CN-122020138-A - Method, device, equipment, medium and product for predicting service life of rotary mechanical part

CN122020138ACN 122020138 ACN122020138 ACN 122020138ACN-122020138-A

Abstract

The application discloses a life prediction method, a device, equipment, a medium and a product of a rotary mechanical component, and relates to the technical field of mechanical equipment state monitoring and fault prediction, wherein the method comprises the steps of obtaining an operation monitoring signal of a target rotary mechanical component in a current time period and extracting preset key characteristics; and respectively inputting the characteristics into corresponding steady-state or unsteady-state working condition life prediction models according to the states, and outputting residual life prediction values. According to the application, computational redundancy is reduced by screening key features, working condition self-adaptive division is realized by utilizing multi-scale wavelet transformation, and a differentiated prediction model is constructed aiming at different health states, so that the accuracy and the robustness of life prediction of the rotating mechanical parts are remarkably improved under complex working conditions, and the lightweight and highly reliable intelligent maintenance decision support is realized.

Inventors

ZHU JUNWEI
Geng Ziheng
ZHU LINFANG
SHAN WEIJIE

Assignees

杭州市滨江区浙工大人工智能创新研究院

Dates

Publication Date: 20260512
Application Date: 20260121

Claims (10)

1. A lifetime prediction method of a rotary machine-like component, the lifetime prediction method comprising: Acquiring an operation monitoring signal of a target rotary mechanical part in a current time period, wherein the operation monitoring signal comprises a vibration signal, a temperature signal and a rotating speed signal; extracting preset key features based on the operation monitoring signals in the current time period; Based on preset key characteristics, carrying out working condition stability analysis by a multi-scale wavelet transformation method to obtain a current steady-state index of the target rotary mechanical part; determining the health state of the target rotary mechanical part in the current time period based on the current steady state index, wherein the health state comprises a normal state, a light fault state and a heavy fault state; If the health state is in a normal state, inputting preset key features into a trained steady-state working condition life prediction model to obtain a residual life prediction value of a target rotating mechanical component in a current time period, wherein the steady-state working condition life prediction model is obtained by carrying out iterative training on a preset first machine learning model by using a first training data set, and the first training data set is constructed based on the preset key features and corresponding residual life true values in a historical time period under the working condition of the normal state; And if the health state is a light fault state or a heavy fault state, a non-steady-state working condition life prediction model trained by preset key features is obtained to obtain a residual life prediction value of the target rotating machinery part in the current time period, wherein the non-steady-state working condition life prediction model is obtained by carrying out iterative training on a preset second machine learning model by using a second training data set, and the second training data set is constructed based on the preset key features and corresponding residual life true values in the historical time period under the light fault state and the heavy fault state.
2. The method for predicting the life of a rotating machine-like component according to claim 1, wherein the method for determining the preset key features specifically comprises: Acquiring operation monitoring signals and corresponding residual life true values of a plurality of rotary mechanical parts in a historical time period; Performing time domain and frequency domain feature calculation operation by a feature extraction method based on operation monitoring signals of a plurality of rotary mechanical parts in a historical time period to obtain a candidate feature set, wherein the candidate feature set comprises maximum values, average values, standard deviations, root mean square shape factors, skewness, kurtosis, crest factors, latitude factors, pulse factors, temperature values and rotating speed values of vibration signals in the vertical direction and the horizontal direction respectively; Calculating a first mutual information value between each candidate feature in the candidate feature set and the corresponding residual life true value and a second mutual information value between any two candidate features in the candidate feature set through a mutual information algorithm; And adopting a greedy search strategy, taking a maximized mRMR value as an optimization target, carrying out feature screening from the candidate feature set, and obtaining a preset key feature when a preset stopping condition is met, wherein the mRMR value is calculated based on a first mutual information value and a second mutual information value, the preset stopping condition comprises that the variation of the mRMR value reaches a first preset maximum iteration number or is smaller than a preset variation threshold value within a continuous first preset number, and the preset key feature comprises the standard deviation, the root mean square and the maximum value of a vibration signal in the vertical direction, the standard deviation, the root mean square and the crest factor of the vibration signal in the horizontal direction, a temperature value and a rotating speed value.
3. The method for predicting the service life of a rotating mechanical component according to claim 1, wherein the method for predicting the service life of the rotating mechanical component according to claim 1 is characterized in that the method for predicting the service life of the rotating mechanical component according to claim 1 comprises the steps of: Based on preset key features, multi-scale wavelet transformation is performed by the following formula: ; ; Wherein, the Representing a first order wavelet transform; Representing preset key characteristics at the time t; Representing a result of performing first-order wavelet transformation on preset key features at the moment t; representing a wavelet basis function at a t moment scale s; Representation of And (3) with Performing convolution operation; representing a second order wavelet transform; Representing a result of second-order wavelet transformation of preset key features at the time t; representing the square of the wavelet basis function at the time scale s; Representation of And (3) with Performing convolution operation; If it is The steady state index at time t of the target rotating machine-like component is 0, wherein, Representing a first preset discrimination threshold; If it is And (2) and The steady state index at time t of the target rotating machine-like component is 1, wherein, Representing a second preset discrimination threshold; If neither of the above conditions is satisfied, the steady state index at time t of the target rotating machinery component is calculated by the following formula: ; Wherein, the A steady state index indicating the time t; A smoothing function representing a t moment scale s; a weight coefficient representing a first order wavelet transform; a weight coefficient representing a second order wavelet transform; and summing and averaging the steady state indexes at all moments in the current time period to obtain the current steady state index.
