturbances within the device. For example, the system can adjust compensation strategies based on real-time feedback to address factors such as robotic arm motion errors and spray valve hysteresis, further improving the stability and accuracy of the spraying process.

[0151] The real-time monitoring and control system for parameters of a water turbine spraying process described below and the real-time monitoring and control method for parameters of a water turbine spraying process described above can correspond to each other.

[0152] Please see the attached Figure 2 , a real-time monitoring and control system for parameters of a turbine spraying process, comprising:

[0153] Data acquisition module, used to collect multimodal data during the spraying process, including trajectory parameters, rheological parameters, thermal physical state and environmental disturbance parameters;

[0154] Tensor construction module, used to construct four-dimensional data structures and perform tensor decomposition to extract core feature tensors and factor matrices;

[0155] Strategy decision module, used to build a control strategy model based on game mechanism and output process control strategy;

[0156] Stability analysis module, which is used to construct a virtual platoon and generate stable control inputs based on the Lyapunov optimization method;

[0157] Execution control module, used to plan the robot arm path and generate spray valve pulse signals based on the strategy results;

[0158] Disturbance compensation module, used to estimate environmental disturbances and perform control feedforward compensation;

[0159] The central control unit is used to coordinate the operation of the above modules and output control instructions in real time to drive the spraying device to complete the spraying operation.

[0160] The system of this embodiment can be used to execute the above method embodiments, and its principles and technical effects are similar, so they will not be repeated here.

[0161] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for real-time monitoring and control of parameters during a water turbine spraying process, characterized in that: The following steps are involved: S1. Collecting multimodal process data during a spraying process, and constructing a four-dimensional data structure based on the multimodal process data; The multimodal process data includes: Spraying trajectory data is obtained through lidar and visual encoder; Material rheological property data, including pressure, viscosity, and shear rate, are obtained through pressure sensors and flow meters; Thermophysical state data, including spray surface temperature and heat flux, are obtained by infrared thermal imaging camera; Environmental disturbance data, including air humidity, wind speed, and workshop temperature, is obtained through environmental sensors; Constructing a four-dimensional data tensor from multimodal process data based on time series ;in: is the dimension of the spray trajectory data, is the dimension of the material rheological data, is the dimension of thermophysical data, The dimension of the environmental perturbation data; S2. Based on the constructed four-dimensional data structure, a tensor decomposition operation is performed to extract a core feature tensor and a factor matrix of each dimension used to reflect the dynamic state of the spraying process; S3. Using the core feature tensor and factor matrix as state input, constructing a reinforcement game strategy model, and outputting a corresponding spraying process parameter control strategy based on the reinforcement game strategy model; The construction of the enhanced game strategy model includes: Design a dual-agent game framework with process quality and energy consumption as optimization objectives respectively; Use multi-round game processes to generate strategic game trajectories and realize the search for equilibrium solutions between different control strategies; The extracted feature tensor and factor matrix are used as the policy state space; The action space includes the process parameters of spraying speed, nozzle temperature and particle size distribution adjustment; The reward function construction follows the comprehensive weighting scheme of indicators. The reward function is constructed based on the comprehensive weighting of spraying quality, resource consumption, spraying pressure gradient and core tensor trace. S4. Based on the output results of the control strategy, a Lyapunov optimization method is introduced to conduct deviation state evaluation by constructing a virtual queue model for describing the stability of the spraying process, and a control input for regulation is generated according to the evaluation results; S5. Based on the control strategy and the control input, the generation of the spraying execution action is completed, specifically including the robot arm path planning and the generation of the pulse control signal of the spraying valve; S6. Based on the environmental disturbance parameters in the multimodal process data, a disturbance observer is used to generate a disturbance estimate, and the disturbance estimate is synthesized with the above-mentioned control strategy result through feedforward compensation to form a control instruction for driving the spray execution end.

