Methods and systems for detecting meshing impact vibration and predicting life of SSS clutches

Abstract

This invention relates to a method and system for detecting engagement impact vibration and predicting the lifespan of an SSS clutch. The method includes: acquiring a clutch model architecture diagram of the SSS clutch to deploy a composite sensor group and collect engagement impact signals; simultaneously acquiring associated signal parameters to obtain a multi-source impact signal set; performing compressed wavelet transform processing on the multi-source impact signal set and conducting joint time-frequency analysis to obtain the joint time-frequency analysis results of the clutch; pre-constructing a lifespan prediction model based on a deep metric learning network, iteratively training the lifespan prediction model based on historical joint time-frequency analysis results, and outputting a converged lifespan prediction model; acquiring the joint time-frequency analysis results in real time and inputting them into the lifespan prediction model, and outputting the lifespan prediction results. Compared with existing technologies, this invention can comprehensively and accurately capture key characteristic information of impact vibration during clutch engagement, achieving precise detection of abnormal impact vibration.

Description

Technical Field

[0001] This invention relates to the field of clutch life prediction technology, and in particular to a method and system for detecting engagement impact vibration and predicting life of SSS clutches. Background Technology

[0002] The SSS clutch (Synchronized Self-Shifting Clutch) is widely used in high-power transmission systems such as marine propulsion systems, large industrial transmission devices, and wind turbine generators. It achieves smooth power transmission and rapid switching through the meshing of a sleeve and gears. As the core transmission components of the SSS clutch, the impact and vibration characteristics of the sleeve and gears during meshing directly affect the clutch's operational stability, transmission efficiency, and service life.

[0003] In actual operation, due to factors such as manufacturing errors, assembly deviations, poor lubrication, or sudden load changes, the gear sleeve and gear are prone to unstable impacts and high-frequency vibrations during meshing. These impacts and vibrations not only exacerbate failure modes such as tooth surface wear and tooth root fatigue cracks, but may also lead to clutch engagement failure or even cause the entire transmission system to malfunction. Therefore, accurately detecting the impact and vibration signals during the meshing process of the gear sleeve and gear, assessing their health status, and predicting their remaining life are of great significance for ensuring the safe operation of the SSS clutch and achieving predictive maintenance.

[0004] In response, the invention disclosed in CN118565813A presents an online method and apparatus for determining the engagement position of an automatic synchronous shifting clutch. The method includes: obtaining a first vibration phase based on a first key phase vibration monitoring system; obtaining a second vibration phase based on a second key phase vibration monitoring system; and calculating the engagement angle between the high-pressure rotor and the low-pressure rotor according to the first vibration phase, the second vibration phase, and the engagement angle calculation formula. The first key phase vibration monitoring system and the second key phase vibration monitoring system are used to detect the vibration phases of the high-pressure rotor and the low-pressure rotor of the target steam turbine, respectively.

[0005] However, the existing methods mentioned above monitor vibration phase data. Phase monitoring alone cannot capture the key characteristics of complex impacts such as tooth surface collision and friction that accompany the meshing process. It is difficult to fully reflect the abnormality of the meshing state, resulting in insufficient strength for clutch fault diagnosis and performance evaluation. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art, which is that it is difficult to fully reflect the abnormality of the meshing state, resulting in insufficient fault diagnosis and performance evaluation of the clutch, and to provide a method and system for detecting meshing impact vibration and predicting life of SSS clutch.

[0007] The objective of this invention can be achieved through the following technical solutions: A method for detecting engagement impact vibration and predicting life of an SSS clutch includes: Obtain the clutch model architecture diagram of the SSS clutch to be predicted, and based on the clutch model architecture diagram, deploy a composite sensor group in the gear sleeve and gear meshing area of ​​the clutch, and synchronously collect meshing impact signals through the composite sensor group. While acquiring meshing impact signals in real time, associated signal parameters are synchronously acquired based on the meshing impact signal acquisition cycle. The associated signal parameters and meshing impact signals are then time-stamped to obtain a multi-source impact signal set. The multi-source impact signal set is subjected to compressed wavelet transform processing and joint time-frequency analysis to obtain the joint time-frequency analysis results of the clutch. The joint time-frequency analysis results are then uploaded to the historical database. A lifetime prediction model based on a deep metric learning network is pre-constructed. This lifetime prediction model is used to extract features based on the joint time-frequency analysis results to obtain the gear sleeve-gear meshing state vector, thereby predicting the probability distribution of the remaining lifetime of the gear sleeve and gear meshing and outputting the final remaining lifetime. Historical joint time-frequency analysis results with remaining lifetime labels are crawled from the historical database. The lifetime prediction model is iteratively trained based on the historical joint time-frequency analysis results and outputs a converged lifetime prediction model. The joint time-frequency analysis results are acquired in real time and input into the lifetime prediction model, and the lifetime prediction results are output.

[0008] Furthermore, the process of determining the arrangement position of the composite sensor group includes: Traverse the clutch model architecture diagram, using the gear sleeve and gear as the topology nodes of the clutch static topology network, extract the connection relationship between the topology nodes, construct the associated edges through the connection relationship between the topology nodes, and abstract the intersection of the associated edges as auxiliary nodes; Identify the topological nodes and auxiliary nodes of the clutch static topology network. Using the number of gear sleeves and gears associated with the topological nodes and auxiliary nodes as a priori conditions, assign weights to the topological nodes and auxiliary nodes based on principal component analysis to obtain node weights. The node selection threshold is dynamically determined based on the node weight. It is then determined whether the node weight exceeds the preset node selection threshold. If the node weight exceeds the preset node selection threshold, the current topology node or auxiliary node is retained. If the node weight does not exceed the preset node selection threshold, the current topology node or auxiliary node is discarded. The composite sensor group is deployed based on the retained topology nodes and auxiliary nodes.

[0009] Furthermore, the node weights are obtained by weighted summation based on the node's association complexity, structural degrees of freedom, and vibration transmission factor variance contribution rate. The association complexity is the result of multiplying the number of gear sleeves and gears associated with the current node by the topological distance between the current node and the center node of the clutch static topology network, and dividing by the total number of nodes in the clutch static topology network. The vibration transmission factor variance contribution rate is the total number of nodes in the clutch static topology network divided by the average distance between the current node and the associated gear sleeve and gear.

