A bearing fault diagnosis method, equipment, and medium based on optimization algorithms.

Abstract

This application discloses a bearing fault diagnosis method, device, and medium based on an optimization algorithm, relating to the field of bearing fault diagnosis technology. The method includes: acquiring a vibration signal sequence of the target bearing during the current monitoring period; inputting it into a pre-calibrated fault signal generation model library; using a Bayesian inference framework, constructing a posterior probability estimate with the set of fault physical parameters as unknowns, using particle filtering in a sequence Monte Carlo optimization algorithm to iteratively solve the posterior probability estimate, generating and updating weighted parameter particles to approximate the probability distribution of the fault physical parameter set at the current moment; calculating the marginal probability of the fault type identifier based on the probability distribution, determining the type corresponding to the maximum marginal probability as the current fault, and extrapolating to obtain the remaining service life probability distribution. This application achieves probabilistic and quantifiable diagnostic results through the above method, rather than relying on a single label.

Description

Technical Field

[0001] This application relates to the technical field of bearing fault diagnosis, and particularly to a bearing fault diagnosis method, device and medium based on an optimization algorithm. Background Art

[0002] With the development of industrial intelligent operation and maintenance, bearing fault diagnosis technology has increasingly become the key to ensuring the safe operation of equipment. The existing diagnosis methods are mainly divided into two categories: one is the method based on signal processing and machine learning, which extracts the features of vibration signals, time-domain statistics, and frequency-domain spectral peaks, and uses support vector machines, random forests or shallow neural networks for classification. This type of method relies on expert experience for feature engineering, and the diagnosis results are mostly static fault labels, which cannot reveal the physical form and evolution process of faults. The other is the method based on deep learning, especially convolutional neural networks, which can automatically learn the deep features of signals and have good classification accuracy under strong noise. However, the diagnosis decision of this type of method lacks the support of physical mechanisms and is difficult to explain; more importantly, it can only distinguish whether there is a fault or the type of fault, and cannot quantitatively evaluate and predict the severity, specific location of the fault and its dynamic evolution process over time.

[0003] Existing research has tried to combine physical models with data-driven methods. For example, by establishing a simplified bearing dynamics model to simulate fault signals, or using an optimization algorithm to calibrate model parameters. However, these methods often limit optimization to the static fitting of model parameters, or only use it for auxiliary feature extraction, and fail to construct a unified inference framework that can inversely calculate the dynamic physical state of faults online in real time. The diagnosis results are still isolated and cannot form a continuous understanding from history, current situation to future prediction, which limits the practical application value in predictive maintenance.

[0004] Through the above analysis, the problems and defects existing in the prior art are as follows:

[0005] The bearing fault diagnosis in the prior art cannot achieve high-precision fault classification, and further cannot inversely calculate the physical parameters of faults online in real time, quantify the severity, trace the historical evolution trajectory and predict the remaining service life. Summary of the Invention

[0006] The embodiments of this application provide a bearing fault diagnosis method, device and medium based on an optimization algorithm, which can solve the problem that the bearing fault diagnosis in the prior art cannot achieve high-precision fault classification, and further cannot inversely calculate the physical parameters of faults online in real time, quantify the severity, trace the historical evolution trajectory and predict the remaining service life.

[0007] In a first aspect, an embodiment of the present application provides a bearing fault diagnosis method based on an optimization algorithm. The method includes: collecting a vibration signal sequence of a target bearing during a current monitoring period; inputting the vibration signal sequence into a pre-calibrated fault signal generation model library, where the input of the fault signal generation model library is a set of fault physical parameters and the output is a simulated vibration signal; based on a Bayesian inference framework, using the vibration signal sequence as evidence to construct a posterior probability estimate with the set of fault physical parameters as the unknown; using particle filtering in the sequential Monte Carlo optimization algorithm to iteratively solve the posterior probability estimate, and in the solving process, generating and updating parameter particles with weights to approximate the probability distribution of the set of fault physical parameters at the current moment; based on the probability distribution, calculating the marginal probability of the fault type identifier, determining the type corresponding to the maximum marginal probability as the current fault, and extrapolating to calculate the remaining useful life probability distribution.

