Method and related apparatus for detecting and optimizing spring operating mechanisms

Abstract

This invention provides a method for detecting and optimizing spring operating mechanisms, as well as related devices. [Solution] The method includes: collecting important data of a spring operating mechanism in real time using stress sensors, velocity sensors, and vibration sensors; constructing a multidimensional data framework and obtaining a multidimensional dataset based on the collected important data; extracting failure features and performing data analysis from the obtained multidimensional dataset, constructing a self-adaptive failure model based on the extracted failure features, then obtaining real-time failure detection results by multimodal data fusion; and optimizing the stress sensors, velocity sensors, and vibration sensors using automatic calibration and optimization of reinforcement learning based on the real-time failure detection results.

Description

[Technical Field]

[0001] This invention belongs to the technical field of fault detection for power equipment, and more specifically, relates to a fault detection and optimization method and related apparatus for spring operating mechanisms. [Background technology]

[0002] In the field of fault detection for CT20 spring operating mechanisms, improving detection accuracy and real-time performance is of great importance. However, conventional fault detection methods generally rely on periodic maintenance and manual inspections. These methods have many limitations and drawbacks. For example, conventional methods make it difficult to monitor the operating status of CT20 spring operating mechanisms in real time. Problems with false alarms or missed reports of faults are common. When detecting the operating status of a spring operating mechanism, simply collecting data from individual components using a single sensor does not provide a comprehensive detection result, making it impossible to accurately identify the type and extent of the fault. Therefore, it is particularly important to improve the fault detection accuracy and self-adaptive capabilities of spring operating mechanisms, realize real-time detection and fault prediction for CT20 spring operating mechanisms, gain a more comprehensive understanding of the equipment's operating status, and reduce false alarms and missed reports of faults. [Overview of the project] [Problems that the invention aims to solve]

[0003] The object of the present invention is to provide a fault detection and optimization method and related apparatus for a spring operating mechanism that solves the technical problems in the prior art, where reliance on a single data source results in insufficient detection results, making it impossible to achieve real-time monitoring and fault prediction, and further leading to false alarms and failures in reporting faults. [Means for solving the problem]

[0004] To achieve the above objectives, the present invention is constructed based on the following technical means.

[0005] According to the first aspect, a method for detecting and optimizing a spring operating mechanism is provided. To collect important data on the spring operating mechanism in real time using stress sensors, velocity sensors, and vibration sensors, Based on the aforementioned important data collected, a multidimensional data framework will be constructed and a multidimensional dataset will be obtained. The process involves extracting failure features and performing data analysis from the obtained multidimensional dataset, constructing a self-adaptive failure model based on the extracted failure features, and then obtaining real-time failure detection results through multimodal data fusion. This includes optimizing the stress sensor, velocity sensor, and vibration sensor by utilizing automatic calibration and optimization of reinforcement learning based on the real-time fault detection results.

[0006] Furthermore, before collecting important data on the spring operating mechanism using the stress sensor, The geometric model of the spring operating mechanism is constructed, the material attributes of each component in the geometric model of the spring operating mechanism are defined, and the stress distribution and deformation behavior of the spring operating mechanism under operating load are simulated using finite element analysis. Based on the actual operating state of the spring operating mechanism, load conditions and boundary conditions are set, and the geometric model of the spring operating mechanism is meshed to generate a basic contour map of the stress distribution. Using a self-adaptive stress sensing algorithm, the stress distribution data on the contour plot of the stress distribution is analyzed to automatically identify stress concentration regions and obtain the gradient value of each stress concentration point and regions where stress changes are significant in the stress distribution. Clustering is performed on the identified stress concentration points, and adjacent high-stress gradient points are combined into a single stress concentration region. The method further includes refining the boundaries of stress concentration regions using a self-adaptive stress sensing algorithm to identify dense stress distribution regions, and determining the placement of stress sensors in the spring operating mechanism using a genetic optimization algorithm.

[0007] Furthermore, before collecting important data on the spring operating mechanism using the speed sensor, Using already placed strain sensors, initial motion data of the spring operating mechanism under different operating conditions is acquired, and based on the initial motion data, a spring motion model is established for the spring operating mechanism, and the dynamic behavior of the spring at rest, in motion, and under different operating conditions is simulated. Using a multimodal motion analysis algorithm (a multimodal motion analysis algorithm is a system of algorithms that comprehensively processes various types of data to accurately analyze, understand, and predict motion-related information), the motion trajectory of the spring in the spring operating mechanism is acquired in real time, visual data is acquired, and important motion features are extracted from the visual data. The method further includes using the established spring motion model to extract the motion parameters of the spring, and fusing the visual data with the motion parameters using a Kalman filter.

[0008] Furthermore, before collecting important data on the spring operating mechanism using the vibration sensor, Using conventional vibration sensors or temporarily placed vibration sensors, vibration data is collected under different operating conditions of the spring operating mechanism. By combining vibration data with stress data collected by a stress sensor and velocity data collected by a velocity sensor, and using multimodal fusion technology (multimodal fusion technology is a method of integrating multiple different types of data such as text, images, audio, and video, which overcomes the limitations of a single data format and enables more comprehensive and accurate information processing), the dynamic behavior of the spring operating mechanism under different operating conditions is analyzed. Using a multimodal motion analysis algorithm, we can identify the vibration modes of a spring under different operating modes, This further includes comparing vibration modes in different operating modes, identifying the operating mode with the most intense vibration, and determining the placement position of the vibration sensor in the spring operating mechanism using a genetic optimization algorithm.

[0009] Furthermore, based on the aforementioned important data collected, constructing a multidimensional data framework and obtaining a multidimensional dataset specifically involves: After the stress sensors, velocity sensors, and vibration sensors have been positioned, a multi-frequency self-adaptive sampling algorithm is used to dynamically adjust the sampling frequency of important data based on the position of the stress sensors, velocity sensors, and vibration sensors, so that multidimensional data of stress, velocity, and vibration are collected simultaneously. A multidimensional data framework is constructed on the multidimensional stress, velocity, and vibration data collected simultaneously, and the collected multidimensional stress, velocity, and vibration data is standardized and integrated using multimodal fusion technology. The standardized multidimensional data is integrated into a three-dimensional data matrix, where the first dimension is time, the second dimension is sensor position, and the third dimension is data type. This includes generating a multidimensional data table based on the aforementioned three-dimensional data matrix and obtaining a multidimensional dataset.

[0010] Furthermore, the process of extracting failure features and performing data analysis from the obtained multidimensional dataset, constructing a self-adaptive failure model based on the extracted failure features, and then obtaining real-time failure detection results through multimodal data fusion, specifically involves: This involves extracting failure features from stress, velocity, and vibration data from a multidimensional dataset, and measuring the linear correlation between different failure feature data using the Pearson correlation coefficient. The correlation calculation results corresponding to all failure features are constructed into a correlation matrix, and interrelated failure features and failure features containing redundant information are identified. Using a random forest algorithm, the importance of each failure feature in failure prediction is evaluated, and based on the evaluation results of the failure feature importance, the most important failure feature is selected to optimize the failure prediction effect. Based on the extracted failure characteristics, a self-adaptive failure model is established, and the probability of failure occurring is calculated in real time using the self-adaptive failure model. This includes obtaining fault detection results and warning information using multimodal fusion technology based on the probability of failure occurrence.

[0011] Furthermore, optimizing the stress sensor, velocity sensor, and vibration sensor using automatic calibration and optimization of reinforcement learning based on the real-time fault detection results means, specifically, performing automatic calibration and optimization of the stress sensor, velocity sensor, and vibration sensor using automatic calibration and optimization of reinforcement learning based on real-time fault detection results and warning information.

