Component abnormal sound residual life prediction method and device, terminal and medium

Abstract

The application provides a component abnormal sound residual life prediction method and device, terminal and medium, the method comprises the following steps: constructing a friction-knock coupling micro-dynamics evolution model based on nonlinear physical priori, and establishing a physical priori state transition equation; collecting high-frequency original waveform of the component under initial state and short-period accelerated endurance load spectrum, and obtaining a vibration characteristic observation vector; establishing a nonlinear observation mapping equation connecting the hidden state vector and the vibration characteristic observation vector; and using an unscented Kalman filter to jointly inversely reconstruct the hidden state vector to obtain a posterior augmented state estimation value of the micro-physical parameters; taking the posterior augmented state estimation value as an initial condition to obtain a set of implicit physical parameters; according to the set of implicit physical quantity data and a multi-modal competition logic mapping function, a residual life prediction value of the component abnormal sound is obtained. The accurate prediction of the residual life of the component abnormal sound under small sample is realized.

Description

Technical Field

[0001] This invention relates to the field of vehicle component failure prediction technology, specifically to a method, device, terminal, and medium for predicting the remaining lifespan of abnormal noises in vehicle components. Background Technology

[0002] In the automotive and high-end manufacturing sectors, component noises (BSR: Buzz, Squeak, and Rattle) are a core issue affecting product NVH (Noise, Vibration, and Harshness, a technical indicator used to measure automotive ride comfort and product quality). Currently, the assessment of the lifecycle of noises mainly relies on two methods: one is pure CAE simulation interference analysis in the early design stage; the other is high-frequency vibration bench durability testing after product prototyping, combined with microphone or engineer subjective listening for "post-event" judgment. In recent years, some technologies have also adopted purely data-driven AI black-box models, attempting to fit lifecycles by collecting a large amount of historical durability data.

[0003] However, existing technical solutions have the following significant drawbacks: First, the passive "post-incident detection" logic leads to extremely long testing cycles. Existing bench tests rely excessively on macroscopic acoustic parameters (such as sound pressure level and psychoacoustic loudness), which are only detected by the system or human ear when internal wear causes macroscopic abnormal noises to "erupt." This lack of means to quantitatively assess "early microscopic latent degradation" makes it impossible to use short-cycle accelerated testing to predict the long-term occurrence points of abnormal noises, greatly increasing the time and economic costs of high-mileage road testing and bench durability testing of the whole vehicle.

[0004] Second, pure black-box AI prediction models suffer from poor generalization, falling into the dilemma of "data hunger" and "violation of physical common sense." Existing prediction models are mostly purely data-driven neural networks (such as LSTM and RNN). Due to the lack of constraints from underlying tribological and dynamic mechanisms, the prediction results are often extremely unstable; moreover, when dealing with parts of different structures and materials, a massive and extremely expensive number of failure samples are required to retrain the model, making it difficult to implement in small-sample engineering practice.

[0005] Third, the feature extraction dimensions are extremely insensitive to structural loosening. Traditional techniques only compare sound pressure or total vibration level (RMS) before and after durability testing. These linear features cannot characterize the "gap nonlinearity" and "stick-slip abrupt change" caused by buckle loosening or contact surface wear. Tiny physical degradation signals are often completely drowned out by the powerful broadband mechanical background noise of the test bench. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention proposes a method, device, terminal, and medium for predicting the remaining life of abnormal noise in components, aiming to achieve accurate prediction of the remaining life of abnormal noise in components with a small sample size.

[0007] In a first aspect, embodiments of this application provide a method for predicting the remaining lifespan of abnormal noise in components, including: A friction-knock coupling microdynamic evolution model based on nonlinear physical priors is constructed. Based on the friction-knock coupling microdynamic evolution model, microscopic physical parameters that cannot be directly measured are obtained. Based on the microscopic physical parameters, hidden state vectors are obtained and physical prior state transition equations are established. The high-frequency raw waveforms of the components are collected under the initial state and under the applied short-period accelerated durability load spectrum. The high-frequency raw waveforms are preprocessed for noise reduction by noise reduction frequency response function estimation. The long-time waveform is sliced ​​by time sliding window, and the acoustic vibration characteristics of each slice are calculated to obtain the acoustic vibration characteristic observation vector. The acoustic vibration characteristic observation vector includes the initial multidimensional observation vector and the degraded characteristic observation vector. A nonlinear observation mapping equation is established to connect the hidden state vector and the acoustic vibration feature observation vector. An unscented Kalman filter is used to deeply fuse the hidden state vector and the acoustic vibration feature observation vector through the nonlinear observation mapping equation. By using blind inference and targeted correction mechanism under sparse observation, the hidden state vector is jointly reconstructed inversely to obtain the posterior augmented state estimate of the microscopic physical parameters. Using the a priori augmented state estimate as the initial condition, the model is continuously extrapolated to future life cycles along the friction-knock coupling microdynamic evolution model to obtain a set of implicit physical parameters. The set of implicit physical parameters includes the microphysical gap and the microcontact stiffness under different life cycles. Based on the implicit physical quantity dataset and the multimodal competition logic mapping function, the maximum Lyapunov exponent, the probability of knocking noise, and the probability of friction noise are obtained. When the maximum Lyapunov exponent, the probability of knocking noise, and the probability of friction noise meet the preset conditions, the remaining life prediction value of the abnormal noise of the component is obtained.

[0008] Optionally, the microscopic physical parameters include microscopic contact stiffness and microscopic physical gap. The construction of a friction-impact coupled microscopic dynamic evolution model based on nonlinear physical priors yields microscopic physical parameters that cannot be directly measured. The process of obtaining hidden state vectors and establishing physical prior state transition equations based on these microscopic physical parameters includes: A frictional microdynamic evolution model incorporating impact friction coupling parameters is constructed, and the microscopic contact stiffness, which cannot be directly measured, is obtained based on the frictional microdynamic evolution model. A micro-dynamic evolution model of percussion noise based on nonlinear physical priors is constructed, and the micro-physical gap that cannot be directly measured is obtained based on the micro-dynamic evolution model of percussion noise. The hidden state vector is obtained based on the microscopic physical parameters and the impact friction coupling parameters; The physical prior state transition equation is established based on the hidden state vector.

[0009] Optionally, the high-frequency raw waveforms of the acquired components in their initial state and under the applied short-period accelerated durability load spectrum are preprocessed by noise reduction using an anti-noise frequency response function estimation. Furthermore, a time sliding window is used to slice the long-term waveform, and the acoustic-vibration characteristics of each slice are calculated to obtain an acoustic-vibration characteristic observation vector, including: High-frequency raw waveform signals of components under initial state and under short-cycle accelerated durability load spectrum are collected; The original high-frequency waveform signal is preprocessed by noise reduction through noise immunity frequency response function estimation to obtain a noise-reduced long-time waveform signal. The noise-reduced long-time waveform signal is sliced ​​using a time sliding window, and the acoustic and vibration characteristics of each slice are calculated. Based on the acoustic vibration characteristics, the initial multidimensional observation vector and the degraded feature observation vector are obtained; Based on the initial multidimensional observation vector and the degraded characteristic observation vector, the acoustic vibration characteristic observation vector is obtained.

[0010] Optionally, the unscented Kalman filter is used to deeply fuse the augmented hidden state vector and the acoustic vibration feature evolution drift through a nonlinear observation mapping equation. A blind inference and targeted correction mechanism under sparse observation is used to jointly inversely reconstruct the hidden state vector, obtaining the posterior augmented state estimate of the microscopic physical parameters, including: Define the triggering logic for sparse observations; When the measured acoustic vibration feature observation vector is obtained, the Sigma point set is extracted by unscented transformation, and the predicted observation feature vector, observation covariance and state observation cross-covariance are obtained based on the augmented state prior prediction of unscented transformation. Based on the measured acoustic vibration feature observation vector, the predicted observation feature vector, the observation covariance, the state observation cross-covariance, the prior augmented state mean, and the hidden state vector, the posterior augmented state estimate and the posterior augmented state covariance are obtained based on feature residual weighted targeted correction and non-negative physical boundary constraints.

[0011] Optionally, when obtaining the measured acoustic vibration feature observation vector, an unscented transform is used to extract the Sigma point set, and based on the augmented state prior prediction of the unscented transform, the predicted observation feature vector, observation covariance, and state-observation cross-covariance are obtained, including: Based on the mean and covariance of the posterior augmented state estimate from the previous time step, 2n+1 Sigma point sets and Sigma point weights are generated using standard unscented transformation. Substitute each Sigma point in the Sigma point set into the physical prior state transition equation for transmission, and calculate the weighted mean and prior covariance of the prior augmented state estimate at the current time. The prior state Sigma point is mapped to the predicted observation Sigma point through the nonlinear observation mapping equation, and the predicted observation eigenvector, observation covariance, and state-observation cross-covariance are obtained by weighted calculation.

