Surface wear volume space-time distribution prediction method combining abrasive particle and vibration monitoring

Abstract

The invention discloses a surface wear volume time-space distribution prediction method combining abrasive particle and vibration monitoring, which comprises the following steps: constructing an abrasive particle size distribution driven Markov process, representing a random wear process, and realizing dynamic and probability representation of real-time wear volume time evolution; a physically constrained double-branch mapping model is constructed, fuzzy mapping of the surface wear volume and the vibration time-frequency characteristics is established, and precise observation of wear volume space distribution is achieved; the two physical mechanisms are fused on the basis of a state space modeling framework, a digital twinborn model combining abrasive particle monitoring and vibration monitoring is constructed, and accurate prediction of the abrasion volume of the friction pair part in the time dimension and the space dimension is achieved. According to the method, the wear volume space-time distribution is predicted by fusing the wear particle information and the physical correlation between the vibration characteristics and the wear process, and the precise monitoring of the wear process is realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of machine wear state monitoring, and particularly relates to a surface wear volume space-time distribution prediction method combining abrasive particles and vibration monitoring. BACKGROUND

[0002] Friction pair components such as bearings inevitably accumulate wear and continue to degrade, making them a sensitive link in the equipment failure chain. This has led to widespread attention to the condition monitoring of key friction pair components. Existing research mainly uses monitoring methods such as vibration, abrasive particles, and temperature to evaluate the fault state based on different physical mechanisms in the wear process. However, there are few reports on the observation and prediction of the wear degradation process. This is mainly because the wear evolution process involves complex behavior coupling such as material damage, topography evolution, and mechanical action, which presents strong randomness, making it impossible to directly observe the space-time degradation process, which is also a fundamental problem in the field of industrial digital twinning.

[0003] The digital twinning model dynamically calibrates the wear degradation model based on monitoring data such as vibration signals, abrasive particle characteristics, and working condition parameters, thereby achieving adaptive modeling of the dynamic degradation process. According to the different representation strategies, it can be divided into indirect representation models and direct prediction models. Indirect representation models construct a wear-sensitive feature vector (such as flow rate, pressure for a gear pump), and evaluate the wear degree according to its evolution process. However, since the feature vector is usually affected by non-wear state factors such as load and speed, such models are usually only suitable for qualitative evaluation at the machine system level. In contrast, direct prediction models introduce wear volume and depth to quantify the wear degradation degree, and have achieved quantitative prediction of the wear degradation trajectory of key friction pair components such as gears and bearings. In particular, by using the Archard-local wear model to establish a dynamic correlation between the wear spatial distribution and the local contact stress, or embedding the wear distribution pattern based on finite element offline analysis as prior knowledge, existing methods have achieved modeling of the wear space-time evolution induced by deterministic factors such as non-uniformly distributed loads.

[0004] For the digital twinning theory, the wear prediction accuracy depends on the physical correlation mechanism between online physical monitoring information and virtual wear process. Existing digital twinning models establish a quantitative mapping relationship between vibration features and surface damage size based on dynamic methods. In addition, existing research has established a quantitative relationship between abrasive particle concentration and dynamic wear rate, achieving prediction of the total wear volume. However, the actual wear process presents a strong randomness-induced space-time degradation process, and these macro-physical correlations fundamentally limit the wear prediction accuracy of the digital twinning model. In addition, wear involves the coupling of multiple mechanisms, and existing models only use a single physical information, further limiting the prediction accuracy in the actual wear process.

[0005] In summary, digital twin technology has become an effective way for real-time prediction of mechanical part performance. However, in the application of wear monitoring, the core physical mechanism still belongs to the macro component level characterization category, and cannot accurately capture the spatio-temporal variability in the wear process. In addition, the existing model uses single physical information, which further limits the accuracy of wear prediction. Therefore, it is urgent to build and integrate multi-modal monitoring information and fine-grained physical association of the wear process, and then build a multi-physical constraint digital twin model to realize the accurate prediction of the spatio-temporal distribution of the surface wear volume of the key friction pair. SUMMARY

[0006] The technical problem to be solved by the present application is to provide a surface wear volume spatio-temporal distribution prediction method combining abrasive particle and vibration monitoring to solve the problem of insufficient observation and prediction of the wear degradation process of the friction pair component in the prior art, especially the problem that the spatio-temporal degradation process cannot be directly observed due to the coupling effect of strong randomness and complex behavior of the wear process, effectively integrating multi-modal monitoring information and fine-grained physical association of the wear process, and realizing accurate prediction of the spatio-temporal distribution of the surface wear volume of the friction pair in the whole cycle wear process.

[0007] The present application adopts the following technical solutions: The surface wear volume spatio-temporal distribution prediction method combining abrasive particle and vibration monitoring comprises the following steps: A random wear characterization model driven by abrasive particle size distribution is constructed to realize dynamic probability characterization of the time series evolution of the wear volume of the friction pair through joint probability characterization and Markov process; A physical constraint double-branch wear-vibration mapping model is constructed to establish a fuzzy mapping between the surface wear volume and the vibration time-frequency characteristics, and to realize accurate observation of the spatial distribution of the surface wear volume of the friction pair; Based on the state space modeling framework, a wear degradation digital twin model combining abrasive particle and vibration monitoring is constructed, the abrasive particle size distribution-Markov process model constructed is taken as the state transition equation of the digital twin model, and the double-branch wear-vibration mapping model constructed is taken as the observation equation constraint condition of the digital twin model, to realize accurate spatio-temporal prediction of the surface wear volume of the friction pair.

[0008] Preferably, the random wear characterization model driven by abrasive particle size distribution is constructed, specifically: The equivalent diameter of the abrasive particle extracted from the online ferrograph image is taken as the input, and the probability density distribution of the abrasive particle volume is estimated through diameter-thickness mapping; The probability density distribution of the abrasive particle volume is sampled multiple times, and the sampling results are accumulated to form a dynamic characterization of the surface wear rate of the friction pair; Based on the abrasive particle driven wear rate characterization, the wear process is modeled as a discrete Markov process; The time evolution of the statistical characteristics of the wear volume of different regions of the surface of the friction pair is predicted by Monte Carlo simulation from the Markov process constructed.

[0009] Preferably, the dynamic characterization of the wear rate of the surface of the friction pair is in particular:

[0010] wherein, is the wear volume of the local region at the time t, is the discrete wear rate over the time interval [t, t + Δt], is the number of abrasive particles generated by the local region, is the volume of the abrasive particles in the first sampling, is the probability density distribution of the volume of the abrasive particles.

[0011] Preferably, the wear process is modeled as a discrete Markov process as follows:

[0012] wherein, is the local wear volume at the time t, is the discrete wear rate over the time interval [t, t + Δt].

[0013] Preferably, a physically constrained dual-branch wear-vibration mapping model is constructed, in particular: the signal segment corresponding to one revolution of the friction pair is intercepted from the real-time vibration signal , the sequence of time-frequency spectrum centroid frequencies is extracted , and the observation index of the wear volume sequence of the friction pair is constituted; a physically constrained wear-vibration mapping model is constructed to realize the fuzzy mapping of the wear volume of the surface of the friction pair and the time-frequency spectrum centroid frequency of the vibration signal, comprising: a dual-granularity prediction framework composed of a pressure prediction branch and an elastic strain prediction branch, and multiple soft / hard physical constraints; the pressure prediction branch uses a convolution network to establish an end-to-end nonlinear mapping between the wear topography and the pressure distribution; the elastic strain prediction branch involves a multi-level mapping of the wear volume-elastic approach- vibration time-frequency spectrum centroid frequency, wherein a Bayesian neural network is used to constitute a fuzzy mapping of the wear volume-elastic approach and measure its cognitive uncertainty, and a physical formula is used to model the mapping of the elastic approach-vibration time-frequency spectrum centroid frequency ; the physical mapping of the elastic approach-vibration time-frequency spectrum centroid frequency in the elastic strain prediction branch; ​​Hard physical constraints transform stress equilibrium conditions into hard constraints, which are then used as a post-processing layer for stress prediction branches. Soft constraints are constructed from deformation compatibility conditions and are used for model parameter optimization.

[0014] Preferably, the physical mapping between the elastic approximation quantity and the centroid frequency of the vibration spectrum is as follows:

[0015] in, This is the predicted value of the centroid frequency in the spectral spectrum during vibration; This is the elastic approximation quantity; To apply a load; For the mass of the friction components; Stiffness index; These are the scaling factor and the translation factor.

[0016] Preferably, the hard constraint is specifically:

[0017] Soft constraints are specifically:

[0018] in, This is the stress distribution matrix; This is the initial gap matrix; The height matrix of the wear morphology; This represents the upper limit of the elastic strain. This is the equivalent elastic modulus of the friction pair; This is a loss function based on physical constraints; This is the compliance matrix; This is the elastic approximation quantity.