[0010] Furthermore, the expression for calculating the node weight is as follows: In the formula, For the first i The node weight of each node. They are nodes The correlation complexity, structural degrees of freedom, and vibration transfer factor, These represent the variance contribution rates of the correlation complexity, structural degrees of freedom, and vibration transfer factor, respectively, obtained based on principal component analysis. They are nodes The number of associated gear sleeves and gears, the topological distance from the central node of the clutch static topology network, and the total number of nodes; For nodes Dynamic response sensitivity, For nodes The average distance to the associated gear sleeve or gear.

[0011] Furthermore, the compressed wavelet transform processing of the multi-source impulse signal set includes: A multi-source signal impulse set is loaded, and compressed wavelet transform processing is performed on the multi-source impulse signal set. A frequency weighting function based on meshing impulse prior is introduced to weight the multi-source impulse signal after compressed wavelet transform processing to obtain a multi-source weighted signal. The multi-source weighted signal is bandpass filtered to output the bandpass-filtered multi-source weighted signal. Based on the acquisition period of the meshing impact signal, the bandpass-filtered multi-source weighted signal is resampled and superimposed to obtain the derived periodic impact characteristics. Time-domain feature analysis, frequency-domain feature analysis, and joint time-frequency-domain feature analysis were performed on the derived periodic impact characteristics and multi-source weighted signals respectively to extract impact pulse characteristics, meshing frequency characteristics, and wavelet packet decomposition characteristics. The impact pulse characteristics, engagement frequency characteristics, and wavelet packet decomposition characteristics are mapped into a low-dimensional normalized space to obtain the joint time-frequency analysis results of the clutch.

[0012] Furthermore, the time-domain features processed by the time-domain feature analysis include: pulse peak value, rise time, decay time, and pulse width; The frequency domain features analyzed include: the meshing frequency and its harmonics and sideband frequencies extracted by Fourier transform; The time-frequency domain features jointly analyzed and processed include: dividing the signal into sub-bands through wavelet packet decomposition, thereby extracting the energy entropy and energy proportion of each sub-band.

[0013] Furthermore, the lifespan prediction model includes an input layer, a deep metric learning network, and an output layer connected in sequence. A feature extraction layer is provided between the input layer and the deep metric learning network, and a lifespan probability distribution layer is provided between the deep metric learning network and the output layer. The feature extraction layer is used to extract features from the joint time-frequency analysis results based on the clutch physical degradation mechanism, and outputs the gear sleeve-gear meshing state vector; The deep metric learning network is used to retrieve the standard impact vibration reference vector based on the gear sleeve-gear meshing state vector, and calculate the state Mahalanobis distance between the gear sleeve-gear meshing state vector and the impact vibration reference vector; based on the state Mahalanobis distance, a multi-stage degradation physical equation considering vibration impact energy and cumulative tooth surface damage is constructed, and the impact vibration index is obtained by solving the multi-stage degradation physical equation; the impact vibration index is dynamically weighted based on a multi-modal attention mechanism, and the state evaluation equation is generated based on the weighted impact vibration index as input and an adaptive particle filter algorithm is used. The lifetime probability distribution layer is used to adaptively adjust the prediction error through the covariance matching method and output the remaining lifetime probability distribution of the gear sleeve-gear meshing. The input layer is used to calculate the remaining life of the gear sleeve-gear meshing based on the probability distribution of the remaining life of the gear sleeve-gear meshing.

[0014] Furthermore, the calculation expression for the remaining service life of the gear sleeve-gear meshing is as follows: In the formula, These represent the remaining life of the gear sleeve-gear meshing, the mean of the remaining life probability distribution, and the variance of the remaining life probability distribution, respectively. is the confidence coefficient.

[0015] The present invention also provides a meshing impact vibration detection and life prediction system for implementing the above-described method for detecting and predicting the meshing impact vibration of an SSS clutch, comprising: The sensor monitoring deployment module is used to obtain the clutch model architecture diagram of the SSS clutch to be predicted. Based on the clutch model architecture diagram, a composite sensor group is deployed in the gear sleeve and gear meshing area of ​​the clutch. The meshing impact signal is collected synchronously through the composite sensor group and uploaded to the edge processing node. The edge processing node is used to acquire the meshing impact signal in real time, and simultaneously acquire related signal parameters based on the meshing impact signal acquisition cycle. It performs timestamp synchronization processing on the related signal parameters and the meshing impact signal to obtain a multi-source impact signal set. The time-frequency analysis module is used to perform compressed wavelet transform processing on the multi-source impact signal set and perform joint time-frequency analysis to obtain the joint time-frequency analysis results of the clutch, and upload the joint time-frequency analysis results to the historical database. The model building module is used to pre-build a lifetime prediction model based on a deep metric learning network. This lifetime prediction model is used to extract features based on the joint time-frequency analysis results to obtain the gear sleeve-gear meshing state vector, thereby predicting the remaining lifetime probability distribution of the gear sleeve and gear meshing and outputting the final remaining lifetime. Historical joint time-frequency analysis results with remaining lifetime labels are crawled from the historical database. The lifetime prediction model is iteratively trained based on the historical joint time-frequency analysis results and outputs a converged lifetime prediction model. The lifetime prediction module is used to input the real-time acquired joint time-frequency analysis results into the lifetime prediction model and output the lifetime prediction results.

[0016] Furthermore, the time-frequency analysis module includes: The wavelet transform unit is used to load the multi-source signal impulse set, perform compressed wavelet transform processing on the multi-source impulse signal set, and introduce a frequency weighting function based on meshing impulse prior to weight the multi-source impulse signal after compressed wavelet transform processing to obtain a multi-source weighted signal. The derived feature unit is used to perform bandpass filtering on the multi-source weighted signal, output the bandpass-filtered multi-source weighted signal, and resample and superimpose the bandpass-filtered multi-source weighted signal based on the acquisition period of the meshing impact signal to obtain the derived periodic impact feature. The time-frequency extraction unit is used to perform time-domain feature analysis, frequency-domain feature analysis, and joint time-frequency-domain feature analysis on the derived periodic impact features and multi-source weighted signals, respectively, to extract impact pulse features, meshing frequency features, and wavelet packet decomposition features; The feature mapping unit is used to map the impact pulse features, engagement frequency features, and wavelet packet decomposition features into a low-dimensional normalized space to obtain the joint time-frequency analysis results of the clutch.

