Health assessment method, device and equipment for rotating part of vehicle running gear and medium

Abstract

The invention discloses a health assessment method, device and equipment for a rotating part of a vehicle running gear, and a medium, and relates to the technical field of vehicle health management, and the method comprises the steps: obtaining a target sensitive fault feature of a to-be-assessed rotating part in the vehicle running gear in a current time period; wherein the target sensitive fault feature is a feature associated with degradation of the rotating part to be evaluated; processing the spliced target sensitive fault features by using a target state level probability mapping model to obtain target probabilities of the to-be-evaluated rotating part under each degradation state level; wherein the target state level probability mapping model is a deep neural network model constructed based on a deep belief network; and quantifying the degradation state of the to-be-evaluated rotating part according to each target probability to obtain a health index of the to-be-evaluated rotating part. According to the scheme, the degradation state of the rotating part of the vehicle running gear can be accurately evaluated.

Description

Technical Field

[0001] This invention relates to the field of vehicle health management technology, and in particular to a method, apparatus, equipment and medium for health assessment of rotating components of a vehicle's running gear. Background Technology

[0002] In industrial production and equipment operation and maintenance scenarios, the stable operation of equipment is crucial for ensuring production continuity and reducing operating costs. With the increasing intelligence of equipment, equipment failures occur frequently under complex operating conditions. Once a failure occurs, it can not only cause production stoppages and economic losses, but also potentially lead to safety risks. Therefore, it is necessary to understand the equipment status in advance through health assessments, realize predictive maintenance, and prevent failures. Thus, conducting equipment health assessments is of great significance.

[0003] Currently, there are two typical approaches to health index assessment: one is based on a single feature, relying solely on a threshold value of a certain operating parameter of the equipment to define its health status. This approach is insufficient to fully reflect the complex degradation process of the equipment and is prone to assessment bias due to one-sided features. The other approach attempts to integrate multiple features, but often uses simple weighting or empirical formulas, failing to fully explore the nonlinear correlations between data and lacking a precise characterization of the probability of different degradation status levels. This makes it unable to adapt to the dynamic changes in the health status of equipment under complex operating conditions, resulting in insufficient accuracy and reliability of the assessment results, and failing to meet the refined requirements of intelligent operation and maintenance for equipment health management.

[0004] In summary, accurately assessing the degradation status of rotating components in a vehicle's running gear is a problem that needs to be solved in this field. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a method, apparatus, device, and medium for health assessment of rotating components of a vehicle's running gear, accurately assessing the degradation state of these components. The specific solution is as follows:

[0006] In a first aspect, this application discloses a health assessment method for rotating components of a vehicle's running gear, comprising:

[0007] Acquire the target sensitive fault features of the rotating component to be evaluated in the vehicle running gear during the current time period; wherein, the target sensitive fault features are features that are associated with the degradation of the rotating component to be evaluated;

[0008] The spliced ​​target sensitive fault features are processed using a target state level probability mapping model to obtain the target probability of the rotating component to be evaluated at each degradation state level; wherein, the target state level probability mapping model is a deep neural network model built based on a deep belief network.

[0009] The degradation state of the rotating component to be evaluated is quantified according to each of the target probabilities to obtain the health index of the rotating component to be evaluated.

[0010] Optionally, acquiring the target sensitive fault characteristics of the rotating component to be evaluated in the vehicle's running gear during the current time period includes:

[0011] Acquire monitoring data for the rotating components to be evaluated in the vehicle's running gear during the current time period;

[0012] Based on the correlation between each monitoring data and the degradation state of the rotating component to be evaluated, target sensitive fault features are screened from each monitoring data.

[0013] Optionally, the step of filtering target sensitive fault features from the monitoring data based on the correlation between each monitoring data and the degradation state of the rotating component to be evaluated includes:

[0014] Obtain degradation-related information of the rotating component to be evaluated; wherein, the degradation-related information includes fault type and historical fault case information;

[0015] Determine the correlation and monotonicity between each of the monitoring data and the degradation-related information;

[0016] Target sensitive fault features with a correlation greater than a first preset threshold and / or a monotonicity greater than a second preset threshold are selected from the monitoring data.

[0017] Optionally, the formula for calculating the correlation is:

[0018] ;

[0019] in, For normalized monitoring data, Where N is the time period and N is the number of monitoring data points.

[0020] The formula for calculating the monotonicity is:

[0021] ;

[0022] in, The data represents the normalized monitoring data, where N is the number of different types of monitoring data. It is a unit step function.

[0023] Optionally, obtain the target state level probability mapping model, including:

[0024] Constructing an initial state level probability mapping model based on deep belief networks;

[0025] The collected training set is divided into batches of training data; wherein, the training set includes full life cycle monitoring data of each rotating component in the vehicle running gear, and the full life cycle monitoring data includes degradation status level data;

[0026] The initial state level probability mapping model is trained using the training data from each batch to obtain the model parameters and model performance index corresponding to each batch.

[0027] Based on the performance index of each model, target model parameters are selected from the model parameters to obtain a target state level probability mapping model based on the target model parameters.

[0028] Optionally, the initial state level probability mapping model includes multiple restricted Boltzmann machines, each of which includes a visible layer and a hidden layer, and the activation function of the hidden layer of the last layer is a sigmoid function. The model parameters include weights, visible layer bias, and hidden layer bias.

[0029] Optionally, dividing the collected training set into batches of training data includes:

[0030] Collect full lifecycle monitoring data for each rotating component in the vehicle's running gear; wherein, the full lifecycle monitoring data refers to historical monitoring data for each historical sampling time period;

[0031] The full lifecycle monitoring data is preprocessed to obtain processed historical data;

[0032] The processed historical data is spliced ​​together along the time dimension to obtain the spliced ​​data corresponding to each historical sampling time period;

[0033] The spliced ​​data is determined as training samples, and a dataset including each training sample is constructed.

[0034] The dataset is divided into batches of training data.

[0035] Optionally, training the initial state level probability mapping model using the training data from each batch to obtain the model parameters and model performance index corresponding to each batch includes:

[0036] The training data for the current batch is determined from the training data from each batch.