[0012] Optionally, based on the measured acoustic vibration feature observation vector, the predicted observation feature vector, the observation covariance, the state observation cross-covariance, the prior augmented state estimate mean, and the hidden state vector, and based on feature residual weighted targeted correction and non-negative physical boundary constraints, the posterior augmented state estimate and the posterior augmented state covariance are obtained, including: Based on the measured acoustic vibration characteristic observation vector and the predicted observation characteristic vector, the characteristic residual is obtained; The augmented Kalman gain matrix is ​​obtained based on the state observation cross-covariance and the observation covariance. Based on the characteristic residuals, the mean of the prior augmented state estimates, and the augmented Kalman gain matrix, the microscopic physical parameters, the mechanistic parameters in the physical prior state transition equation, and the impact friction coupling parameters are jointly calibrated and updated, and the pre- and post-constraint augmented state estimates are obtained. The a posteriori augmented state estimates are obtained by processing the pre- and post-constraint augmented state estimates through non-negative physical boundary constraint operations.

[0013] Optionally, the step of obtaining the maximum Lyapunov exponent, the probability of knocking noise, and the probability of friction noise based on the implicit physical quantity dataset and the multimodal competition logic mapping function, and obtaining the remaining life prediction value of the component noise when the maximum Lyapunov exponent, the probability of knocking noise, and the probability of friction noise satisfy preset conditions, includes: Based on the dataset of implicit physical quantities, the magnitude of the perturbation of the phase space trajectory is calculated using an evolutionary algorithm based on the Jacobian variational equation, and the maximum Lyapunov exponent is obtained. Based on the implicit physical quantity dataset and the multimodal competition logic mapping function, a coupling competition risk function is constructed in the two-dimensional physical phase space to obtain the probability of knocking noise and friction noise. When the maximum Lyapunov index, the probability of knocking noise, and the probability of friction noise meet the preset conditions, the remaining life prediction value of the component noise is obtained.

[0014] Secondly, embodiments of this application provide a device for predicting the remaining lifespan of components with abnormal noise, comprising: The model building module is used to construct a friction-knock coupling microdynamic evolution model based on nonlinear physical priors, obtain microscopic physical parameters that cannot be directly measured based on the friction-knock coupling microdynamic evolution model, obtain hidden state vectors based on microscopic physical parameters, and establish physical prior state transition equations. The observation vector determination module is used to acquire the high-frequency raw waveforms of components in their initial state and under short-period accelerated durability load spectrum, perform noise reduction preprocessing on the high-frequency raw waveforms through noise-resistant frequency response function estimation, and slice the long-time waveforms using a time sliding window to calculate the acoustic and vibration characteristics of each slice, thereby obtaining the acoustic and vibration characteristic observation vectors; the acoustic and vibration characteristic observation vectors include the initial multidimensional observation vectors and the degraded characteristic observation vectors; The estimation module is used to establish a nonlinear observation mapping equation connecting the hidden state vector and the acoustic vibration feature observation vector; and to use an unscented Kalman filter to deeply fuse the augmented hidden state vector and the acoustic vibration feature evolution drift through the nonlinear observation mapping equation. By using blind inference and targeted correction mechanisms under sparse observation, the hidden state vector is jointly inversely reconstructed to obtain the posterior augmented state estimate of the microscopic physical parameters. The parameter set determination module is used to continuously extrapolate the friction-knock coupling microdynamic evolution model to future life cycles using the a priori augmented state estimate as the initial condition, to obtain the implicit physical parameter set; the implicit physical parameter set includes the microphysical gap and the microcontact stiffness under different life cycles. The life prediction module is used to obtain the maximum Lyapunov exponent, the probability of knocking noise, and the probability of friction noise based on the implicit physical quantity dataset and the multimodal competition logic mapping function. When the maximum Lyapunov exponent, the probability of knocking noise, and the probability of friction noise meet the preset conditions, the remaining life prediction value of the component noise is obtained.

[0015] Thirdly, embodiments of this application provide a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the component noise remaining life prediction method as described in any of the first aspects above.

[0016] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the component noise remaining life prediction method as described in any of the first aspects above.

[0017] Fifthly, embodiments of this application provide a computer program product that, when run on a terminal device, causes the terminal device to execute the component noise remaining life prediction method described in any of the first aspects above.

[0018] This application constructs a friction-impact coupling microdynamic evolution model based on nonlinear physical priors, and obtains microscopic physical parameters that cannot be directly measured based on the friction-impact coupling microdynamic evolution model; it obtains hidden state vectors and establishes physical prior state transition equations based on the microscopic physical parameters; it collects high-frequency raw waveforms of components in their initial state and under short-period accelerated durability load spectrum, performs noise reduction preprocessing on the high-frequency raw waveforms through noise-resistant frequency response function estimation, and slices the long-time waveforms using a time sliding window to calculate the acoustic vibration characteristics of each slice, obtaining an acoustic vibration characteristic observation vector; the acoustic vibration characteristic observation vector includes an initial multidimensional observation vector and a degraded characteristic observation vector; it establishes a nonlinear observation mapping equation connecting the hidden state vector and the acoustic vibration characteristic observation vector; and it uses an unscented Kalman filter to augment the hidden state vector. The acoustic vibration feature evolution drift is deeply fused with the nonlinear observation mapping equation. Using a blind inference and targeted correction mechanism under sparse observation, the hidden state vector is jointly reconstructed inversely to obtain the posterior augmented state estimate of the microscopic physical parameters. Using the posterior augmented state estimate as the initial condition, the model is continuously extrapolated to future lifecycles along the friction-impact coupling microdynamic evolution model to obtain a set of implicit physical parameters. This set includes microscopic physical gaps and microscopic contact stiffness at different lifecycles. Based on the implicit physical quantity dataset and the multimodal competitive logic mapping function, the maximum Lyapunov exponent, the probability of impact noise, and the probability of friction noise are obtained. When the maximum Lyapunov exponent, the probability of impact noise, and the probability of friction noise meet preset conditions, the remaining lifespan prediction of the component noise is obtained. This achieves accurate prediction of the remaining lifespan of component noise under small sample conditions. Attached Figure Description

[0019] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0020] Figure 1 This is a flowchart illustrating the first embodiment of the method for predicting the remaining life of abnormal noise of components provided in this application; Figure 2 This is a flowchart of the method for predicting the remaining life of abnormal noise of components provided in an embodiment of this application; Figure 3 This is a topology diagram of the physical prior unscented Kalman filter state reconstruction and observation update data flow in the component abnormal noise remaining life prediction method provided in an embodiment of this application; Figure 4 This is a microscopic evolution drift diagram of high-order harmonics and wavelet energy entropy characteristics before and after short-term durability in the component abnormal noise remaining life prediction method provided in an embodiment of this application.

[0021] Figure 5 This is a topological mapping of phase space noise risk and a multimodal lifetime prediction diagram in the component noise remaining life prediction method provided in an embodiment of this application. Among them, Figure 5 (a) is a mapping diagram of the physical parameter degradation phase space and the BSR risk topology: Figure 5 (b) is the probability evolution prediction diagram of independent abnormal noise modes; Figure 5 (c) is the trajectory diagram of PI-UKF implicit physical parameters reconstruction and blind inference.

[0022] Figure 6 This is a schematic diagram of the component noise and remaining life prediction device provided in the embodiments of this application; Figure 7 This is a schematic diagram of the structure of the terminal device provided in the embodiments of this application. Detailed Implementation

[0023] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0024] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0025] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0026] Figure 1 The flowchart shown is a first embodiment of the component noise remaining life prediction method provided in this application. It is provided as an example and not as a limitation. Figure 1 and Figure 2 As shown, the method may include: S10, Construct a friction-knock coupling microdynamic evolution model based on nonlinear physical priors, obtain microscopic physical parameters that cannot be directly measured based on the friction-knock coupling microdynamic evolution model, and establish physical prior state transition equations based on the microscopic physical parameters as hidden state vectors. This application proposes a gray-box abnormal noise lifetime prediction architecture that deeply integrates "white-box nonlinear physical mechanism data" and "black-box acoustic and vibration feature data." The "white-box" refers to the underlying physical mechanism equations constructed based on nonlinear friction and contact dynamics, used to provide prior predictions of the system's implicit physical state evolution; the "black-box" refers to the data-driven features (such as total harmonic distortion rate and wavelet packet energy entropy) extracted from macroscopic acoustic and vibration signals collected from the test bench, used to provide observation anchors for the actual degradation state. The two are deeply fused using a filtering algorithm.

[0027] The system acquires high-frequency raw waveforms from sensors during short-cycle bench tests and extracts feature clusters by time-sliding window slicing. These clusters are then input as observations into an "unscented Kalman filter (PI-UKF)" jointly constructed by the LuGre dynamic friction model and the Hunt-Crossley contact dissipation model. Through a "blind push and targeted correction" mechanism under sparse observations and non-negative physical boundary constraints, the system reconstructs the microscopic stiffness and clearance states that cannot be directly measured in real time. Finally, in the 2D physical phase space, the system combines Logistic mapping to output the probability of noise outbreaks for two independent modes (i.e., Rattle noise dominated by clearance loosening and Squeak noise dominated by stiffness decay), achieving accurate prediction of the residual life (RUL) of component noise under small sample conditions.