[0019] Preferably, the wear degradation digital twin model combining abrasive and vibration monitoring includes a physical entity and a virtual entity. The physical entity involves the basic friction pair, as well as online abrasive and vibration monitoring. The virtual entity is a state-space model, using a Markov process driven by abrasive size distribution to construct the state transition equation, and a wear-vibration mapping model with physical constraints to construct the observation equation. The physical entity derives real-time abrasive size distribution and vibration characteristics to update the wear rate and vibration observations in the virtual entity online. The virtual entity derives wear volume sequence prediction results to achieve accurate monitoring of the actual friction pair. In the real-time prediction process, online parameter updates are introduced, updating the observation equation parameters at the start time of each running interval. By maximizing and Inter-correlation coefficient determines phase shift ; combined with the above data interaction and parameter update, a particle filter is used to predict the wear volume sequence online , to realize the spatiotemporal distribution prediction of surface wear volume.

[0020] Preferably, the phase shift is:

[0021] wherein, is a cyclic phase shift processing, that is the cyclic phase shift applied to , and is a time-frequency spectrum centroid frequency sequence.

[0022] In a second aspect, an embodiment of the present application provides a surface wear volume spatiotemporal distribution prediction system combining abrasive particles and vibration monitoring, comprising: a probability module, which constructs a random wear characterization model driven by abrasive particle size distribution, realizes dynamic probability characterization of wear volume time series evolution of a friction pair through joint probability characterization and Markov process; a mapping module, which constructs a physical constraint double-branch wear-vibration mapping model, establishes a fuzzy mapping between surface wear volume and vibration time-frequency features, and realizes observation of surface wear volume spatial distribution of the friction pair; a prediction module, which constructs a wear degradation digital twin model combining abrasive particles and vibration monitoring based on a state space modeling framework, takes the constructed abrasive particle size distribution-Markov process model as a state transition equation of the digital twin model, takes the constructed double-branch wear-vibration mapping model as a constraint condition of an observation equation of the digital twin model, and realizes accurate spatiotemporal prediction of surface wear volume of the friction pair.

[0023] In a third aspect, a computer device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor realizes the steps of the above-mentioned surface wear volume spatiotemporal distribution prediction method combining abrasive particles and vibration monitoring when executing the computer program.

[0024] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium comprising a computer program, and the computer program realizes the steps of the above-mentioned surface wear volume spatiotemporal distribution prediction method combining abrasive particles and vibration monitoring when executed by a processor.

[0025] In a fifth aspect, a chip comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor realizes the steps of the above-mentioned surface wear volume spatiotemporal distribution prediction method combining abrasive particles and vibration monitoring when executing the computer program.

[0026] In a sixth aspect, an electronic device is provided, which comprises a computer program, and the computer program, when executed by the electronic device, implements the steps of the surface wear volume spatiotemporal distribution prediction method of joint abrasive particle and vibration monitoring.

[0027] Compared with the prior art, the present application has at least the following beneficial effects: The surface wear volume spatiotemporal distribution prediction method of joint abrasive particle and vibration monitoring integrates Markov model and double-branch model based on a state space modeling framework to form a digital twin model, combines abrasive particle and vibration data, improves the comprehensiveness and accuracy of wear prediction, captures the time sequence evolution law of wear through Markov process, adapts to uncertainty, and combines physical formula and machine learning through double-branch model to ensure the physical rationality of prediction, realizes real-time monitoring and prediction of the wear process, and supports online updating and closed-loop optimization; through multi-source data fusion (abrasive particle + vibration), physical constraint driving (hard / soft constraint), and probabilistic modeling (Markov process + Bayesian network), a high-precision and strong-robustness wear prediction framework is constructed, the black box problem of pure data model is avoided, the explainability of prediction is ensured, risk warning and life prediction are supported through probabilistic representation and Bayesian network, and the digital twin framework combines online updating to realize continuous monitoring of equipment health status and maintenance decision support.

[0028] Further, through joint probabilistic representation and Markov process, an abrasive particle driven wear representation is constructed, which provides an effective solution for representing random wear volume evolution.

[0029] Further, by fusing physical constraints and wear morphology samples, a wear-vibration fine-grained mapping is constructed, which provides an effective solution for observing spatial heterogeneity in the wear process.

[0030] Further, the complementary fusion of abrasive particle and vibration multi-source physical information is realized, and the accurate prediction of wear volume spatiotemporal distribution is realized.

[0031] Further, the uncertainty is quantified through the probability distribution of abrasive particle volume, avoiding the limitation of a single numerical value, and random sampling is used to quickly simulate complex wear processes, which is suitable for high-dimensional problems.

[0032] Further, complex relationships are decomposed into manageable sub-modules to improve model robustness, Bayesian networks capture the fuzziness of wear-elasticity approaching quantity to support risk assessment, hard constraints are forced to meet physical laws, and soft constraints optimize parameters to avoid model deviation from reality.

[0033] Further, phase errors caused by sensor or system delay are eliminated to improve the consistency of prediction and actual measurement, phase parameters are dynamically adjusted to enhance the generalization ability of the model under different working conditions.

[0034] It can be understood that the beneficial effects of the above-mentioned second aspect to the sixth aspect can be referred to the relevant description in the above-mentioned first aspect, which will not be repeated here.

[0035] In summary, the present application realizes the accurate spatiotemporal prediction of the wear volume of the friction pair by establishing the wear characterization driven by the abrasive particles and the fine-grained mapping of wear-vibration.

[0036] The technical solutions of the present application will be further described in detail below with the help of the drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 The present application is a general flowchart; Figure 2 The wear-vibration mapping model structure diagram is provided; Figure 3 The digital twin model structure diagram is provided; Figure 4 The wear volume sequence prediction result diagram of the present application is provided; Figure 5 The schematic diagram of the computer equipment provided by an embodiment of the present application is provided; Figure 6 The block diagram of a chip provided by the present application according to an embodiment is provided.

[0038] Among them, 60. Computer equipment; 61. Processor; 62. Memory; 63. Computer program; 600. Electronic equipment; 610. Processing unit; 620. Storage unit; 6201. Random access storage unit; 6202. Cache storage unit; 6203. Read-only storage unit; 6204. Program / utility; 6205. Program module; 630. Bus; 640. Display unit; 650. Input / output interface; 660. Network adapter; 700. External equipment. DETAILED DESCRIPTION

[0039] The present application provides a surface wear volume spatiotemporal distribution prediction method combining abrasive particle and vibration monitoring, constructs a Markov process driven by abrasive particle size distribution, characterizes a random wear process, realizes dynamic and probabilistic characterization of real-time wear volume temporal evolution, constructs a physically constrained double-branch mapping model, establishes a fuzzy mapping of surface wear volume and vibration time-frequency features, realizes accurate observation of wear volume spatial distribution, and based on the state space modeling framework, fuses the aforementioned two physical mechanisms, constructs a digital twin model combining abrasive particle monitoring and vibration monitoring, and realizes accurate prediction of the wear volume of the friction pair component in the time and space dimensions. The present application predicts the spatiotemporal distribution of the wear volume by fusing the physical correlation between the abrasive particle information and the vibration features and the wear process, realizes accurate monitoring of the wear process.

[0040] Please see Figure 1 The present invention provides a method for predicting the spatiotemporal distribution of surface wear volume using combined abrasive and vibration monitoring, comprising the following steps: S1, combining probabilistic characterization and Markov process, constructs a stochastic wear characterization model driven by abrasive size distribution, realizing dynamic and probabilistic characterization of real-time wear rate.

[0041] S101. Using the equivalent diameter of the abrasive grains extracted from the online ferrography image as input, and estimating the probability density distribution of the abrasive grain volume through diameter-thickness mapping; Current online ferrography monitoring still relies on two-dimensional image analysis and cannot directly obtain abrasive grain volume information. Therefore, a diameter-thickness mapping based on a logarithmic function is constructed to obtain the volume of each abrasive grain, thereby forming the probability density distribution of abrasive grain volume.

[0042] S102. By simplifying the wear process into the cumulative peeling behavior of surface material, the wear volume can be represented by the abrasive volume per unit time, i.e., the rate of change of the surface wear volume. Considering the spatial difference of the wear volume, the wear surface is divided into discrete units, and the wear process is quantified by the wear volume sequence. For a single discrete unit, its wear rate can be represented as the cumulative volume of peeled abrasive grains within a specific time, and these abrasive grain volumes follow the probability density distribution obtained in S101. Therefore, multiple sampling and accumulation of the probability density distribution of abrasive grain volume can be used to construct a dynamic probabilistic characterization of the local wear rate of the friction pair surface, specifically:

[0043] in, for Discrete wear rate over time The number of abrasive particles generated in a localized area. No. abrasive grain volume of the second sample. This represents the probability density distribution of abrasive particle volume.

[0044] S103. Wear rate characterization based on abrasive-driven processes: The wear process is modeled as a discrete Markov process, specifically as follows:

[0045] in, for Wear volume over time S104. Based on the aforementioned Markov process, Monte Carlo sampling is introduced, which can predict the temporal evolution of the statistical characteristics of wear volume of different discrete units on the friction pair surface from the abrasive volume information. For example, the average wear level can be quantified by the mean, or the discreteness of the wear degradation degree can be quantified by the standard deviation.