[0037] The initial state level probability mapping model is trained using the current batch of training data to obtain the model parameters corresponding to the current batch of training data;

[0038] Based on the model parameters corresponding to the current batch of training data, obtain the current state level probability mapping model, and obtain the model performance index of the current state level probability mapping model.

[0039] Optionally, the model performance index is the average loss value;

[0040] Accordingly, the step of selecting target model parameters from the model parameters based on the model performance indices includes:

[0041] Select the batch with the smallest average loss value from all batches as the first target batch;

[0042] The model parameters of the first target batch are determined as the target model parameters.

[0043] Optionally, the model performance index includes the average loss value and the average accuracy.

[0044] Accordingly, the step of selecting target model parameters from the model parameters based on the model performance indices includes:

[0045] The second target batch is determined from each batch based on the average loss value and the average accuracy.

[0046] The model parameters of the second target batch are determined as the target model parameters.

[0047] Optionally, obtaining the average loss value of the current state level probability mapping model includes:

[0048] The current batch of training data is input into the current state level probability mapping model so that the current state level probability mapping model outputs the predicted probability of each training sample in the current batch of training data under different degradation state levels.

[0049] Obtain the actual probability of each training sample in the degradation state level data under different degradation state levels;

[0050] The loss value between the predicted probability and the actual probability is quantified using the cross-entropy loss function to obtain the loss value of each training sample in the current batch of training data, and the average loss value of each loss value is obtained.

[0051] Optionally, obtaining the average accuracy of the current state level probability mapping model includes:

[0052] The current batch of training data is input into the current state level probability mapping model so that the current state level probability mapping model outputs the predicted probability of each training sample in the current batch of training data under different degradation state levels.

[0053] Obtain the actual probability of each training sample in the degradation state level data under different degradation state levels;

[0054] The accuracy of each training sample in the current batch of training data is determined based on the predicted probability and the actual probability in the degradation state level data, and the average accuracy of each accuracy is obtained.

[0055] Optionally, quantifying the degradation state of the rotating component to be evaluated based on each of the target probabilities to obtain a health index of the rotating component to be evaluated includes:

[0056] Obtain the weighted summation of the probabilities of each target; wherein the weighted summation represents the comprehensive index of the current deviation of the rotating component to be evaluated from its healthy state;

[0057] Invert the weighted summation result to obtain the current health tendency index of the rotating component to be evaluated;

[0058] If the health tendency index is greater than the preset health baseline value, then the health tendency index is determined as the health index of the rotating component to be evaluated.

[0059] If the health tendency index is not greater than the preset health baseline value, then the preset health baseline value is determined as the health index of the rotating component to be evaluated.

[0060] Secondly, this application discloses a health assessment device for a rotating component of a vehicle's running gear, comprising:

[0061] The fault feature acquisition module is used to acquire the target sensitive fault features of the rotating component to be evaluated in the vehicle running gear within the current time period; wherein, the target sensitive fault features are features that are associated with the degradation of the rotating component to be evaluated;

[0062] The level probability mapping module is used to process the spliced ​​target sensitive fault features using the target state level probability mapping model to obtain the target probability of the rotating component to be evaluated under each degradation state level; wherein, the target state level probability mapping model is a deep neural network model built based on a deep belief network.

[0063] The health index acquisition module is used to quantify the degradation state of the rotating component to be evaluated based on each of the target probabilities, so as to obtain the health index of the rotating component to be evaluated.

[0064] Thirdly, this application discloses an electronic device, including:

[0065] Memory, used to store computer programs;

[0066] A processor is used to execute the computer program to implement the steps of the aforementioned disclosed method for health assessment of rotating components of a vehicle running gear.

[0067] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the aforementioned disclosed method for health assessment of rotating components of a vehicle running gear.

[0068] The beneficial effects of this application are as follows: It obtains the target sensitive fault features of the rotating component to be evaluated in the vehicle's running gear within the current time period; wherein the target sensitive fault features are features associated with the degradation of the rotating component to be evaluated; it processes the stitched target sensitive fault features using a target state level probability mapping model to obtain the target probability of the rotating component to be evaluated at each degradation state level; wherein the target state level probability mapping model is a deep neural network model constructed based on a deep belief network; and it quantifies the degradation state of the rotating component to be evaluated according to each target probability to obtain the health index of the rotating component to be evaluated. Therefore, this application, by acquiring target-sensitive fault features related to the degradation of the rotating component to be evaluated, can avoid interference from irrelevant or weakly correlated features and improve the ability to represent the true degradation state of the component. By using a target state level probability mapping model based on a deep belief network to process the spliced ​​target-sensitive fault features, it can automatically learn the complex mapping relationship between fault features and degradation states, leveraging the powerful nonlinear feature extraction capabilities of the deep belief network. This eliminates the need for manual experience in designing feature combinations, thereby enhancing the model's adaptability and generalization ability to complex operating conditions. Simultaneously, it outputs the target probability at each degradation state level, which is significantly higher than traditional methods. The assessment method can more meticulously depict the complete degradation process of components from health to failure, providing accurate decision-making basis for predictive maintenance. Furthermore, a health index is obtained by quantifying the probability of each target, transforming discrete degradation state level probabilities into continuous quantitative indicators. This facilitates real-time monitoring of component degradation trends and predicts failure risks through the health index, realizing a shift from passive maintenance to proactive prevention. Moreover, by splicing features, multi-source monitoring data can be flexibly integrated to adapt to the health assessment needs of different types of rotating components, exhibiting strong versatility and scalability. Ultimately, this provides technical support for the operation and maintenance of rotating components in vehicle running gear, improving operation and maintenance efficiency and safety. Attached Figure Description

[0069] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0070] Figure 1 This application discloses a flowchart of a health assessment method for rotating components of a vehicle running gear.