[0012] According to a second aspect, a fault detection and optimization system for a spring operating mechanism is provided. A data collection module for collecting important data on the spring operating mechanism in real time, Based on the aforementioned important data collected, a framework construction module is provided to build a multidimensional data framework and obtain a multidimensional dataset. A failure feature extraction module for extracting failure features from the obtained multidimensional dataset, A data analysis module for performing data analysis on multidimensional datasets, A model building module for constructing a self-adaptive failure model, Includes a reinforcement learning module for optimizing stress sensors, velocity sensors, and vibration sensors.

[0013] According to a third aspect, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the steps of the fault detection and optimization method for a spring operating mechanism described above are realized when the processor executes the computer program.

[0014] According to a fourth aspect, a computer-readable storage medium is provided, a computer program is stored in the computer-readable storage medium, and the fault detection and optimization method for the spring operating mechanism described above is realized when the computer program is executed by a processor. [Effects of the Invention]

[0015] Compared to the prior art, the present invention has the following beneficial effects.

[0016] 1. When performing fault detection on a spring operating mechanism, operating data of the spring operating mechanism can be collected in real time using stress sensors, velocity sensors, and vibration sensors closely related to the operation of the spring operating mechanism, enabling rapid extraction of fault data. This enables rapid detection and warning of spring operating mechanism failures. Next, by combining three different types of collected data during the fault detection process, the comprehensiveness and accuracy of the detection results are further improved, avoiding problems of false alarms and missed reports. Finally, by combining a self-adaptive failure model, a multimodal motion analysis algorithm, and automatic calibration and optimization based on reinforcement learning, important data during the operation of the equipment can be acquired and analyzed in real time. This significantly improves the timeliness of fault detection, while simultaneously optimizing the sensors and ensuring reliability during the data collection process.

[0017] 2. By constructing an accurate geometric model of the spring operating mechanism and defining the material attributes of each component in detail, a solid foundation is provided for subsequent finite element analysis. This ensures the accuracy and reliability of the simulation results, and allows for the simulation of stress distribution and deformation behavior under operating loads of the spring operating mechanism using finite element analysis, intuitively illustrating the stress state under different operating conditions of the mechanism. This contributes to the advance detection of potential stress concentration areas and deformation trends, providing important reference for failure prevention. By setting load and boundary conditions based on the actual operating state of the spring operating mechanism, the simulation results are brought closer to the actual situation, improving the directionality and practicality of the analysis. Furthermore, a self-adaptive stress sensing algorithm deeply analyzes data on contour plots of the stress distribution, automatically identifying stress concentration areas and areas of significant stress change. This not only improves the accuracy and efficiency of stress identification but also reduces human intervention and errors, providing accurate target areas through subsequent sensor placement.

[0018] 3. The spring motion model established based on initial motion data can simulate the dynamic behavior of the spring at rest, in motion, and under different operating conditions, contributing to a deep understanding of the spring's motion characteristics and providing theoretical support for subsequent data collection and analysis. By employing a multimodal motion analysis algorithm, the motion trajectory of the spring in the spring operating mechanism can be acquired in real time, along with abundant visual data. Important motion characteristics such as velocity and acceleration can be extracted from the visual data, providing multidimensional information for data analysis. By fusing the visual data and motion parameters extracted by the spring motion model using a Kalman filter, data noise and errors can be effectively reduced, improving the accuracy and reliability of the data.

[0019] 4. By using conventional vibration sensors or temporarily placed sensors, vibration data is collected under different operating conditions of the spring operating mechanism, providing a rich data source for subsequent data analysis and making the analysis results more comprehensive and accurate. By combining vibration data with stress data and velocity data and analyzing it using multimodal fusion technology, the interaction and influence between different physical quantities can be comprehensively considered, thereby more accurately representing the dynamic behavior of the spring operating mechanism. Multimodal data fusion improves the depth and breadth of the analysis and makes the results more reliable.

[0020] 5. After the stress sensors, velocity sensors, and vibration sensors have been positioned, a multi-frequency self-adaptive sampling algorithm is used to dynamically adjust the sampling frequency of important data based on the sensor placement. This ensures that data sampling is more concentrated in critical areas or under critical operating conditions, thereby allowing for the acquisition of more detailed information. Simultaneously, the sampling frequency is appropriately reduced in non-critical areas or under stable operating conditions, thereby reducing the load on data storage and processing. This dynamic adjustment mechanism improves the flexibility and efficiency of data acquisition.

[0021] 6. Multimodal data fusion and failure feature extraction can more comprehensively reflect the operating status of equipment, improving the accuracy and reliability of failure detection. The self-adaptive failure model can calculate the probability of failure in real time and provide early warnings before failures occur, reducing equipment downtime and contributing to improved production efficiency and economic benefits. Subsequently, based on the importance assessment of failure features, a more scientific and rational maintenance policy can be formulated, allowing for focused monitoring and maintenance of critical failure features, extending the service life of equipment and reducing maintenance costs. The self-adaptive failure model can dynamically adjust parameters based on real-time data, thereby adapting to changes in equipment status, strengthening the system's adaptability and robustness, and enabling it to better handle complex and volatile operating environments.

[0022] 7. Automatic calibration and optimization ensure that sensors output accurate data under different operating conditions, reducing false alarms and missed reports, and contributing to improved fault detection accuracy. The reinforcement learning model can adaptively adjust sensor parameters to respond to changes in system state, enhancing system robustness and enabling it to better handle complex and volatile operating environments. Automatic calibration and optimization reduce the need for manual intervention, lower maintenance costs, and simultaneously improve the accuracy of sensor data, thus reducing unnecessary downtime and maintenance due to false alarms. [Brief explanation of the drawing]

[0023] To more clearly explain the technical configuration of the embodiments of the present invention, the drawings used in the embodiments are briefly described below. Note that the following drawings are merely illustrative of some embodiments of the present invention and do not limit the scope of the invention. Those skilled in the art can create other relevant drawings based on these drawings without requiring any creative effort. [Figure 1] This is a flowchart of the fault detection and optimization method for a spring operating mechanism according to the present invention. [Figure 2] This is a geometric model diagram of the CT20 spring operating mechanism in the fault detection and optimization method for spring operating mechanisms according to the present invention. [Figure 3] This is a contour plot of the stress distribution in the fault detection and optimization method for a spring operating mechanism according to the present invention. [Figure 4] This is a time-domain vibration signal diagram for a fault detection and optimization method for a spring operating mechanism according to the present invention. [Figure 5] This is a schematic diagram illustrating the principle of the fault detection and optimization system for a spring operating mechanism according to the present invention. [Modes for carrying out the invention]

[0024] To further clarify the objectives, technical configurations, and advantages of the embodiments of the present invention, the following provides a clear and complete description of the technical configurations of the embodiments of the present invention, in conjunction with the drawings of the embodiments. Clearly, the embodiments described are only some, not all, embodiments of the present invention. Generally, the assemblies of the embodiments of the present invention shown in the drawings may be arranged and designed in a variety of different configurations.

[0025] Therefore, the detailed description of embodiments of the present invention provided in the drawings below is not intended to limit the scope of protection of the present invention, but merely to represent selected embodiments of the present invention. All other embodiments obtained based on the embodiments of the present invention without requiring creative effort from those skilled in the art are all within the scope of protection of the present invention.

[0026] Furthermore, similar symbols and letters in the drawings indicate the same or corresponding constituent elements; therefore, once a constituent element is defined in one drawing, it is not necessary to define and interpret it again in subsequent drawings.

[0027] In describing embodiments of the present invention, when directions or positional relationships indicated by terms such as "up," "down," "horizontal," and "inside" appear as matters to be described, these are based on the directions or positional relationships shown in the drawings, or the directions or positional relationships that are generally used when the product of this invention is in use. These are merely for the convenience and simplification of the description of the present invention and do not indicate or imply that the mentioned device or element has a specific direction or must be configured and operated in a specific direction, and should not be understood as limitations on the present invention. Furthermore, terms such as "first," "second," etc., are used solely for the purpose of description and should not be understood as indicating or implying relative importance.

[0028] Furthermore, when the term "horizontal" appears, it is not required that the part be absolutely horizontal; it may be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," and does not necessarily mean that the structure is perfectly horizontal; it may be slightly tilted.