[0008] In an implementation manner of the present application, calculating the marginal probability of the fault type identifier based on the probability distribution, determining the type corresponding to the maximum marginal probability as the current fault, and extrapolating to calculate the remaining useful life probability distribution specifically includes: performing fixed-interval smoothing optimization based on the parameter particles and weight history, and inversely calculating the evolution trajectory of the set of fault physical parameters within a past time window; extrapolating to calculate the remaining useful life probability distribution of the target bearing based on the probability distribution and the parameter change pattern characterized in the evolution trajectory.

[0009] In an implementation manner of the present application, before inputting the vibration signal sequence into the pre-calibrated fault signal generation model library, the method further includes: obtaining a bearing periodic vibration signal sequence of the target bearing evolving from a normal state to a fault state, and the corresponding fault identifier; initializing the fault signal generation model library and setting initial values for the basic physical parameters in the fault signal generation model library; constructing a parameter optimization problem and an optimization objective, where the optimization variable of the parameter optimization problem is the time series of the basic physical parameters, and the optimization objective is to minimize the difference value between the simulated vibration signal and the measured periodic vibration signal; using a gradient-based optimization algorithm to solve the parameter optimization problem and updating the basic physical parameters to complete the calibration.

[0010] In an implementation manner of the present application, the optimization objective is to minimize the difference value between the simulated vibration signal and the measured periodic vibration signal, which specifically includes: calculating the root mean square error of the measured signal and the simulated vibration signal in the time domain, and calculating the correlation coefficient of the envelope spectrum in the frequency domain; weighting and summing the root mean square error and the correlation coefficient to obtain a joint loss function.

[0011] In one implementation of this application, a sequential Monte Carlo optimization algorithm is used to iteratively solve the posterior probability estimate. Specifically, this includes: randomly sampling within the domain of the fault physical parameters to generate an initial particle swarm; predicting and propagating the parameter state of each particle at the next moment according to the state transition equation of the bearing system to obtain a parameter set; inputting the parameter set into a fault signal generation model library to obtain a simulated vibration signal, calculating the likelihood of the simulated vibration signal with the current actual observed signal, and updating the particle weights accordingly; resampling based on the updated particle weights, removing particles with weights below a threshold, copying particles with weights above a threshold, generating a new equally weighted particle set, until all data for the current monitoring period is processed to obtain the posterior probability distribution of the fault physical parameter set.

[0012] In one implementation of this application, a fixed interval smoothing optimization is performed based on parameter particles and weight history. Specifically, this includes: obtaining the forward filter particle set at all times based on the forward filter particle set generated at each time step in the particle filtering process; using forward and backward smoothing algorithms, starting from the last time step, traversing the time window in reverse, and calculating the smoothed particle distribution at each historical time step based on the information of the state transition model and the forward filter particle set to obtain the evolution trajectory.

[0013] In one implementation of this application, the probability distribution of the remaining service life of the target bearing is extrapolated based on the probability distribution and the parameter change pattern represented in the evolution trajectory. Specifically, this includes: extracting the sequence of fault equivalent parameters changing over time in the evolution trajectory; fitting the sequence to obtain an empirical model that evolves over time; using the probability distribution of the current set of fault physical parameters as initial conditions, performing Monte Carlo extrapolation using the empirical model to simulate the evolution path of the fault equivalent parameters over a preset future time; recording the corresponding future time point when the simulated fault equivalent parameters reach a failure threshold predefined based on the bearing Hertzian contact stress analysis; and generating the probability distribution of the remaining service life through multiple simulations.

[0014] In one implementation of this application, resampling is performed based on the weights of the particles, specifically including: calculating the variance of the position values ​​of all particles; if the variance is lower than a preset threshold, injecting a small-amplitude uniformly distributed disturbance that follows the rotation period of the bearing.

[0015] Secondly, embodiments of this application also provide a bearing fault diagnosis device based on an optimization algorithm. The device includes at least one processor and a memory communicatively connected to the at least one processor. The memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform any one of the steps of a bearing fault diagnosis method based on an optimization algorithm.

[0016] Thirdly, embodiments of this application also provide a non-volatile computer storage medium for bearing fault diagnosis based on an optimization algorithm, which stores computer-executable instructions, wherein the computer-executable instructions are configured to execute any one of the steps of a bearing fault diagnosis method based on an optimization algorithm.