[0071] Figure 2 This is a schematic diagram of a specific neural network model structure disclosed in this application;

[0072] Figure 3 This is a schematic diagram of a specific model training disclosed in this application;

[0073] Figure 4 This is a schematic diagram illustrating the development of a specific health index as disclosed in this application;

[0074] Figure 5 This is a schematic diagram of the structure of a health assessment device for a rotating component of a vehicle running gear disclosed in this application;

[0075] Figure 6 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0076] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0077] In industrial production and equipment operation and maintenance scenarios, the stable operation of equipment is crucial for ensuring production continuity and reducing operating costs. With the increasing intelligence of equipment, equipment failures occur frequently under complex operating conditions. Once a failure occurs, it can not only cause production stoppages and economic losses, but also potentially lead to safety risks. Therefore, it is necessary to understand the equipment status in advance through health assessments, realize predictive maintenance, and prevent failures. Thus, conducting equipment health assessments is of great significance.

[0078] Currently, there are two typical approaches to health index assessment: one is based on a single feature, relying solely on a threshold value of a certain operating parameter of the equipment to define its health status. This approach is insufficient to fully reflect the complex degradation process of the equipment and is prone to assessment bias due to one-sided features. The other approach attempts to integrate multiple features, but often uses simple weighting or empirical formulas, failing to fully explore the nonlinear correlations between data and lacking a precise characterization of the probability of different degradation status levels. This makes it unable to adapt to the dynamic changes in the health status of equipment under complex operating conditions, resulting in insufficient accuracy and reliability of the assessment results, and failing to meet the refined requirements of intelligent operation and maintenance for equipment health management.

[0079] Therefore, this application provides a health assessment scheme for rotating components of a vehicle running gear, which accurately assesses the degradation status of the rotating components of the vehicle running gear.

[0080] See Figure 1 As shown in the embodiment of this application, a health assessment method for a rotating component of a vehicle's running gear is disclosed, including:

[0081] Step S11: Obtain the target sensitive fault features of the rotating component to be evaluated in the vehicle running gear during the current time period; wherein, the target sensitive fault features are features that are associated with the degradation of the rotating component to be evaluated.

[0082] In this embodiment, obtaining the target sensitive fault characteristics of the rotating component to be evaluated in the vehicle running gear within the current time period includes: obtaining various monitoring data of the rotating component to be evaluated in the vehicle running gear within the current time period; and filtering out the target sensitive fault characteristics from the monitoring data based on the correlation between each monitoring data and the degradation state of the rotating component to be evaluated.

[0083] First, the original monitoring data of the rotating components to be evaluated in the vehicle's running gear are obtained from the ground system within the current time period. The ground system refers to the operation and maintenance management system, which stores data primarily including component conclusion data, historical data, maintenance system data, and real-time monitoring data of components issued by the onboard system. Basic information on online / disassembled rotating components stored in the ground system, along with the original monitoring data for the current time period, is also obtained. For example, the original monitoring data for the most recent week can be used, with the time period set according to the specific application scenario. The categories of original monitoring data include alarm trend data, SV trend data, db trend data, vibration trend data, temperature trend data, out-of-roundness runout trend data, and health assessment data. The vehicle running gear includes multiple rotating components. When the rotating component to be evaluated is an axle box bearing, the specific original monitoring data is shown in the table below:

[0084] Table 1 Monitoring Data Table

[0085]

[0086] It should be noted that due to hardware or operating condition abnormalities or other reasons, the original monitoring data may contain missing, duplicate, or abnormal phenomena, so data repair is required.

[0087] For handling missing values: There are roughly three reasons for missing data. The first is insufficient in-vehicle storage space, leading to data loss due to a limited data retention period. The second is data loss due to in-vehicle program shutdown / startup or malfunction. The third is data loss due to data packet loss. Based on these causes, three measures are implemented: First, for missing values ​​caused by data retention strategies, a normal distribution is fitted to the data, and the missing time is used to generate the missing data. Second, for data loss due to in-vehicle program malfunctions, which often results in fewer missing values, mean imputation is used to fill in the missing data. Third, for missing values ​​caused by data packet loss, which often involves longer periods of missing data, the trend of data before and after the loss is fitted using nonlinear equations, and the resulting nonlinear function is used for data filling.

[0088] Regarding the handling of duplicate values: The original monitoring data of the vehicle-mounted system is stored in the form of data packets. To store this data on the ground system, the common method is to import the data packets, parse their contents, and then store them. Therefore, most duplicate data obtained from the ground system is due to duplicate imports of data packets. Therefore, the solution for handling duplicate data is to directly delete the duplicate values.

[0089] For handling outliers: Outliers generally arise from two causes: first, data storage errors due to program reliability issues during storage; second, errors in the vehicle's computing program leading to incorrect data storage. Therefore, there are two handling measures for outlier data: first, data exceeding thresholds are directly removed, such as dB data outside the range [0, 100dB], temperatures outside the range [-125°, 125°], and alarm data outside the range [0, 1, 2]. Second, short-term data is used for mean / mode calculations to fill in the gaps, primarily for dB and temperature data.

[0090] After repairing the original monitoring data, it is necessary to normalize the repaired monitoring data. This involves normalizing each type of data after repair, one by one, and performing maximum and minimum normalization on each structured data type to obtain the normalized monitoring data. The normalization formula is as follows:

[0091] ;

[0092] In the formula, i represents the i-th type of monitoring data. This indicates that at time j, This represents the normalized index value of the i-th monitored data type at time j. , These are the maximum and minimum values ​​of the i-th monitored data type, respectively.

[0093] Next, based on the correlation between the normalized monitoring data and the degradation state of the rotating component to be evaluated, target sensitive fault features are screened from each monitoring data. It is understandable that some features in the monitoring data can reflect the degradation state of the rotating component, while others are not related to the degradation state; however, the sensitive fault features may differ for different rotating components.

[0094] In this embodiment, the step of filtering target sensitive fault features from the monitoring data based on the correlation between each monitoring data and the degradation state of the rotating component to be evaluated includes: obtaining degradation-related information of the rotating component to be evaluated; wherein, the degradation-related information includes fault type and historical fault case information; determining the correlation and monotonicity between each monitoring data and the degradation-related information; and filtering target sensitive fault features from the monitoring data whose correlation is greater than a first preset threshold and / or whose monotonicity is greater than a second preset threshold.