[0029] In describing embodiments of the present invention, unless otherwise specifically defined and limited, the terms “installation,” “mounting,” “connection,” and “connection” should be understood in a broad sense. For example, a connection may be fixed, removable, or integral; a connection may be mechanical or electrical; a connection may be direct or indirect through an intermediate medium; or communication between two elements. Those skilled in the art will be able to understand the specific meaning of these terms in the present invention depending on the specific situation.

[0030] In the field of fault detection for CT20 spring operating mechanisms, improving detection accuracy and real-time performance is of great importance. Conventional fault detection methods generally rely on periodic maintenance and manual inspections. However, these methods have many limitations and drawbacks. For example, conventional methods make it difficult to monitor the operating status of CT20 spring operating mechanisms in real time, and problems such as false alarms or missed reports of faults are likely to occur. Furthermore, when detecting the operating status of a spring operating mechanism, simply collecting data from individual components using a single sensor does not provide a comprehensive detection result, and it is not possible to accurately identify the type and level of the fault. Therefore, it is particularly important to improve the fault detection accuracy and self-adaptive capability of spring operating mechanisms, achieve real-time detection and fault prediction for CT20 spring operating mechanisms, gain a more comprehensive understanding of the operating status of the equipment, and reduce false alarms and missed reports of faults.

[0031] To solve the above technical problems, the inventor provides a method for detecting and optimizing a spring operating mechanism and related apparatus.

[0032] A first embodiment of the present invention provides a method for fault detection and optimization of a spring operating mechanism, which includes the following steps as shown in Figure 1.

[0033] In S101, stress sensors, velocity sensors, and vibration sensors are used to collect important data on the spring operating mechanism in real time. Exemplarily, a geometric model of the CT20 spring operating mechanism is first constructed. As shown in Figure 2, this geometric model includes the main components of the spring operating mechanism, specifically the spring, operating lever, link mechanism, and support structure. Each component needs to be accurately depicted in the geometric model to ensure accurate results in subsequent stress analysis. As shown in the table below, material attributes of each component in the geometric model of the spring operating mechanism are defined to perform accurate stress analysis. These attributes include elastic modulus, yield strength, and density. It should be noted that the selection of material attributes is generally determined based on the requirements of the mechanical design and the actual operating environment of the CT20 spring operating mechanism.

[0034] [Table 1]

[0035] By defining the geometric model and material attributes of each component, the fundamental data necessary for subsequent stress analysis can be provided. After defining the material attributes of each component, the stress distribution and deformation behavior of the CT20 spring operating mechanism under operating load were simulated using finite element analysis (FEA) on the material attribute data of each component shown in the table above. This provides important information for determining the mounting position of the stress sensor. Finite element analysis is a numerical analysis technique that simulates a true physical system (geometric and load conditions) using mathematical approximation methods.

[0036] Furthermore, load and boundary conditions were set based on the actual operating state of the CT20 spring operating mechanism. The load conditions include the forces applied during the spring's energy storage and release, while the boundary conditions include fixed points and support points. Specifically, in the fully compressed state, a load force of approximately 500 N is generally applied to the end of the spring. This force acts along the axial direction of the spring and is used to simulate the spring's energy storage and release processes. The operating lever needs to be moved against an operating force of approximately 300 N when operating the switch, and this force is applied along the operating direction of the lever. In the boundary conditions, one end of the spring is fixed inside the case, so this end is set to be completely fixed. That is, both the displacement and rotation at this point are constrained. At the same time, support boundary conditions are set at the connection point between the link mechanism and the support structure. The displacement at these points is limited, but rotational degrees of freedom are allowed, thereby accurately simulating the actual operating support conditions.

[0037] Next, a mesh was created for the geometric model of the spring operating mechanism, and stress analysis was performed using a finite element analysis (FEA) tool to calculate the stress distribution under the operating load of the entire mechanism. As shown in Figure 3, a basic contour plot of the stress distribution was generated. Subsequently, the stress distribution data was analyzed by applying an Adaptive Stress Sensing Algorithm (ASSA) to automatically identify stress concentration regions. The Adaptive Stress Sensing Algorithm (ASSA) is an algorithm for analyzing and processing stress-related data. Based on the contour plot of the stress distribution, the data in the figure is extracted in matrix form, and the stress value σ(i,j,k) at each grid point is used as input data to calculate the stress gradient at each point.

number

[0038] Here, σ is the stress value, and x, y, and z are the spatial coordinate axes, respectively.

[0039] The discrete form of the stress gradient can be expressed as follows:

number

[0040] These equations allow us to obtain the gradient value at each point in the stress distribution and identify regions where stress changes are significant. Next, we apply the threshold method to obtain the threshold value ∇σ of the stress gradient. threshold A threshold is set, and when the stress gradient value ∇σ(x,y,z) at a certain point exceeds this threshold, that point is marked as a stress concentration point: If|∇σ(x,y,z)|>, ∇σ threshold , then mark as stress concentration point.

[0041] Next, clustering was performed on the identified stress concentration points, and adjacent high-stress gradient points were combined into a single stress concentration region. These regions represent potential sensor placement areas. Next, the ASSA algorithm is applied to perform multiple iterative optimizations, and the stress gradient threshold ∇σ is determined. threshold The stress concentration region is repeatedly calculated while making gradual adjustments to maximize the coverage of stress information within this region. Simultaneously, the ASSA algorithm is used to gradually refine the boundary of the stress concentration region, ultimately identifying the area where the stress distribution is most densely concentrated and narrowing down the most suitable location for attaching the stress sensor. Within the identified stress concentration region, a genetic optimization algorithm is applied to determine the optimal placement of the sensor. The goal of this algorithm is to minimize measurement redundancy in the sensor coverage area while maximizing stress data acquisition. The target function for optimizing sensor position is as follows:

number

[0042] Here, σ(P i ) is the stress value at the sensor placement point, σ avg This represents the average stress value in this region.

[0043] Genetic optimization algorithms are global optimization search algorithms inspired by natural selection and genetic mechanisms in organisms. They derive the optimal or approximate optimal solution to a problem by simulating operations such as heredity, mutation, and selection in the evolutionary process of organisms.

[0044] Finally, a proposed arrangement of stress sensors will be created, including the mounting position, direction, and precautions for each sensor, as shown in the table below. The entire process and verification results will also be recorded to provide justification for subsequent implementations.

[0045] [Table 2]

[0046] Before collecting critical data on the spring operating mechanism using velocity sensors, it is necessary to acquire and record initial motion data under different operating conditions of the spring operating mechanism using further positioned strain sensors. This initial motion data includes changes in position, displacement, time, and velocity. Based on the initial motion data, a motion model of the spring in the spring operating mechanism is established, and the dynamic behavior of the spring at rest, in motion, and under different operating conditions is simulated. Next, a multimodal motion analysis algorithm is performed, and the motion trajectories of the spring and its associated components are acquired in real time using high-speed cameras and computer vision technology, and visual data is obtained. Furthermore, important motion features, such as the spatial displacement, velocity, and acceleration of components such as the spring and links, are extracted from this visual data. The specific steps include the following:

[0047] Ensure that visual data and rudimentary motion model data are synchronized in time and space. Generally, visual data (e.g., image sequences from a high-speed camera) and motion models (e.g., motion trajectories generated based on rudimentary sensor data) may have different time sampling frequencies and spatial resolutions, requiring time alignment and spatial registration. This allows the two sets of data to be compared and merged on the same coordinate system and timeline.

[0048] Next, we extract key motion features (e.g., displacement, velocity, acceleration) from the visual data and use the optical flow method to extract the object's motion vector field:

number

[0049] Here, I x and I y These are the gradients of the image in the x and y directions, respectively, and I t is the rate of change of the image over time, and u and v are the horizontal and vertical velocity components (i.e., motion vectors) of the image point. The corresponding motion parameters are extracted from the motion model, and generally, these parameters have already been roughly estimated from the sensor data, such as position change, velocity, and acceleration.