[0046] S2, based on the physical contact mechanism, a physical constraint double-branch wear-vibration mapping model is constructed, a fuzzy mapping between the surface wear volume and the vibration time-frequency characteristics is established, and the accurate observation of the spatial distribution of the wear volume is realized; S201, the spatial change of the wear volume induces the corresponding frequency change of the vibration signal. Therefore, the time-frequency characteristics are preferably selected to quantify the instantaneous change of the vibration characteristics. Specifically, according to the rotational speed of the friction pair, the signal segment corresponding to one rotation of the friction pair is intercepted from the real-time vibration signal , and then the short-time Fourier transform is used to calculate the time-frequency spectrum, and the time-frequency spectrum centroid frequency sequence is extracted therefrom , which constitutes the observation index of the wear volume sequence of the friction pair; S202, by fusing the physical constraint and the wear topography sample, a physical constraint wear-vibration mapping model is constructed, and a fuzzy mapping between the wear volume and the vibration signal time-frequency spectrum centroid frequency is realized. The model is shown in Figure 2 , which adopts a double-branch prediction framework involving an elastic strain prediction branch and a pressure prediction branch. The elastic strain prediction branch constitutes the main line of the wear-vibration mapping, and the pressure prediction branch is designed to constrain the previous branch. In addition, the stress balance condition and the deformation coordination condition in the physical mechanism are introduced to construct multiple physical constraints for model prediction post-processing and parameter optimization.

[0047] S203, the pressure prediction branch , adopts a convolution network to establish an end-to-end nonlinear mapping between the wear topography and the pressure distribution; S204, the elastic strain prediction branch , involves a multi-level mapping of wear volume-elastic approach- vibration time-frequency spectrum centroid frequency. Considering the cognitive uncertainty introduced by different wear topographies, a Bayesian neural network is used to constitute a fuzzy mapping between the wear volume and the elastic approach , and effectively measure this uncertainty.

[0048] S205, according to the elastic approach-intrinsic frequency-vibration frequency causal chain, a physical formula is used to constitute the mapping between the elastic approach and the vibration time-frequency spectrum centroid frequency , specifically:

[0049] wherein, is the predicted value of the vibration time-frequency spectrum centroid frequency; is the elastic approach; is the applied load; is the mass of the friction component; is the stiffness index; is the scaling factor and the translation factor.

[0050] 206、Hard physical constraints, transforming stress balance conditions into hard constraints, used as a post-processing layer of stress prediction branch, specifically:

[0051] where, is the stress distribution matrix; is the initial gap matrix; is the height matrix of wear topography; is the upper limit of elastic strain; is the equivalent elastic modulus of friction pair; is the loss function based on physical constraints; is the compliance matrix.

[0052] S207、Soft constraints, constructing soft constraints in the form of penalty function from deformation compatibility conditions, specifically:

[0053] In the model training process, by minimizing this soft constraint, unsupervised optimization of model parameters can be achieved.

[0054] S3、Based on the state space modeling framework, the joint wear and vibration monitoring wear degradation digital twin model is constructed by integrating step S1 and step S2, and the accurate spatiotemporal prediction of the wear volume of the friction pair is realized.

[0055] Please refer to Figure 3 , the wear and vibration driven digital twin model is composed of physical entities and virtual entities, wherein the physical entities involve core friction components, and online wear particle monitoring and vibration monitoring.

[0056] The virtual entity is a state space model, which adopts a Markov process driven by wear particle size distribution to construct the state transition equation, and a wear-vibration mapping model based on physical constraints to construct the observation equation, specifically:

[0057] where, represents the cyclic displacement processing; is the observation noise of vibration features, which is assumed to follow a normal distribution with zero mean and variance.

[0058] A bidirectional data interaction mechanism is introduced in the digital twin model. The physical entity derives the real-time wear particle size distribution and vibration features to update the wear rate and vibration observation values in the virtual entity online. The virtual entity derives the wear volume sequence prediction results to realize accurate monitoring of the actual friction pair.

[0059] In the real-time prediction process, online parameter updates are introduced; in response to changes in vibration response characteristics, the parameters of the observation equations are updated at the start time of each operating interval. In the real-time prediction process, by maximizing and Inter-correlation coefficient to determine phase shift Based on the above data interaction and parameter updates, particle filtering is used to predict the wear volume sequence online. .

[0060] Observation equation parameters Specifically:

[0061] Phase shift Specifically:

[0062] In another embodiment of the present invention, a spatiotemporal distribution prediction system for surface wear volume based on combined abrasive and vibration monitoring is provided. This system can be used to implement the above-mentioned spatiotemporal distribution prediction method for surface wear volume based on combined abrasive and vibration monitoring. Specifically, the spatiotemporal distribution prediction system for surface wear volume based on combined abrasive and vibration monitoring includes a probability module, a mapping module, and a prediction module.

[0063] Among them, the probability module constructs a stochastic wear characterization model driven by abrasive particle size distribution, and realizes dynamic probabilistic characterization of the temporal evolution of wear volume of friction pair through joint probabilistic characterization and Markov process; The mapping module constructs a physically constrained dual-branch wear-vibration mapping model, establishes a fuzzy mapping between surface wear volume and vibration time-frequency characteristics, and realizes the observation of the spatial distribution of wear volume on the friction pair surface; The prediction module, based on the state-space modeling framework, constructs a wear degradation digital twin model that combines abrasive and vibration monitoring. The constructed abrasive size distribution-Markov process model is used as the state transition equation of the digital twin model, and the constructed dual-branch wear-vibration mapping model is used as the observation equation constraint condition of the digital twin model, thereby achieving accurate spatiotemporal prediction of the wear volume of the friction pair surface.

[0064] The application provides a terminal device, which comprises a processor and a memory for storing a computer program, wherein the computer program comprises program instructions, and the processor is used for executing the program instructions stored in the computer storage medium. The processor can be a central processing unit (CPU), and can also be other general-purpose processors, graphics processing units (GPUs), tensor processing units (TPUs), digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components and the like, which are the computing core and control core of the terminal, and are suitable for implementing one or more instructions, and are specifically suitable for loading and executing one or more instructions to implement a corresponding method flow or a corresponding function; the processor in the embodiment of the application can be used for the operation of the surface wear volume spatiotemporal distribution prediction method based on joint abrasive particle and vibration monitoring, comprising: a random wear characterization model driven by abrasive particle size distribution is constructed, dynamic probability characterization of the time sequence evolution of the wear volume of the friction pair is realized through joint probability characterization and Markov process; a physical constraint double-branch wear-vibration mapping model is constructed, fuzzy mapping between the surface wear volume and the vibration time-frequency characteristics is established, and accurate observation of the spatial distribution of the surface wear volume of the friction pair is realized; based on a state space modeling framework, a wear degradation digital twin model based on joint abrasive particle and vibration monitoring is constructed, the abrasive particle size distribution-Markov process model constructed is taken as a state transition equation of the digital twin model, and the double-branch wear-vibration mapping model constructed is taken as a constraint condition of an observation equation of the digital twin model, and accurate spatiotemporal prediction of the surface wear volume of the friction pair is realized.

[0065] Please refer to Figure 5 , the terminal device is a computer device, the computer device 60 of the embodiment comprises a processor 61, a memory 62 and a computer program 63 stored in the memory 62 and capable of running on the processor 61, and the computer program 63 realizes the method for estimating the concentration of radioactive iodine species in the post-accident containment when executed by the processor 61, and details are not repeated here. Alternatively, the computer program 63 realizes the functions of various models / units in the surface wear volume spatiotemporal distribution prediction system based on joint abrasive particle and vibration monitoring when executed by the processor 61, and details are not repeated here.

[0066] The computer device 60 can be a desktop computer, a notebook computer, a palm computer, a cloud server, and the like. The computer device 60 can include, but is not limited to, a processor 61, a memory 62. Those skilled in the art can understand that the processor 61 and the memory 62 can be connected through a bus, and the bus can be a peripheral component interconnect (PCI) bus, a serial advanced technology attachment (SATA) bus, a universal serial bus (USB), or the like. Figure 5 The computer device 60 is only an example and does not constitute a limitation on the computer device 60, and can include more or fewer components than shown, or combine certain components, or include different components, for example, the computer device can also include an input / output device, a network access device, a bus, and the like.

[0067] The processor 61 can be a central processing unit (CPU), and can also be other general-purpose processors, graphics processing units (GPUs), tensor processing units (TPUs), digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, and the like. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0068] The memory 62 can be an internal storage unit of the computer device 60, such as a hard disk or a memory of the computer device 60. The memory 62 can also be an external storage device of the computer device 60, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, and the like.

[0069] Further, the memory 62 can include both an internal storage unit and an external storage device of the computer device 60. The memory 62 is used to store computer programs and other programs and data required by the computer device. The memory 62 can also be used to temporarily store data that has been output or will be output.

[0070] Please refer to Figure 6The terminal device is an electronic device 600, which is in the form of a general computing device. Components of the electronic device can include, but are not limited to, at least one processing unit 610, at least one storage unit 620, a bus 630 that connects the various platform components, including the storage unit 620 and the processing unit 610, a display unit 640, and the like.