[0095] Obtain degradation-related information for the rotating component to be evaluated, specifically the failure types and historical failure case information of the rotating component. This includes identifying the types of failures present in the rotating component and the failure cases that have occurred previously. Analyze the correlation between each monitoring data point and the failure types and historical failure case information; that is, determine the relationship between different monitoring data points and different failure types, and whether different monitoring data points can produce different failures and varying degrees of failure. Therefore, determine the correlation and monotonicity between each monitoring data point and the degradation-related information. The correlation index reflects the relationship between degradation characteristics and the actual degradation state of the component, while the monotonicity index reflects the consistency between the characteristics and the component's performance degradation state. In other words, correlation, monotonicity, and the degree of degradation are directly proportional. Therefore, by setting a first preset threshold and a second preset threshold, target sensitive fault features with correlation greater than the first preset threshold, or target sensitive fault features with monotonicity greater than the second preset threshold, or target sensitive fault features with both correlation greater than the first preset threshold and monotonicity greater than the second preset threshold, can be accurately located. These features can more effectively reflect the fault evolution and degradation process of the component and reduce the influence of irrelevant or interfering features on the model.

[0096] In this embodiment, the formula for calculating the correlation is:

[0097] ;

[0098] in, For normalized monitoring data, Where N is the time period and N is the number of monitoring data points.

[0099] The formula for calculating the monotonicity is:

[0100] ;

[0101] in, The data represents the normalized monitoring data, where N is the number of different types of monitoring data. It is a unit step function.

[0102] The correlation index is calculated using the correlation formula. The correlation value is in the range of [0,1]. The closer the value is to 1, the stronger the connection between the feature and the degradation state of the component, indicating that the feature can better reflect the degradation state of the component. The monotonicity index is calculated using the monotonicity formula. The monotonicity value is in the range of [0,1]. The closer the value is to 1, the better the monotonicity trend of the feature, and the better it can measure the degradation state of the component.

[0103] Furthermore, the selected target sensitive fault features are concatenated over time to obtain the concatenated target sensitive fault features. For the acquired data of a specific axle box bearing to be evaluated, full lifecycle feature data is concatenated over time to form dynamic evaluation data. This data consists of multiple independent variables and one dependent variable (health level) for a specific axle box bearing. Specifically, a set of independent variable data and one health level data are acquired daily according to the actual operating dates of the vehicle. The target sensitive fault features acquired in the current time period belong to different sub-time periods; for example, the current time period is the last three days, with each day being a sub-time period.

[0104] Step S12: Process the spliced ​​target sensitive fault features using the target state level probability mapping model to obtain the target probability of the rotating component to be evaluated under each degradation state level; wherein, the target state level probability mapping model is a deep neural network model built based on a deep belief network.

[0105] The spliced ​​target sensitive fault features are input into the target state level probability mapping model. The target state level probability mapping model processes the spliced ​​target sensitive fault features to predict the target probability of the rotating component to be evaluated under each degradation state level. There are a total of five degradation state levels: normal state level, sub-healthy state level, minor fault state level, moderate fault state level, and severe fault state level. The output of the target state level probability mapping model can be the target probability corresponding to these five levels, or the target probability corresponding to some of the levels.

[0106] In this embodiment, obtaining the target state level probability mapping model includes: constructing an initial state level probability mapping model based on a deep belief network; dividing the collected training set into batches of training data; wherein the training set includes full life cycle monitoring data of each rotating component in the vehicle running gear, and the full life cycle monitoring data includes degradation state level data; training the initial state level probability mapping model using the training data of each batch to obtain the model parameters and model performance index corresponding to each batch; selecting target model parameters from the model parameters based on the model performance index, so as to obtain the target state level probability mapping model based on the target model parameters.

[0107] Collect full lifecycle monitoring data of each rotating component in the vehicle's running gear over a historical period, including degradation status level data. Understandably, when collecting this data, it is also necessary to perform data repair and normalization on these raw data to obtain data that can be used as a training set. Then, the training set is divided into multiple batches of training data, for example, batch_size=64 to divide the data into n batches, and then each batch of training data is used to train the model.

[0108] An initial state level probability mapping model is constructed based on a Deep Belief Network (DBN). The model is trained sequentially using batches of training data, for example, a total of 1000 batches. First, the model is trained using the first batch of training data to optimize its parameters, resulting in the model parameters and performance index for the first batch. Then, the model trained using the second batch of training data is trained again, yielding the model parameters and performance index for the second batch. This process continues until all batches of training data have been used to complete the model training. This process yields the model parameters and performance index for each batch. The model performance index characterizes the effectiveness of the model's level probability mapping. Therefore, based on the model performance index, target model parameters are selected from the model parameters to achieve the best level probability mapping effect, thus obtaining the target state level probability mapping model.

[0109] In this embodiment, the initial state level probability mapping model includes multiple restricted Boltzmann machines, each of which includes a visible layer and a hidden layer, and the activation function of the hidden layer of the last layer is a sigmoid function. The model parameters include weights, visible layer bias, and hidden layer bias.

[0110] For example Figure 2 The diagram illustrates a specific neural network model structure. The state-level probability mapping model includes multiple Restricted Boltzmann Machines (RBMs). In one specific case, the model includes 5 network layers and 4 sets of RBMs. Each RBM includes one visible layer and one hidden layer. The activation function of the hidden layer in the last layer is the sigmoid function. For example, the dimensions of each network layer are 32, 8, 4, 2, and 5, the sampling mode parameter is set to "bernoulli", the Gibbs sampling step parameter is 5, the number of visible layers is [32, 8, 4, 2], and the number of hidden layers is [8, 4, 2, 5]. That is, the first set of visible layers... The Restricted Boltzmann Machine (RBM1) consists of 32 visible layers and 8 hidden layers; the second group of Restricted Boltzmann Machines (RBM2) consists of 8 visible layers and 4 hidden layers; the third group of Restricted Boltzmann Machines (RBM3) consists of 4 visible layers and 2 hidden layers; and the fourth group of Restricted Boltzmann Machines (RBM4) consists of 2 visible layers and 5 hidden layers. The learning rate can be set to 0.0005, the number of iterations to 100, the batch size to 128, the number of early stop training rounds to 10, and the optimizer type to adam. The model training is completed using multiple full-lifecycle datasets to obtain the weights, hidden layer biases, and visible layer biases.