[0017] Compared with the prior art, the present invention has the following advantages: (1) The acquisition and fusion method of multi-source data such as meshing impact signals and associated signal parameters in this invention can comprehensively and accurately capture key characteristic information of impact vibration during clutch meshing. Furthermore, the life prediction model can achieve accurate detection of impact vibration anomalies based on real-time joint time-frequency analysis results and dynamic weighting using a multimodal attention mechanism, and further perform life prediction analysis to accurately obtain the probability distribution of the remaining life of the gear sleeve-gear meshing, thereby calculating the remaining life and making the life prediction results more accurate and reliable. This method, which combines physical models and data-driven approaches, fully leverages the advantages of both and improves the accuracy and adaptability of life prediction.

[0018] (2) When the composite sensor group is deployed in the gear sleeve and gear meshing area of ​​the clutch based on the clutch model architecture diagram, the screening threshold is dynamically adjusted based on the distribution of node weights. This can avoid the problem of a one-size-fits-all approach with a fixed threshold. The dynamic threshold screening minimizes the number of sensors, reduces hardware costs, wiring complexity and data processing volume, and improves the real-time performance and reliability of the monitoring system while ensuring that key vibration information is not lost. At the same time, the optimized sensor layout can concentrate the sensors on the key path and sensitive point of vibration transmission, resulting in a higher signal-to-noise ratio and richer impact characteristics.

[0019] (3) When performing compressed wavelet transform on a multi-source impact signal set, this invention reduces the decomposition calculation of redundant frequency bands by compressing the wavelet basis function or decomposition scale, while preserving the time-varying details of the impact signal, thereby improving computational efficiency and feature resolution. The design of the frequency weighting function strengthens the energy proportion of the impact feature frequency band and suppresses noise interference from irrelevant frequency bands. The combination of compressed wavelet transform and prior weighting makes the time-frequency energy distribution of the multi-source impact signal clearer, allowing key features to be highlighted. The joint extraction of multi-dimensional features can fully characterize the non-stationary characteristics of the impact signal, avoiding the one-sidedness of single-domain analysis. This allows the integration of time-domain, frequency-domain, and time-frequency-domain features of multi-source signals to cover the multi-dimensional information such as the intensity, period, and energy distribution of the impact signal, avoiding misjudgment due to the lack of a single feature.

[0020] (4) When the life prediction model of the present invention performs impact vibration detection on the clutch based on the joint time-frequency analysis results, the feature extraction guided by the physical degradation mechanism ensures the high correlation between the state vector and the actual degradation process of the clutch, avoids the interference of false features, and enables the subsequent model to capture the degradation trend more accurately. Based on the Mahalanobis distance and physical mechanism, a staged degradation equation is constructed, which can accurately describe the damage evolution rate of different degradation stages. The impact vibration index obtained by physical equation inversion can transform the abstract "state deviation" into a quantifiable degradation index. The introduction of Mahalanobis distance improves the robustness of the state deviation measurement, and the integration of physical equation constrains the degradation logic of the model. The combination of the two enables the model to accurately track the degradation trajectory under complex working conditions.

[0021] (5) When performing life prediction analysis on the impact vibration detection results based on the life prediction model, this invention assigns weights to different impact vibration indicators according to their dynamic changes in contribution to degradation, avoiding misjudgment of indicator importance caused by fixed weights. Furthermore, by using a particle filtering algorithm to deduce the probability distribution of future states, combined with an adaptive mechanism, the model's ability to model nonlinear and non-Gaussian degradation processes is improved. Finally, the probabilistic output and adaptive adjustment mechanism make the remaining life prediction results more reliable. Accurate remaining life prediction can provide early warning of potential clutch failures, helping maintenance personnel to plan maintenance schedules in advance. This preventative maintenance method can reduce downtime and maintenance costs caused by sudden failures, improving equipment operating efficiency and economic benefits. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating a method for detecting engagement impact vibration and predicting life of an SSS clutch provided in Embodiment 1 of the present invention. Figure 2 This is a schematic diagram illustrating the implementation process of a method for deploying a composite sensor group in the gear sleeve and gear meshing area of ​​a clutch based on a clutch model architecture diagram, as provided in Embodiment 1 of the present invention. Figure 3 This is a schematic diagram of the implementation process of a method for compressing wavelet transform of a multi-source impulse signal set provided in Embodiment 1 of the present invention; Figure 4 This is a schematic diagram illustrating the implementation process of a life prediction model for impact vibration detection of a clutch based on joint time-frequency analysis results, as provided in Embodiment 1 of the present invention. Figure 5 This is a schematic diagram of the implementation process of a method for life prediction analysis of impact vibration detection results based on a life prediction model provided in Embodiment 1 of the present invention; Figure 6 This is a schematic diagram of the structure of an SSS clutch engagement impact vibration detection and life prediction system provided in Embodiment 2 of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0024] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0025] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0026] Example 1 Existing methods monitor vibration phase data. Phase monitoring alone cannot capture the key characteristics of complex impacts such as tooth surface collisions and friction accompanying the meshing process, making it difficult to comprehensively reflect abnormalities in the meshing state. This results in insufficient capacity for clutch fault diagnosis and performance evaluation. To address these issues, this embodiment proposes a method for detecting meshing impact vibration and predicting the lifespan of an SSS clutch. Figure 1 As shown, it includes: S1: Obtain the clutch model architecture diagram of the SSS clutch to be predicted, and deploy a composite sensor group in the gear sleeve and gear meshing area of ​​the clutch based on the clutch model architecture diagram, and synchronously collect meshing impact signals through the composite sensor group. S2: While acquiring meshing impact signals in real time, associated signal parameters are synchronously acquired based on the meshing impact signal acquisition cycle. The associated signal parameters and meshing impact signals are time-stamped and synchronized to obtain a multi-source impact signal set. S3: Perform compressed wavelet transform processing on the multi-source impact signal set and perform joint time-frequency analysis to obtain the joint time-frequency analysis results of the clutch. Upload the joint time-frequency analysis results to the historical database. S4: A lifetime prediction model based on a deep metric learning network is pre-constructed. This lifetime prediction model is used to extract features based on the joint time-frequency analysis results to obtain the gear sleeve-gear meshing state vector, thereby predicting the remaining lifetime probability distribution of the gear sleeve and gear meshing and outputting the final remaining lifetime. Historical joint time-frequency analysis results with remaining lifetime labels are crawled from the historical database. The lifetime prediction model is iteratively trained based on the historical joint time-frequency analysis results and outputs a converged lifetime prediction model. S5: Real-time acquisition of joint time-frequency analysis results, input into the lifetime prediction model, and output lifetime prediction results.