[0050] Using an established spring motion model, we extract the spring motion parameters and fuse them with the visual data and the parameters from the motion model using a Kalman filter. The Kalman filter is a recursive algorithm that can combine data from different sources and take into account the noise level of each data source, thereby producing the optimal estimate. Key equations of the Kalman filter include the following:

[0051] State prediction:

number

[0052] This step is to predict the state (such as position, velocity, etc.) of the system at the current time t. JPEG0007868240000008.jpg2929 is the predicted state vector, which is obtained through the transformation of the state transition matrix A from the state vector x at the previous time. Here, B is the control input matrix, and u t-1 is the control input (such as external force or acceleration) applied at time t - 1. t-1

[0053] State prediction is used to estimate the future state (such as position, velocity) of the spring operating mechanism based on the previous state and control input (such as external force or operating action).

[0054] Error covariance prediction:

Number

[0055] This step is the predicted error covariance matrix JPEG0007868240000010.jpg2725. The covariance matrix represents the uncertainty of state estimation. The covariance matrix P at the previous time t‐1 is transformed through the state transition matrix A and further added with the process noise covariance matrix Q to obtain the predicted error covariance matrix JPEG0007868240000011.jpg2725 at the current time, which can contribute to the determination for reducing uncertainty by arranging the sensor in an area with relatively large error.

[0056] Kalman gain calculation:

Number

[0057] The Kalman gain K tis a weight matrix for finding the optimal combination between the predicted state and the actual observation. Here, H is the observation matrix, which maps the state vectors to the observation space, and R is the observation-noise covariance matrix. The role of the Kalman gain is to measure the relative importance of the uncertainty of the prediction and the uncertainty of the observation. A relatively large gain indicates that the observation is more certain, and a relatively small gain indicates that the prediction is more certain.

[0058] By calculating the Kalman gain, it is possible to determine how to trade off the contributions of visual data (actual observations) and motion models (predictions) in sensor placement. The magnitude of the gain directly affects the accuracy of the final motion characteristic estimation and further influences the decision-making process for optimizing sensor placement.

[0059] Status update:

number

[0060] The state update formula is used to update the current state estimate by combining the predicted state and the actual observed values. Here, z t This is the observed value actually measured at time t, JPEG0007868240000014.jpg2040 is the predicted observed value obtained after mapping the predicted state to the observation space. The state update equation is the Kalman gain K t By adjusting the predicted state and bringing it closer to actual observations, the accuracy of state estimation is improved.

[0061] Update of error covariance:

number

[0062] The error covariance update formula is the covariance matrix P at the current time. tThis is used to update the state estimate, which is a correction for the uncertainty of the state estimate after combining it with the observation. As new information is introduced into the observation, the error covariance generally decreases, meaning that the state estimate becomes more accurate.

[0063] By using sensor data after Kalman filtering, estimation errors in motion characteristics can be reduced, thereby improving the accuracy of the motion model. Continuously updating the error covariance matrix allows for dynamic adjustment of the sensor placement to adapt to the constantly changing motion environment.

[0064] Motion model analysis identifies regions where velocity changes are significant during the motion of the spring and its related components. These regions are generally locations where stress or dynamic effects are concentrated and where sensitivity to velocity measurements is highest. The specific steps include the following:

[0065] Calculation of velocity change rate: Using the fused motion model, calculate the velocity change rate over the entire motion process, i.e., the derivative of velocity with respect to time:

number

[0066] In the equation, v(t) represents the velocity at a certain time t. JPEG0007868240000017.jpg4046 represents the derivative of velocity with respect to time, i.e., the instantaneous acceleration a(t) at time t. Used to analyze the dynamic behavior of the CT20 spring operating mechanism, by calculating the rate of velocity change, it is possible to identify the time and region in the system where velocity changes are significant, which generally corresponds to the acceleration or deceleration process of the system.

[0067] Velocity gradient analysis: This method calculates the gradient in a velocity field, i.e., the rate of change of velocity in space. A region with a large velocity gradient indicates that the velocity change in that area is relatively rapid and generally corresponds to important events in the motion process. The formula for calculating the velocity gradient is as follows:

number

[0068] In the equation, ∇ v (x,y,z) is the velocity gradient vector, which includes the rate of change of velocity in the x, y, and z directions, respectively. JPEG0007868240000019.jpg3616, JPEG0007868240000020.jpg3116, It is represented as JPEG0007868240000021.jpg2916.

[0069] In this patent, the calculation of the velocity gradient is used to identify regions where velocity changes are significant during the motion process of the CT20 spring operating mechanism. Areas with a large velocity gradient generally indicate a rapid change in velocity in that region. This could be due to significant events such as sudden spring release, stopping, or steering. These regions are generally areas where dynamic effects are concentrated and are suitable for mounting velocity sensors to accurately capture this rapidly changing data.

[0070] The discrete form of the velocity gradient is as follows:

number

[0071] The velocity gradient was approximated using the finite difference method. The above equations represent the discrete forms of the velocity gradient in the three directions x, y, and z, respectively.

[0072] for example, The discrete form of JPEG0007868240000023.jpg3620 is obtained by calculating the rate of change of velocity in the x-direction, which is obtained by dividing the velocity difference between two adjacent points (i+1,j,k) and (i-1,j,k) in the x-direction by the distance between them, 2Δx.

[0073] Similarly, JPEG0007868240000024.jpg3116 and The discrete form of JPEG0007868240000025.jpg2916 calculates the velocity gradients in the y and z directions, respectively.

[0074] The key points in a motion process where accurate velocity measurement is most crucial are determined. These key points generally appear at the limits of the motion trajectory, in regions of maximum acceleration, or in regions of relatively high vibration frequencies. The objective of this invention is to determine the placement of the velocity sensor in the spring operating mechanism using a genetic optimization algorithm, ensuring that the velocity sensor covers the most critical motion regions. The goal of this algorithm is to minimize errors in velocity sensor position and maximize the acquisition of critical motion data. The target function is set as follows:

number

[0075] Here, v(P i ) is the velocity value at the sensor placement point, v avg This is the average velocity value in this region.

[0076] As shown in the table below, a proposed placement of the speed sensors was created based on the MMMAA analysis results.

[0077] [Table 3]

[0078] Similarly, before collecting critical data on the spring operating mechanism using vibration sensors, the process further includes collecting vibration data under different operating conditions of the spring operating mechanism using conventional or temporarily placed vibration sensors. These data include time-domain vibration signals under spring compression, release, and other operating states, as shown in Figure 4. The time-domain vibration signals are then converted to the frequency domain using a Fourier transform (Fast Fourier Transform, FFT):

number

[0079] Here, F(f) represents the signal in the frequency domain, V[n] is the nth sampling point in the time domain, N is the total number of sampling points, k is the frequency index, and j is the imaginary unit.

[0080] And obtain the amplitude |F(f)|:

number

[0081] Here, Re(F(f)) 2 and Im(F(f)) 2 These are the real and imaginary parts of the frequency domain signal, respectively.

[0082] Next, the amplitude data is combined with the previously collected stress and velocity data, and the dynamic behavior of the spring operating mechanism under different operating conditions is further analyzed using multimodal fusion technology. The specific steps are as follows:

[0083] Vibration, stress, and velocity data are time-synchronized and standardized so that they are analyzed on the same time axis. Each data point is standardized to a zero-mean and unit variance.

number

[0084] Here, V(t), σ(t), and v(t) represent vibration, stress, and velocity data, respectively, while μ and σ are the mean and standard deviation of the corresponding data.