[0111] In this embodiment, dividing the collected training set into batches of training data includes: collecting full lifecycle monitoring data of each rotating component in the vehicle's running gear; wherein the full lifecycle monitoring data is historical monitoring data under each historical sampling time period; preprocessing the full lifecycle monitoring data to obtain processed historical data; splicing the processed historical data along the time dimension to obtain spliced ​​data corresponding to each historical sampling time period; determining the spliced ​​data as training samples and constructing a dataset including each training sample; and dividing the dataset into batches of training data.

[0112] Understandably, in the process of acquiring training data, the first step is to collect full lifecycle monitoring data for each rotating component in the vehicle's running gear, including historical monitoring data for each historical sampling period. Next, the full lifecycle monitoring data is preprocessed to obtain processed historical data. Preprocessing includes missing value imputation, duplicate value removal, outlier repair, normalization, and feature extraction. Feature extraction involves selecting sensitive fault features from each monitoring data point, determining the correlation and monotonicity between each historical monitoring data point and degradation-related information, and then selecting data points from each historical monitoring data point whose correlation is greater than a first preset threshold and / or whose monotonicity is greater than... The second preset threshold is used to identify sensitive fault characteristics. Then, the processed historical data is spliced ​​along the time dimension to obtain the spliced ​​data corresponding to each historical sampling time period. Based on the extracted sensitive fault feature indicators, the feature data of the entire life cycle of multiple disassembled rotating parts are spliced ​​along the time dimension to form model training data, including multiple independent variable data and one health level dependent variable data. That is, according to the actual operating date of the vehicle, a set of independent variable data and one health level data are obtained every day. In this way, the spliced ​​data is determined as training samples, and these training samples constitute the dataset. Finally, the dataset is divided into batches of training data according to the preset batch size.

[0113] In this embodiment, training the initial state level probability mapping model using the training data from each batch to obtain the model parameters and model performance index corresponding to each batch includes: determining the current batch of training data from the training data from each batch; training the initial state level probability mapping model using the current batch of training data to obtain the model parameters corresponding to the current batch of training data; obtaining the current state level probability mapping model based on the model parameters corresponding to the current batch of training data, and obtaining the model performance index of the current state level probability mapping model.

[0114] The following explanation uses the first batch of training data from each batch as an example to illustrate model training. For example... Figure 3The diagram illustrates a specific model training process. The current batch of training data is determined from each batch of training data. This current batch of training data is used to train the initial state level probability mapping model. This batch of data is then transformed into a prediction output through the network's hierarchical structure. The activation function of the last hidden layer maps the output data to probability values ​​between 0 and 1 (i.e., the probability value under each degradation state level). This yields the prediction results and model parameters for this batch of data, thus obtaining the model parameters corresponding to the current batch of training data. Then, based on the model parameters corresponding to the current batch of training data, the current state level probability mapping model is obtained, and the model performance index of the current state level probability mapping model is acquired. The model performance index is mainly determined by the difference between the model's prediction results and the actual results.

[0115] In the first specific embodiment, the model performance index is the average loss value. Correspondingly, target model parameters are selected from each model parameter based on the model performance index, including: selecting the batch with the smallest average loss value from each batch as the first target batch; and determining the model parameters of the first target batch as the target model parameters. When the model performance index is the average loss value, the loss value of each training data in each batch is first calculated to obtain the average loss value of each batch. The smaller the loss value, the better the model performance. Therefore, the batch with the smallest average loss value in each batch is selected as the first target batch, and the model parameters of the first target batch are then determined as the target model parameters, ensuring model performance.

[0116] In a second specific embodiment, the model performance index includes the average loss value and the average accuracy; correspondingly, the step of selecting target model parameters from the model parameters based on each of the model performance indices includes: determining a second target batch from each batch based on the average loss value and the average accuracy; and determining the model parameters of the second target batch as the target model parameters. To select model parameters with better performance, both the average loss and average accuracy can be used as model performance indices. This involves combining the average loss and average accuracy to select a second target batch. In the first selection scenario, candidate batches are first selected based on the average loss, meaning the candidate batches are a predetermined number of batches with the smallest average loss. Since the average loss values ​​of the candidate batches are very similar, the second target batch can be selected from these candidate batches based on the average accuracy; that is, the second target batch is the batch with the highest average accuracy among all candidate batches. In the second selection scenario, the average accuracy of each batch is inverted to obtain the average error rate of each batch. Selection weights are set for the average loss and average error rate, and the weighted sum of the average loss and average error rate for each batch is calculated. The batch with the smallest weighted sum is determined as the second target batch. After determining the second target batch, its model parameters are set as the target model parameters. This results in a model with a lower loss rate and higher accuracy.

[0117] In this embodiment, obtaining the average loss value of the current state level probability mapping model includes: inputting the current batch of training data into the current state level probability mapping model so that the current state level probability mapping model outputs the predicted probability of each training sample in the current batch of training data under different degradation state levels; obtaining the actual probability of each training sample in the degradation state level data under different degradation state levels; quantifying the loss value between the predicted probability and the actual probability using the cross-entropy loss function to obtain the loss value of each training sample in the current batch of training data, and obtaining the average loss value of each loss value.

[0118] Taking the current batch of training data as an example, the current batch of training data is first input into the current state level probability mapping model. The current state level probability mapping model outputs the predicted probability of each training sample in the current batch of training data under different degradation state levels, and obtains the actual probability of each training sample in the degradation state level data under different degradation state levels. Then, the cross-entropy loss function is used to quantify the loss value between the predicted probability and the actual probability, that is, to obtain the loss value of each training sample in the current batch. Then, the loss is averaged for each batch of data, that is, the cumulative loss rate in each batch of data is calculated, and the quotient of the cumulative loss rate and the current batch size is used as the average loss value.

[0119] Obtaining the average accuracy of the current state level probability mapping model includes: inputting the current batch of training data into the current state level probability mapping model so that the current state level probability mapping model outputs the predicted probability of each training sample in the current batch of training data under different degradation state levels; obtaining the actual probability of each training sample in the degradation state level data under different degradation state levels; determining the accuracy of each training sample in the current batch of training data based on the predicted probability and the actual probability in the degradation state level data, and obtaining the average accuracy of each accuracy.