[0028] This application is based on a multi-degree-of-freedom high-frequency abnormal noise test bench, a micro accelerometer, a near-field microphone, and a computing node that includes state estimation matrix operations.

[0029] Among them, the friction-impact coupling microdynamic evolution model includes a friction microdynamic evolution model that incorporates impact-friction coupling parameters and a nonlinear physical prior impact abnormal noise microdynamic evolution model.

[0030] This application establishes underlying micro-dynamic evolution models for friction noise and impact noise respectively. In order to characterize the mutual excitation of the two failure modes at the physical level, a cross-coupling mechanism is established between friction and impact dynamics, which serves as a priori constraint for the state transition of the subsequent filtering algorithm.

[0031] As one implementation method, the microscopic physical parameters include microscopic contact stiffness. and microscopic physical gaps A friction-knock coupling microdynamic evolution model based on nonlinear physical priors is constructed. Based on this model, microscopic physical parameters that cannot be directly measured are obtained. Hidden state vectors are derived from these microscopic physical parameters, and physical prior state transition equations are established. Specifically, this includes: A1. Construct a frictional microdynamic evolution model incorporating impact friction coupling parameters, and obtain the microscopic contact stiffness, which cannot be directly measured, based on the frictional microdynamic evolution model. ; As one implementation method, a frictional microdynamic evolution model incorporating bidirectional cross-coupling terms is constructed. Based on this model, the microscopic contact stiffness, which cannot be directly measured, is obtained. Specifically, it includes: B1, Obtain the transient temperature of the contact surface. Ambient temperature Relative sliding speed Component rotation speed and component forces; B2, based on ambient temperature , and the transient temperature of the contact surface ,Establish Dynamic friction of the interface The accompanying temperature evolution equation; B3, based on ambient temperature and contact surface transient temperature Thus, the microscopic contact stiffness of the thermal-stiffness coupling mapping is obtained; B4, based on relative sliding speed Establish dynamic friction at the interface Microscopic elastic bristle deformation Microscopic contact stiffness The equation for dynamic friction at the interface; B5, based on relative sliding speed Establishing micro-elastic bristle deformation Microscopic contact stiffness Mapping function with nonlinear friction velocity The Stribeck constraint equations; B6, given initial microdeformation Based on the Stribek constraint equations, the time derivative of the microscopic deformation is calculated in the time domain using the fourth-order Runge-Kutta method. Perform numerical integration and update the average bristle deformation in real time. ; and the time derivative Average deformation of bristles Relative sliding speed Substituting the interface dynamic friction equation, the Stribeck effect equation, and the Stribeck constraint equation, the microscopic contact stiffness is calculated. ; To address frictional noise, a dynamic LuGre brush friction model considering temperature field coupling is introduced to characterize the stick-slip degradation of the contact interface. Frequent friction at the contact interface leads to microscopic heat generation, and an accompanying temperature evolution equation is established:

[0032] in, The transient temperature of the contact surface; The ambient temperature (obtained directly by a temperature and humidity sensor of the test environment); Frictional heat conversion rate; Here, represents the heat convection dissipation coefficient (both are conventional physical parameters, obtainable through standard tribothermodynamic bench tests in existing technologies or by consulting material handbooks). Considering the high-temperature softening effect of polymer materials, a thermo-stiffness coupling mapping is constructed:

[0033] in, The initial microscopic contact stiffness at the reference temperature (in N / m, this initial parameter is obtained by equivalent conversion after measuring the material modulus using a nanoindentation tester or an existing material tensile testing machine). This is the temperature-stiffness softening coefficient of the material. By substituting the friction mechanism equation, the cross-scale degradation transfer from thermodynamics to kinetics is realized.

[0034] A disturbance term affecting the frictional slip velocity due to impact is introduced into the tribological model. Real impact vibrations alter the normal pressure distribution at the contact interface and induce localized micro-slippage. The corrected effective slip velocity... Represented as:

[0035] in, The coefficient of friction due to impact. This represents the transient collision embedding quantity. Replace the original equation Substituting into the nonlinear Stribeck friction constraint equation forms .

[0036] Define the micro-elastic bristle deformation Interface dynamic friction Expressed as:

[0037] Among them, the deformation evolution equation is affected by the nonlinear Stribeck effect. constraint:

[0038] The above The evolution process is as follows: given the initial microscopic deformation Based on the above ordinary differential equation, the fourth-order Runge-Kutta method is used in the time domain to... Perform numerical integration to update the bristle deformation in real time. The characters in the above formula have the following meanings: : Dynamic frictional force generated at the contact interface due to friction; : The average deformation state variable of the microscopic elastic bristles; The time derivative of microscopic deformation; The relative sliding velocity of the friction interface; Microscopic contact stiffness (a core state parameter, in N / m, with an initial value determined by material properties, which will implicitly decay as the durability cycle progresses; its hidden state as a filter is adaptively estimated by the algorithm). : Microscopic bristle damping coefficient (a conventional parameter, in N·s / m, obtained through basic friction calibration tests or set as an initial empirical value); Macroscopic viscous friction coefficient (a conventional parameter, in N·s / m, obtained in the same way as above); : A nonlinear friction velocity mapping function that includes the Stribeck effect.

[0039] A2. Construct a micro-dynamic evolution model of percussion noise based on nonlinear physical priors, and obtain the micro-physical gaps that cannot be directly measured based on the micro-dynamic evolution model of percussion noise. ; As one implementation method, a microscopic dynamic evolution model of percussion noise based on nonlinear physical priors is constructed. Based on this model, the microscopic physical gaps that cannot be directly measured are obtained. Specifically, it includes: C1, obtain the physical clearance of the connection point (contact point) of the contact component. Excitation acceleration Relative velocity of collision Local nonlinear collision force , Basic relaxation gap and contact feature length ; C2, based on the relative velocity of the collision and local nonlinear collision force The transient collision embedding amount is obtained. ; C3, based on excitation acceleration The absolute displacements (first displacement and second displacement) at both ends of the contact point are calculated; based on the transient collision embedment amount... The first displacement and the second displacement are used to obtain the physical gap at the connection point (contact point) of the contact components. (Basic relaxation gap caused by pure mechanical wear) ); C4, based on the basic relaxation gap , characteristic length of contact component Ambient temperature and the transient temperature of the contact surface To obtain the microscopic physical gap .

[0040] For Rattle: A Hunt-Crossley contact model considering hysteresis dissipation is introduced. This is when the physical gap at the connection point... As vibration relaxation increases, local nonlinear impact force for:

[0041] A feedback mechanism based on the thermal deformation of the gap is introduced into the impact dynamics model, incorporating frictional temperature rise. The high temperatures generated by intense interfacial friction cause thermal expansion and contraction of the layout materials, dynamically altering the mating gap. The corrected physical gap is then determined. Express:

[0042] in, is the nonlinear correction coefficient for thermal expansion, and is a constant for conventional materials; For interface temperature rise; The loosening gap in the base is caused by pure mechanical wear; The coefficient of linear expansion of the component material (refer to the material handbook). This refers to the feature length of the contact element. The meanings of the characters in the above formula are as follows: Local nonlinear collision force; : Contact stiffness coefficient (initial value obtained through static mechanical indentation test); Nonlinear exponent (a common parameter, typically taken as 1.5 for point contact based on Hertz contact theory); Hysteresis damping dissipation coefficient (obtained by conversion using the material restitution coefficient formula); Transient collision embedding quantity, given by the formula Calculated; The relative velocity of the collision; , : Absolute displacement at both ends of the collision contact point (obtained by quadratic integration derivation through input excitation acceleration in the dynamic simulation model); Time-varying physical gap (core state parameter, which will gradually increase as durability relaxes).

[0043] A3. Based on the microscopic physical parameters and the impact friction coupling parameters, the hidden state vector is obtained. ; The micro-contact stiffness Microscopic physical gaps The hidden state vector of the system is obtained from the impact friction coupling parameters. .

[0044] The core microscopic physical parameters of the above evolution (i.e., the microscopic physical gaps) Microscopic contact stiffness and impact friction coupling parameters Extracted as the hidden state vector of the system This hidden state vector represents the essential physical quantity that actually degrades within the system but cannot be directly measured by sensors.

[0045] A4, based on the hidden state vector Establish the physical prior state transition equations.

[0046] Based on the system hidden state vector Establish the prior state transition equation; and use nonnegative physical boundary constraints to define the hidden state vector of the system. Apply constraints and store the state variables from the previous time step. and current equivalent test time / mileage Substituting the continuous differential equations of the dynamic LuGre brush friction model and the Hunt-Crossley contact model, the time step is determined using the fourth-order Runge-Kutta method. The discretized numerical integral calculation yields the microscopic gap prior prediction value at the current time. and the prior prediction of microscopic stiffness at the current moment .

[0047] That is, to establish the physical prior state transition equations: ; To achieve dynamic deduction at discrete time steps, the above state transition function The specific integral form is expressed as: ; The characters in the above formula have the following meanings: :time The hidden state vector; :time The hidden state vector; , : respectively time The physical gap and microscopic contact stiffness; Based on the aforementioned derivation of the nonlinear state transition function using the LuGre and Hunt-Crossley mechanisms, the specific processing logic is as follows: The state variables from the previous time step... and current equivalent test time / mileage Substituting into the LuGre and Hunt-Crossley continuity differential equations, the fourth-order Runge-Kutta method is used to determine the time step. The discretized numerical integral calculation yields the prior predictions of the micro gap and stiffness at the current moment. Physical time or equivalent accelerated test mileage; Process noise represents unmodeled dynamics and random disturbances.