[0027] In short, the above method first deploys a composite sensor group in the gear sleeve and gear meshing area of ​​the clutch based on the clutch model architecture diagram. The associated signal parameters and meshing impact signals are time-stamped and synchronized. The multi-source impact signal set is processed by compressed wavelet transform and joint time-frequency analysis is performed to obtain the joint time-frequency analysis results of the clutch. The life prediction model performs impact vibration detection on the clutch based on the joint time-frequency analysis results. Based on the life prediction model, the life prediction analysis is performed on the impact vibration detection results to obtain the remaining life probability distribution of the gear sleeve-gear meshing. The remaining life of the gear sleeve-gear meshing is calculated based on the remaining life probability distribution of the gear sleeve-gear meshing.

[0028] The above-mentioned method for detecting and predicting the life of SSS clutch engagement impact vibration specifically includes: S10, Obtain the clutch model architecture diagram. Based on the clutch model architecture diagram, deploy a composite sensor group in the gear sleeve and gear meshing area of ​​the clutch. The composite sensor group synchronously collects meshing impact signals and uploads them to the edge processing node 200. The edge processing node 200 is a computing node deployed near the data acquisition end, used for preliminary processing and analysis of the collected raw data. In this embodiment, the edge processing node 200 is responsible for receiving the meshing impact signals collected by the composite sensor group and synchronously collecting associated signal parameters. It performs timestamp synchronization processing on these data to generate a multi-source impact signal set. The composite sensor group is a sensor array composed of various types of sensors, used to simultaneously collect multiple physical quantities. The composite sensor group includes, but is not limited to: accelerometers (used to measure vibration acceleration during clutch engagement and capture impact vibration signals); strain sensors (measure strain generated in clutch components during engagement, reflecting structural stress); temperature sensors (monitor temperature changes in the clutch working area; abnormal temperatures may indicate overheating or wear of components); and pressure sensors (measure contact pressure between components during engagement to understand the engagement state). Displacement sensor: Measures the relative displacement between the gear sleeve and the gear to assess meshing accuracy. Acoustic sensor: Captures acoustic signals during meshing to analyze noise levels and abnormal sounds. S20, the edge processing node 200 acquires the meshing impact signal in real time, and synchronously acquires the associated signal parameters based on the meshing impact signal acquisition cycle. It performs time-stamp synchronization processing on the associated signal parameters and the meshing impact signal to obtain a multi-source impact signal set. The meshing impact signal refers to the vibration signal generated during the meshing process of the clutch sleeve and gear, while the associated signal parameters include, but are not limited to, speed signal, torque signal, temperature signal, pressure signal, and displacement signal. S30, load the multi-source impact signal set, perform compressed wavelet transform processing on the multi-source impact signal set, and perform joint time-frequency analysis to obtain the joint time-frequency analysis results of the clutch, and upload the joint time-frequency analysis results to the historical database; S40: A life prediction model based on a deep metric learning network is pre-built, historical time-frequency analysis results are retrieved from a historical database, the life prediction model is iteratively trained based on the historical time-frequency analysis results, and a converged life prediction model is output. The S50 acquires the joint time-frequency analysis results in real time. The life prediction model performs impact vibration detection on the clutch based on the joint time-frequency analysis results and outputs the impact vibration detection results.

[0029] S60, Load the impact vibration test results, perform life prediction analysis on the impact vibration test results based on the life prediction model, obtain the remaining life probability distribution of the gear sleeve-gear meshing, and calculate the remaining life of the gear sleeve-gear meshing based on the remaining life probability distribution of the gear sleeve-gear meshing.

[0030] This invention provides a method for deploying a composite sensor array in the gear sleeve and gear meshing area of ​​a clutch based on a clutch model architecture diagram. Figure 2 This document illustrates a flowchart of a method for deploying a composite sensor array in the gear sleeve and gear meshing area of ​​a clutch based on a clutch model architecture diagram. The method specifically includes: S101, Traverse the clutch model architecture diagram, take the gear sleeve-gear as the topology node of the clutch static topology network, extract the connection relationship between the topology nodes, construct the associated edge through the connection relationship between the topology nodes, and abstract the intersection of the associated edge as auxiliary node. The intersection of the associated edge is the superposition area of ​​multi-source vibration or the weak area of ​​the structure. Abstracting it as an auxiliary node can cover these key positions that are easily ignored, and avoid blind spots of vibration information. S102: Identify the topological nodes and auxiliary nodes of the clutch static topology network. Using the number of gear sleeves / gears associated with the topological nodes and auxiliary nodes as a priori conditions, assign weights to the topological nodes and auxiliary nodes based on principal component analysis to obtain node weights. The node weights quantify the contribution of each node to vibration monitoring, avoiding the inefficient deployment of average force and ensuring that high-weight nodes (nodes with high association complexity, high structural freedom, and strong vibration transmission factor) are preferentially retained, thereby improving the cost-effectiveness of signal acquisition. S103, dynamically determine the node filtering threshold based on the node weight, and determine whether the node weight exceeds the preset node filtering threshold. The formula for calculating node weights is as follows: In the formula, Represents a node The node weights, They are nodes The associated complexity, structural degrees of freedom, and vibration transfer factor, while These are the correlation complexity, structural degrees of freedom, and vibration transfer factor variance contribution rates obtained based on principal component analysis. in, Representing nodes respectively The number of associated gear sleeves / gears, the topological distance from the center node of the clutch static topology network, and the total number of nodes. Represents a node Dynamic response sensitivity, Represents a node The average distance to the associated gear sleeve / gear; S104, If the node weight exceeds the preset node filtering threshold, retain the current topology node / auxiliary node; S105, If the node weight does not exceed the preset node filtering threshold, discard the current topology node / auxiliary node; S106, deploy composite sensor groups on the retained topology nodes / auxiliary nodes.