[0120] Taking the current batch of training data as an example, the current batch of training data is first input into the current state level probability mapping model. The current state level probability mapping model outputs the predicted probability of each training sample in the current batch of training data under different degradation state levels, and obtains the actual probability of each training sample in the degradation state level data under different degradation state levels. Next, based on the predicted probability and the actual probability in the degradation state level data, the accuracy of each training sample in the current batch of training data is determined, that is, the deviation between the predicted result and the actual result is compared, the cumulative accuracy in each batch of data is calculated, and then the quotient of the cumulative accuracy and the current batch size is used as the average accuracy.

[0121] It is important to note that after inputting the current batch of training data into the current state level probability mapping model, the model will output the predicted probability of each training sample under different degradation state levels. Taking one training sample as an example, the output prediction result is the predicted probability of the training sample under the normal state level, sub-healthy state level, minor fault state level, moderate fault state level, and severe fault state level, respectively. Specifically, the normal state level, sub-healthy state level, minor fault state level, and moderate fault state level can be represented by numbers, such as 0, 1, 2, 3, and 4 representing the normal state level, sub-healthy state level, minor fault state level, and moderate fault state level, respectively. Then the output prediction result is the number corresponding to the degradation state level and the predicted probability corresponding to that number. Similarly, the actual probability under different degradation state levels can also be in the form of a combination of numbers and actual probabilities. This number is used to represent the normal state level, sub-healthy state level, minor fault state level, and moderate fault state level.

[0122] Step S13: Quantify the degradation state of the rotating component to be evaluated according to each of the target probabilities to obtain the health index of the rotating component to be evaluated.

[0123] In this embodiment, quantifying the degradation state of the rotating component to be evaluated based on each target probability to obtain a health index of the rotating component to be evaluated includes: obtaining a weighted summation result of each target probability; wherein the weighted summation result represents a comprehensive index of the rotating component to be evaluated's current deviation from a healthy state; inverting the weighted summation result to obtain a current health tendency index of the rotating component to be evaluated; if the health tendency index is greater than a preset health baseline value, then the health tendency index is determined as the health index of the rotating component to be evaluated; if the health tendency index is not greater than the preset health baseline value, then the preset health baseline value is determined as the health index of the rotating component to be evaluated.

[0124] Obtain the weighted sum of the probabilities of each target; this weighted sum represents the comprehensive index indicating the current deviation of the rotating component from its healthy state. Invert the weighted sum to obtain the current health tendency index of the rotating component. However, the current health tendency index may be less than 0. Therefore, if the health tendency index is greater than the preset health baseline value of 0, the health tendency index is determined as the health index of the rotating component; if the health tendency index is not greater than the preset health baseline value of 0, the preset health baseline value of 0 is determined as the health index of the rotating component. Specifically, the formula for obtaining the health index is as follows:

[0125] ;

[0126] In the formula, The current health index of the rotating component to be evaluated. These represent the target probabilities of the sub-health state level, minor fault state level, moderate fault state level, and severe fault state level output by the model, respectively.

[0127] Understandably, when a rotating component is in a normal state, the probability of other fault levels is very low, and the health index naturally reflects its good condition. Therefore, there is no need to include the normal state health level in the formula. This design can concisely and effectively reflect the degree of component degradation through the comprehensive calculation of fault-related probabilities, realizing the quantitative representation of the component's degradation process throughout its entire life cycle through the health index. In other words, by obtaining the weighted sum of the probabilities of each target, the comprehensive index representing the current deviation of the rotating component from its healthy state is represented. Then, the inverse of this index is used to obtain the health tendency index. Combined with the preset health baseline value, the health index is determined. This can reasonably quantify the degree of component deviation from its healthy state and accurately reflect its health tendency. At the same time, by limiting the preset health baseline value, the health index can be prevented from deviating excessively from the actual healthy state, ensuring that the health index can more accurately and reliably reflect the current health status of the component. This provides more effective data support for component condition monitoring and maintenance strategy formulation, improving the rationality and practicality of the assessment of the component degradation process.

[0128] Health indices are primarily used to quantify and identify the degradation process of a system, serving as a crucial step in predicting remaining useful life (UPS) and significantly impacting the accuracy of UPS predictions. Health indices are typically continuous functions within the range [0, 1]. They can be divided into different health intervals to represent the health level of equipment, thereby evaluating the current system operating status. Different health levels correspond to different maintenance strategies. Dynamically adjusting maintenance plans and strategies based on health index results maximizes maintenance efficiency and reduces costs, achieving optimal condition-based maintenance and ensuring system safety. Figure 4 The diagram illustrates a specific health index development. As the usage time of rotating components increases, the health index of the rotating components gradually decreases. The lower the health index, the worse the operating condition of the rotating components, that is, the higher the degree of degradation.