[0085] The main features were extracted from the standardized data, and the following were obtained:

[0086] Vibration characteristics F V (t):

number

[0087] Here, the peak value F V,peak (t) represents the maximum amplitude of the vibration signal at time t, and V'(t) is the standardized vibration signal. max(V'(t)) represents the maximum value of the vibration signal within a given time, and is the root mean square value F. V,RMS (t) reflects the RMS value of the vibration signal, and N is the total number of sampling points for signal sampling. V'(t i V'(t) represents the vibration signal value at the i-th sampling point, and FFT(V'(t)) is the Fourier transform result of the standardized vibration signal V'(t), which is used to obtain features in the frequency domain and to select the frequency component with the highest amplitude as a feature in the spectrogram.

[0088] Stress characteristics Fσ(t):

number

[0089] Here, σ'(t) is the standardized stress signal. max(σ'(t)) and min(σ'(t)) represent the maximum and minimum values ​​of the stress signal, respectively. ∇σ'(t) is the gradient of the stress signal and represents the rate of change of stress in the spatial coordinate system.

[0090] Speed ​​characteristic F v (t):

number

[0091] Here, JPEG0007868240000034.jpg4259 is the derivative with respect to time t. max(v'(t)) is the maximum value of the normalized velocity signal. std(v'(t)) is the standard deviation of the velocity signal and is used to represent the breadth of the velocity distribution at different time points.

[0092] Vibration characteristics F V (t), stress characteristic Fσ(t) and velocity characteristic F v Fusion of (t):

number

[0093] Here, w V , w σ , w v These are weighting coefficients corresponding to vibration, stress, and velocity features, respectively, and these weights reflect the importance of each feature to the overall dynamic behavior. Fusion contributes to determining the relationship between vibration and other motion features (e.g., velocity, stress), thereby identifying the most vibration-prone areas.

[0094] Next, we use MMMAA to identify the vibration modes in different operating modes of the spring, and the specific steps are as follows:

[0095] Features related to the operating mode are extracted from vibration, stress, and velocity characteristics to form a feature vector:

number

[0096] Here, F V,peak (t) represents the peak value of the vibration signal at time t, max(V'(t)) represents the maximum value of the vibration signal within a predetermined time, F V,RMS (t) is the root mean square (RMS) of the vibration signal, JPEG0007868240000037.jpg61117 represents the average of the squared values ​​of the vibration signals at all sampling points, F V,FFT (t) is the spectral feature of the vibration signal, obtained by the Fourier transform FFT(V'(t)), and represents the amplitude of the vibration signal in the frequency domain, F σ,max (t) is the maximum value of the stress signal, max(σ'(t)) represents the maximum value of the stress signal within a given time, F v,rate(t) is the rate of change of the velocity signal, i.e., acceleration. Represented as JPEG0007868240000038.jpg3759, it reflects the rate of change of velocity over time. V'(t) represents the standardized vibration signal, σ'(t) represents the standardized stress signal, and v'(t) represents the standardized velocity signal.

[0097] These features are classified using K-mean clustering, operating modes are grouped, different vibration modes are identified, and the main vibration characteristics are determined by analyzing the frequency, amplitude, and duration T of each mode:

number

[0098] Here, f is the spectral amplitude |F V,FFT The main oscillation frequency identified by maximizing (f')| is argmax. f´ is, |F V,FFT This indicates that the frequency that maximizes (f')| has been identified, and A=F V,peak (t) represents the maximum amplitude of the vibration signal and corresponds to the peak value of the vibration signal, while T is the signal duration, which is generally determined by calculating the time interval at which the signal exceeds a certain threshold. By comparing vibration modes under different operating modes, the operating conditions with the most severe vibration are identified, and the parts with the highest sensitivity to vibration are determined by combining stress and velocity characteristics. Based on the analysis results, high-amplitude or high-frequency sensitivity regions are marked as candidate locations for sensor placement.

[0099] Finally, to further improve the accuracy of the sensor's position distribution, a genetic optimization algorithm is used to determine the placement of the vibration sensors within the spring operating mechanism. This algorithm minimizes measurement errors in the vibration signal and maximizes the coverage of critical vibration data from the sensors, enabling the acquisition of the most important vibration information in these highly sensitive regions.

number

[0100] Here, P i This is the position of the i-th sensor, and a(P i ) is the vibration acceleration value at this position, and a max This represents the maximum vibration acceleration value in this region.

[0101] As shown in the table below, a proposed placement of vibration sensors was created based on the MMMAA analysis results.

[0102] [Table 4]

[0103] When fault detection is performed on a spring operating mechanism using the above method, operating data of the spring operating mechanism is collected in real time using stress sensors, velocity sensors, and vibration sensors closely related to the operation of the spring operating mechanism, allowing for rapid extraction of fault data. This enables rapid detection and warning of spring operating mechanism failures. Furthermore, by combining three different types of collected data during the fault detection process, the comprehensiveness and accuracy of the detection results are further improved, avoiding problems of false alarms and missed reports. Finally, by combining a self-adaptive failure model, a multimodal motion analysis algorithm, and automatic calibration and optimization based on reinforcement learning, important data during the operation of the equipment can be acquired and analyzed in real time, significantly improving the timeliness of fault detection, while simultaneously optimizing the sensors and ensuring reliability during the data collection process. By constructing an accurate geometric model of the spring operating mechanism and defining the material attributes of each component in detail, a solid foundation is provided for subsequent finite element analysis, ensuring the accuracy and reliability of simulation results. Furthermore, by using finite element analysis to simulate the stress distribution and deformation behavior of the spring operating mechanism under operating load, the stress state under different operating conditions of the mechanism can be intuitively represented, contributing to the prior detection of potential stress concentration areas and deformation trends, and providing important reference for failure prevention. Load conditions and boundary conditions are set based on the actual operating conditions of the spring operating mechanism, bringing the simulation results closer to the actual situation and improving the directionality and practicality of the analysis. In addition, a self-adaptive stress sensing algorithm deeply analyzes data on contour plots of the stress distribution, automatically identifying stress concentration areas and areas with significant stress changes, improving the accuracy and efficiency of stress identification, reducing human intervention and errors, and providing accurate target areas through subsequent sensor placement.The spring motion model established based on initial motion data can simulate the dynamic behavior of the spring at rest, in motion, and under different operating conditions, contributing to a deep understanding of the spring's motion characteristics and providing theoretical support for subsequent data collection and analysis. By employing a multimodal motion analysis algorithm, the spring's motion trajectory in the spring operating mechanism can be acquired in real time, along with rich visual data. Important motion features such as velocity and acceleration can be extracted from this visual data, providing multidimensional information for data analysis. By fusing the visual data and motion parameters extracted by the spring motion model using a Kalman filter, data noise and errors can be effectively reduced, improving data accuracy and reliability. Using conventional vibration sensors or temporarily placed sensors, vibration data under different operating conditions of the spring operating mechanism can be collected, providing a rich data source for subsequent data analysis and making the analysis results more comprehensive and accurate. By combining vibration data with stress and velocity data and analyzing it using multimodal fusion technology, the interactions and influences between different physical quantities can be comprehensively considered, thereby more accurately representing the dynamic behavior of the spring operating mechanism. Multimodal data fusion improves the depth and breadth of the analysis and makes the results more reliable. After the placement of stress sensors, velocity sensors, and vibration sensors is complete, a multi-frequency self-adaptive sampling algorithm is used to dynamically adjust the sampling frequency of important data based on the sensor placement positions. This ensures that data sampling is more concentrated in critical areas or critical operating conditions, thereby obtaining more detailed information. At the same time, the sampling frequency is appropriately reduced in non-critical areas or stable operating conditions, reducing the load on data storage and processing. Such a dynamic adjustment mechanism improves the flexibility and efficiency of data acquisition.

[0104] In S102, a multidimensional data framework is constructed based on the collected important data, and a multidimensional dataset is obtained. Exemplarily, after the placement of stress sensors, velocity sensors, and vibration sensors is completed, a multi-frequency self-adaptive sampling algorithm is used to dynamically adjust the sampling frequency of important data based on the placement positions of the stress sensors, velocity sensors, and vibration sensors, so that multidimensional data of stress, velocity, and vibration are collected simultaneously.