[0031] In this embodiment of the invention, when deploying composite sensor groups in the gear sleeve and gear meshing area of ​​the clutch based on the clutch model architecture diagram, the screening threshold is dynamically adjusted based on the distribution of node weights. This avoids the problem of a one-size-fits-all approach with a fixed threshold. Dynamic threshold screening minimizes the number of sensors while ensuring that key vibration information is not lost, thereby reducing hardware costs, wiring complexity, and data processing volume. At the same time, it improves the real-time performance and reliability of the monitoring system. Furthermore, optimizing the sensor layout allows the sensors to be concentrated on the critical path and sensitive points of vibration transmission, resulting in a higher signal-to-noise ratio and richer impact characteristics in the collected signals.

[0032] This invention provides a method for compressed wavelet transform processing of multi-source impulse signal sets. Figure 3 This diagram illustrates the implementation flow of a compressed wavelet transform method for processing multi-source impact signal sets. The method specifically includes: S201: Load a multi-source signal impact set, perform compressed wavelet transform on the multi-source impact signal set, and introduce a frequency weighting function based on meshing impact prior to weight the multi-source impact signal after compressed wavelet transform to obtain a multi-source weighted signal. In this process, by compressing the wavelet basis function or decomposition scale, the decomposition calculation of redundant frequency bands is reduced, while the time-varying details of the impact signal are preserved, thus improving the computational efficiency and feature resolution. The frequency weighting function is designed to enhance the energy proportion of the impact characteristic frequency bands and suppress noise interference from irrelevant frequency bands. The combination of compressed wavelet transform and prior weighting makes the time-frequency energy distribution of the multi-source impact signal clearer, thus highlighting the key features. S202 performs bandpass filtering on the multi-source weighted signal, outputs the bandpass-filtered multi-source weighted signal, and resamples and averages the filtered multi-source weighted signal based on the acquisition period of the meshing impact signal to obtain the derived periodic impact characteristics. Among them, resampling the filtered signal based on the acquisition period of the meshing impact signal and averaging the sampled signals of multiple periods can significantly improve the signal-to-noise ratio of the periodic impact characteristics. S203: Obtain the derived periodic impact characteristics and multi-source weighted signals. Perform time-domain feature analysis, frequency-domain feature analysis, and joint time-frequency domain feature analysis on the periodic impact characteristics and multi-source weighted signals respectively. Extract the impact pulse characteristics, meshing frequency characteristics, and wavelet packet decomposition characteristics of the periodic impact characteristics and multi-source weighted signals. Among them, the time-domain features include pulse peak value, rise time, decay time, pulse width, etc., which directly reflect the intensity (peak value) and duration of the impact. The frequency-domain features extract the meshing frequency and its harmonics and sideband frequencies through Fourier transform, reflecting the periodicity and frequency modulation characteristics of the impact. The time-frequency domain features divide the signal into more refined sub-frequency bands through wavelet packet decomposition, thereby extracting the energy entropy and energy proportion of each sub-frequency band, reflecting the dynamic distribution of impact energy in different frequency bands. In this embodiment, the joint extraction of multi-dimensional features can completely characterize the non-stationary characteristics of the impact signal and avoid the one-sidedness of single-domain analysis. S204, load the impact pulse features, meshing frequency features and wavelet packet decomposition features, map the impact pulse features, meshing frequency features and wavelet packet decomposition features into a low-dimensional normalized space, and obtain the joint time-frequency analysis results of the clutch.

[0033] In this embodiment of the invention, when performing compressed wavelet transform on a multi-source impact signal set, the compression of wavelet basis functions or decomposition scale reduces the decomposition calculation of redundant frequency bands while preserving the time-varying details of the impact signal, thus improving computational efficiency and feature resolution. A frequency weighting function is designed to enhance the energy proportion of impact characteristic frequency bands and suppress noise interference from irrelevant frequency bands. The combination of compressed wavelet transform and prior weighting makes the time-frequency energy distribution of the multi-source impact signal clearer, highlighting key features. The joint extraction of multi-dimensional features can fully characterize the non-stationary characteristics of the impact signal, avoiding the one-sidedness of single-domain analysis. This allows the fusion of time-domain, frequency-domain, and time-frequency-domain features of the multi-source signals to cover multi-dimensional information such as the intensity, period, and energy distribution of the impact signal, avoiding misjudgments caused by the lack of a single feature.

[0034] In this embodiment of the invention, the lifespan prediction model includes an input layer, a deep metric learning network, and an output layer. A feature extraction layer is provided between the input layer and the deep metric learning network, and a lifespan probability distribution layer is provided between the deep metric learning network and the output layer.

[0035] The feature extraction layer extracts features from the joint time-frequency analysis results based on the physical degradation mechanism of the clutch, and outputs the gear sleeve-gear meshing state vector; A deep metric learning network is used to obtain the gear sleeve-gear meshing state vector, and based on the gear sleeve-gear meshing state vector, a standard impact vibration reference vector is retrieved. The state Mahalanobis distance between the gear sleeve-gear meshing state vector and the impact vibration reference vector is calculated. Based on the state Mahalanobis distance, a multi-stage degradation physical equation considering vibration impact energy and cumulative tooth surface damage is constructed. The impact vibration index is obtained by solving the multi-stage degradation physical equation. The lifetime probability distribution layer dynamically assigns weights to the shock vibration index based on the multimodal attention mechanism. Using the weighted shock vibration index as input, it generates the state evaluation equation based on the adaptive particle filter algorithm, and adaptively adjusts the prediction error through the covariance matching method to output the remaining lifetime probability distribution of the gear sleeve-gear meshing. The remaining life of the gear sleeve-gear meshing is calculated based on the probability distribution of the remaining life of the gear sleeve-gear meshing. The life prediction model provides an accurate and reliable prediction tool for the intelligent operation and maintenance of SSS clutches through hierarchical architecture design, feature extraction guided by physical degradation mechanism, deep metric learning that integrates state Mahalanobis distance and physical equations, and probabilistic evaluation with multimodal attention and adaptive particle filtering. This significantly reduces the risk of sudden failure and operation and maintenance costs.