[0129] The beneficial effects of this application are as follows: It obtains the target sensitive fault features of the rotating component to be evaluated in the vehicle's running gear within the current time period; wherein the target sensitive fault features are features associated with the degradation of the rotating component to be evaluated; it processes the stitched target sensitive fault features using a target state level probability mapping model to obtain the target probability of the rotating component to be evaluated at each degradation state level; wherein the target state level probability mapping model is a deep neural network model constructed based on a deep belief network; and it quantifies the degradation state of the rotating component to be evaluated according to each target probability to obtain the health index of the rotating component to be evaluated. Therefore, this application, by acquiring target-sensitive fault features related to the degradation of the rotating component to be evaluated, can avoid interference from irrelevant or weakly correlated features and improve the ability to represent the true degradation state of the component. By using a target state level probability mapping model built on a deep belief network to process the spliced ​​target-sensitive fault features, it can automatically learn the complex mapping relationship between fault features and degradation states, leveraging the powerful nonlinear feature extraction capabilities of the deep belief network. This eliminates the need for manual experience in designing feature combinations, thereby enhancing the model's adaptability and generalization ability to complex operating conditions. Simultaneously, it outputs the target probability at each degradation state level, which is significantly higher than traditional evaluation methods. This estimation method can more meticulously depict the complete degradation process of components from health to failure, providing accurate decision-making basis for predictive maintenance. Furthermore, by quantifying the probability of each target, a health index is obtained, transforming discrete degradation state level probabilities into continuous quantitative indicators. This facilitates real-time monitoring of component degradation trends and predicts failure risks through the rate of index change, achieving a shift from passive maintenance to proactive prevention. Moreover, by splicing features, multi-source monitoring data can be flexibly integrated to adapt to the health assessment needs of different types of rotating components, exhibiting strong versatility and scalability. Ultimately, it provides technical support for the operation and maintenance of rotating components in vehicle running gear, improving operational efficiency and safety. This application constructs a DBN-based state level probability mapping model based on the full lifecycle data of components stored in the ground system and the correctly labeled degradation state level data of rotating components. This model can dynamically update model parameters and dynamically output health indices. It can also continuously and automatically update and iterate model parameters based on the disassembled full lifecycle of components, optimizing the model and exhibiting strong robustness in measuring component degradation states. Utilizing existing correctly labeled health level data improves data reliability and reduces the model's dependence on the original collected data. This application also repairs data based on the causes of data anomalies. By analyzing different data anomalies and their causes, data repair is performed, improving data reliability and the rationality of the repaired data. Furthermore, the network model constructed by integrating component degradation information from diverse data sources can dynamically update model parameters to provide a component health index, offering data-driven support for measuring component status. The data used in this application is conclusion data from the original sampling data, reducing reliance on the original data and the requirements for sampling equipment, making it more engineering-ready.

[0130] See Figure 5 As shown in the figure, this application discloses a health assessment device for a rotating component of a vehicle running gear, comprising:

[0131] The fault feature acquisition module 11 is used to acquire the target sensitive fault features of the rotating component to be evaluated in the vehicle running gear during the current time period; wherein, the target sensitive fault features are features that are associated with the degradation of the rotating component to be evaluated;

[0132] The level probability mapping module 12 is used to process the spliced ​​target sensitive fault features using the target state level probability mapping model to obtain the target probability of the rotating component to be evaluated under each degradation state level; wherein, the target state level probability mapping model is a deep neural network model built based on a deep belief network.

[0133] The health index acquisition module 13 is used to quantify the degradation state of the rotating component to be evaluated according to each of the target probabilities, so as to obtain the health index of the rotating component to be evaluated.

[0134] The beneficial effects of this application are as follows: It obtains the target sensitive fault features of the rotating component to be evaluated in the vehicle's running gear within the current time period; wherein the target sensitive fault features are features associated with the degradation of the rotating component to be evaluated; it processes the stitched target sensitive fault features using a target state level probability mapping model to obtain the target probability of the rotating component to be evaluated at each degradation state level; wherein the target state level probability mapping model is a deep neural network model constructed based on a deep belief network; and it quantifies the degradation state of the rotating component to be evaluated according to each target probability to obtain the health index of the rotating component to be evaluated. Therefore, this application, by acquiring target-sensitive fault features related to the degradation of the rotating component to be evaluated, can avoid interference from irrelevant or weakly correlated features and improve the ability to represent the true degradation state of the component. By using a target state level probability mapping model built on a deep belief network to process the spliced ​​target-sensitive fault features, it can automatically learn the complex mapping relationship between fault features and degradation states, leveraging the powerful nonlinear feature extraction capabilities of the deep belief network. This eliminates the need for manual experience in designing feature combinations, thereby enhancing the model's adaptability and generalization ability to complex operating conditions. Simultaneously, it outputs the target probability at each degradation state level, which is significantly higher than traditional evaluation methods. The estimation method can more meticulously depict the complete degradation process of components from health to failure, providing accurate decision-making basis for predictive maintenance. Then, based on the probability quantification of each target, a health index is obtained, transforming discrete degradation state level probabilities into continuous quantitative indicators. This facilitates real-time monitoring of component degradation trends and predicts failure risks through the rate of index change, realizing a shift from passive maintenance to proactive prevention. Furthermore, by splicing features, multi-source monitoring data can be flexibly integrated to adapt to the health assessment needs of different types of rotating components, exhibiting strong versatility and scalability. Ultimately, this provides technical support for the operation and maintenance of rotating components in vehicle running gear, improving operation and maintenance efficiency and safety.

[0135] Furthermore, embodiments of this application also provide an electronic device. Figure 6 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0136] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Specifically, it may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the health assessment method for rotating components of a vehicle running gear, as disclosed in any of the foregoing embodiments.

[0137] In this embodiment, the power supply 23 is used to provide operating voltage for various hardware devices on the electronic device; the communication interface 24 can create a data transmission channel between the electronic device and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0138] The processor 21 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 21 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 21 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 21 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0139] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored on it include operating system 221, computer program 222 and data 223, etc., and the storage method can be temporary storage or permanent storage.

[0140] The operating system 221 manages and controls the various hardware devices and computer programs 222 on the electronic device to enable the processor 21 to perform calculations and processing on the massive amounts of data 223 in the memory 22. It can be Windows, Unix, Linux, etc. The computer program 222, in addition to including a computer program capable of performing the health assessment method for the rotating components of the vehicle running gear as disclosed in any of the foregoing embodiments, may further include computer programs capable of performing other specific tasks. The data 223 may include data received by the electronic device from external devices, as well as data collected by its own input / output interface 25.

[0141] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned health assessment method for the rotating components of a vehicle running gear. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0142] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0143] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application. The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly in hardware, software modules executed by a processor, or a combination of both. The software module may be located in random access memory (RAM), memory, read-only memory (ROM), electrically programmable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), register, hard disk, removable disk, CD-ROM (Compact Disc Read-Only Memory), or any other form of storage medium known in the art.

[0144] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0145] The above provides a detailed description of the health assessment method, apparatus, equipment, and medium for a rotating component of a vehicle's running gear provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for health assessment of rotating components of a vehicle's running gear, characterized in that, include: Acquire the target sensitive fault features of the rotating component to be evaluated in the vehicle running gear during the current time period; wherein, the target sensitive fault features are features that are associated with the degradation of the rotating component to be evaluated; The spliced ​​target sensitive fault features are processed using a target state level probability mapping model to obtain the target probability of the rotating component to be evaluated at each degradation state level; wherein, the target state level probability mapping model is a deep neural network model built based on a deep belief network. The degradation state of the rotating component to be evaluated is quantified according to each of the target probabilities to obtain the health index of the rotating component to be evaluated.