[0071] The storage unit stores program code that can be executed by the processing unit 610 to cause the processing unit 610 to perform the steps described in the method portion of the specification above in accordance with the various example embodiments of the present application. For example, the processing unit 610 can perform the steps shown in FIG. 6. Figure 1

[0072] The storage unit 620 can include a readable medium in the form of volatile storage such as a random access memory (RAM) 6201 and / or cache memory 6202, and can further include a non-volatile storage such as read only memory (ROM) 6203.

[0073] The storage unit 620 can also include a program / utility 6204 having a set of program modules 6205 such as an operating system, one or more application programs, other program modules, and program data, each of which can give rise to the implementation of a network environment in one or a combination of the examples.

[0074] The bus 630 can be representative of one or more of several types of bus structures, including a storage bus or bus controller, a peripheral bus, a graphics bus, a processing bus, or a local bus using any of a variety of bus architectures.

[0075] The electronic device 600 can also communicate with one or more external devices 700 such as a keyboard or pointing device, a Bluetooth device, and / or other devices not shown by way of the input / output interface 650. Furthermore, the electronic device 600 can communicate with one or more devices that enable a user to interact with the electronic device 600, and / or one or more devices that enable the electronic device 600 to communicate with one or more other computing devices. Such communication can be via the input / output interface 650. The electronic device 600 can also communicate with one or more networks (such as a local area network, a wide area network, and / or the public switched telephone network) via the network adapter 660. The network adapter 660 can communicate with the other components of the electronic device 600 via the bus 630. It should be appreciated that, although not shown explicitly, other hardware and / or software components could be used in conjunction with the electronic device 600. Such components include, but are not limited to, microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archival storage systems, etc.

[0076] ​Embodiment 4 The present application further provides a storage medium, specifically a computer readable storage medium, which is a memory device in the terminal device, used for storing programs and data. It can be understood that the computer readable storage medium herein can include the built-in storage medium in the terminal device, and of course can include the expansion storage medium supported by the terminal device, and can be any tangible medium containing or storing programs, which can be used by or in combination with the instruction execution system, device or apparatus. The computer readable storage medium provides a storage space, which stores the operating system of the terminal. Moreover, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, which can be one or more computer programs (including program codes). It should be noted that more specific examples of the computer readable storage medium include an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory, an optical fiber, a portable compact disk read-only memory, an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0077] The computer readable storage medium further includes a data signal carried in the baseband or as a part of a carrier wave, in which readable program codes are borne. Such a propagated data signal can take various forms, including but not limited to electro-magnetic signal, optical signal or any suitable combination of the above. The readable storage medium can also be any readable medium other than the readable storage medium, which can send, propagate or transmit programs for use by or in combination with the instruction execution system, device or apparatus. The program codes contained in the readable storage medium can be transmitted by any suitable medium, including but not limited to wireless, wired, optical cable, radio frequency, etc. or any suitable combination of the above.

[0078] The program code for performing the operations of the present application can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, C++, etc., and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device or entirely on the remote computing device or server. In the latter scenario, the remote computing device can be connected to the user's computing device through any type of network, including a local area network or a wide area network, or the connection can be made to an external computing device, such as through the Internet using an Internet Service Provider.

[0079] The one or more instructions stored in the computer readable storage medium can be loaded and executed by the processor to implement the corresponding steps of the surface wear volume spatiotemporal distribution prediction method related to joint abrasive particle and vibration monitoring in the above embodiments; the one or more instructions stored in the computer readable storage medium are loaded and executed by the processor to implement the following steps: A random wear representation model driven by abrasive particle size distribution is constructed, dynamic probability representation of the temporal evolution of the wear volume of the friction pair is realized through joint probability representation and Markov process; a physically constrained double-branch wear-vibration mapping model is constructed, fuzzy mapping between the surface wear volume and the vibration time-frequency features is established, and accurate observation of the surface wear volume spatial distribution of the friction pair is realized; based on a state space modeling framework, a wear degradation digital twin model of joint abrasive particle and vibration monitoring is constructed, the abrasive particle size distribution-Markov process model constructed is taken as a state transition equation of the digital twin model, and the double-branch wear-vibration mapping model constructed is taken as a constraint condition of an observation equation of the digital twin model, and accurate spatiotemporal prediction of the surface wear volume of the friction pair is realized.

[0080] The database involved in each of the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a blockchain, and the like, without being limited thereto. The processor involved in each of the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, and the like, without being limited thereto.

[0081] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work are within the scope of protection of the present application.

[0082] In the prior art, the wear degradation monitoring of friction pair components mainly relies on single physical information such as vibration signals or wear particle characteristics. Existing models establish the correlation between vibration characteristics and surface damage size through dynamic methods, or predict the total wear volume by using the quantitative relationship between wear particle concentration and wear rate. However, the actual wear process involves complex coupling of material failure, topography evolution and mechanical action, showing strong randomness, which makes it difficult for existing methods to capture the spatial and temporal differences. In addition, the macro component-level physical correlation mechanism cannot reflect the local wear dynamics, limiting the prediction accuracy.

[0083] To solve the above problems, the prior art cannot accurately characterize the random-induced spatio-temporal degradation process, and single physical information cannot cover the wear behavior of multi-mechanism coupling. To address this contradiction, it is necessary to integrate multi-modal monitoring data and build a fine-grained physical correlation model. First, the wear particle size distribution can reflect the local wear dynamics, and the wear rate can be represented by probabilistic modeling. Second, the vibration signal time-frequency characteristics and surface topography have physical correlation, and a multi-branch mapping model needs to be established to capture the spatial distribution. Finally, integrate multi-source information into the digital twin framework to achieve dynamic calibration and spatio-temporal prediction.

[0084] Therefore, the present application proposes a method for predicting the spatio-temporal distribution of surface wear volume based on joint wear particle and vibration monitoring, comprising the following steps: constructing a random wear characterization model driven by wear particle size distribution, realizing dynamic characterization of wear volume time evolution by joint probability representation and Markov process; constructing a physically constrained dual-branch wear-vibration mapping model, establishing a fuzzy mapping between wear volume and vibration time-frequency characteristics; based on the state space modeling framework, constructing a wear degradation digital twin model based on joint wear particle and vibration monitoring, realizing the spatio-temporal prediction of surface wear volume.

[0085] Among them, the random wear characterization model driven by wear particle size distribution refers to the construction of a dynamic wear rate model through wear particle equivalent diameter and volume probability distribution, which can be realized by Monte Carlo simulation and discrete Markov process, and is used to quantify the random evolution of local wear volume. The physically constrained dual-branch wear-vibration mapping model refers to a dual-granularity framework combining pressure prediction branch and elastic strain prediction branch, which can be realized by convolution network and Bayesian neural network for nonlinear mapping, and physical constraints are imposed through stress balance and deformation coordination conditions to establish the correlation between wear volume spatial distribution and vibration characteristics. The wear degradation digital twin model refers to a state space model integrating the state transition equation driven by wear particles and the observation equation driven by vibration, which can be realized by particle filtering algorithm for online parameter updating and prediction, and is used to integrate multi-source monitoring data and dynamically calibrate the prediction results.

[0086] Specifically, first, the equivalent diameter of the abrasive particle is extracted based on the online ferrography image, the volume probability distribution is estimated through the diameter-thickness mapping, and the discrete wear rate sequence is constructed by using Monte Carlo sampling. Subsequently, the wear process is modeled as a Markov chain to predict the time evolution of the local wear volume. Second, the time-frequency spectrum centroid frequency sequence is extracted from the vibration signal to construct a double-branch mapping model: the pressure prediction branch establishes a nonlinear mapping between the wear morphology and the pressure distribution through a convolution network, the elastic strain prediction branch realizes a fuzzy mapping from the wear volume to the elastic approach through a Bayesian network, and the correlation between the approach and the vibration features is derived by combining the physical formula. Finally, in the digital twin model, the Markov process driven by the abrasive particle is taken as the state transition equation, and the vibration feature mapping is taken as the observation equation, and the model parameters are dynamically updated and the spatio-temporal distribution of the wear volume is predicted by fusing the real-time monitoring data through the particle filtering algorithm.

[0087] Compared with the prior art, the existing method only uses a single physical information to establish a macro component-level correlation, while the present scheme constructs a multi-granularity physical constraint model by combining abrasive particle and vibration monitoring data, which can capture the local wear dynamics. The existing model relies on deterministic physical relationships, while the present scheme introduces probabilistic representation and fuzzy mapping to effectively handle the spatio-temporal differences induced by randomness. The existing digital twin framework lacks a dynamic calibration mechanism, while the present scheme realizes adaptive adjustment of wear prediction through bidirectional data interaction and online parameter updating.

[0088] Through the above technical scheme, the present application can accurately represent the spatio-temporal distribution of the wear volume on the surface of the friction pair, overcoming the limitations of single physical information models. By fusing abrasive particle size distribution and vibration time-frequency features, dynamic modeling of multi-mechanism coupled wear processes is achieved. By combining physical constraints and probabilistic modeling, the prediction accuracy of the random degradation process is improved. Through the online calibration function of the digital twin framework, the dynamic changes of the actual working conditions are adapted.