[0017] This application provides a bearing fault diagnosis method, device, and medium based on optimization algorithms. By constructing the diagnosis problem as a posterior probability estimation problem based on a Bayesian framework and a fault signal generative model library, and solving it using a sequential Monte Carlo optimization algorithm, it can output the full probability distribution of fault physical parameters, realizing the probabilistic and quantifiable nature of the diagnosis results rather than a single label, significantly improving the reliability and information richness of the results. Secondly, by introducing fixed interval smoothing optimization, the continuous evolution trajectory of the fault over a past period can be inferred from the current data, realizing the leap from static diagnosis to dynamic process tracing. Furthermore, based on this trajectory and the current state, the remaining service life of the bearing and its probability distribution can be extrapolated and predicted, providing a direct and reliable decision-making basis for predictive maintenance. Finally, the entire method is based on an interpretable physical model, tightly coupling observational data and mechanistic knowledge through optimization algorithms, making the diagnosis process transparent and the results reliable, while possessing strong noise resistance and online adaptive capabilities. Attached Figure Description

[0018] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0019] Figure 1 A flowchart of a bearing fault diagnosis method based on an optimization algorithm is provided for an embodiment of this application;

[0020] Figure 2 This is a schematic diagram of the internal structure of a bearing fault diagnosis device based on an optimization algorithm, provided in an embodiment of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] The embodiments of the present application provide a bearing fault diagnosis method, device and medium based on an optimization algorithm, which solve the problems that the bearing fault diagnosis in the prior art cannot achieve high-precision fault classification, and further cannot online and real-time invert the physical parameters of the fault, quantify the severity, trace the historical evolution trajectory and predict the remaining service life.

[0023] The technical solutions proposed in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0024] Figure 1 It is a flowchart of a bearing fault diagnosis method based on an optimization algorithm provided by the embodiments of the present application. As Figure 1 shown, a bearing fault diagnosis method based on an optimization algorithm provided by the embodiments of the present application specifically includes the following steps:

[0025] Step 10: Collect the vibration signal sequence of the target bearing during the current monitoring period.

[0026] In this step, the vibration acceleration signal of the target bearing is collected in real time, and the precise rotational speed pulse signal is collected synchronously. First, the vibration signal is preprocessed by denoising and standardization, and then angular domain resampling is performed using the rotational speed signal to convert the non-stationary time-domain signal into an angular domain stationary signal, so as to eliminate the influence of rotational speed fluctuation on fault feature extraction.

[0027] Step 20: Input the vibration signal sequence into a pre-calibrated fault signal generation model library. The input of the fault signal generation model library is a set of fault physical parameters, and the output is a simulated vibration signal.

[0028] In this step, the angular domain stationary signal is input into the calibrated fault signal generation model library, that is, a parameterized forward simulator constructed based on the bearing system dynamics principle, which defines a deterministic or probabilistic mapping relationship from the fault physical space to the vibration signal space. The input is a set of interpretable physical parameters characterizing the bearing fault, including fault type identifiers: inner race fault, outer race fault, rolling element fault or cage fault; fault geometric parameters: the circumferential position of the fault on the raceway, which can be represented by an angle, the equivalent length along the rolling direction, the depth, etc.; and the shape factor characterizing the sharpness of the fault. The output is a simulated vibration signal, usually an acceleration signal, calculated based on the above physical parameters and the known bearing system characteristics, such as mass, stiffness, damping, etc., which will reproduce the vibration response characteristics generated by the real bearing in the corresponding fault state.

[0029] Furthermore, within the Bayesian inference framework, this model library essentially acts as a likelihood function model. Given a specific set of fault physical parameter assumptions, the model library can calculate what kind of vibration signal is expected to be observed under these assumptions. By comparing the simulated signal output by the model library with the actual observed signal obtained in step 10, the likelihood of the current set of parameter assumptions explaining the actual observed data can be quantitatively evaluated.

[0030] Understandably, pre-calibration refers to the fact that the model library has undergone parameter optimization in the offline phase. Using full-lifecycle experimental data with known fault conditions, optimization algorithms are used to adjust the internal physical parameters of the model library, such as contact stiffness coefficient and structural damping coefficient. This allows the model library to generate simulated signals that highly match real-world conditions for a specific type of bearing. Therefore, the model library used during online diagnostics is a targeted physical model with high-fidelity simulation capabilities, ensuring the accuracy of subsequent state inversion.

[0031] As an optional embodiment, before inputting the vibration signal sequence into a pre-calibrated fault signal generation model library, the method may further include: Step 01: Obtaining the bearing periodic vibration signal sequence of the target bearing evolving from a normal state to a fault state, and the corresponding fault identifier.