4. A method of predicting the life of a rotating machine-like component according to claim 3, wherein determining the state of health of the target rotating machine-like component for a current time period based on the current steady state index comprises: If the current steady state index is greater than or equal to a first preset steady state threshold, the health state is a normal state; If the current steady state index is smaller than the first preset steady state threshold and larger than or equal to the second preset steady state threshold, the health state is a mild fault state; and if the current steady state index is smaller than the second preset steady state threshold value, the health state is a severe fault state.
5. The method for predicting the life of a rotating machine component according to claim 1, wherein the training process of the steady-state working condition life prediction model specifically comprises: Based on the first training data set, clustering each preset key feature through a fuzzy clustering algorithm to obtain a clustering center value of each preset key feature; Based on the clustering center value of each preset key feature, a membership function of a preset type is respectively constructed for each preset key feature, and the center parameter of the membership function of each preset key feature is initialized to be a corresponding clustering center value, wherein the preset type comprises a Gaussian type or a triangular type; based on a first training data set and a membership function initialized by each preset key feature, identifying the linear parameters of a preset first machine learning model by a least square method to obtain initial linear parameters of a steady-state working condition life prediction model; And carrying out iterative optimization on the central parameter of the membership function of each preset key feature and the linear parameter of the steady-state working condition life prediction model by taking the root mean square error between the residual life prediction value output by the minimized first machine learning model and the corresponding residual life true value as a target until the second preset maximum iteration number or the root mean square error variation in the continuous second preset number is smaller than the preset root mean square error variation threshold value, thereby obtaining the trained steady-state working condition life prediction model.
6. The method for predicting the life of a rotating machine component according to claim 1, wherein the training process of the unsteady state working condition life prediction model specifically comprises: based on the second training data set, clustering each preset key feature through a fuzzy clustering algorithm to obtain a clustering center value of each preset key feature; Based on the clustering central value of each preset key feature, respectively constructing an interval type II membership function for each preset key feature, and initializing the central parameter of the interval type II membership function of each preset key feature to be a corresponding clustering central value, wherein the interval type II membership function comprises an uncertainty interval formed by surrounding a lower membership function and an upper membership function; And carrying out global optimization on parameters of a second machine learning model by adopting a genetic algorithm based on a second training data set and a section two-type membership function initialized by each preset key feature until the third preset maximum iteration number or the change of an adaptability function value in the continuous third preset number is smaller than a preset adaptability function value change threshold value, so as to obtain a trained unsteady working condition life prediction model, wherein the adaptability function value is a negative value of root mean square error between a residual life prediction value output by the second machine learning model and a corresponding residual life true value, and the second machine learning model is a section two-type fuzzy system model.
7. A lifetime prediction device for a rotary machine component, wherein lifetime prediction of the rotary machine component is applied to the lifetime prediction method for a rotary machine component according to any one of claims 1 to 6, the lifetime prediction device for a rotary machine component comprising: the data acquisition module is used for acquiring operation monitoring signals of the target rotary mechanical component in the current time period, wherein the operation monitoring signals comprise vibration signals, temperature signals and rotating speed signals; the feature extraction module is used for extracting preset key features based on the operation monitoring signals in the current time period; The working condition analysis module is used for carrying out working condition stability analysis through a multi-scale wavelet transformation method based on preset key characteristics to obtain the current steady-state index of the target rotary mechanical component; The state judging module is used for determining the health state of the target rotating machinery part in the current time period based on the current steady state index, wherein the health state comprises a normal state, a light fault state and a heavy fault state; The steady state prediction module is used for inputting preset key features into a trained steady state working condition life prediction model to obtain a residual life prediction value of a target rotating mechanical component in a current time period if the health state is in a normal state, wherein the steady state working condition life prediction model is obtained by carrying out iterative training on a preset first machine learning model by using a first training data set, and the first training data set is constructed based on the preset key features and corresponding residual life true values in a historical time period under the normal state working condition; The unsteady state service life prediction module is used for obtaining a residual service life prediction value of the target rotating machinery part in the current time period by training an unsteady state service life prediction model with preset key features if the health state is in a light fault state or a heavy fault state, wherein the unsteady state service life prediction model is obtained by performing iterative training on a preset second machine learning model by using a second training data set, and the second training data set is constructed based on preset key features and corresponding residual service life true values in a historical time period in the light fault state and the heavy fault state.
8. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor executes the computer program to implement the method of life prediction of a rotating machinery-like component according to any of claims 1-6.
9. A computer-readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements a method for predicting the life of a rotating machine-like component according to any one of claims 1-6.
10. A computer program product comprising a computer program, characterized in that the computer program, when executed by a processor, implements a method for predicting the life of a rotating machine-like component according to any one of claims 1-6.