[0089] The present application further proposes specific steps for constructing a random wear representation model driven by abrasive particle size distribution, including: taking the equivalent diameter of the abrasive particle extracted from the online ferrography image as the input, and estimating the probability density distribution of the abrasive particle volume through diameter-thickness mapping; dynamically representing the wear rate of the friction pair surface by sampling the probability density distribution of the abrasive particle volume multiple times and accumulating the sampling results; modeling the wear process as a discrete Markov process based on the abrasive particle-driven wear rate representation; and predicting the time evolution of the statistical characteristics of the wear volume in different regions of the friction pair surface using Monte Carlo simulation.

[0090] The equivalent diameter of the abrasive particle refers to the diameter parameter of the equivalent circle obtained by converting the abrasive particle contour through image processing technology, which can be realized by using morphological edge detection combined with the minimum circumscribed circle algorithm to quantify the geometric size of the abrasive particle.

[0091] wherein the diameter-thickness mapping refers to an empirical relationship between two-dimensional diameter and thickness based on statistical laws of three-dimensional topography of abrasive particles, and can be specifically realized by fitting historical abrasive particle data using a regression analysis method, and is used to calculate the volume of abrasive particles from two-dimensional images.

[0092] wherein the Monte Carlo simulation refers to a numerical calculation method for generating random samples through probability density distribution, and can be specifically realized by using a Markov chain Monte Carlo algorithm, and is used to simulate random evolution characteristics of the wear process.

[0093] wherein the Markov process refers to a random process model with no memory characteristics, and can be specifically described by using a transition probability matrix to describe the change law of the wear state at adjacent time points, and is used to represent the time sequence correlation of wear evolution.

[0094] Specifically, after obtaining abrasive particle image data through an online ferrography monitoring system, the equivalent diameter parameter of the abrasive particle is first extracted, and the volume value of each abrasive particle is calculated based on the pre-established diameter-thickness mapping relationship. The probability density distribution function of the abrasive particle volume is constructed by using a kernel density estimation method, and the Latin hypercube sampling technique is used for multiple independent sampling, and the abrasive particle volume of each sampling is added to obtain the wear rate estimation value of the discrete time period. The wear rate of each time period is used as a state variable of the Markov chain, and a state transition probability matrix is established to describe the evolution law of the wear state. Finally, a large number of possible wear evolution paths are generated by using the Monte Carlo method, and the mean, variance and other statistical quantities of the wear volume in different regions under each path are counted, so as to realize the probabilistic prediction of the wear spatio-temporal evolution.

[0095] Compared with the prior art, the existing wear prediction model usually establishes a wear equation based on deterministic parameters, and cannot represent the randomness of the abrasive particle generation process. By introducing the probability density distribution and the Markov process, the statistical characteristics of the abrasive particle size are converted into the dynamic evolution of the wear rate distribution, so that the uncertainty of material shedding in the actual wear process can be reflected. The fixed wear coefficient used in the existing method is difficult to adapt to the dynamic changes under complex working conditions, and the present scheme continuously corrects the probability model parameters by updating the abrasive particle distribution data online, thereby improving the adaptability of the model.

[0096] Through the above technical scheme, the present application can effectively solve the problem that the existing wear prediction model ignores the randomness of abrasive particle generation, and by establishing a probabilistic wear representation model, the dynamic tracking of the time sequence evolution of the wear volume of the friction pair surface is realized. By fusing the abrasive particle distribution data monitored online, the adaptability of the model to random factors in the actual wear process is improved, and a probabilistic quantitative basis is provided for predicting the wear spatio-temporal distribution.

[0097] The application further proposes to construct a joint abrasive particle and vibration monitoring wear degradation digital twin model, specifically: the digital twin model is composed of a physical entity and a virtual entity, the physical entity involves a basic friction pair and online abrasive particle monitoring and vibration monitoring; the virtual entity is a state space model, a Markov process driven by abrasive particle size distribution is used to constitute a state transition equation, and an observation equation is constructed by a physically constrained wear-vibration mapping model; a two-way data interaction mechanism is introduced, the physical entity derives real-time abrasive particle size distribution and vibration characteristics to update the wear rate and vibration observation values in the virtual entity, the virtual entity derives the wear volume sequence prediction result to realize accurate monitoring of the actual friction pair; online parameter updating is introduced in the real-time prediction process, the observation equation parameters are updated at the beginning of each running interval in response to changes in vibration response characteristics; the phase shift is determined by maximizing the correlation coefficient between the predicted value and the actual value; combined with data interaction and parameter updating, particle filtering is used to predict the wear volume sequence online.

[0098] The state space model refers to a dynamic system model composed of a state transition equation and an observation equation, which can specifically describe the time sequence evolution of the wear volume using a Markov process driven by abrasive particles, and construct an observation equation combined with vibration characteristics, thereby realizing dynamic modeling of the wear degradation process. The two-way data interaction mechanism refers to real-time data transmission and updating between the physical entity and the virtual entity, which can specifically collect abrasive particle distribution and vibration signals through sensors and synchronize them to the virtual model for parameter calibration, while feeding back the prediction results to the physical entity, thereby forming a closed-loop monitoring. The phase shift determination refers to adjusting the time alignment of the prediction sequence and the actual observation sequence through correlation analysis, which can specifically calculate the correlation coefficient using a sliding window and select the offset corresponding to the maximum value, thereby eliminating the time lag error between signal collection and model prediction.

[0099] Specifically, the digital twin model constructs a state space framework with physical constraints by fusing multi-modal data of abrasive particle monitoring and vibration monitoring. A Markov process driven by abrasive particle size distribution is used to describe the random evolution characteristics of the wear volume, while the vibration signal establishes a mapping relationship with the wear volume through the time-frequency centroid frequency. During operation, the physical entity real-time collects the abrasive particle equivalent diameter and vibration signal segments, extracts the abrasive particle volume distribution and centroid frequency sequence, and inputs these data into the virtual entity. The virtual entity updates the wear rate prediction value through the state transition equation, while using the observation equation to convert the predicted wear volume into vibration characteristics and compare it with the actual observation value. The online parameter updating mechanism adjusts the scaling factor and translation factor in the observation equation to adapt to the changes in vibration response characteristics with the wear state. The determination of the phase shift eliminates the prediction error caused by time asynchronization by calculating the correlation coefficient of the prediction sequence and the actual sequence using a sliding window. Finally, the particle filtering algorithm realizes dynamic prediction of the wear volume sequence combined with the state space model and real-time data.

[0100] Compared with the prior art, the existing digital twin model usually only relies on single physical information, such as wear prediction only through vibration signals or abrasive particle concentration, and lacks dynamic modeling of random wear processes. However, the present scheme can capture the randomness of material damage and the dynamic changes of vibration response by fusing abrasive particle size distribution and vibration time-frequency features and constructing a two-way interaction mechanism with physical constraints. In addition, the existing method does not consider the time offset problem of signal acquisition and model prediction, while the present scheme significantly improves the accuracy of time series prediction through phase offset correction and online parameter updating. Compared with the model based on macro component level dynamics mapping, the present scheme realizes the spatiotemporal distribution prediction of local wear volume of the friction pair surface through fine-grained state space modeling.

[0101] Through the above technical scheme, the present application solves the problem that the existing model cannot accurately capture the spatiotemporal difference of wear due to the dependence on single physical information. By fusing the random wear model driven by abrasive particle distribution and the observation model constrained by vibration signals, dynamic calibration of wear volume evolution is realized. The two-way data interaction mechanism combined with online parameter updating can adapt to the changes of vibration response under different working conditions, thereby improving the prediction robustness. The phase offset correction method effectively eliminates the cumulative error in time series prediction, so that the digital twin model can accurately reflect the real-time wear state of the actual friction pair.

[0102] The present application further proposes to construct a physically constrained two-branch wear-vibration mapping model, specifically including intercepting the signal segment corresponding to one rotation of the friction pair from the real-time vibration signal, extracting the time-frequency spectrum centroid frequency sequence to form the observation index; constructing a dual-granularity prediction framework including a pressure prediction branch and an elastic strain prediction branch, and applying multiple soft and hard physical constraints; the pressure prediction branch uses a convolutional network to establish an end-to-end nonlinear mapping between wear morphology and pressure distribution; the elastic strain prediction branch constructs a fuzzy mapping between wear volume and elastic approach by a Bayesian neural network, and combines physical formulas to establish a mapping between elastic approach and vibration time-frequency spectrum centroid frequency; the stress balance condition is converted into a hard constraint acting on the post-processing layer of the pressure prediction branch, and a soft constraint is constructed through the deformation compatibility condition for model parameter optimization.