[0032] In this step, a specific accelerated life test is pre-calibrated on the target model bearing in a controlled experimental environment. Vibration acceleration and speed sensors are used to simultaneously collect vibration signal sequences throughout the bearing's entire lifecycle, from its normal state through initial fault development and eventual complete failure. Simultaneously, through periodic disassembly and inspection or high-performance monitoring, fault indicators corresponding to key time points are recorded, such as the minute in which pitting occurs on the inner ring, or the minute in which peeling and expansion occur on the outer ring.

[0033] Step 02: Initialize the fault signal generation model library by setting initial values ​​for the basic physical parameters in the fault signal generation model library.

[0034] In this step, the parameterized bearing system dynamic equations are first embedded. The input is a set of physical parameters characterizing the fault, such as the fault type identifier, the circumferential position of the fault on the component, and the equivalent length and depth of the fault along the raceway direction. The output is a simulated vibration acceleration signal.

[0035] During initialization, theoretical or empirical initial values ​​are set for the basic physical parameters in the model, such as contact stiffness and system damping coefficient, based on the bearing's geometric parameters, such as the number of rollers, pitch circle diameter, contact angle, etc., and material properties.

[0036] Step 03: Construct the parameter optimization problem and optimization objective. The optimization variable of the parameter optimization problem is the time series of the basic physical parameters, and the optimization objective is to minimize the difference between the simulated vibration signal and the measured periodic vibration signal.

[0037] In this step, the time series of the underlying physical parameters to be calibrated in the model library means that these parameters are allowed to change slowly as the bearing condition deteriorates.

[0038] As an optional embodiment, the optimization objective is to minimize the difference between the simulated vibration signal and the measured periodic vibration signal, which may specifically include: Step 031: Calculate the root mean square error between the measured signal and the simulated vibration signal in the time domain, and calculate the correlation coefficient of the envelope spectrum in the frequency domain; Step 032: Weighted summation of the root mean square error and the correlation coefficient to obtain the joint loss function.

[0039] In practice, the optimization objective is to minimize the overall difference between the simulated signal generated by the model library and the measured signal, which is measured using a joint loss function. First, the root mean square error (RMSE) of the simulated and measured signals is calculated in the time domain to capture the overall energy and waveform matching. In the frequency domain, Hilbert transforms are performed on both signals to obtain their envelope spectra, and the correlation coefficient between the two envelope spectra is calculated to assess the consistency of the fault characteristic frequency components. The minimum RMSE is then weighted and summed with the correlation coefficient to form the final joint loss function; the weights can be adjusted according to the emphasis on time-domain or frequency-domain fidelity.

[0040] Step 04: Use a gradient-based optimization algorithm to solve the parameter optimization problem and update the basic physical parameters to complete the calibration.

[0041] In this step, since the optimization variable is a time series, involving a high-dimensional parameter space, this embodiment employs an efficient adjoint method. By constructing and solving the adjoint equation, the gradient of the loss function with respect to parameters at all time points can be efficiently calculated in a single forward-backward propagation. This gradient information is then used to iteratively update the basic physical parameters of the model library until the loss function converges. Ultimately, a high-fidelity fault signal generation model library calibrated for this bearing model is obtained, which can be stored offline for subsequent online diagnostic calls.

[0042] Step 30: Based on the Bayesian inference framework, the vibration signal sequence is used as evidence to construct a posterior probability estimate with the set of fault physical parameters as unknowns.

[0043] In this step, under the Bayesian inference framework, the stationary signal in the angular domain of the vibration signal sequence obtained after the acquisition and preprocessing in step 10 is regarded as known observational evidence, and the set of fault physical parameters hidden behind the signal that are expected to be known, namely fault type, location, size, depth, etc., are regarded as hidden states or unknowns to be estimated.

[0044] Specifically, what needs to be solved is the posterior probability distribution of the set of fault physical parameters, that is, P(state | observation), given the current and historical observation data; according to Bayes' theorem, this posterior distribution is proportional to the product of the likelihood function P(observation | state) and the prior distribution P(state). Among them: Prior distribution: It represents the initial knowledge of the fault state before obtaining the current observation data, such as whether it may occur in the inner ring or the outer ring and the approximate range of dimensions, which can be set based on historical data or general knowledge; Likelihood function: It quantifies the possibility of generating the currently observed vibration signal under a specific fault state hypothesis. This model library is the physical implementation of the likelihood function P(observation | state), which can calculate and generate corresponding simulated signals for any set of input fault physical parameters, so as to evaluate the quality of this set of parameters in explaining the current observation data.