2. A method for real-time monitoring and control of parameters during a water turbine spraying process according to claim 1, characterized in that: In step S2, the tensor decomposition operation includes: Perform normalization and denoising preprocessing on the constructed four-dimensional data structure; Extract the core feature tensor and the corresponding dimensional factor matrix based on the tensor decomposition algorithm Tucker; Use the sliding time window mechanism to incrementally update newly collected data to improve dynamic response capabilities; The decomposition results are used to represent the main dynamic modes of the spraying process status and serve as input for subsequent strategies.

3. The method for real-time monitoring and control of parameters during a water turbine spraying process according to claim 2, characterized in that: The tensor decomposition operation satisfies the following mathematical expression: ； in: is the original four-dimensional data tensor, representing the multimodal spraying process data; is the decomposed core tensor, which represents the main dynamic characteristics of the spraying process; is the factor matrix of the spray trajectory data dimension; is the factor matrix of the material rheological property data dimension; is the factor matrix of the thermophysical state data dimension; is the factor matrix of the environmental disturbance data dimension; For tensors and matrices in The multiplication operation in the mode, that is, along the Dimensions are linearly mapped.

4. The method for real-time monitoring and control of parameters during a water turbine spraying process according to claim 1, characterized in that: In step S4, constructing a virtual queue model for describing the stability of the spraying process includes: Define the control parameter drift as the basis for virtual queue update; Set the threshold term and control penalty function, and combine the Lyapunov function to determine the deviation stability boundary; Establish system stability indicators based on spray pressure changes and material deposition rate; According to the queue status feedback control input generation rules, the output rate is dynamically adjusted.

5. The method for real-time monitoring and control of parameters during the spraying process of a water turbine according to claim 4, characterized in that: The Lyapunov stability control satisfies the following optimization function: ； in: For the current moment Control inputs include spray speed, path correction, and nozzle temperature; is the drift of the virtual queue, which measures the change of queue status between consecutive moments; For the virtual queue at the current moment Status; is a positive scalar, called the drift-penalty balance factor, which is used to adjust the weight relationship between control performance and queue stability; is the expected value of the loss function, which represents the core tensor The stability or error of the current state is measured.

6. The method for real-time monitoring and control of parameters during a water turbine spraying process according to claim 1, characterized in that: In step S5, the generation of the spray execution action is completed, including: According to the control strategy optimization output, a set of spraying path points of the robot arm is generated; Use spline interpolation for trajectory smoothing and acceleration limit control; Convert the generated path into a servo control instruction and combine it with a pulse modulation signal to drive the spraying actuator; The spray valve control signal adopts a multi-level modulation method, and the frequency and switch duty cycle are dynamically adjusted according to the pressure demand.

7. The method for real-time monitoring and control of parameters during a water turbine spraying process according to claim 1, characterized in that: In step S6, the feedforward compensation synthesis of the disturbance estimation amount and the above control strategy result includes: Construct a disturbance observer to model the disturbance sources of temperature fluctuations and sudden changes in air velocity from the external environment; Use a combination of exponential filtering and Kalman filtering to improve the accuracy of disturbance estimation; Use disturbance estimators for feedforward compensation, including gain adjustment, delay correction, and error fitting; The final compensation control instruction is used to drive the execution end in real time to maintain process stability.

8. A real-time monitoring and control system for parameters of a water turbine spraying process, applied to a real-time monitoring and control method for parameters of a water turbine spraying process according to any one of claims 1 to 7, characterized in that: include: Data acquisition module, used to collect multimodal data during the spraying process, including trajectory parameters, rheological parameters, thermal physical state and environmental disturbance parameters; Tensor construction module, used to construct four-dimensional data structures and perform tensor decomposition to extract core feature tensors and factor matrices; Strategy decision module, used to build a control strategy model based on game mechanism and output process control strategy; Stability analysis module, which is used to construct a virtual platoon and generate stable control inputs based on the Lyapunov optimization method; Execution control module, used to plan the robot arm path and generate spray valve pulse signals based on the strategy results; Disturbance compensation module, used to estimate environmental disturbances and perform control feedforward compensation; The central control unit is used to coordinate the operation of the above modules and output control instructions in real time to drive the spraying device to complete the spraying operation.

Images