[0103] The double-granularity prediction framework refers to a parallel network structure for jointly modeling the contact pressure distribution prediction and the elastic approach prediction, and can be specifically implemented by a double-branch convolutional neural network to realize multi-dimensional feature fusion through parallel processing of different physical quantities. The pressure prediction branch refers to a contact pressure distribution prediction module established based on a convolutional neural network, and can specifically extract the non-linear relationship between the wear morphology and the pressure distribution by using multiple convolution kernels, and realize the spatial distribution prediction of the local contact stress through end-to-end training. The elastic approach refers to the approach displacement of the surface of the friction pair due to elastic deformation, and can specifically establish a fuzzy mapping between the wear volume and the elastic approach by using a Bayesian neural network, and output the cognitive uncertainty through a probability distribution. The hard physical constraint refers to a forced satisfaction condition based on a stress balance equation, and can specifically embed the stress balance condition into the network output layer through matrix operation, and correct the prediction result through numerical calculation to meet the physical law. The soft physical constraint refers to a loss function constraint term based on a deformation coordination condition, and can specifically convert the coordination condition into a regularization term by using the Lagrange multiplier method, and optimize the network parameters through back propagation.

[0104] Specifically, in the vibration signal processing stage, the vibration signal segment corresponding to the single-cycle operation of the friction pair is intercepted, and the time-frequency spectrum centroid frequency sequence is extracted as an observation index. In the model construction stage, a double-branch network architecture is used to predict the contact pressure distribution and the elastic approach: the pressure prediction branch extracts the wear morphology features through the convolution layer and outputs the contact pressure distribution matrix; the elastic strain prediction branch establishes a probability mapping between the wear volume and the elastic approach through the Bayesian network, and deduces the quantitative relationship between the vibration features and the elastic approach based on the physical formula. In the constraint application stage, the hard constraint performs numerical correction on the pressure prediction result through the stress balance equation to ensure that the prediction result meets the mechanical balance condition; the soft constraint constructs a regularization term through the deformation coordination equation, and optimizes the network parameters in the training process to reduce the degree of violation of the physical law. Thus, a wear-vibration mapping model with physical interpretability is formed.

[0105] Compared with the prior art, the traditional method usually directly establishes the mapping relationship between the wear volume and the vibration features by using a single network model, and lacks the modeling capability of intermediate physical quantities such as contact pressure and elastic deformation, resulting in limited model generalization. The present scheme separates the prediction processes of the contact pressure and the elastic strain by using a double-branch architecture, processes the cognitive uncertainty by using a Bayesian network, and introduces soft and hard constraints to ensure that the prediction result meets the mechanical law, thereby effectively improving the model robustness under complex working conditions.

[0106] By the technical solution, the technical problem that a traditional single mapping model cannot accurately reflect the coupling effect of contact pressure distribution and elastic deformation is solved, the high-precision mapping between the wear volume and the vibration feature is realized by decomposing the physical quantity prediction process and applying multiple types of physical constraints, and reliable observation input is provided for accurate prediction of a subsequent digital twin model.

[0107] The application further proposes that the physical mapping of the elastic approach amount-vibration time-frequency spectrum centroid frequency in the elastic strain prediction branch is specifically:

[0108] Among them, is a predicted value of the vibration time-frequency spectrum centroid frequency; is an elastic approach amount; is an applied load; is a friction pair mass; is a stiffness index; is a scaling factor and a translation factor.

[0109] The elastic approach amount refers to the elastic deformation amount of the friction pair surface due to wear, which can be estimated by using the Hertz contact model in the contact mechanics theory, and is used to quantify the influence of the wear volume on the contact stiffness. The centroid frequency refers to the energy distribution center position of the vibration signal time-frequency spectrum, which can be extracted by using the short-time Fourier transform, and is used to represent the shift rule of the frequency domain feature of the vibration signal. The scaling factor and the translation factor refer to linear parameters used to adjust the dimension matching of the physical formula, which can be calibrated by using the least square method on experimental data, and are used to eliminate the inherent deviation between the theoretical model and the actual system.

[0110] Specifically, in the elastic strain prediction branch, the mapping relationship between the elastic approach amount and the vibration time-frequency spectrum centroid frequency based on the physical formula is established, so as to associate the elastic deformation caused by the wear volume with the frequency domain feature of the vibration signal. The elastic approach amount is mapped with the wear volume by using the Bayesian neural network, and the centroid frequency is calculated by using the physical formula containing the applied load, the friction component mass and the stiffness index. The physical mapping linearly corrects the theoretical model by introducing the scaling factor and the translation factor, so that the predicted value can adapt to the dynamic characteristics of the actual system. During operation, the elastic approach amount affects the system vibration response through the change of the contact stiffness, and the predicted value of the centroid frequency directly reflects the dynamic change of the elastic approach amount through the physical formula, so as to realize the multi-level correlation modeling between the wear volume and the vibration feature.

[0111] Compared with the prior art, the existing vibration feature and wear volume mapping method usually only relies on a data-driven neural network modeling, lacking explicit expression of physical mechanisms such as contact stiffness and elastic deformation. The present scheme introduces a physical formula based on contact mechanics theory, uses the elastic approach as an intermediate variable to construct the quantitative correlation between the wear volume and the vibration feature, retains the adaptability of the data-driven model, and improves the explainability and stability of the mapping relationship through physical constraints.

[0112] Through the above technical scheme, the present application can effectively eliminate the vibration feature deviation error caused by the nonlinear change of the contact stiffness, and explicitly express the dynamic correlation between the elastic deformation and the vibration frequency domain feature through the physical formula, so as to realize high-precision mapping of the wear volume and the vibration time-frequency spectrum centroid frequency under complex working conditions, and provide reliable observation input for subsequent time-space prediction of the digital twin model.

[0113] The present application further proposes to construct a wear degradation digital twin model combining abrasive particles and vibration monitoring. The model is composed of a physical entity and a virtual entity. The physical entity involves a basic friction pair and online abrasive particle monitoring and vibration monitoring. The virtual entity is a state space model. A Markov process driven by abrasive particle size distribution is used to construct a state transition equation, and an observation equation is constructed by a physically constrained wear-vibration mapping model. A bidirectional data interaction mechanism is introduced. The physical entity derives real-time abrasive particle size distribution and vibration features to update the wear rate and vibration observation values in the virtual entity online. The virtual entity derives the wear volume sequence prediction result to realize accurate monitoring of the actual friction pair. Online parameter updating is introduced in the real-time prediction process. The observation equation parameters are updated at the beginning of each running interval according to the change of the vibration response characteristics. The phase offset is determined by maximizing the correlation coefficient. Particle filtering is used to predict the wear volume sequence online by combining data interaction and parameter updating.

[0114] The state space model refers to a framework that describes the dynamic behavior of a system through mathematical equations. It can be implemented by combining a state transition equation and an observation equation. The state transition equation is constructed by a Markov process driven by abrasive particle size distribution, which is used to represent the time evolution law of the wear volume. The observation equation is constructed by a physically constrained wear-vibration mapping model, which is used to correlate the vibration feature with the wear volume. The bidirectional data interaction mechanism refers to the real-time data exchange channel between the physical entity and the virtual entity. It can be realized through a bidirectional transmission interface of sensor data acquisition and model parameters. The physical entity transmits the online monitored abrasive particle size distribution and vibration features to the virtual entity. The virtual entity feeds back the predicted wear volume sequence to the physical entity. Online parameter updating refers to the process of dynamically adjusting the model parameters according to real-time monitoring data. It can be realized through a phase offset optimization algorithm and a correlation coefficient maximization criterion, which is used to eliminate the time deviation between the vibration signal and the wear volume, and ensure the dynamic matching of the observation equation with the actual state of the physical entity.

[0115] Specifically, the digital twin model collects the size distribution of abrasive particles and vibration signals in real time through the physical entity, the state transition equation in the virtual entity simulates the random evolution of the wear volume based on the Markov process driven by abrasive particles, and the observation equation generates the predicted value based on the physical mapping relationship between the vibration features and the wear volume. The bidirectional data interaction mechanism inputs the monitoring data of the physical entity into the virtual entity, and at the same time, the prediction results are fed back to the physical entity to form a closed loop. The online parameter updating module adjusts the observation equation parameters at the beginning of each running period, determines the optimal phase offset by calculating the correlation coefficient of the vibration feature sequence and the predicted sequence, and eliminates the timing error between signal acquisition and model prediction. The particle filtering algorithm combines the state transition equation and the updated observation equation to probabilistically predict the wear volume sequence, and outputs the spatiotemporal distribution results with statistical characteristics.

[0116] Compared with the prior art, the existing digital twin model usually only uses a single physical information to construct macro component-level correlation, such as relying only on vibration signals or abrasive particle concentration for qualitative evaluation, and cannot capture the spatiotemporal randomness of the wear process. The present scheme fuses two types of physical information, abrasive particle size distribution and vibration time-frequency spectrum centroid frequency, to construct a state space model with bidirectional data interaction and dynamic parameter updating, establishes fine-grained correlation between wear volume and multi-source monitoring data at the microscopic scale, and effectively solves the problem of limited prediction accuracy caused by single physical mechanism in the prior art.