[0045] Step 40: Use particle filtering in the sequential Monte Carlo optimization algorithm to iteratively solve the posterior probability estimation. The solving process approximates the probability distribution of the set of fault physical parameters at the current moment by generating and updating parameter particles with weights.

[0046] In this step, due to the highly nonlinear and non-Gaussian characteristics of the bearing system dynamics and fault evolution, the Bayesian posterior distribution is often difficult to solve by analytical methods. Therefore, the sequential Monte Carlo optimization algorithm, especially particle filtering, is used to numerically approximate this complex distribution. The core idea of particle filtering is to use a set of random samples, called particles and their weights, to represent the probability distribution.

[0047] As an optional embodiment, the sequential Monte Carlo optimization algorithm is used to iteratively solve the posterior probability estimation, which may specifically include: Step 401: Randomly sample within the domain of definition of the fault physical parameters to generate an initial particle swarm.

[0048] In this step, within the domain of definition of the set of fault physical parameters: type enumeration, position angle range, and reasonable dimension interval, random sampling is performed to generate an initial particle swarm {x0 (i) , w0 (i)} i=1 N . Among them, each particle x0 (i) represents a complete set of fault parameter hypotheses. For example: {type = inner ring, position = 45 degrees, length = 2 mm, depth = 0.5 mm}). The initial weights w0(i) are all set to 1 / N, indicating that all hypotheses are considered equally likely at the initial moment.

[0049] Step 402: According to the state transition equation of the bearing system, predict and propagate the parameter state of each particle at the next moment to obtain a set of parameters.

[0050] In this step, for each particle The predicted state at time k is obtained by predicting the state based on a state transition equation that describes how the fault state evolves over time. The state transition equations embody prior knowledge of the dynamic behavior of the fault. For example, the fault size parameter can be modeled as a slowly increasing stochastic process, L. k =L k 1+ ,in It produces small Gaussian noise; for the inner or outer ring at the fault location, the variation strictly follows the rotational kinematics of the bearing; this step realizes the time recursion of the particle swarm from the previous moment to the next moment.

[0051] Step 403: Input the parameter set into the fault signal generation model library to obtain the simulated vibration signal, calculate the likelihood of the simulated vibration signal with the current actual observed signal, and update the particle weight accordingly.

[0052] In this step, each predicted particle The assumed fault parameters are input into a pre-calibrated fault signal generation model library, and the corresponding simulated vibration signals are output. The simulated signal is compared with the current actual observation signal obtained in step 10, and the likelihood is calculated. The likelihood measures the probability that the current observation data is generated by the fault state described by the particle. This likelihood can be measured by calculating the similarity of the envelope spectra of the two near the bearing fault characteristic frequency.

[0053] Step 404: Resample according to the updated particle weights, remove particles with weights below the threshold, copy particles with weights above the threshold, generate a new set of equally weighted particles, and process all data in the current monitoring period to obtain the posterior probability distribution of the fault physical parameter set.

[0054] In this step, the distribution of the particle swarm after updating and resampling is a Monte Carlo approximation of the posterior probability distribution of the fault physical parameter set, which contains all the statistical information of the fault state.

[0055] As an optional embodiment, resampling based on the particle weights may specifically include: step 4041: calculating the variance of the position value distribution of all particles.

[0056] In this step, a new set of equally weighted particles is generated during a standard resampling operation. Then, the algorithm calculates the variance of the distribution of all particles in the dimension of the fault circumferential position. Variance is an indicator of the degree of data dispersion. The smaller the value, the more concentrated the particle swarm's estimate of the fault position and the higher the consensus. Conversely, it indicates that there is still a large divergence among particles in their position estimates.

[0057] Step 4042: If the distribution variance is lower than the preset threshold, inject a small-amplitude uniformly distributed disturbance that follows the rotation period of the bearing.

[0058] In this step, the algorithm presets a variance threshold, which is a small positive number that can be determined based on experience or through previous experiments. This threshold is used to determine whether the particle swarm may converge to a potentially inaccurate local consensus prematurely and without fully exploring the solution space due to insufficient evidence, such as weak fault signals.