Description

Method, device, equipment, medium and product for predicting service life of rotary mechanical part Technical Field The application relates to the technical field of mechanical equipment state monitoring and fault prediction, in particular to a service life prediction method, a device, equipment, a medium and a product of a rotary mechanical part. Background The rolling and rotating mechanical component is used as a key component of mechanical equipment, is widely applied to the industrial field, and the performance of the rolling and rotating mechanical component directly influences the reliability and the operation efficiency of the equipment. The method for accurately predicting the residual service life of the rotary mechanical parts is a core technology for realizing predictive maintenance and avoiding unplanned shutdown. The service life prediction method of the rotating machinery part in the related art mainly depends on analysis of signals such as vibration, temperature and the like generated in the operation process of the rotating machinery part. However, under complex working conditions (such as load fluctuation, rotation speed change or noise interference), such as statistical models based on steady-state signal characteristics or prediction methods adopting a single fuzzy system (such as a Type-1 Fuzzy System,T1 FS) model), there are obvious disadvantages that, on one hand, high-dimensional characteristics extracted from original signals have a large amount of redundancy, the calculation load of the model is increased, the real-time requirement is difficult to meet, on the other hand, signal noise and uncertainty under the complex working conditions can significantly interfere with the prediction precision of the single model, and the model itself is difficult to adaptively process the data characteristic differences under different working conditions. Therefore, a life prediction method for a rotary mechanical component is needed, which can effectively process feature redundancy, adapt to working condition changes and properly cope with data uncertainty, so as to improve prediction accuracy and practicality and provide reliable support for intelligent maintenance decision. Disclosure of Invention The application aims to provide a life prediction method, a device, equipment, a medium and a product of a rotary mechanical part, which can effectively remove high-dimensional redundancy characteristics, automatically identify stable and unstable working conditions of the rotary mechanical part and model the working condition uncertainty, so that a high-precision and light residual life prediction result can be output under a complex running condition, and reliable support is provided for intelligent maintenance decision. In order to achieve the above object, the present application provides the following solutions: In a first aspect, the present application provides a method for predicting the life of a rotating machine-like component, comprising: Acquiring an operation monitoring signal of a target rotary mechanical part in a current time period, wherein the operation monitoring signal comprises a vibration signal, a temperature signal and a rotating speed signal; extracting preset key features based on the operation monitoring signals in the current time period; Based on preset key characteristics, carrying out working condition stability analysis by a multi-scale wavelet transformation method to obtain a current steady-state index of the target rotary mechanical part; determining the health state of the target rotary mechanical part in the current time period based on the current steady state index, wherein the health state comprises a normal state, a light fault state and a heavy fault state; If the health state is in a normal state, inputting preset key features into a trained steady-state working condition life prediction model to obtain a residual life prediction value of a target rotating mechanical component in a current time period, wherein the steady-state working condition life prediction model is obtained by carrying out iterative training on a preset first machine learning model by using a first training data set, and the first training data set is constructed based on the preset key features and corresponding residual life true values in a historical time period under the working condition of the normal state; And if the health state is a light fault state or a heavy fault state, a non-steady-state working condition life prediction model trained by preset key features is obtained to obtain a residual life prediction value of the target rotating machinery part in the current time period, wherein the non-steady-state working condition life prediction model is obtained by carrying out iterative training on a preset second machine learning model by using a second training data set, and the second training data set is constructed based on the preset key features and corresponding residual