[0036] This invention provides a method for detecting impact vibration in clutches based on a life prediction model and joint time-frequency analysis results. Figure 4 This diagram illustrates the implementation flow of a clutch impact vibration detection method based on joint time-frequency analysis results using a life prediction model. The method specifically includes: S301, obtain the joint time-frequency analysis results. The feature extraction layer extracts features from the joint time-frequency analysis results based on the physical degradation mechanism of the clutch and outputs the gear sleeve-gear meshing state vector. The feature extraction guided by the physical degradation mechanism ensures a high correlation between the state vector and the actual degradation process of the clutch, avoids false feature interference, and enables the subsequent model to capture the degradation trend more accurately. S302, obtain the gear sleeve-gear meshing state vector, retrieve the standard impact vibration reference vector based on the gear sleeve-gear meshing state vector, and calculate the state Mahalanobis distance between the gear sleeve-gear meshing state vector and the impact vibration reference vector. The phased degradation equation is constructed based on the Mahalanobis distance and physical mechanism, which can accurately describe the damage evolution rate of different degradation stages. The impact vibration index obtained by inverting the physical equation can transform the abstract "state deviation" into a quantifiable degradation index. The introduction of Mahalanobis distance improves the robustness of the state deviation measurement, and the integration of the physical equation constrains the degradation logic of the model. The combination of the two enables the model to accurately track the degradation trajectory under complex working conditions. The formula for calculating the Mahalanobis distance between the gear sleeve-gear meshing state vector and the impact vibration reference vector is expressed as follows: in, Represents the gear sleeve-gear meshing state vector With the impact vibration reference vector The state Mahalanobis distance, Let be the covariance matrix of the gear sleeve-gear meshing state vector; S303, based on the state Mahalanobis distance, a multi-stage degradation physical equation considering vibration impact energy and cumulative tooth surface damage is constructed, and the impact vibration index is obtained by solving the multi-stage degradation physical equation. The multi-stage degradation physical equation is expressed as follows: in, Let these represent the current state estimation vector and the previous state estimation vector, respectively. For Kalman gain, These are the vibration index estimation vector and observation matrix at the current moment, respectively.

[0037] In this embodiment of the invention, when the life prediction model performs impact vibration detection on the clutch based on the joint time-frequency analysis results, the feature extraction guided by the physical degradation mechanism ensures a high correlation between the state vector and the actual degradation process of the clutch, avoids spurious feature interference, and enables the subsequent model to capture the degradation trend more accurately. The phased degradation equation constructed based on Mahalanobis distance and physical mechanism can accurately describe the damage evolution rate of different degradation stages. The impact vibration index obtained by inverting the physical equation can transform the abstract "state deviation" into a quantifiable degradation index. Furthermore, the introduction of Mahalanobis distance improves the robustness of the state deviation measurement, and the integration of the physical equation constrains the degradation logic of the model. The combination of the two enables the model to accurately track the degradation trajectory under complex working conditions.

[0038] This invention provides a method for predicting and analyzing the lifespan of impact vibration test results based on a lifespan prediction model. Figure 5 This diagram illustrates the implementation flow of a method for life prediction analysis of impact vibration test results based on a life prediction model. The method specifically includes: S401 dynamically assigns weights to the shock vibration index based on a multimodal attention mechanism, and generates a state evaluation equation based on an adaptive particle filter algorithm using the weighted shock vibration index as input. S402 adaptively adjusts the prediction error using the covariance matching method and outputs the probability distribution of the remaining life of the gear sleeve-gear meshing. S403 calculates the remaining life of the gear sleeve-gear meshing based on the probability distribution of the remaining life of the gear sleeve-gear meshing. The calculation of the remaining life is essentially a Monte Carlo simulation. Utilizing the uncertainty of the current state (N particles) given by particle filtering, the physical model is used to extrapolate every possible future, thereby obtaining the entire probability distribution picture of the RUL, rather than a single guess. The formula for calculating the remaining service life of the gear sleeve-gear meshing is expressed as follows: In the formula, Let $\mathbf$ represent the remaining life of the gear sleeve-gear meshing, the mean of the remaining life probability distribution, and the variance of the remaining life probability distribution, respectively. is the confidence coefficient.

[0039] In this embodiment of the invention, when performing life prediction analysis on impact vibration detection results based on the life prediction model, different impact vibration indicators are weighted according to their dynamic changes in current contribution to degradation, avoiding misjudgment of indicator importance caused by fixed weights. Furthermore, the probability distribution of future states is inferred through a particle filtering algorithm, combined with an adaptive mechanism, improving the model's ability to model nonlinear and non-Gaussian degradation processes. Finally, the probabilistic output and adaptive adjustment mechanism make the remaining life prediction results more reliable. Accurate remaining life prediction can provide early warning of potential clutch failures, helping maintenance personnel to plan maintenance schedules in advance. This preventative maintenance method can reduce downtime and repair costs caused by sudden failures, improving equipment operating efficiency and economic benefits.

[0040] Example 2 This embodiment also provides an SSS clutch engagement impact vibration detection and life prediction system. Figure 6 This diagram illustrates the structure of the SSS clutch engagement impact vibration detection and life prediction system. The SSS clutch engagement impact vibration detection and life prediction system specifically includes: The sensor monitoring deployment module 100 is used to acquire the clutch model architecture diagram and deploy a composite sensor group in the gear sleeve and gear meshing area of ​​the clutch based on the clutch model architecture diagram. The composite sensor group synchronously collects the meshing impact signal and uploads the meshing impact signal to the edge processing node 200. Edge processing node 200 acquires meshing impact signals in real time and synchronously acquires associated signal parameters based on the meshing impact signal acquisition cycle. It performs timestamp synchronization processing on the associated signal parameters and meshing impact signals to obtain a multi-source impact signal set. The time-frequency analysis module 300 is used to load a multi-source impact signal set, perform compressed wavelet transform processing on the multi-source impact signal set, and perform joint time-frequency analysis to obtain the joint time-frequency analysis results of the clutch, and upload the joint time-frequency analysis results to the historical database. The time-frequency analysis module 300 includes: Wavelet transform unit 310 is used to load a multi-source signal impulse set, perform compressed wavelet transform processing on the multi-source impulse signal set, and introduce a frequency weighting function based on meshing impulse prior to weight the multi-source impulse signal after compressed wavelet transform processing to obtain a multi-source weighted signal. The derived feature unit 320 is used to perform bandpass filtering on the multi-source weighted signal, output the bandpass filtered multi-source weighted signal, and resample and superimpose the filtered multi-source weighted signal based on the acquisition period of the meshing impact signal to obtain the derived periodic impact feature. The time-frequency extraction unit 330 is used to acquire the derived periodic impact features and multi-source weighted signals, and to perform time-domain feature analysis, frequency-domain feature analysis, and joint time-frequency-domain feature analysis on the periodic impact features and multi-source weighted signals, respectively, to extract the impact pulse features, meshing frequency features, and wavelet packet decomposition features of the periodic impact features and multi-source weighted signals. The feature mapping unit 340 is used to load the impact pulse features, meshing frequency features and wavelet packet decomposition features, and map the impact pulse features, meshing frequency features and wavelet packet decomposition features into a low-dimensional normalized space to obtain the joint time-frequency analysis results of the clutch.