[0059] If the calculated distribution variance is lower than a preset threshold, a perturbation mechanism is triggered. The perturbation is not blind random noise, but rather based on the physical periodicity of the bearing rotation. Specifically, it is triggered for each particle's fault location parameter θ. (j) : , where δ (j) From the interval [ The random values ​​sampled uniformly on [a, a], where a is a small amplitude value set according to one revolution of the bearing. This perturbation aims to artificially and moderately re-expand the distribution range of particles on the position parameters without disrupting the established basic position consensus.

[0060] This solves the particle depletion problem that particle filtering often encounters in the early stages of diagnosis or when fault signals are weak. When fault characteristics are not yet obvious, the observed data has low discriminative power for location, and the likelihood of all particles may be similar, leading to a rapid loss of particle diversity after resampling, causing the algorithm to prematurely determine a potentially inaccurate location. The perturbation applied in this application is physically based and conditional, triggering only when particles become too clustered, forcing the particle swarm to maintain the necessary diversity in the location dimension. This significantly improves the algorithm's ability to explore location estimation in the fault initiation stage, as well as its adaptability to external abrupt changes or intermittent faults.

[0061] Step 50: Based on the probability distribution, calculate the marginal probability of the fault type identifier, determine the type corresponding to the maximum marginal probability as the current fault, and extrapolate to calculate the remaining service life probability distribution.

[0062] In this step, the marginal probability of the fault type identifier is calculated, and the sum of the weights of particles belonging to each preset fault type in the particle set is counted. This sum is the posterior probability of the fault type. The fault type corresponding to the maximum marginal probability is determined as the most important fault mode of the current bearing.

[0063] For continuous physical parameters, a weighted statistic of the posterior distribution is calculated, and the weighted mean is used as a point estimate of the parameter. The weighted variance or confidence interval is then calculated to characterize the uncertainty of the estimate. Subsequently, the estimated fault size parameter is compared with a severity threshold range preset based on the bearing model and operating conditions to determine the severity level of the current fault.

[0064] As an optional embodiment, based on the probability distribution, the marginal probability of the fault type identifier is calculated, the type corresponding to the maximum marginal probability is determined as the current fault, and the remaining service life probability distribution is extrapolated. Specifically, it may include: Step 501: Based on the parameter particles and weight history, fixed interval smoothing optimization is performed, and the evolution trajectory of the fault physical parameter set in the past time window is obtained by inversion calculation.

[0065] In this step, to understand the development process of the fault, the forward filter particle set and weight history generated at each time point by the particle filtering process are saved. Through fixed interval smoothing optimization, the continuous evolution trajectory of the fault physical parameters within the past time window is inverted.

[0066] As an optional implementation, fixed-interval smoothing optimization is performed based on parameter particles and weight history, specifically including: Step 5011: Obtain the forward-filtered particle set for all time moments based on the forward-filtered particle set generated at each time moment during the particle filtering process. Step 5012: Using forward and backward smoothing algorithms, starting from the last time moment, the time window is traversed in reverse, and the smoothed particle distribution for each historical time moment is calculated based on the state transition model and the information of the forward-filtered particle set to obtain the evolution trajectory.

[0067] In this step, the forward-filtered particle set is the set of particles generated and updated at each time point during the particle filtering process in step 40, and it naturally exists in the historical record of the filtering process. A forward-backward smoothing algorithm is employed, iterating backward from the last time point T. For a historical time point k, it comprehensively utilizes: the forward-filtered particle set at that time; all posterior observation information from time point k+1 to time point T, passed through reverse recursion; and a state transition model describing the state evolution law. By reweighting or resampling the forward particles, a more accurate smoothed particle distribution for that historical time point is calculated. Connecting the central trend of the smoothed distribution at each historical time point yields the continuous evolution trajectory of the key fault parameters, clearly showing how the fault has developed from the past to the current state.

[0068] Step 502: Based on the probability distribution and the parameter change patterns represented in the evolution trajectory, extrapolate the probability distribution of the remaining service life of the target bearing.

[0069] As an optional embodiment, the probability distribution of the remaining service life of the target bearing is extrapolated based on the probability distribution and the parameter change pattern characterized in the evolution trajectory. Specifically, it may include: Step 5021: Extracting the sequence of fault equivalent parameters changing over time in the evolution trajectory.

[0070] In this step, key physical parameters that characterize the severity of the fault are extracted from the smooth evolution trajectory obtained by the above inversion, namely the equivalent length or equivalent area of ​​the fault and its corresponding timestamp or cumulative number of revolutions, to form a time series of the parameter.