[0105] First, the Multi-Frequency Adaptive Sampling Algorithm (MFASA) dynamically adjusts the sampling frequency of each sensor based on its position and the characteristics it monitors. Specifically, for sensors that detect significant state changes, MFASA increases the sampling frequency to ensure the acquisition of high-resolution data. The Multi-Frequency Adaptive Sampling Algorithm is an intelligent sampling method used in signal processing. Its core feature is its ability to dynamically adjust the sampling policy in multiple frequency dimensions according to the signal characteristics. The sampling frequency f of each sensor at time tt s (t) can be expressed as follows:

number

[0106] V(t) where F0 is the initial sampling frequency, α is the adjustment coefficient, and ΔV(t) is the rate of change of the sensor output signal at time t. max This represents the maximum rate of change of the sensor output signal.

[0107] Next, the synchronous data acquisition system records each sampling time t i Data is collected simultaneously from all sensors, ensuring that stress, velocity, and vibration data are collected at the same time. Stress data σ(t i ), velocity data v(t i ) and vibration acceleration data a(t iThese values ​​are obtained from the stress sensor, velocity sensor, and vibration sensor, respectively. Each data value can be expressed as follows:

number

[0108] Here, S(x,y,z,t i ), V(x,y,z,t i ) and A(x,y,z,t i ) are, respectively, time t i This represents stress, velocity, and vibration acceleration data located at the (x,y,z) coordinates.

[0109] Finally, multidimensional data synchronization processing, using time alignment techniques, enables the comparison and analysis of data collected at different frequencies on the same time axis. The multidimensional data D(t) after data synchronization is organized into a unified data structure, facilitating subsequent storage and processing.

number

[0110] In this data structure, D(t) represents the integrated data collected at time t and includes three parts: stress σ(t), velocity v(t), and vibration acceleration a(t). This step ensures coordination and consistency between the data and provides high-quality data input for subsequent analysis and modeling.

[0111] A multidimensional data framework is constructed on the multidimensional stress, velocity, and vibration data collected simultaneously, and the collected multidimensional stress, velocity, and vibration data is standardized and integrated using multimodal fusion technology. It synchronizes and stores data from different sensors, ensuring compatibility and coordination between different types of data, and provides complete and accurate data input for subsequent data analysis and prediction.

[0112] First, time alignment and standardization processes are performed on various data so that all data can be analyzed synchronously on the same time axis. Specifically, t i At this point in time, σ(t i ) is used as stress data, v(t i ) as velocity data, a(t i ) is used as vibration data, T(t i Set the temperature data to ). Standardize this data to a zero mean and unit variance to eliminate the effects of dimensional differences:

number

[0113] Here, μ σ , μ v , μ a , μ T These are the average values ​​of stress, velocity, vibration, and temperature data, respectively, and σ σ , σ v , σ a , σ T This is its standard deviation.

[0114] Next, the standardized data is integrated into a single unified three-dimensional data matrix, where the first dimension is time, the second dimension is sensor position, and the third dimension is data type. i position(x j ,y k ,z l The data points of ) can be represented as follows:

number

[0115] Here, D(t i ,x j ,y k ,z l ) is time t i and position (x j ,y k ,z l This is a multidimensional dataset in which standardized stress, velocity, vibration, and temperature data are included.

[0116] Finally, a multidimensional data table is created to structure and store this data, preparing it for subsequent analysis and processing. This tabular format can be considered as records in a database, with each record containing time, location coordinates, and various data values. The final tabular format is as follows:

[0117] [Table 5]

[0118] In S103, failure features are extracted and data analysis is performed from the obtained multidimensional dataset, and a self-adaptive failure model is constructed based on the extracted failure features. Then, real-time failure detection results are obtained by multimodal data fusion. For example, failure features of stress data, velocity data, and vibration data are extracted from the multidimensional dataset. It should be noted that by extracting features from these three types of data, it is possible to extract the most valuable information for failure prediction from a complex dataset, and this includes, specifically, the following.

[0119] Stress feature extraction: By analyzing stress data, the stress peak value σ is extracted. peak Features such as the stress gradient ∇σ are extracted. The stress peak feature represents the maximum stress the system experiences during the operation, and the stress gradient represents the rate of change of stress at different locations.

number

[0120] Here, σ(t) represents the stress value over time, and ∇σ represents the rate of change of stress in each spatial direction.

[0121] Velocity feature extraction: Analyze velocity data to extract the rate of change of velocity a(t) and the velocity peak value v. PEAKFeatures such as the following are extracted: The rate of change of velocity can identify the dynamic response characteristics of the system, and the velocity peak value represents the maximum velocity the system reaches during operation:

number

[0122] Here, v(t) is the velocity data, and a(t) represents the rate of change of velocity over time (acceleration).

[0123] Vibration feature extraction: Root mean square value a from vibration signal RMS And the frequency domain feature F(f) is extracted. The vibration mean square value a RMS It is generally used to evaluate the vibration intensity of a system, and the frequency domain feature F(f) can be used to analyze the frequency components of the signal:

number

[0124] Here, a(t) is vibration data, and a RMS F(f) represents the RMS value of the signal, and F(f) is the frequency domain feature of the vibration signal, which is obtained by calculating it using the Fourier transform (FFT).

[0125] This process involves extracting various features such as stress, velocity, and vibration from a multidimensional data framework, further selecting and optimizing these features to identify the most important features for failure prediction, thereby improving the accuracy and efficiency of the prediction model. First, correlation analysis is performed to identify which features are most closely related to failure occurrence. The main task of this step is to identify features highly relevant to the system's health, particularly failure prediction. Specific steps include:

[0126] Calculation of Pearson correlation coefficient: Pearson correlation coefficient r xy We use this method to measure the linear correlation between different failure characteristic data.

number

[0127] Here, r xy This represents the correlation coefficient between features x and y. JPEG0007868240000052.jpg127 and JPEG0007868240000053.jpg147 represents the average values ​​of each of these features. Features with values ​​close to 1 or -1 show a strong correlation, while values ​​close to 0 show a weak correlation.

[0128] A correlation calculation structure corresponding to all failure features is constructed in the correlation matrix, identifying interrelated failure features and failure features containing redundant information. Then, after completing the correlation analysis, the importance of each feature in failure prediction is further evaluated. This step is mainly completed by a feature importance algorithm, ensuring that the most predictive features are retained.

[0129] We will use the Random Forest algorithm to evaluate the importance of each failure feature in failure prediction:

number

[0130] Here, Importance(X j ) is feature X j This shows the importance of ΔI Gini This is a feature X for the t-th tree. j This is the Gini coefficient contribution, where T is the total number of trees in the random forest.

[0131] Principal Component Analysis (PCA): PCA is used to perform dimensionality reduction analysis by projecting high-dimensional feature data into a lower-dimensional space, extracting principal components, which reflect the main patterns of change in the data. Z=XW

[0132] Here, Z is the data matrix after dimensionality reduction, X is the original feature matrix, and W is the weight vector matrix of the feature matrix. PCA identifies the most important combination of features by maximizing the variance of the projected data.

[0133] Finally, based on the evaluation of the importance of the features, the most important features are selected and further optimized to improve the effectiveness of failure prediction.

[0134] Redundant feature removal: Based on correlation analysis and feature importance scoring, redundant features that are highly correlated with other features are removed to reduce model complexity and the risk of overfitting.

[0135] Feature set optimization: By combining feature importance and the results of principal component analysis, the final feature set is optimized to ensure that only the most predictive features remain. The optimized feature set can not only improve the model's predictive performance but also reduce computational costs.

[0136] Based on extracted failure characteristics, an Adaptive Fault Prediction Model (AFPM) is established. This model is then used to calculate the probability of failure in real time. By combining Bayesian networks and fuzzy logic techniques, AFPM can accurately assess failure risk in complex and uncertain environments and provide reliable probability outputs for subsequent intelligent alarms. The core task of AFPM is to extract and optimize key features through real-time processing to calculate the probability of system failure. These key features may include parameters such as stress, velocity, and vibration. AFPM performs probabilistic inference using a Bayesian network and combines it with fuzzy logic to process the fuzzyness of the input data, ultimately obtaining a probability value for failure.