[0048] To ensure the rigor of the physical meaning and the convergence of the subsequent covariance matrix decomposition, nonnegative physical boundary constraints are introduced into the state transition equation: ,in It is an extremely small positive number, ensuring that the physical clearance and stiffness are always greater than zero.

[0049] S20: Acquire the high-frequency raw waveforms of components in their initial state and under short-cycle accelerated durability load spectrum. Perform noise reduction preprocessing on the high-frequency raw waveforms using noise-resistant frequency response function estimation. Then, slice the long-time waveform using a time sliding window, calculate the acoustic vibration characteristics of each slice, and obtain the acoustic vibration characteristic observation vector. ; Among them, the acoustic vibration feature observation vector The degraded feature observation vector With the initial multidimensional observation vector The difference; As one implementation method, high-frequency raw waveforms of components in their initial state and under short-period accelerated durability load spectrum are acquired. Noise reduction preprocessing is performed on the high-frequency raw waveforms using noise-resistant frequency response function estimation. Furthermore, a time sliding window is used to slice the long-term waveform, and the acoustic-vibration characteristics of each slice are calculated to obtain the acoustic-vibration characteristic observation vector. Specifically, it includes: D1, acquires high-frequency raw waveform signals of components in their initial state and under short-cycle accelerated durability load spectrum; The new components were fixed to the test bench, and broadband random vibration and sinusoidal sweep signals were input. A fixed-duration high-frequency raw waveform was continuously acquired using an accelerometer and a near-field microphone. The high-frequency raw waveform signal included the vibration signal acquired by the accelerometer and the sound pressure signal acquired by the near-field microphone.

[0050] D2, the original high-frequency waveform signal is preprocessed for noise reduction by estimating the noise resistance frequency response function to obtain a noise-reduced long-time waveform signal; To eliminate the background noise of the test bench itself, the system uses the input excitation reference acceleration and the measurement point response acceleration to calculate the noise immunity frequency response function estimate. :

[0051] The characters in the above formula have the following meanings: : Estimated value of frequency response function; Input excitation reference signal (corresponding signal) ) and the measurement point response signal obtained through the accelerometer (corresponding signal) The cross-power spectral density of the original high-frequency waveform signal; Input excitation reference signal (corresponding signal) The self-power spectral density of ).

[0052] in, and The data is obtained by windowing (e.g., Hanning window) and performing Fast Fourier Transform calculations on the acquired time-series signal using the Welch method. Primarily used as a preprocessing technique for subsequent feature extraction, the subsequent feature calculations will be directly based on the data obtained from the previous process. The reconstructed response signal after filtering and noise reduction is processed.

[0053] D3, using a time sliding window with a fixed step size to slice the noise-reduced long-time waveform signal, and calculating the acoustic and vibration characteristics of each slice; To overcome the vulnerability of single-point features to mechanical noise and transient changes, the system applies a fixed-step time sliding window to slice the denoised long-time waveform. For each slice, a representation is calculated separately, that is, the feature scatter clusters of each slice are extracted.

[0054] D4. Based on the acoustic vibration characteristics, the initial multidimensional observation vector is obtained. and degenerate feature observation vector .

[0055] Among them, the initial multidimensional macroscopic observation vector This includes the initial total harmonic distortion rate and the initial wavelet packet energy entropy. Among these, the degraded feature observation vector... This includes the total harmonic distortion rate after degradation and the wavelet packet energy entropy after degradation.

[0056] To ensure the repeatability and statistical reliability of the initial features, the system acquires at least three independent waveform signals before durability testing. After slicing and calculating, the statistical mean is taken to form the initial multidimensional macroscopic observation vector. The initial multidimensional macroscopic observation vector includes the initial total harmonic distortion rate. and the initial wavelet packet energy entropy ; Among them, the initial total harmonic distortion rate and the degraded total harmonic distortion The total harmonic distortion rate (THD) of the nonlinearity of the impact gap can be used to determine this. The formula for calculating ) is: ; in, Total harmonic distortion (THD); The amplitude of the fundamental frequency (excitation frequency) of the response signal; The first response signal Second-highest order harmonic amplitude; This represents the highest calculated harmonic order.

[0057] The total harmonic distortion rate (THD), characterizing the nonlinearity of the striking gap, is calculated by performing an FFT (Fast Fourier Transform) on the signal slice. If the excitation is a sinusoidal sweep, the fundamental frequency... This is the amplitude at the current transient frequency of the frequency sweep excitation; if the excitation is broadband random, the fundamental frequency is... Defined as the amplitude at the first principal resonant frequency of the component, which is measured in advance through modal testing. For the first Second harmonic amplitude Take 5.

[0058] Among them, the initial wavelet packet energy entropy and the degenerate wavelet packet energy entropy The wavelet packet energy entropy, which characterizes the increase in high-frequency transient disorder caused by the deterioration of the stick-slip friction interface, can be used to represent this. The formula for calculating ) is: ; in, The wavelet packet energy entropy; For the first The probability ratio of the energy of a high-frequency wavelet packet frequency band to the total energy of all high-frequency frequency bands; The total number of high-frequency sub-bands.

[0059] The method for calculating the wavelet packet energy entropy, which characterizes the increase in high-frequency transient disorder caused by the deterioration of the stick-slip friction interface, is as follows: The system uses Daubechies wavelets for three-level wavelet packet decomposition, dividing the high-frequency band into... Each sub-band. Calculate the energy of each sub-band. probability ratio Among them, the Dobesh wavelet can be db4. The Dobesh wavelet is one of the most commonly used orthogonal wavelet bases in the field of wavelet analysis.

[0060] D5, based on the initial multidimensional macroscopic observation vector and degenerate feature observation vector The acoustic vibration characteristic observation vector is obtained. .

[0061] Among them, acoustic vibration characteristic observation vector That is, the multidimensional feature evolution drift. (The characteristic scatter plot micro-evolution drift between the initial state and the short-period endurance state is as follows) Figure 3 (As shown), this is used to capture the early implicit degradation rate of the structure. Multidimensional feature evolution drift. It can be used as the acoustic vibration feature observation vector for the actual update of the unscented Kalman filter.

[0062] Calculate the multidimensional feature evolution drift The aim is to eliminate the static feature dispersion caused by manufacturing tolerances and initial assembly stresses in parts. By anchoring the relative dynamic drift rate of features, background manufacturing noise can be stripped away, making subsequent filters sensitive only to implicit degradation caused by actual wear and relaxation, thus ensuring robustness of cross-batch part prediction under small sample conditions.

[0063] S30, Establish a connection between the hidden state vectors With the acoustic vibration feature observation vector The nonlinear observation mapping equation is obtained; an unscented Kalman filter is used to combine the physical prior state transition equation with the acoustic vibration feature observation vector. Deep fusion is performed, and blind inference and targeted correction mechanisms under sparse observation are used to jointly inversely reconstruct the hidden state vector and output the posterior estimate of the microscopic physical parameters. Based on pre-collected bench calibration data, a system is constructed that can extract the hidden state vector. With the acoustic vibration feature observation vector The nonlinear observation mapping equation is described. This equation can be implemented using a data-driven surrogate modeling method. Its internal parameters are obtained through offline supervised learning training and remain fixed during the online prediction phase. It is used to simulate the complex nonlinear relationship between physical degradation processes and measurable response signals. The role of the nonlinear observation mapping equation is to represent the implicit physical states... ( Mapped to measurable acoustic vibration features (acoustic vibration feature observation vector) ( The nonlinear observation mapping equation employs a radial basis function neural network (RBF) or multinomial regression to construct a surrogate model. Proxy model The bridge is obtained through offline training using historical calibration data from the test bench. Its internal parameters (center, width, weight) are fixed and not updated during runtime, serving only as a static mapping bridge.

[0064] As one implementation method, a connection hidden state vector is established. Acoustic vibration characteristic observation vector The nonlinear observation mapping equations include:

[0065] in, For a moment The acoustic vibration characteristic observation vector; This refers to the measurement and observation noise of sensors and the environment.

[0066] This is a nonlinear observation function that connects microscopic wear physical quantities with macroscopic acoustic and vibration characteristics. A surrogate model based on radial basis function neural networks or multinomial regression is adopted, which is trained from historical calibration data of the test bench, and establishes a model from physical parameters (i.e., hidden state vectors). ( , )) to acoustic vibration characteristic observation vector The mapping; where the function for:

[0067] in, This represents the total number of hidden layer nodes in the RBF network. For the first The center vector of each basis function (with the same dimension as the hidden state vector); For the first The width of each basis function; This is the weight matrix from the hidden layer to the output layer; This is the bias vector. The internal parameters of the above model ( , , , All of these were acquired during the bench calibration phase before system operation by collecting known degraded samples and their acoustic and vibration feature datasets, and then using offline supervised learning training to solidify the data.