[0041] The model building module 400 is used to pre-build a life prediction model based on a deep metric learning network, retrieve historical time-frequency analysis results from a historical database, iteratively train the life prediction model based on the historical time-frequency analysis results, and output a converged life prediction model. The life prediction module 500 is used to acquire the joint time-frequency analysis results in real time. The life prediction model performs impact vibration detection on the clutch based on the joint time-frequency analysis results and outputs the impact vibration detection results. The impact vibration detection results are used to load the impact vibration detection results. The life prediction model performs life prediction analysis on the impact vibration detection results to obtain the probability distribution of the remaining life of the gear sleeve-gear meshing. The remaining life of the gear sleeve-gear meshing is calculated based on the probability distribution of the remaining life of the gear sleeve-gear meshing.

[0042] In this embodiment of the invention, the lifetime prediction module 500 includes: The vibration detection unit 510 is used to acquire the joint time-frequency analysis results in real time. The life prediction model performs impact vibration detection on the clutch based on the joint time-frequency analysis results and outputs the impact vibration detection results. The predictive analysis unit 520 is used to load the impact vibration test results, perform life prediction analysis on the impact vibration test results based on the life prediction model, obtain the remaining life probability distribution of the gear sleeve-gear meshing, and calculate the remaining life of the gear sleeve-gear meshing based on the remaining life probability distribution of the gear sleeve-gear meshing.

[0043] In summary, this invention provides a method and system for detecting and predicting the life of an SSS clutch engagement impact vibration. In the embodiments of this invention, the acquisition and fusion of multi-source data on engagement impact signals and associated signal parameters can comprehensively and accurately capture key characteristic information of impact vibration during clutch engagement. Furthermore, the life prediction model can achieve precise detection of impact vibration anomalies based on real-time joint time-frequency analysis results combined with a multimodal attention mechanism for dynamic weighting, and further perform life prediction analysis to accurately obtain the probability distribution of the remaining life of the gear sleeve-gear meshing, thereby calculating the remaining life and making the life prediction results more accurate and reliable. This method, combining physical models and data-driven approaches, fully leverages the advantages of both, improving the accuracy and adaptability of life prediction.

[0044] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for detecting engagement impact vibration and predicting life of an SSS clutch, characterized in that, include: Obtain the clutch model architecture diagram of the SSS clutch to be predicted, and based on the clutch model architecture diagram, deploy a composite sensor group in the gear sleeve and gear meshing area of ​​the clutch, and synchronously collect meshing impact signals through the composite sensor group. While acquiring meshing impact signals in real time, associated signal parameters are synchronously acquired based on the meshing impact signal acquisition cycle. The associated signal parameters and meshing impact signals are then time-stamped to obtain a multi-source impact signal set. The multi-source impact signal set is subjected to compressed wavelet transform processing and joint time-frequency analysis to obtain the joint time-frequency analysis results of the clutch. The joint time-frequency analysis results are then uploaded to the historical database. A lifetime prediction model based on a deep metric learning network is pre-constructed. This lifetime prediction model is used to extract features based on the joint time-frequency analysis results to obtain the gear sleeve-gear meshing state vector, thereby predicting the probability distribution of the remaining lifetime of the gear sleeve and gear meshing and outputting the final remaining lifetime. Historical joint time-frequency analysis results with remaining lifetime labels are crawled from the historical database. The lifetime prediction model is iteratively trained based on the historical joint time-frequency analysis results and outputs a converged lifetime prediction model. The joint time-frequency analysis results are acquired in real time and input into the lifetime prediction model, and the lifetime prediction results are output.

2. The method for detecting engagement impact vibration and predicting life of an SSS clutch according to claim 1, characterized in that, The process of determining the arrangement position of the composite sensor group includes: Traverse the clutch model architecture diagram, using the gear sleeve and gear as the topology nodes of the clutch static topology network, extract the connection relationship between the topology nodes, construct the associated edges through the connection relationship between the topology nodes, and abstract the intersection of the associated edges as auxiliary nodes; Identify the topological nodes and auxiliary nodes of the clutch static topology network. Using the number of gear sleeves and gears associated with the topological nodes and auxiliary nodes as a priori conditions, assign weights to the topological nodes and auxiliary nodes based on principal component analysis to obtain node weights. The node selection threshold is dynamically determined based on the node weight. It is then determined whether the node weight exceeds the preset node selection threshold. If the node weight exceeds the preset node selection threshold, the current topology node or auxiliary node is retained. If the node weight does not exceed the preset node selection threshold, the current topology node or auxiliary node is discarded. The composite sensor group is deployed based on the retained topology nodes and auxiliary nodes.

3. The method for detecting engagement impact vibration and predicting life of an SSS clutch according to claim 2, characterized in that, The node weights are obtained by weighted summation based on the node's association complexity, structural degrees of freedom, and vibration transmission factor variance contribution rate. The association complexity is the result of multiplying the number of gear sleeves and gears associated with the current node by the topological distance between the current node and the center node of the clutch static topology network, and dividing by the total number of nodes in the clutch static topology network. The vibration transmission factor variance contribution rate is the total number of nodes in the clutch static topology network divided by the average distance between the current node and the associated gear sleeve and gear.

4. The method for detecting engagement impact vibration and predicting life of an SSS clutch according to claim 3, characterized in that, The expression for calculating the node weight is: In the formula, For the first i The node weight of each node. They are nodes The correlation complexity, structural degrees of freedom, and vibration transfer factor, These represent the variance contribution rates of the correlation complexity, structural degrees of freedom, and vibration transfer factor, respectively, obtained based on principal component analysis. They are nodes The number of associated gear sleeves and gears, the topological distance from the central node of the clutch static topology network, and the total number of nodes; For nodes Dynamic response sensitivity, For nodes The average distance to the associated gear sleeve or gear.