[0137] First, perform Bayesian network modeling. The Bayesian network is a core component of AFPM and is used to estimate the probability of system failures based on historical data.

[0138] Node definition: In the Bayesian network, the important features extracted by S3 (such as the stress peak value σ PEAK , the rate of change of speed a(t), and the root mean square value of vibration a RMS , etc.) are regarded as input nodes, and the failure state is regarded as the output node.

[0139] Construction of the Conditional Probability Table (CPT): Use historical data to construct the Conditional Probability Table (CPT). These tables show the probability of failure under different feature conditions. For example, when a certain stress peak value σ PEAK exceeds a threshold, the probability of the system entering the failure state can be obtained from statistical data.

[0140] Bayesian inference: Calculate the posterior probability of failure occurrence using the Bayesian formula.

Number

[0141] Here, P(fault|data) is the probability of observing these feature data in a given failure state, P(fault) is the prior probability of failure occurrence, and P(data) is the total probability of observing these data.

[0142] Next, integrate the fuzzy logic system. To handle the uncertainty of input data, AFPM combines with the fuzzy logic system to process fuzzy data and improve the accuracy of failure probability evaluation.

[0143] Fuzzy variable definition: Convert important features into fuzzy variables. For example, the stress peak value σ PEAKConvert it into fuzzy sets (e.g., "low", "medium", "high"), and set the fuzzy membership function based on the distribution of the actual data.

[0144] Fuzzy rule base: Construct a fuzzy rule base and define the failure probability in different fuzzy states. For example, the rule may be set as "when σ PEAK is 'high' and a RMS is 'high', the failure probability is 'high'".

[0145] Fuzzy inference: Map the fuzzy variables to the failure probability by a fuzzy inference mechanism. Combine the output of the fuzzy inference with the probability result of the Bayesian network to obtain a more accurate failure probability estimation.

[0146] Finally, AFPM calculates the overall probability of system failure by integrating the inference results of the Bayesian network and the fuzzy logic system:

Number

[0147] Here, α is a weight coefficient, and its value range is 0 ≤ α ≤ 10, which is used to balance the influence on the final results of the Bayesian network and the fuzzy logic system. P BN (fault) is the failure probability inferred by the Bayesian network. P FL (fault) is the failure probability by the fuzzy logic system.

[0148] This failure probability value P AFPM (fault) is an important input for subsequent decisions and the early warning system, and is used to generate specific warning information and maintenance suggestions in subsequent steps.

[0149] Multimodal fusion technology comprehensively analyzes stress data, velocity data, and vibration data to obtain fault detection results and warning information, and adaptively adjusts warning levels based on different fault types, providing important reference information for system maintenance and optimization.

[0150] First, the failure probability P calculated by AFPM AFPM The (fault) is transmitted to the Intelligent Warning Algorithm (MDLSWA, Multi-Dimensional Learning with Sparse Representation Warning Algorithm). MDLSWA analyzes this probability and determines whether or not it is necessary to generate warning information. The criterion for this decision is generally a predetermined threshold P where the probability of failure is certain. threshold It is about exceeding: IfP AFPM (fault)> P threshold , then generate a warning.

[0151] If the failure probability exceeds a threshold, MDLSWA generates a preliminary warning message, displaying the type and severity of the potential failure. Next, regarding multimodal data fusion, to improve the accuracy and real-time nature of the warning message, MDLSWA uses cross-modal data fusion technology to comprehensively analyze multimodal data from different sensors (e.g., stress, velocity, vibration, etc.). This process includes the following:

[0152] Data Synchronization and Fusion: MDLSWA uses cross-modal data fusion technology to synchronize data from different modal systems (stress, velocity, vibration, etc.) in time and space, and integrates them with their respective feature weights.

[0153] Deep Learning Analysis: The fused data is input into a deep learning model, which has already been trained and can identify characteristic modes in different failure modes. By analyzing the fused data, the model generates more accurate failure predictions and warning information.

[0154] Next, regarding self-adaptive warning level adjustment, MDLSWA not only generates basic warning information, but also adaptively adjusts the warning level based on the type and severity of the fault.

[0155] Warning Level Setting: Based on the type of failure, possible impact, and current system state output from the deep learning model, MDLSWA categorizes warning information into different levels (e.g., "Low", "Medium", "High").

[0156] Real-time adjustment: During system operation, MDLSWA dynamically adjusts warning levels based on real-time data changes. If the system condition deteriorates or a more serious failure mode is detected, the warning level is automatically increased.

[0157] Ultimately, the smart alert information generated by MDLSWA includes the following:

[0158] Failure type identification: This specifies the type of failure that is likely to occur (e.g., stress fatigue, vibration anomaly, etc.).

[0159] Failure probability: Overall failure probability P AFPM Display (fault).

[0160] Warning Level: Displays the severity of the warning (e.g., "Low Risk," "Medium Risk," "High Risk").

[0161] In S104, the stress sensor, velocity sensor, and vibration sensor are optimized using automatic calibration and optimization of reinforcement learning based on the real-time fault detection results. For example, automatic calibration and optimization are performed on the stress sensor, velocity sensor, and vibration sensor using automatic calibration and optimization of reinforcement learning based on real-time fault detection results and warning information.

[0162] Based on the warning information provided by MDLSWA, Reinforcement-Learning-based Automatic Calibration and Optimization (RLABO) automatically calibrates the sensor to ensure its accuracy in future operations. RLABO further optimizes the calibration policy through reinforcement learning, providing assurance for data collection and analysis in the next operating cycle, thereby reinforcing the long-term stable operation of the CT20 spring actuation mechanism.

[0163] After MDLSWA generates warning information, it automatically initiates the sensor calibration flow based on the warning level and type. RLABO uses real-time learning to incrementally adjust the sensor calibration parameters in the operating environment. The state space of the reinforcement learning model includes the sensor's current calibration state, environmental conditions, and historical data, while the operation space concerns the adjustable calibration parameters, such as the sensor's zero offset and sensitivity adjustment. The effectiveness of the calibration is measured by a reward function, R(s, a), which is determined by the accuracy of the sensor data and the accuracy of system detection after calibration. The formula is as follows:

number

[0164] Here, D i actual This is the sensor data after calibration, D iexpected is ideal sensor data, c(a) is the cost of performing calibration operation a, and λ is a coefficient that trades off accuracy and cost.

[0165] Based on the decision results of reinforcement learning, RLABO automatically adjusts sensor parameters, ensures high precision in its future operations, records the updated parameters, and provides reference for the operations of the next cycle. With the continuous collection of data, RLABO can continuously optimize the calibration policy and keep the system in an optimal state under different operating conditions.

[0166] Closed-loop intelligent optimization: After the maintenance is completed, the Closed-Loop Intelligent Optimization Algorithm (CLIOA) module is activated after the automatic calibration of the sensor is completed, and global optimization is performed on the entire fault detection system. CLIOA automatically adjusts algorithm parameters and detection flow by reviewing historical data and predicting future data. First, a genetic algorithm is applied, where each individual represents a parameter combination of one fault detection model, such as feature weights and model hyperparameters, and the fitness function is determined by the fault detection accuracy and operation efficiency of the model. The formula is as follows:

Equation

[0167] Here, x represents the parameter set of the model, and β is a coefficient that trades off accuracy and computational cost. Next, CLIOA further adjusts the model parameters using the Particle Swarm Optimization (PSO) algorithm. Each particle represents one possible parameter setting, and the update rule is as follows:

Equation

[0168] Here, v i (t) is the velocity of particle i, and x i (t) is the position of the particle, p i is the individual optimal position of the particle, g is the global optimal position, ω, c1, and c2 are control parameters, and r1, r2 are random numbers. CLIOA combines a genetic algorithm and particle swarm optimization, performs multiple iterations to identify optimal parameter settings, and enables the fault detection system to maintain the most efficient operation under different operating conditions.