[0068] Among them physical parameters Including microscopic physical gaps Microscopic contact stiffness .

[0069] like Figure 3 As shown, as one implementation method, an unscented Kalman filter is used to deeply fuse the physical prior state transition equation with the acoustic vibration feature observation vector. Utilizing a blind inference and targeted correction mechanism under sparse observations, a joint inverse reconstruction is performed to output posterior estimates of the microscopic physical parameters. Specifically, this includes: The unscented Kalman filter model employs a "blind inference and targeted correction" mechanism under sparse observations: During long-period test bench operation without acoustic measurements (the time interval of sparse observations is set according to the accelerated equivalent mileage, such as one measurement every 1000 kilometers), the filter only performs a priori prediction steps (i.e., blind inference), at which time the state covariance matrix only propagates with the Jacobian matrix of the physical differential equation; when the discrete measurement nodes are reached and features (measured acoustic vibration feature observation vectors) are obtained... When this occurs, targeted correction is triggered (its physical prior state reconstruction and observation update data stream topology are as follows). Figure 3 (As shown).

[0070] Given the highly nonlinear nature of friction and collision, the system employs unscented transform (UT) to extract the Sigma point set. The nonlinear probability distribution is propagated through weight allocation. The number of Sigma points is... ( (To augment the state dimension). Its weights are calculated using the standard UT scaling formula: ,in The scaling parameter is used. To achieve self-learning evolution of the mechanistic model, this system constructs a joint estimation and adaptive update mechanism for states and parameters. Conventional filtering algorithms only consider the core states (gap states) that need to be solved. With stiffness As a physical state vector (physical state parameter), but in the complex tribodynamics, the microscopic brush damping coefficient... Hysteresis damping dissipation coefficient Furthermore, the coupling coefficient is not a constant parameter but evolves nonlinearly with the polishing or peeling of the contact surface. Treating the extended parameter part as a first-order Markov process, a process evolution equation with Gaussian white noise is established. The time-varying model parameters in the mechanistic model are concatenated with the core state vector to extend it to the augmented hidden state vector. The latest augmented hidden state vector is defined as... The aforementioned auxiliary physical parameters and coupling coefficients are assumed to follow a Gaussian random walk evolution model: ,in . : respectively time With time The auxiliary physical parameter vector (which includes the stiffness degradation parameter in this embodiment) With damping degradation parameters wait); : The noise vector of the random walk process of auxiliary parameters during the evolution process; : indicates that the mean is The covariance matrix is The Gaussian (normal) distribution; The auxiliary parameter is the process noise covariance matrix (the size of the diagonal elements of this matrix is ​​used to control the drift amplitude or permissible range of variation of the mechanistic parameters during self-learning updates). The mechanistic parameters include stiffness degradation parameters (microscopic bristle damping coefficients). With damping degradation parameters (hysteresis damping dissipation coefficient) .

[0071] Calculate the augmented Kalman gain matrix And calculate the actual measurable acoustic vibration eigenvector. With model-predicted feature vectors The characteristic residuals between them are used. These residuals are weighted using an augmented Kalman gain matrix, and the physical state and mechanistic parameters are simultaneously corrected in reverse.

[0072] To ensure physical meaning, after the Kalman gain update is completed, a non-negative physical boundary constraint operation is introduced on the posterior state: This ensures that the clearance and stiffness do not become negative.

[0073] Through the aforementioned joint self-learning mechanism, the mechanistic equation not only corrects the current degradation damage value during each posterior update, but also adaptively adjusts the attenuation characteristics of the friction and damping model itself, greatly reducing the dependence of the physical prior model on the empirical calibration of the initial fixed parameters.

[0074] The characters in the above formula have the following meanings: :time The posterior state estimate that integrates physical mechanisms and measured data; :time Prior prediction of augmented state values ​​based on pure mechanistic model deduction; The augmented Kalman gain matrix is ​​calculated dynamically. : Predicted feature vectors mapped from prior state estimation.

[0075] As one implementation method, an unscented Kalman filter is used to deeply fuse the hidden state vector and the acoustic vibration feature observation vector through a nonlinear observation mapping equation. Utilizing a blind inference and targeted correction mechanism under sparse observations, the hidden state vector is jointly inversely reconstructed to output posterior augmented state estimates of the microscopic physical parameters. Specifically, this includes: E1 sets the triggering logic for sparse observations; During the time interval when the acoustic vibration characteristic observation vector is not obtained, only blind inference based on the physical prior state transition equation is performed; when the preset discrete measurement node is reached and the measured acoustic vibration characteristic observation vector is obtained, At that time, the targeted correction process is triggered to calculate the Kalman gain and update the posterior state estimate; The unscented Kalman filter model employs a "blind inference and targeted correction" mechanism under sparse observations: During long-period test bench operation without acoustic measurements (the time interval of sparse observations is set according to the accelerated equivalent mileage, such as one measurement every 1000 kilometers), the filter only performs a priori prediction steps (i.e., blind inference), at which time the state covariance matrix only propagates with the Jacobian matrix of the physical differential equation; when the discrete measurement nodes are reached and features (measured acoustic vibration feature observation vectors) are obtained... When this occurs, targeted correction is triggered (its physical prior state reconstruction and observation update data stream topology are as follows). Figure 4 (As shown).

[0076] Measured acoustic vibration characteristic observation vector The measured degraded feature observation vector With the initial multidimensional observation vector The difference is the measured acoustic vibration characteristic observation vector. .

[0077] E2, obtain the measured acoustic vibration characteristic observation vector. At that time, the Sigma point set is extracted using unscented transformation, and the predicted observation feature vector is obtained based on the augmented state prior prediction of the unscented transformation. Observation covariance and state observation cross-covariance ; Measured acoustic vibration characteristic observation vector The measured degraded feature observation vector With the initial multidimensional observation vector The difference is the measured acoustic vibration characteristic observation vector. .

[0078] As one implementation method, an unscented transform is used to extract the Sigma point set, and the predicted observation feature vector is obtained based on the augmented state prior prediction of the unscented transform. Observation covariance and state observation cross-covariance Specifically, it includes: F1, based on the mean of the posterior augmented state estimate from the previous time step. Covariance The lower triangular square root matrix of the covariance matrix is ​​calculated using Cholesky decomposition. Generate 2n+1 Sigma point sets and Sigma point weights ; in, ; ; ; ; ; in, Represents the lower triangular square root matrix The Column direction, To increase the dimension of the state vector, To determine the scaling parameter for the divergence of the Sigma point distribution, and The hyperparameters are preset UT transformation parameters; the augmented state vector includes at least the microscopic contact stiffness. Microscopic physical gaps And auxiliary physical parameters.

[0079] F2, substitute each Sigma point in the Sigma point set into the physical prior state transition equation for propagation, and calculate the weighted mean of the prior augmented state estimate at the current time. and prior covariance ; ; ; in, The process noise covariance matrix for the augmented state; F3 maps the prior state Sigma point to the predicted observation Sigma point through the nonlinear observation mapping equation, and then calculates the predicted observation feature vector by weighting. Observation covariance and state observation cross-covariance ; in, ; ; ; in, To measure the observation noise covariance matrix; E3, based on the measured acoustic vibration characteristic observation vector The predicted observation feature vector The observation covariance The state observation cross-covariance Mean of prior augmented state estimates and the hidden state vector Based on feature residual weighted targeted correction and non-negative physical boundary constraints, the posterior augmented state estimate is obtained. and posterior augmented state covariance .

[0080] As one implementation method, based on the measured acoustic vibration characteristic observation vector The predicted observation feature vector The observation covariance The state observation cross-covariance Mean of prior augmented state estimates and the hidden state vector Based on feature residual weighted targeted correction and non-negative physical boundary constraints, the posterior augmented state estimate is obtained. and posterior augmented state covariance Specifically, it includes: G1, based on the measured acoustic vibration characteristic observation vector and the predicted observation feature vector The characteristic residuals are obtained. ; As one implementation method, based on the measured acoustic vibration characteristic observation vector and the predicted observation feature vector The characteristic residuals are obtained. Specifically, it includes: ; in, For characteristic residuals, This represents the acoustic vibration characteristic observation vector; To predict the observed feature vector.

[0081] G2, based on the state observation cross-covariance and the observed covariance The augmented Kalman gain matrix is ​​obtained. ; As one implementation method, based on the state observation cross-covariance and the observed covariance The augmented Kalman gain matrix is ​​obtained. Specifically, it includes: ; in, To augment the Kalman gain matrix, For state observation cross-covariance; To observe the covariance.

[0082] G3, based on characteristic residuals Mean of prior augmented state estimates and augmented Kalman gain matrix The microscopic physical parameters, the mechanistic parameters in the physical prior state transition equation, and the impact-friction coupling parameters are jointly calibrated and updated to obtain the constrained and prior a prior augmented state estimates. In the time update step of the unscented Kalman filter, the covariance of the state and parameters is simultaneously transmitted using the joint Jacobian matrix or the unscented transformation. In the measurement update step, the measured acoustic vibration characteristic residuals are used. Through the Kalman gain matrix The weighting is assigned, and the physical state (wear amount) is calculated in one step. Mechanism parameters (friction properties) and impact friction coupling parameters Joint calibration and updates are performed to achieve two-way closed-loop collaboration between physical and data processing. The calculation formula is as follows: ; in, To constrain the a priori and post-hoc augmented state estimates; For characteristic residuals; The mean of the prior augmented state estimate; To augment the Kalman gain matrix.