[0117] Through the above technical scheme, the present application realizes dynamic and accurate prediction of the spatiotemporal distribution of the wear volume of the friction pair surface, specifically: the random characteristics of the wear volume evolution are captured through the Markov process driven by the size distribution of abrasive particles, the spatial distribution observation constraint is provided through the vibration feature mapping model, the state deviation between the model and the entity is eliminated by combining the bidirectional data interaction, the observation equation is dynamically calibrated by using the online parameter updating mechanism, and finally the wear volume prediction sequence with probabilistic statistical characteristics is output by fusing multi-source information through the particle filtering.

[0118] The application further proposes to construct a physically constrained double-branch wear-vibration mapping model, specifically: the signal segment corresponding to one rotation of the friction pair is intercepted from the real-time vibration signal, the time-frequency spectrum centroid frequency sequence is extracted, and the observation index of the friction pair wear volume sequence is constituted; a physically constrained wear-vibration mapping model is constructed to realize the fuzzy mapping of the friction pair surface wear volume and the vibration signal time-frequency spectrum centroid frequency, including a double-granularity prediction framework composed of a pressure prediction branch and an elastic strain prediction branch, and multiple soft and hard physical constraints; the physical mapping of the elastic approach amount-vibration time-frequency spectrum centroid frequency in the elastic strain prediction branch is specifically: the predicted value of the vibration time-frequency spectrum centroid frequency is calculated by the elastic approach amount, the applied load, the friction component mass, the stiffness index, the scaling factor and the translation factor through a physical formula; the hard physical constraint converts the stress balance condition into a hard constraint, which is used as the post-processing layer of the stress prediction branch; the soft constraint is constructed from the deformation coordination condition and is used for model parameter optimization.

[0119] Wherein, the elastic approach amount refers to the displacement amount of the friction pair surface caused by elastic deformation, which can be calculated by combining the contact mechanics model with the material elastic parameters, and is used to represent the deformation degree caused by local wear. The applied load refers to the external force borne by the friction pair during operation, which can be measured in real time by a pressure sensor and used as an input parameter for physical mapping. The stiffness index refers to the elastic stiffness characteristics of the friction pair material, which can be obtained by material performance testing and used to quantify the relationship between elastic strain and vibration characteristics. The hard constraint refers to the mathematical limitation based on the stress balance condition, which can be constructed by the physical relationship between the stress distribution matrix and the initial gap matrix, and is used to modify the output results of the prediction branch. The soft constraint refers to the optimization target based on the deformation coordination condition, which can be used to improve the generalization ability of the mapping model by iteratively adjusting the model parameters through the construction of a loss function.

[0120] Specifically, in the elastic strain prediction branch, a Bayesian neural network is first established to realize the fuzzy mapping of the wear volume and the elastic approach amount, and the cognitive uncertainty is quantified. Then, the elastic approach amount is converted into the predicted value of the vibration time-frequency spectrum centroid frequency based on a physical formula, which includes measurable parameters such as the applied load and the friction component mass, as well as the stiffness index and scaling translation factors obtained through experimental calibration. The hard constraint processes the output of the prediction branch through the stress balance equation to ensure that the stress distribution matrix and the initial gap matrix satisfy the physical balance condition. The soft constraint constructs a loss function through the deformation coordination equation to optimize the neural network parameters during model training, so that the prediction results meet the elastic deformation coordination rules. Through the double-branch framework, the pressure distribution and the elastic strain information are processed simultaneously, and the soft and hard constraints are used in conjunction to realize the multi-level physical correlation between the wear volume and the vibration characteristics.

[0121] Compared with the prior art, the existing method generally establishes a wear-vibration mapping relationship by using a single physical feature, for example, only a linear relationship between vibration amplitude and wear depth is constructed based on a dynamic equation, without considering the coupling effect of elastic deformation and contact stress. The scheme synchronously captures two types of physical mechanisms of pressure distribution and elastic strain through a double-branch structure, processes the nonlinear mapping relationship by combining a Bayesian neural network, and introduces physical constraints based on stress balance and deformation coordination, effectively solving the error accumulation problem caused by single physical information modeling. At the same time, the hard constraint directly corrects the prediction result through post-processing, and the soft constraint improves the adaptability of the model through parameter optimization, and the synergistic effect of the two significantly improves the physical consistency of the mapping model.

[0122] Through the above technical scheme, the application can accurately establish the physical correlation between the surface wear volume and the vibration time-frequency spectrum centroid frequency. Under the condition of measurement noise and working condition fluctuation, the stability of the wear volume spatial distribution observation result can still be maintained. By fusing multiple-level physical mapping relationships and constraint conditions, the cognitive uncertainty caused by single information source modeling is effectively reduced, providing an accurate observation data basis for subsequent digital twin model spatiotemporal prediction.

[0123] The application further proposes a physical mapping of elastic approach amount-vibration time-frequency spectrum centroid frequency in the elastic strain prediction branch, specifically, the predicted value of the vibration time-frequency spectrum centroid frequency is calculated by the elastic approach amount, the applied load, the friction component mass, the stiffness index, the scaling factor and the translation factor through a physical formula.

[0124] The elastic approach amount refers to the elastic deformation amount of the friction pair surface due to wear, which can be realized by establishing a fuzzy mapping between the wear volume and the elastic approach amount using a Bayesian neural network, for quantifying the influence degree of the wear volume on the elastic deformation. The applied load refers to the external pressure borne by the friction pair during operation, which can be obtained by real-time measurement through a pressure sensor, for reflecting the influence of the system force state on the vibration characteristics. The friction component mass refers to the inertia parameter of the moving component of the friction pair, which can be calculated by the product of the material density and the volume, for characterizing the modulation effect of the system inertia on the vibration frequency. The stiffness index refers to the stiffness characteristic parameter of the contact area of the friction pair, which can be calculated by the ratio of the material elastic modulus and the contact area, for describing the influence law of the system stiffness on the vibration frequency. The scaling factor and the translation factor refer to parameters for adjusting the matching degree of the physical model output and the measured data, which can be determined by least squares optimization, for eliminating the inherent deviation between the theoretical model and the actual system.

[0125] Specifically, in the elastic strain prediction branch, by establishing a physical formula containing the elastic approach amount, the applied load, the friction component mass, and the stiffness index, the elastic deformation amount caused by the wear volume is converted into the predicted value of the vibration time-frequency spectrum centroid frequency. When the wear occurs on the surface of the friction pair, the elastic approach amount increases with the increase of the wear volume, resulting in the change of the contact stiffness. The applied load and the friction component mass jointly affect the inertia effect of the system, and the stiffness index reflects the effect of the material characteristics on the contact stiffness. The scaling factor and the translation factor adjust the output of the physical model, so that the predicted centroid frequency value matches the actual vibration signal characteristics. The physical mapping relationship realizes the deterministic conversion from the elastic deformation to the vibration frequency by quantifying the physical action mechanism of the wear volume on the vibration characteristics.

[0126] Compared with the prior art, the existing model usually establishes the relationship between wear and vibration characteristics by using a single physical quantity or an empirical formula, for example, only the vibration frequency offset is derived by the change of the contact stress. However, the present scheme introduces the elastic approach amount as an intermediate variable, combines the applied load, the friction component mass, and other system parameters to construct a vibration frequency prediction model coupled with multiple physical quantities. This mapping relationship based on the physical mechanism can more comprehensively reflect the dynamic influence of the wear volume change on the vibration characteristics, and overcome the defects of the traditional empirical model in the adaptability under complex working conditions.

[0127] Through the above technical scheme, the present application realizes the physical mechanism correlation between the vibration time-frequency spectrum centroid frequency and the wear volume, and effectively eliminates the influence of non-wear factors on the vibration characteristics. The physical mapping model can accurately capture the quantitative relationship between the elastic deformation caused by wear and the vibration frequency change, and provides a reliable physical basis for the subsequent accurate inversion of the wear volume.

[0128] The present application further proposes that, in the real-time prediction process, the phase offset is determined by maximizing the cross-correlation coefficient, specifically: The phase offset refers to the relative displacement difference of the vibration signal and the wear volume sequence on the time axis, and can be specifically realized by calculating the similarity of the two signals at different time delays by using the cross-correlation function, and the phase offset amount is determined by finding the time delay corresponding to the maximum correlation coefficient. The correlation coefficient refers to the linear correlation degree of the two signal sequences in statistics, and can be specifically calculated by using the Pearson correlation coefficient or the normalized cross-correlation function, and is used to quantify the synchronization between the vibration characteristics and the wear volume sequence.

[0129] Specifically, during the operation of the friction pair, there is a dynamic shift in time dimension between the vibration signal and the evolution process of the wear volume. By real-time collection of the vibration feature sequence and the wear volume sequence predicted by the virtual entity, the correlation coefficients of the two at different phase shifts are calculated, and the shift that makes the correlation coefficient reach the maximum is selected as the optimal phase compensation value. This process intercepts the real-time vibration signal fragment through a sliding window, matches it with the predicted wear volume sequence point by point, uses the normalized cross-correlation algorithm to traverse all possible shifts, and finally determines the optimal phase alignment parameter. Real-time correction of phase shift can eliminate the timing mismatch caused by factors such as speed fluctuation and signal transmission delay, thereby improving the adaptability of the digital twin model to dynamic working conditions.