5. The method for detecting engagement impact vibration and predicting life of an SSS clutch according to claim 1, characterized in that, The compressed wavelet transform processing of the multi-source impulse signal set includes: A multi-source signal impulse set is loaded, and compressed wavelet transform processing is performed on the multi-source impulse signal set. A frequency weighting function based on meshing impulse prior is introduced to weight the multi-source impulse signal after compressed wavelet transform processing to obtain a multi-source weighted signal. The multi-source weighted signal is bandpass filtered to output the bandpass-filtered multi-source weighted signal. Based on the acquisition period of the meshing impact signal, the bandpass-filtered multi-source weighted signal is resampled and superimposed to obtain the derived periodic impact characteristics. Time-domain feature analysis, frequency-domain feature analysis, and joint time-frequency-domain feature analysis were performed on the derived periodic impact characteristics and multi-source weighted signals respectively to extract impact pulse characteristics, meshing frequency characteristics, and wavelet packet decomposition characteristics. The impact pulse characteristics, engagement frequency characteristics, and wavelet packet decomposition characteristics are mapped into a low-dimensional normalized space to obtain the joint time-frequency analysis results of the clutch.

6. The method for detecting engagement impact vibration and predicting life of an SSS clutch according to claim 5, characterized in that, The time-domain features analyzed include: pulse peak value, rise time, decay time, and pulse width. The frequency domain features analyzed include: the meshing frequency and its harmonics and sideband frequencies extracted by Fourier transform; The time-frequency domain features jointly analyzed and processed include: dividing the signal into sub-bands through wavelet packet decomposition, thereby extracting the energy entropy and energy proportion of each sub-band.

7. The method for detecting engagement impact vibration and predicting life of an SSS clutch according to claim 1, characterized in that, The lifespan prediction model includes an input layer, a deep metric learning network, and an output layer connected in sequence. A feature extraction layer is provided between the input layer and the deep metric learning network, and a lifespan probability distribution layer is provided between the deep metric learning network and the output layer. The feature extraction layer is used to extract features from the joint time-frequency analysis results based on the clutch physical degradation mechanism, and outputs the gear sleeve-gear meshing state vector; The deep metric learning network is used to retrieve the standard impact vibration reference vector based on the gear sleeve-gear meshing state vector, and calculate the state Mahalanobis distance between the gear sleeve-gear meshing state vector and the impact vibration reference vector; based on the state Mahalanobis distance, a multi-stage degradation physical equation considering vibration impact energy and cumulative tooth surface damage is constructed, and the impact vibration index is obtained by solving the multi-stage degradation physical equation; the impact vibration index is dynamically weighted based on a multi-modal attention mechanism, and the state evaluation equation is generated based on the weighted impact vibration index as input and an adaptive particle filter algorithm is used. The lifetime probability distribution layer is used to adaptively adjust the prediction error through the covariance matching method and output the remaining lifetime probability distribution of the gear sleeve-gear meshing. The input layer is used to calculate the remaining life of the gear sleeve-gear meshing based on the probability distribution of the remaining life of the gear sleeve-gear meshing.

8. The method for detecting engagement impact vibration and predicting life of an SSS clutch according to claim 7, characterized in that, The formula for calculating the remaining service life of the gear sleeve-gear meshing is as follows: In the formula, These represent the remaining life of the gear sleeve-gear meshing, the mean of the remaining life probability distribution, and the variance of the remaining life probability distribution, respectively. is the confidence coefficient.

9. A meshing impact vibration detection and life prediction system for implementing the meshing impact vibration detection and life prediction method for an SSS clutch as described in any one of claims 1-8, characterized in that, include: The sensor monitoring deployment module is used to obtain the clutch model architecture diagram of the SSS clutch to be predicted. Based on the clutch model architecture diagram, a composite sensor group is deployed in the gear sleeve and gear meshing area of ​​the clutch. The meshing impact signal is collected synchronously through the composite sensor group and uploaded to the edge processing node. The edge processing node is used to acquire the meshing impact signal in real time, and simultaneously acquire related signal parameters based on the meshing impact signal acquisition cycle. It performs timestamp synchronization processing on the related signal parameters and the meshing impact signal to obtain a multi-source impact signal set. The time-frequency analysis module is used to perform compressed wavelet transform processing on the multi-source impact signal set and perform joint time-frequency analysis to obtain the joint time-frequency analysis results of the clutch, and upload the joint time-frequency analysis results to the historical database. The model building module is used to pre-build a lifetime prediction model based on a deep metric learning network. This lifetime prediction model is used to extract features based on the joint time-frequency analysis results to obtain the gear sleeve-gear meshing state vector, thereby predicting the remaining lifetime probability distribution of the gear sleeve and gear meshing and outputting the final remaining lifetime. Historical joint time-frequency analysis results with remaining lifetime labels are crawled from the historical database. The lifetime prediction model is iteratively trained based on the historical joint time-frequency analysis results and outputs a converged lifetime prediction model. The lifetime prediction module is used to input the real-time acquired joint time-frequency analysis results into the lifetime prediction model and output the lifetime prediction results.

10. The system according to claim 9, characterized in that, The time-frequency analysis module includes: The wavelet transform unit is used to load the multi-source signal impulse set, perform compressed wavelet transform processing on the multi-source impulse signal set, and introduce a frequency weighting function based on meshing impulse prior to weight the multi-source impulse signal after compressed wavelet transform processing to obtain a multi-source weighted signal. The derived feature unit is used to perform bandpass filtering on the multi-source weighted signal, output the bandpass-filtered multi-source weighted signal, and resample and superimpose the bandpass-filtered multi-source weighted signal based on the acquisition period of the meshing impact signal to obtain the derived periodic impact feature. The time-frequency extraction unit is used to perform time-domain feature analysis, frequency-domain feature analysis, and joint time-frequency-domain feature analysis on the derived periodic impact features and multi-source weighted signals, respectively, to extract impact pulse features, meshing frequency features, and wavelet packet decomposition features; The feature mapping unit is used to map the impact pulse features, engagement frequency features, and wavelet packet decomposition features into a low-dimensional normalized space to obtain the joint time-frequency analysis results of the clutch.

Images