[0083] G4, augmented state estimates before and after constraints through nonnegative physical boundary constraint operations. The process is performed to obtain the posterior augmented state estimate. ; Wherein, is the non-negative physical boundary constraint operation. ; Among them, for microscopic physical gaps : ; Among them, for micro-contact stiffness : .

[0084] Among them, for the impact friction coupling parameters : .

[0085] G5 propagates and updates the error covariance matrix, yielding the quantified value of the uncertainty.

[0086] If the current time is the blind inference stage, the covariance is represented by the state transition Jacobian matrix. Propagation yields a quantified value of uncertainty. .

[0087] ; in, Let k be the prior error covariance matrix at time k; The state transition Jacobian matrix; The posterior error covariance matrix of the previous time step; This is the transpose of the Jacobian matrix; Let be the process noise covariance matrix.

[0088] Wherein, if the current time is the correction phase, then based on the Kalman gain and observation covariance A posterior update of the covariance yields the quantified value of the uncertainty. Uncertainty quantification value for: ; in, Let be the posterior augmented state covariance matrix at time k; The mean of the prior augmented state; Kalman gain; To observe the covariance.

[0089] Then, the updated posterior augmented state estimates are... The posterior augmented state covariance matrix at time k As input to the next time step filtering iteration, it provides a latent state trajectory containing confidence intervals for subsequent remaining lifetime prediction.

[0090] S40, A posteriori augmented state estimate Using these as initial conditions, the friction-knock coupling microdynamic evolution model is continuously extrapolated to future lifecycles to obtain a set of implicit physical parameters; this set of implicit physical parameters includes microphysical gaps under different lifecycles. Microscopic contact stiffness under different life cycles ; A posteriori augmented state estimate Using the initial conditions, the friction-knock coupling microdynamic evolution model is continuously extrapolated to future lifecycles along the extrapolated time step as the time span, resulting in a set of implicit physical parameters. This set of implicit physical parameters includes the microphysical gaps at different lifecycles. Microscopic contact stiffness under different life cycles ; Based on the differential equations (LuGre+Hunt-Crossley) from step S10, the fourth-order Runge-Kutta method (RK4) is used to numerically integrate the state from the previous time step to deduce the new microscopic gap at the current time step. and stiffness Simultaneously, the error covariance matrix is ​​propagated forward to calculate the uncertainty (confidence interval). This is based on the microscopic physical gaps at several time points. and micro-contact stiffness This yields the set of implicit physical parameters.

[0091] That is, by continuously extrapolating the mechanistic evolution equation to the future life cycle (the implicit physical parameter reconstruction and blind extrapolation trajectory of PI-UKF, as shown in...), Figure 4 (As shown in subgraph (c)). The extrapolated time step is... Set the time span to an equivalent distance of 100 kilometers.

[0092] S50, based on the implicit physical quantity dataset and the multimodal competitive logic mapping function, obtains the maximum Lyapunov exponent λ and the probability of knocking abnormal sounds. and the probability of friction noise occurring When the maximum Lyapunov index, the probability of knocking noise, and the probability of friction noise meet the preset conditions, the remaining life prediction value of the abnormal noise of the component is obtained.

[0093] To break free from the subjective reliance of human hearing in traditional NVH testing and establish rigorous mathematical and engineering failure thresholds, this application combines pure mathematical chaos criteria with an engineering probability mapping model.

[0094] As one implementation method, based on the implicit physical quantity dataset and the multimodal competitive logic mapping function, the maximum Lyapunov exponent λ and the probability of knocking abnormal sounds are obtained. and the probability of friction noise occurring When the maximum Lyapunov index, the probability of knocking noise, and the probability of friction noise meet preset conditions, the remaining life prediction value of the component noise is obtained, specifically including: H1, based on the dataset of hidden physical quantities, yields the maximum Lyapunov exponent; Introducing the maximum Lyapunov exponent This serves as the underlying mathematical criterion for determining the occurrence of broadband anomalies (macroscopic failure) in the system. Failure is determined when the system evolves from a periodic response bifurcation to a high-frequency chaotic knocking state.

[0095] The characters in the above formula have the following meanings: The maximum Lyapunov exponent of a dynamical system. The calculation is based on the system's relative motion state variables (such as relative displacement in the Hunt-Crossley model). and its speed); Time variable; The magnitude of the small perturbation vector in the initial state of the system, with the initial perturbation magnitude chosen as... Microvectors at the meter level; Time The magnitude of the phase space trajectory perturbation of the evolved system. This is to calculate the magnitude of the phase space trajectory perturbation in discrete time-domain data. The system employs an evolutionary algorithm based on the Jacobian variational equation, and the specific calculation steps are as follows: ① Using the reconstructed posterior physical state .

[0096] ② Regarding the state transition function Solve for the state vector The partial derivatives are used to obtain the transient Jacobian matrix of the system on the current trajectory. .

[0097] ③Establish the linear variational equation for the perturbation evolution .

[0098] ④ At the initial moment, a very small orthogonal perturbation vector is assigned. (its Euclidean norm modulus) ).

[0099] ⑤ The fourth-order Runge-Kutta method (RK4) is used to compare the original system equations with the variational equations on the time scale. Perform synchronous numerical integration to obtain Perturbation vector at time step The required trajectory perturbation magnitude can be obtained by taking its Euclidean norm. The maximum Lyapunov exponent is then calculated by substituting it into the limit formula, which is used to quantitatively determine whether the system has experienced a high-frequency chaotic bifurcation (abnormal burst).

[0100] To solve the limiting formula for the Lyapunov exponent, a variational equation for the system's state evolution needs to be constructed. In calculating the partial derivatives of the augmented Jacobian matrix corresponding to this variational equation, the equivalent relative velocity, including the perturbation term, is considered. The following expression is parsed.

[0101]

[0102] in, ; It is a symbolic function; , is the derivative of the nonlinear Stribeck function with respect to velocity; The partial derivative terms will be introduced into the augmented Jacobian matrix. Location; The input is a numerical integration operator. The input is the state variable and time step of the previous time step. The output is the predicted value of the micro gap at the current time step.

[0103] H2, based on the implicit physical quantity dataset and the multimodal competition logic mapping function, yields the probability of knocking noise and the probability of friction noise. In two-dimensional physical space ( In this model, the two failure modes compete with each other, creating a competitive relationship. A multimodal competitive logic mapping (such as the Softmax function) is used to output the final competitive failure probability distribution, rather than two independent probabilities. The multimodal competitive logic mapping function is a Logistic regression mapping function.

[0104] To address the rattle noise dominated by increased clearance and the squeak noise dominated by decreased stiffness, a coupling competition risk function is constructed: Tap risk function: Friction risk function: The Softmax function is used to output the probability of contention failure.

[0105]

[0106] The characters in the above formula have the following meanings: The probability of a rattling sound occurring (range: ); : Impact noise sensitivity coefficient (the slope of the surge in control probability as the gap changes) (obtained by fitting and calibration based on prior engineering experience or historical material bench test data); The current microscopic physical gap parameter is not calculated directly from a single static formula, but rather from the posterior state estimate output by the PI-UKF filter in step S4. As initial conditions, in the blind extrapolation stage detached from the actual test data, the data is dynamically obtained by continuous numerical integration (i.e. extrapolation) in the time domain, relying purely on the state transition equation in step S1. The critical gap threshold for high-frequency chaos triggered by knocking noises corresponds to the maximum Lyapunov exponent of the system. (Gap state value when first crossing zero). Single physical quantity The mapping method is as follows: in the aforementioned only containing During the integral extrapolation process, the real-time computing system When representing the overall dynamic response of the system. When the system first changes from negative to positive (a chaotic bifurcation occurs when crossing zero), it immediately records the physical gap value corresponding to the current simulation moment. and use it as the critical gap threshold. .

[0107] The probability of occurrence of frictional noise (squeak) (range: ); Friction noise sensitivity coefficient (obtained based on prior engineering experience or friction material test calibration); : Current microscopic contact elastic stiffness. Similarly, the algorithm for obtaining this parameter starts with the posterior state output of the filter in step S4, and follows the physical prior mechanism equation established in step S1 to continuously integrate and extrapolate over future time steps; The critical stiffness threshold for frictional noise to trigger a sudden stick-slip transition (corresponding to the system's maximum Lyapunov exponent) Stiffness state value when first crossing zero.

[0108] This model ensures the probability conservation of the total system failure risk and accurately identifies the dominant abnormal noise modes that cause degradation.