[0130] Compared with the prior art, the traditional method usually assumes that the vibration signal and the wear process maintain a fixed time synchronization relationship, without considering the timing deviation caused by mechanical transmission gap, sensor response delay and other factors in actual operation. In the prior art, fixed phase compensation or ignoring the timing matching problem leads to systematic deviation between the model prediction results and the actual wear state. However, the present scheme effectively solves the problem of prediction error accumulation caused by timing mismatch by dynamically optimizing the phase shift.

[0131] Through the above technical scheme, the present application realizes accurate alignment of the vibration monitoring signal and the wear volume prediction result in time dimension, eliminates the timing deviation caused by mechanical transmission gap or sensor delay, and enables the digital twin model to adaptively adjust the timing correspondence between the observation data and the prediction result, thereby improving the prediction accuracy of the dynamic evolution process of the friction pair surface wear volume in space and time.

[0132] To verify the effectiveness of the method of the present application, wear tests were carried out based on a basic rolling and sliding test rig, and the prediction accuracy of the wear volume in space and time was evaluated by comparing the prediction results of the digital twin model with the detection results of the wear surface.

[0133] The test data set used involves the whole cycle wear degradation process, and it is completely independent of the training data set. The prediction results of part of the wear volume are as shown in Figure 4 It can be observed that the wear volume curves predicted at different times are highly consistent with the measured values, and the relative average error is less than 10%, and the structural similarity is higher than 50%. This shows that the digital twin model constructed can accurately predict the spatio-temporal evolution of the wear volume.

[0134] In summary, the wear surface severity evaluation method, system, medium and equipment of the present application have the following characteristics: (1) The present application establishes the relationship between the abrasive particle size distribution and the wear rate probability representation, which can represent the real-time wear rate through online abrasive particle size information, and realizes accurate modeling of the wear process.

[0135] (2) The application combines the contact mechanism and the wear morphology sample, constructs a physically constrained wear-vibration mapping, constitutes a fine mapping between the wear volume sequence and the vibration feature sequence, realizes the online accurate characterization of the spatial heterogeneity of the wear process, and significantly improves the fine granularity and accuracy of vibration analysis.

[0136] (2) The application combines the online abrasive particles and vibration monitoring information, constructs a digital twin model, efficiently and complementarily utilizes the two kinds of physical information, can realize the accurate prediction of the spatio-temporal evolution of the wear volume, and significantly improves the accuracy of online wear monitoring.

[0137] The above content only illustrates the technical idea of the application, and cannot limit the protection scope of the application. Any modification made according to the technical idea of the application on the basis of the technical scheme falls within the protection scope of the claims of the application.

Claims

1. A method of surface wear volume spatio-temporal distribution prediction in conjunction with abrasive particles and vibration monitoring, characterized by, The method comprises the following steps: A random wear characterization model driven by abrasive particle size distribution is constructed, and dynamic probability representation of time sequence evolution of wear volume of the friction pair is realized through joint probability representation and Markov process. A physical constraint double-branch wear-vibration mapping model is constructed, and fuzzy mapping between the surface wear volume and vibration time-frequency features is established, and accurate observation of spatial distribution of the surface wear volume of the friction pair is realized. Based on a state space modeling framework, a wear degradation digital twin model combined with abrasive particle and vibration monitoring is constructed, the abrasive particle size distribution-Markov process model constructed is taken as a state transition equation of the digital twin model, and the double-branch wear-vibration mapping model constructed is taken as a constraint condition of an observation equation of the digital twin model, and accurate space-time prediction of the surface wear volume of the friction pair is realized.

2. The method of claim 1, wherein, The random wear characterization model driven by abrasive particle size distribution is constructed in particular as follows: The equivalent diameter of the abrasive particle is extracted from an online ferrograph image as input, and the probability density distribution of the abrasive particle volume is estimated through diameter-thickness mapping. The probability density distribution of the abrasive particle volume is sampled multiple times, and the sampling results are accumulated to constitute dynamic representation of the surface wear rate of the friction pair. Based on the abrasive particle driven wear rate representation, the wear process is modeled as a discrete Markov process. Monte Carlo simulation is adopted to predict time sequence evolution of statistical characteristics of wear volumes of different regions of the surface of the friction pair from the constructed Markov process.

3. The method of claim 2, wherein, The dynamic representation of the surface wear rate of the friction pair is in particular as follows: wherein, is a discrete wear rate over time, is the number of abrasive particles generated in a local area, the volume of the abrasive particles of the nth subsample, is the probability density distribution of the volume of the abrasive particles.

4. The method of claim 2, wherein, The wear process is modeled as a discrete Markov process as follows: wherein, is the local wear volume at the time instant, is the discrete wear rate over the time.

5. The method of claim 1, wherein, The physical constraint double-branch wear-vibration mapping model is constructed in particular as follows: The signal segment corresponding to one rotation of the friction pair is intercepted from the real-time vibration signal , a time-frequency spectrum centroid frequency sequence is extracted , an observation index of a friction pair wear volume sequence is constituted The physical constraint wear-vibration mapping model is constructed to realize fuzzy mapping between the surface wear volume of the friction pair and the time-frequency spectrum centroid frequency of the vibration signal, and comprises a double-granularity prediction framework composed of a pressure prediction branch and an elastic strain prediction branch, and multiple soft / hard physical constraints. Pressure prediction branch An end-to-end nonlinear mapping between wear topography and pressure distribution is established using convolutional networks. Elastic strain prediction branch , involving multi-level mapping of wear volume-elastic approaching quantity-vibration time-frequency spectrum centroid frequency, wherein a Bayesian neural network is used to constitute a fuzzy mapping of wear volume-elastic approaching quantity , and measure its cognitive uncertainty, and a physical formula is used to model the mapping of elastic approaching quantity-vibration time-frequency spectrum centroid frequency ; The physical mapping of the elastic approach amount-vibration time-frequency spectrum centroid frequency in the elastic strain prediction branch. The hard constraint converts the stress balance condition into a hard constraint, which is used as a post-processing layer of the stress prediction branch. The soft constraint is constructed from the deformation coordination condition and is used for model parameter optimization.

6. The method of claim 5, wherein, The physical mapping of the elastic approach amount-vibration time-frequency spectrum centroid frequency is in particular as follows: wherein, is a predicted value of the vibration time-frequency spectrum centroid frequency; is a spring approach amount; is an applied load; is a friction component mass; is a stiffness index; is a scaling factor and a translation factor.

7. The combined abrasive particle and vibration monitored surface wear volume spatio-temporal distribution prediction method of claim 5, wherein, The hard constraint is in particular as follows: The soft constraint is in particular as follows: wherein, is a stress distribution matrix; is an initial gap matrix; is a height matrix of the wear topography; is an upper limit value of the elastic strain; is an equivalent elastic modulus of the friction pair; is a loss function based on physical constraints; is a compliance matrix; is an elastic approach quantity.

8. The method of claim 1, wherein, The wear degradation digital twin model combined with abrasive particle and vibration monitoring comprises a physical entity and a virtual entity, the physical entity relates to a basic friction pair, and online abrasive particle monitoring and vibration monitoring, the virtual entity is a state space model, the state transition equation is composed of the abrasive particle size distribution driven Markov process, and the observation equation is constructed from the physical constraint wear-vibration mapping model; the real-time abrasive particle size distribution and vibration features derived from the physical entity are used to update the wear rate and vibration observation values in the virtual entity online; the virtual entity derives the wear volume sequence prediction result to realize accurate monitoring of the actual friction pair; In the real-time prediction process, online parameter updates are introduced, updating the observation equation parameters at the start time of each running interval. By maximizing and Inter-correlation coefficient determines phase shift Combining the above data interaction and parameter updates, particle filtering is used to predict the wear volume sequence online. This enables the prediction of the spatiotemporal distribution of surface wear volume.

9. The method of claim 8, wherein, Phase offset Is: wherein is a cyclic phase shift process, i.e. is applied to the cyclic phase shift is a sequence of time-frequency centroid frequencies.​ 10. A system for spatio-temporal distribution prediction of surface wear volume in conjunction with abrasive particle and vibration monitoring, characterized by, It comprises: A probability module is constructed to construct a random wear characterization model driven by abrasive particle size distribution, and dynamic probability representation of time sequence evolution of wear volume of the friction pair is realized through joint probability representation and Markov process. The mapping module constructs a double-branch wear-vibration mapping model of physical constraints, establishes a fuzzy mapping between the surface wear volume and the vibration time-frequency characteristics, and realizes observation of the spatial distribution of the surface wear volume of the friction pair; The prediction module, based on a state space modeling framework, constructs a wear degradation digital twin model combined with abrasive particle and vibration monitoring, takes the abrasive particle size distribution-Markov process model constructed as a state transition equation of the digital twin model, takes the double-branch wear-vibration mapping model constructed as a constraint condition of the observation equation of the digital twin model, and realizes accurate spatiotemporal prediction of the surface wear volume of the friction pair.

Images