[0169] Finally, CLIOA updates all system parameters and saves the optimization results, enabling the entire system to operate with the highest efficiency and accuracy in future operating cycles.

[0170] The second aspect of the embodiment of the present invention provides a fault detection and optimization system for a spring operating mechanism, which, as shown in FIG. 5, includes the following: A collection module for collecting important data of the spring operating mechanism in real time, and A framework construction module for constructing a multi-dimensional data framework and obtaining a multi-dimensional data set based on the collected important data, and A fault feature extraction module for performing fault feature extraction from the obtained multi-dimensional data set, and A data analysis module for performing data analysis on the multi-dimensional data set, and A model construction module for constructing a self-adaptive fault model, and A reinforcement learning module for optimizing a stress sensor, a speed sensor, and a vibration sensor, characterized by a fault detection and optimization system for a spring operating mechanism.

[0171] A third embodiment of the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor performs the steps of the fault detection and optimization method for the spring operating mechanism described above when executing the computer program.

[0172] A fourth embodiment of the present invention provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the fault detection and optimization method for the spring operating mechanism described above is realized.

[0173] Finally, it should be noted that the above embodiments are merely for illustrating the technical configuration of the present invention and do not limit the scope of protection of the present invention. Although the present invention has been described in detail based on the above embodiments, it will be obvious to those skilled in the art that various changes, modifications, or equivalent substitutions can be made to specific embodiments based on the disclosure of the present invention. All of these changes, modifications, or equivalent substitutions fall within the scope of protection of the present invention as described in the claims.

Claims

1. A method for detecting and optimizing a spring operating mechanism, Using stress sensors, velocity sensors, and vibration sensors, we collect important data on the spring operating mechanism, such as stress data, velocity data, and vibration data, in real time. Based on the collected important data, a multidimensional data framework is constructed that includes a three-dimensional data matrix with time, sensor position, and data type as axes, and a multidimensional dataset is obtained that consists of a multidimensional data table generated based on the three-dimensional data matrix. The method involves extracting failure features and performing data analysis from the obtained multidimensional dataset, constructing a self-adaptive failure model that estimates the probability of failure based on the extracted failure features, and then obtaining real-time failure detection results by multimodal data fusion that integrates the feature quantities of the stress data, velocity data, and vibration data. Based on the real-time fault detection results, the stress sensor, velocity sensor, and vibration sensor are optimized using automatic calibration and optimization through reinforcement learning. Based on the collected important data, constructing a multidimensional data framework and obtaining a multidimensional dataset involves, after the placement of stress sensors, velocity sensors, and vibration sensors is complete, utilizing a multi-frequency self-adaptive sampling algorithm to dynamically adjust the sampling period of important data based on the placement positions of the stress sensors, velocity sensors, and vibration sensors, so that multidimensional data of stress, velocity, and vibration are collected simultaneously. A multidimensional data framework is constructed on the multidimensional stress, velocity, and vibration data collected simultaneously, and the collected multidimensional stress, velocity, and vibration data is standardized using multimodal fusion technology. The standardized multidimensional data is integrated into a three-dimensional data matrix, where the first dimension is time, the second dimension is sensor position, and the third dimension is data type. Based on the aforementioned three-dimensional data matrix, a multi-dimensional data table is generated, and a multi-dimensional dataset is obtained. Includes, A method for detecting and optimizing a spring operating mechanism, characterized in that optimization here refers to adjusting the entire process, from data acquisition to failure feature extraction, self-adaptive failure model construction, and sensor operation, to the best possible state through reinforcement learning.

2. Before collecting important data on the spring operating mechanism using the aforementioned stress sensor, The geometric model of the spring operating mechanism is constructed, the material attributes of each component in the geometric model of the spring operating mechanism are defined, and the stress distribution and deformation behavior of the spring operating mechanism under operating load are simulated using finite element analysis. Based on the actual operating state of the spring operating mechanism, load conditions and boundary conditions are set, and the geometric model of the spring operating mechanism is meshed to generate a basic contour map of the stress distribution. Using a self-adaptive stress sensing algorithm, the stress distribution data on the contour plot of the stress distribution is analyzed to automatically identify stress concentration regions and obtain the gradient value of each stress concentration point and regions where stress changes are significant in the stress distribution. Clustering is performed on the identified stress concentration points, and adjacent high-stress gradient points are combined into a single stress concentration region. The process involves applying a self-adaptive stress sensing algorithm to refine the boundaries of stress concentration regions, identifying areas of high stress distribution, and determining the placement of the stress sensor in the spring operating mechanism using a genetic optimization algorithm. The method for detecting and optimizing a spring operating mechanism according to claim 1, further comprising the above.

3. Before collecting important data on the spring operating mechanism using the aforementioned speed sensor, Using already placed strain sensors, initial motion data of the spring operating mechanism under different operating conditions is acquired, and based on the initial motion data, a spring motion model of the spring operating mechanism is established, and the dynamic behavior of the spring at rest, in motion, and under different operating conditions is simulated. The system employs a multimodal motion analysis algorithm to acquire the motion trajectory of the spring in the spring operating mechanism in real time, acquire visual data, and extract important motion features from the visual data. Using the established spring motion model, the motion parameters of the spring are extracted, and the visual data and the motion parameters are fused using a Kalman filter. The method for detecting and optimizing a spring operating mechanism according to claim 1, further comprising the above.

4. Before collecting important data on the spring operating mechanism using the aforementioned vibration sensor, Using conventional vibration sensors or temporarily placed vibration sensors, vibration data is collected under different operating conditions of the spring operating mechanism. By combining vibration data with stress data collected by a stress sensor and velocity data collected by a velocity sensor, the dynamic behavior of the spring operating mechanism under different operating conditions is analyzed using multimodal fusion technology. Using a multimodal motion analysis algorithm, we can identify the vibration modes of a spring under different operating modes, This involves comparing vibration modes in different operating modes, identifying the operating mode with the most intense vibration, and determining the placement of the vibration sensor in the spring operating mechanism using a genetic optimization algorithm. The method for detecting and optimizing a spring operating mechanism according to claim 1, further comprising the above.

5. The process involves extracting failure features and performing data analysis from the obtained multidimensional dataset, constructing a self-adaptive failure model based on the extracted failure features, and then obtaining real-time failure detection results through multimodal data fusion. This involves extracting failure features from stress, velocity, and vibration data from a multidimensional dataset, measuring the linear correlation between different failure feature data using the Pearson correlation coefficient, constructing a correlation matrix from the correlation calculation results corresponding to all failure features, and identifying interrelated failure features and failure features containing redundant information. Using a random forest algorithm, the importance of each failure feature in failure prediction is evaluated, and based on the evaluation results of the failure feature importance, the most important failure feature is selected to optimize the failure prediction effect. Based on the extracted failure characteristics, a self-adaptive failure model is established, and the probability of failure occurring is calculated in real time using the self-adaptive failure model. Based on the probability of failure, fault detection results and warning information are obtained using multimodal fusion technology. A method for detecting and optimizing a spring operating mechanism according to claim 1, characterized by including the following:

6. Based on the real-time fault detection results, the stress sensor, velocity sensor, and vibration sensor are optimized using automatic calibration and optimization of reinforcement learning. The fault detection and optimization method for a spring operating mechanism according to claim 5, characterized in that automatic calibration and optimization of a stress sensor, a velocity sensor, and a vibration sensor are performed using automatic calibration and optimization of reinforcement learning based on real-time fault detection results and warning information.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor performs the steps of the fault detection and optimization method for a spring operating mechanism described in any one of claims 1 to 6 when executing the computer program.

8. A computer-readable storage medium wherein a computer program is stored in the computer-readable storage medium, and the computer program, when executed by a processor, realizes the steps of the fault detection and optimization method for a spring operating mechanism described in any one of claims 1 to 6.

Images