[0109] Final determination and output: The physical time of the system extrapolating the mechanistic equations. The system maps the set test load spectrum or vehicle operating profile to an equivalent mileage. Simultaneously, to quantify the uncertainty of the prediction results, the system utilizes the error covariance matrix synchronously transmitted during the blind inference phase using the PI-UKF method. Error interval extrapolation is performed. During the blind extrapolation phase, the covariance matrix is ​​calculated according to... Continued transmission (of which) (This is the linearized Jacobian matrix of the nonlinear state transition equation). Let the covariance matrix at the final extrapolation to the failure point be... Extract the diagonal elements of the variance corresponding to the gap or stiffness. The system assumes that the error follows a Gaussian distribution, through... _Calculate the dynamic confidence band for predicted lifetime, where is the quantile of the standard normal distribution.

[0110] H3, when the maximum Lyapunov index, the probability of knocking noise, and the probability of friction noise meet the preset conditions, the remaining life prediction value of the component noise is obtained.

[0111] The preset conditions are: the maximum Lyapunov exponent λ>0, the probability of knocking noise is greater than the preset knocking probability threshold, or the probability of friction noise is greater than the preset friction probability threshold.

[0112] When the maximum Lyapunov exponent λ>0, the probability of knocking noise is greater than the preset knocking probability threshold, or the probability of friction noise is greater than the preset friction probability threshold, the current equivalent mileage is taken as the equivalent mileage when the failure red line is triggered, and the remaining life prediction value of the component noise is obtained based on the equivalent mileage when the failure red line is triggered and the equivalent mileage that has been experienced.

[0113] Among them, the remaining life prediction value RUL for abnormal noise of components = equivalent mileage at the time of triggering the failure red line - equivalent mileage already experienced.

[0114] That is, the system synchronously plots independent failure probability evolution curves (such as those for different types of abnormal noise) on the host computer. Figure 5 (b) shown), and 2D phase space risk topology cloud map (as shown in the image). Figure 5 (a) is shown). And the implicit parameter trajectory plot with a 95% confidence region (as shown). Figure 5 (c) shows this). In the continuous extrapolation evolution, when the system's maximum Lyapunov exponent crosses zero ( When the probability of occurrence of any independent mode exceeds the high-risk warning line set by the project (such as 85%), the system will automatically stop the simulation and output to the user a precise multimodal abnormal noise outbreak mileage (i.e., RUL distribution band) containing confidence intervals.

[0115] This application constructs a friction-impact coupling microdynamic evolution model based on nonlinear physical priors, and obtains microscopic physical parameters that cannot be directly measured based on the friction-impact coupling microdynamic evolution model; it obtains hidden state vectors and establishes physical prior state transition equations based on the microscopic physical parameters; it collects high-frequency raw waveforms of components in their initial state and under short-period accelerated durability load spectrum, performs noise reduction preprocessing on the high-frequency raw waveforms through noise-resistant frequency response function estimation, and slices the long-time waveforms using a time sliding window to calculate the acoustic vibration characteristics of each slice, obtaining an acoustic vibration characteristic observation vector; the acoustic vibration characteristic observation vector includes an initial multidimensional observation vector and a degraded characteristic observation vector; it establishes a nonlinear observation mapping equation connecting the hidden state vector and the acoustic vibration characteristic observation vector; and it uses an unscented Kalman filter to augment the hidden state vector. The acoustic vibration feature evolution drift is deeply fused with the nonlinear observation mapping equation. Using a blind inference and targeted correction mechanism under sparse observation, the hidden state vector is jointly reconstructed inversely to obtain the posterior augmented state estimate of the microscopic physical parameters. Using the posterior augmented state estimate as the initial condition, the model is continuously extrapolated to future lifecycles along the friction-impact coupling microdynamic evolution model to obtain a set of implicit physical parameters. This set includes microscopic physical gaps and microscopic contact stiffness at different lifecycles. Based on the implicit physical quantity dataset and the multimodal competitive logic mapping function, the maximum Lyapunov exponent, the probability of impact noise, and the probability of friction noise are obtained. When the maximum Lyapunov exponent, the probability of impact noise, and the probability of friction noise meet preset conditions, the remaining lifespan prediction of the component noise is obtained. This achieves accurate prediction of the remaining lifespan of component noise under small sample conditions.

[0116] This application combines a nonlinear dynamic contact / friction mechanism model with the early weak acoustic and vibration characteristic evolution rate, and uses a physical constraint filtering algorithm to reconstruct the latent degradation state of component noise (BSR) and predict the remaining life in advance.

[0117] This application overcomes the pain point of having to run the entire process before drawing conclusions. By extracting minute acoustic and vibration drift data from the very early stage (e.g., 1 / 10 of the lifespan), the occurrence points of abnormal noise throughout the entire lifespan can be extrapolated, shortening the bench verification cycle by more than 70%. This drastically reduces the verification cycle and cost.

[0118] This application overturns the heavy reliance of pure black-box AI on massive amounts of failure data. It transforms physical mechanisms into prior constraints for filters, combining a small number of measured features to achieve self-correction, exhibiting excellent physical interpretability and cross-material generalization ability. It achieves extremely high prediction robustness with small sample sizes.

[0119] This application focuses on high harmonic distortion (THD) and wavelet energy entropy (WPEE), which reflect nonlinearity, and uses mathematical methods to invert microscopic contact stiffness from macroscopic vibration, thus identifying potential problems missed by conventional acoustic testing. This achieves a deep insight into latent degradation.

[0120] This application uses the chaotic bifurcation point where the maximum Lyapunov exponent turns from negative to positive as the mathematical criterion for anomalies, achieving 100% objective and standardized evaluation. It also eliminates the subjectivity of the evaluation indicators.

[0121] For those consistent with the above, please refer to Figure 6 , Figure 6 This application provides a schematic diagram of a device for predicting the remaining life of a component with abnormal noise. For example... Figure 6 As shown, the device includes: The model building module 601 is used to construct a friction-knock coupling microdynamic evolution model based on nonlinear physical priors, obtain microscopic physical parameters that cannot be directly measured based on the friction-knock coupling microdynamic evolution model, obtain hidden state vectors based on microscopic physical parameters, and establish physical prior state transition equations. The observation vector determination module 602 is used to acquire the high-frequency original waveform of the component in its initial state and under the applied short-period accelerated durability load spectrum, perform noise reduction preprocessing on the high-frequency original waveform through noise reduction frequency response function estimation, and slice the long-time waveform using a time sliding window to calculate the acoustic vibration characteristics of each slice to obtain the acoustic vibration characteristic observation vector; the acoustic vibration characteristic observation vector includes an initial multidimensional observation vector and a degraded characteristic observation vector; The estimation determination module 603 is used to establish a nonlinear observation mapping equation connecting the hidden state vector and the acoustic vibration feature observation vector; and to use an unscented Kalman filter to deeply fuse the hidden state vector and the acoustic vibration feature observation vector through the nonlinear observation mapping equation, and to use a blind inference and targeted correction mechanism under sparse observation to jointly inversely reconstruct the hidden state vector to obtain the posterior augmented state estimate of the microscopic physical parameters. The parameter set determination module 604 is used to continuously extrapolate along the friction-knock coupling microdynamic evolution model to future life cycles, using the a priori augmented state estimate as the initial condition, to obtain a set of implicit physical parameters; the set of implicit physical parameters includes microphysical gaps and microcontact stiffness under different life cycles. The life prediction module 605 is used to obtain the maximum Lyapunov exponent, the probability of knocking noise, and the probability of friction noise based on the implicit physical quantity dataset and the multimodal competition logic mapping function. When the maximum Lyapunov exponent, the probability of knocking noise, and the probability of friction noise meet the preset conditions, the remaining life prediction value of the component noise is obtained.

[0122] like Figure 7 As shown, this application embodiment also provides a terminal device 2, which includes: at least one processor 20, a memory 21, and a computer program 22 stored in the memory 21 and executable on the at least one processor. The processor 20 and the memory 21 are connected. When the processor 20 executes the computer program 22, it implements the steps in the robot motion planning method embodiment.

[0123] This application also provides a computer storage medium storing a computer program for electronic data interchange, which causes a computer to perform some or all of the steps of any of the component noise remaining life prediction methods described in the above method embodiments.

[0124] This application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program that causes a computer to perform some or all of the steps of any of the component noise remaining life prediction methods described in the above method embodiments.

[0125] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable storage medium can include at least: any entity or device capable of carrying computer program code to a device / terminal equipment, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks. In some jurisdictions, according to legislation and patent practice, computer-readable storage media cannot be electrical carrier signals or telecommunication signals.