[0071] Step 5022: Fit the sequence to obtain an empirical model that evolves over time.

[0072] In this step, the extracted parameter time series is fitted using curve fitting methods. Based on the physical mechanism of fault propagation, power law models, exponential models, or linear models are often used for fitting to obtain mathematical expressions describing the changes of parameters over time.

[0073] Step 5023: Using the probability distribution of the current set of physical parameters of the fault as the initial condition, Monte Carlo extrapolation is performed using an empirical model to simulate the evolution path of the fault equivalent parameters over a preset time.

[0074] In this step, the fault parameters obtained in step 40 at the current moment, especially the probability distribution of the equivalent size, are used as the initial conditions for prediction. A large number of Monte Carlo simulations are performed; in each simulation, a set of size samples is randomly drawn from the initial distribution, and considering the uncertainties of the model itself and the process, numerical integration or iteration is performed using the fitted empirical model to deduce the possible growth path of the fault size in the future.

[0075] Step 5024: When the simulated fault equivalent parameters reach the predefined failure threshold based on the bearing Hertz contact stress analysis, record the corresponding future time point.

[0076] In this step, a critical fault size threshold that would cause bearing failure is pre-set based on the bearing's rated load and material properties, using Hertzian contact stress theory or engineering standards and experience. In each Monte Carlo simulation, when the simulated fault size first reaches or exceeds this threshold, the corresponding future time point is recorded, and this time point is the predicted failure time for this simulation.

[0077] Step 5025: Generate the probability distribution of remaining useful life through multiple simulations.

[0078] In this step, the failure times obtained from all simulations are collected to form a sample set of remaining service life. By performing statistical analyses such as kernel density estimation on these samples, the probability density function and cumulative distribution function of the remaining service life can be generated. This not only provides point estimates of the remaining service life, such as the expected value and median, but also provides a prediction interval with confidence, achieving truly probabilistic and risk-quantitative service life prediction, and providing accurate basis for predictive maintenance decisions.

[0079] The above are embodiments of the method proposed in this application. Based on the same inventive concept, embodiments of this application also provide a bearing fault diagnosis device based on an optimization algorithm, the structure of which is as follows: Figure 2 As shown.

[0080] Figure 2 This is a schematic diagram of the internal structure of a bearing fault diagnosis device based on an optimization algorithm, provided as an embodiment of this application. Figure 2 As shown, the device includes:

[0081] At least one processor 201;

[0082] And a memory 202 that is communicatively connected to at least one processor;

[0083] The memory 202 stores instructions that can be executed by at least one processor. The instructions are executed by at least one processor 201 to enable at least one processor 201 to: perform any one of the steps of a bearing fault diagnosis method based on an optimization algorithm.

[0084] Some embodiments of this application provide corresponding to Figure 1 A non-volatile computer storage medium for bearing fault diagnosis based on an optimization algorithm stores computer-executable instructions, which are configured to execute any one of the steps of a bearing fault diagnosis method based on an optimization algorithm.

[0085] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments for IoT devices and media are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0086] The systems, media, and methods provided in this application are one-to-one correspondences. Therefore, the systems and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and media will not be repeated here.

[0087] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0088] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0089] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0090] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0091] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0092] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0093] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0094] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0095] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A bearing fault diagnosis method based on an optimization algorithm, characterized in that, The method includes: Collect the vibration signal sequence of the target bearing during the current monitoring period; The vibration signal sequence is input into a pre-calibrated fault signal generation model library, where the input to the fault signal generation model library is a set of fault physical parameters and the output is a simulated vibration signal. Based on the Bayesian inference framework, the vibration signal sequence is used as evidence to construct a posterior probability estimate with the set of fault physical parameters as unknowns. The posterior probability estimate is iteratively solved using particle filtering in the sequential Monte Carlo optimization algorithm. The solution process involves generating and updating parameter particles carrying weights to approximate the probability distribution of the set of fault physical parameters at the current moment. Based on the probability distribution, the marginal probabilities of the fault type identifier are calculated, the type corresponding to the maximum marginal probability is determined as the current fault, and the remaining service life probability distribution is extrapolated to obtain the remaining service life probability distribution, specifically including: Based on the parameter particles and weight history, a fixed interval smoothing optimization is performed, and the evolution trajectory of the fault physical parameter set within the past time window is obtained by inversion calculation. Based on the probability distribution and the parameter change patterns characterized in the evolution trajectory, the remaining service life probability distribution of the target bearing is extrapolated and calculated. Before inputting the vibration signal sequence into a pre-calibrated fault signal generation model library, the method further includes: acquiring the bearing periodic vibration signal sequence of the target bearing evolving from a normal state to a fault state, and the corresponding fault identifier; Initialize the fault signal generation model library and set initial values ​​for the basic physical parameters in the fault signal generation model library; A parameter optimization problem and an optimization objective are constructed. The optimization variable of the parameter optimization problem is the time series of the basic physical parameters. The optimization objective is to minimize the difference between the simulated vibration signal and the measured periodic vibration signal. The parameter optimization problem is solved using a gradient-based optimization algorithm to update the basic physical parameters and complete the calibration.