[0126] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0127] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0128] In the embodiments provided in this application, it should be understood that the disclosed devices / terminal equipment and methods can be implemented in other ways. For example, the device / terminal equipment embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling or direct coupling or communication connection may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0129] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

Claims

1. A method for predicting the remaining life of abnormal noise in components, characterized in that, include: A friction-knock coupling microdynamic evolution model based on nonlinear physical priors is constructed, and microscopic physical parameters that cannot be directly measured are obtained from the friction-knock coupling microdynamic evolution model. The hidden state vector is obtained based on the microscopic physical parameters, and the physical prior state transition equation is established. High-frequency raw waveforms of components under initial state and short-cycle accelerated durability load spectrum are collected. The high-frequency raw waveforms are preprocessed for noise reduction by noise reduction frequency response function estimation. Long-term waveforms are sliced ​​by time sliding window and the acoustic vibration characteristics of each slice are calculated to obtain the acoustic vibration characteristic observation vector. Establish a nonlinear observation mapping equation connecting the hidden state vector and the acoustic vibration feature observation vector; An unscented Kalman filter is used to deeply fuse the hidden state vector and the acoustic vibration feature observation vector through a nonlinear observation mapping equation. By using blind inference and targeted correction mechanism under sparse observation, the hidden state vector is jointly inversely reconstructed to obtain the posterior augmented state estimate of the microscopic physical parameters. Using the a priori augmented state estimate as the initial condition, the model is continuously extrapolated to future life cycles along the friction-knock coupling microdynamic evolution model to obtain a set of implicit physical parameters. The set of implicit physical parameters includes the microphysical gap and the microcontact stiffness under different life cycles. Based on the implicit physical quantity dataset and the multimodal competition logic mapping function, the maximum Lyapunov exponent, the probability of knocking noise, and the probability of friction noise are obtained. When the maximum Lyapunov exponent, the probability of knocking noise, and the probability of friction noise meet the preset conditions, the remaining life prediction value of the abnormal noise of the component is obtained.

2. The method for predicting the remaining lifespan of abnormal noise in components according to claim 1, characterized in that, The microscopic physical parameters include microscopic contact stiffness and microscopic physical gap. The aforementioned construction is based on a nonlinear physical prior to the friction-impact coupling microscopic dynamic evolution model. Based on the friction-impact coupling microscopic dynamic evolution model, microscopic physical parameters that cannot be directly measured are obtained. The hidden state vector is obtained based on the microscopic physical parameters, and the physical prior state transition equation is established, including: A frictional microdynamic evolution model incorporating impact friction coupling parameters is constructed, and the microscopic contact stiffness, which cannot be directly measured, is obtained based on the frictional microdynamic evolution model. A micro-dynamic evolution model of percussion noise based on nonlinear physical priors is constructed, and the micro-physical gap that cannot be directly measured is obtained based on the micro-dynamic evolution model of percussion noise. The hidden state vector is obtained based on the microscopic physical parameters and the impact friction coupling parameters; The physical prior state transition equation is established based on the hidden state vector.

3. The method for predicting the remaining lifespan of abnormal noise in components according to claim 1, characterized in that, The high-frequency raw waveforms of the acquired components under initial state and short-period accelerated durability load spectrum are preprocessed for noise reduction using noise-resistant frequency response function estimation. Furthermore, a time sliding window is used to slice the long-time waveform, and the acoustic-vibration characteristics of each slice are calculated to obtain the acoustic-vibration characteristic observation vector, including: High-frequency raw waveform signals of components under initial state and under short-cycle accelerated durability load spectrum are collected; The original high-frequency waveform signal is preprocessed by noise reduction through noise immunity frequency response function estimation to obtain a noise-reduced long-time waveform signal. The noise-reduced long-time waveform signal is sliced ​​using a time sliding window, and the acoustic and vibration characteristics of each slice are calculated. Based on the acoustic vibration characteristics, the initial multidimensional observation vector and the degraded feature observation vector are obtained; Based on the initial multidimensional observation vector and the degraded characteristic observation vector, the acoustic vibration characteristic observation vector is obtained.

4. The method for predicting the remaining lifespan of abnormal noise in components according to claim 1, characterized in that, The method employs an unscented Kalman filter to deeply fuse the hidden state vector and the acoustic vibration feature observation vector through a nonlinear observation mapping equation. Utilizing a blind inference and targeted correction mechanism under sparse observations, the hidden state vector is jointly inversely reconstructed to obtain posterior augmented state estimates of the microscopic physical parameters, including: Define the triggering logic for sparse observations; When the measured acoustic vibration feature observation vector is obtained, the Sigma point set is extracted by unscented transformation, and the predicted observation feature vector, observation covariance and state observation cross-covariance are obtained based on the augmented state prior prediction of unscented transformation. Based on the measured acoustic vibration feature observation vector, the predicted observation feature vector, the observation covariance, the state observation cross-covariance, the prior augmented state mean, and the hidden state vector, the posterior augmented state estimate and the posterior augmented state covariance are obtained based on feature residual weighted targeted correction and non-negative physical boundary constraints.

5. The method for predicting the remaining life of abnormal noise in components according to claim 4, characterized in that, When obtaining the measured acoustic vibration feature observation vector, an unscented transform is used to extract the Sigma point set, and based on the augmented state prior prediction of the unscented transform, the predicted observation feature vector, observation covariance, and state-observation cross-covariance are obtained, including: Based on the mean and covariance of the posterior augmented state estimate from the previous time step, 2n+1 Sigma point sets and Sigma point weights are generated using standard unscented transformation. Substitute each Sigma point in the Sigma point set into the physical prior state transition equation for transmission, and calculate the weighted mean and prior covariance of the prior augmented state estimate at the current time. The prior state Sigma point is mapped to the predicted observation Sigma point through the nonlinear observation mapping equation, and the predicted observation eigenvector, observation covariance, and state-observation cross-covariance are obtained by weighted calculation.

6. The method for predicting the remaining lifespan of abnormal noise in components according to claim 4, characterized in that, Based on the measured acoustic vibration feature observation vector, the predicted observation feature vector, the observation covariance, the state observation cross-covariance, the prior augmented state estimate mean, and the hidden state vector, and using feature residual weighted targeted correction and non-negative physical boundary constraints, the posterior augmented state estimate and posterior augmented state covariance are obtained, including: Based on the measured acoustic vibration characteristic observation vector and the predicted observation characteristic vector, the characteristic residual is obtained; The augmented Kalman gain matrix is ​​obtained based on the state observation cross-covariance and the observation covariance. Based on the characteristic residuals, the mean of the prior augmented state estimates, and the augmented Kalman gain matrix, the microscopic physical parameters, the mechanistic parameters in the physical prior state transition equation, and the impact friction coupling parameters are jointly calibrated and updated, and the pre- and post-constraint augmented state estimates are obtained. The a posteriori augmented state estimates are obtained by processing the pre- and post-constraint augmented state estimates through non-negative physical boundary constraint operations.

7. The method for predicting the remaining life of abnormal noise in components according to claim 1, characterized in that, The method involves obtaining the maximum Lyapunov exponent, the probability of knocking noise, and the probability of friction noise based on the implicit physical quantity dataset and the multimodal competitive logic mapping function. When these parameters meet preset conditions, the remaining life prediction value for the component noise is obtained, including: Based on the dataset of implicit physical quantities, the magnitude of the perturbation of the phase space trajectory is calculated using an evolutionary algorithm based on the Jacobian variational equation, and the maximum Lyapunov exponent is obtained. Based on the implicit physical quantity dataset and the multimodal competition logic mapping function, a coupling competition risk function is constructed in the two-dimensional physical phase space to obtain the probability of knocking noise and friction noise. When the maximum Lyapunov index, the probability of knocking noise, and the probability of friction noise meet the preset conditions, the remaining life prediction value of the component noise is obtained.

8. A device for predicting the remaining lifespan of abnormal noises in components, characterized in that, include: The model building module is used to construct a friction-knock coupling microdynamic evolution model based on nonlinear physical priors, obtain microscopic physical parameters that cannot be directly measured based on the friction-knock coupling microdynamic evolution model, obtain hidden state vectors based on microscopic physical parameters, and establish physical prior state transition equations. The observation vector determination module is used to collect the high-frequency raw waveforms of components in the initial state and under the applied short-period accelerated durability load spectrum, perform noise reduction preprocessing on the high-frequency raw waveforms by estimating the noise resistance frequency response function, and slice the long-time waveforms using a time sliding window to calculate the acoustic vibration characteristics of each slice, thereby obtaining the acoustic vibration characteristic observation vector. The acoustic vibration feature observation vector includes an initial multidimensional observation vector and a degraded feature observation vector; The estimation module is used to establish a nonlinear observation mapping equation connecting the hidden state vector and the acoustic vibration feature observation vector; and to use an unscented Kalman filter to deeply fuse the hidden state vector and the acoustic vibration feature observation vector through the nonlinear observation mapping equation. By using blind inference and targeted correction mechanisms under sparse observation, the hidden state vector is jointly inversely reconstructed to obtain the posterior augmented state estimate of the microscopic physical parameters. The parameter set determination module is used to continuously extrapolate the friction-knock coupling microdynamic evolution model to future life cycles using the a priori augmented state estimate as the initial condition, to obtain the implicit physical parameter set; the implicit physical parameter set includes the microphysical gap and the microcontact stiffness under different life cycles. The life prediction module is used to obtain the maximum Lyapunov exponent, the probability of knocking noise, and the probability of friction noise based on the implicit physical quantity dataset and the multimodal competition logic mapping function. When the maximum Lyapunov exponent, the probability of knocking noise, and the probability of friction noise meet the preset conditions, the remaining life prediction value of the component noise is obtained.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the component noise remaining life prediction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the component noise remaining life prediction method as described in any one of claims 1 to 7.

Images