2. The bearing fault diagnosis method based on optimization algorithm according to claim 1, characterized in that, The optimization objective is to minimize the difference between the simulated vibration signal and the measured periodic vibration signal, specifically including: The root mean square error between the measured signal and the simulated vibration signal is calculated in the time domain, and the correlation coefficient of the envelope spectrum is calculated in the frequency domain. The joint loss function is obtained by weighted summation of the root mean square error and the correlation coefficient.

3. The bearing fault diagnosis method based on optimization algorithm according to claim 1, characterized in that, The sequential Monte Carlo optimization algorithm is used to iteratively solve the posterior probability estimate, specifically including: An initial particle swarm is generated by randomly sampling within the domain of the defined fault physical parameters. Based on the state transition equation of the bearing system, the parameter state of each particle at the next moment is predicted and propagated to obtain the parameter set; The parameter set is input into the fault signal generation model library to obtain the simulated vibration signal. The likelihood of the simulated vibration signal and the current actual observed signal is calculated, and the weight of the particles is updated accordingly. Resampling is performed based on the updated particle weights. Particles with weights below the threshold are removed, and particles with weights above the threshold are copied to generate a new set of equally weighted particles. This process continues until all data for the current monitoring period is processed, resulting in the posterior probability distribution of the fault physical parameter set.

4. The bearing fault diagnosis method based on optimization algorithm according to claim 1, characterized in that, The fixed-interval smoothing optimization based on the parameter particles and weight history specifically includes: Based on the forward filter particle sets generated at each time step in the particle filtering process, the forward filter particle sets at all times are obtained. Using forward and backward smoothing algorithms, starting from the last moment, the time window is traversed in reverse. Based on the state transition model and the information of the forward filtered particle set, the smoothed particle distribution at each historical moment is calculated to obtain the evolution trajectory.

5. The bearing fault diagnosis method based on optimization algorithm according to claim 1, characterized in that, The extrapolation calculation of the probability distribution of the remaining service life of the target bearing based on the probability distribution and the parameter change patterns characterized in the evolution trajectory specifically includes: In the evolution trajectory, the sequence of fault equivalent parameters changing over time is extracted; The sequence is fitted to obtain an empirical model that evolves over time. Using the probability distribution of the current set of physical parameters of the fault as the initial condition, Monte Carlo extrapolation is performed using the empirical model to simulate the evolution path of the fault equivalent parameters over a future preset time. When the simulated fault equivalent parameters reach the failure threshold predefined based on bearing Hertz contact stress analysis, the corresponding future time point is recorded. Through multiple simulations, a probability distribution of the remaining useful life is generated.

6. The bearing fault diagnosis method based on optimization algorithm according to claim 4, characterized in that, The resampling based on the weights of the particles specifically includes: Calculate the variance of the position values ​​distribution for all particles; If the distribution variance is lower than a preset threshold, a small, uniformly distributed disturbance is injected into the bearing rotation period.

7. A bearing fault diagnosis device based on an optimization algorithm, characterized in that, The device includes: At least one processor; And, a memory communicatively connected to the at least one processor; The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to: Perform as claimed in claim 1 The steps of any one of the following bearing fault diagnosis methods based on optimization algorithms.

8. A non-volatile computer storage medium for bearing fault diagnosis based on an optimization algorithm, storing computer-executable instructions, characterized in that, The computer-executable instructions are set as follows: Perform as claimed in claim 1 The steps of any one of the following bearing fault diagnosis methods based on optimization algorithms.

Images