2. The health assessment method for rotating components of a vehicle running gear according to claim 1, characterized in that, The acquisition of the target sensitive fault characteristics of the rotating component to be evaluated in the vehicle running gear within the current time period includes: Acquire monitoring data for the rotating components to be evaluated in the vehicle's running gear during the current time period; Based on the correlation between each monitoring data and the degradation state of the rotating component to be evaluated, target sensitive fault features are screened from each monitoring data.

3. The health assessment method for rotating components of a vehicle running gear according to claim 2, characterized in that, The step of filtering out target sensitive fault features from the monitoring data based on the correlation between each monitoring data and the degradation state of the rotating component to be evaluated includes: Obtain degradation-related information of the rotating component to be evaluated; wherein, the degradation-related information includes fault type and historical fault case information; Determine the correlation and monotonicity between each of the monitoring data and the degradation-related information; Target sensitive fault features with a correlation greater than a first preset threshold and / or a monotonicity greater than a second preset threshold are selected from the monitoring data.

4. The health assessment method for rotating components of a vehicle running gear according to claim 3, characterized in that, The formula for calculating the correlation is: ； in, For normalized monitoring data, Where N is the time period and N is the number of monitoring data points. The formula for calculating the monotonicity is: ； in, The data represents the normalized monitoring data, where N is the number of different types of monitoring data. It is a unit step function.

5. The health assessment method for rotating components of a vehicle running gear according to claim 1, characterized in that, Obtain the target state level probability mapping model, including: Constructing an initial state level probability mapping model based on deep belief networks; The collected training set is divided into batches of training data; wherein, the training set includes full life cycle monitoring data of each rotating component in the vehicle running gear, and the full life cycle monitoring data includes degradation status level data; The initial state level probability mapping model is trained using the training data from each batch to obtain the model parameters and model performance index corresponding to each batch. Target model parameters are selected from the model parameters based on the model performance indexes, and a target state level probability mapping model is obtained based on the target model parameters.

6. The health assessment method for rotating components of a vehicle running gear according to claim 5, characterized in that, The initial state level probability mapping model includes multiple restricted Boltzmann machines, each of which includes a visible layer and a hidden layer, and the activation function of the hidden layer of the last layer is a sigmoid function. The model parameters include weights, visible layer bias, and hidden layer bias.

7. The health assessment method for rotating components of a vehicle running gear according to claim 5, characterized in that, The process of dividing the collected training set into batches of training data includes: Collect full lifecycle monitoring data for each rotating component in the vehicle's running gear; wherein, the full lifecycle monitoring data refers to historical monitoring data for each historical sampling time period; The full lifecycle monitoring data is preprocessed to obtain processed historical data; The processed historical data is spliced ​​together along the time dimension to obtain the spliced ​​data corresponding to each historical sampling time period; The spliced ​​data is determined as training samples, and a dataset including each training sample is constructed. The dataset is divided into batches of training data.

8. The health assessment method for rotating components of a vehicle running gear according to claim 5, characterized in that, The step of training the initial state level probability mapping model using the training data from each batch to obtain the model parameters and model performance index corresponding to each batch includes: The training data for the current batch is determined from the training data from each batch. The initial state level probability mapping model is trained using the current batch of training data to obtain the model parameters corresponding to the current batch of training data; Based on the model parameters corresponding to the current batch of training data, obtain the current state level probability mapping model, and obtain the model performance index of the current state level probability mapping model.

9. The health assessment method for rotating components of a vehicle running gear according to claim 8, characterized in that, The model performance index is the average loss value; Accordingly, the step of selecting target model parameters from the model parameters based on the model performance indices includes: Select the batch with the smallest average loss value from all batches as the first target batch; The model parameters of the first target batch are determined as the target model parameters.

10. The health assessment method for rotating components of a vehicle running gear according to claim 8, characterized in that, The model performance indices include average loss and average accuracy. Accordingly, the step of selecting target model parameters from the model parameters based on the model performance indices includes: The second target batch is determined from each batch based on the average loss value and the average accuracy. The model parameters of the second target batch are determined as the target model parameters.

11. The health assessment method for rotating components of a vehicle running gear according to claim 9 or 10, characterized in that, Obtaining the average loss value of the current state level probability mapping model includes: The current batch of training data is input into the current state level probability mapping model so that the current state level probability mapping model outputs the predicted probability of each training sample in the current batch of training data under different degradation state levels. Obtain the actual probability of each training sample in the degradation state level data under different degradation state levels; The loss value between the predicted probability and the actual probability is quantified using the cross-entropy loss function to obtain the loss value of each training sample in the current batch of training data, and the average loss value of each loss value is obtained.

12. The health assessment method for rotating components of a vehicle running gear according to claim 10, characterized in that, Obtaining the average accuracy of the current state level probability mapping model includes: The current batch of training data is input into the current state level probability mapping model so that the current state level probability mapping model outputs the predicted probability of each training sample in the current batch of training data under different degradation state levels. Obtain the actual probability of each training sample in the degradation state level data under different degradation state levels; The accuracy of each training sample in the current batch of training data is determined based on the predicted probability and the actual probability in the degradation state level data, and the average accuracy of each accuracy is obtained.

13. The health assessment method for rotating components of a vehicle running gear according to claim 1, characterized in that, The step of quantifying the degradation state of the rotating component to be evaluated based on each of the target probabilities to obtain a health index of the rotating component to be evaluated includes: Obtain the weighted summation of the probabilities of each target; wherein the weighted summation represents the comprehensive index of the current deviation of the rotating component to be evaluated from its healthy state; Invert the weighted summation result to obtain the current health tendency index of the rotating component to be evaluated; If the health tendency index is greater than the preset health baseline value, then the health tendency index is determined as the health index of the rotating component to be evaluated. If the health tendency index is not greater than the preset health baseline value, then the preset health baseline value is determined as the health index of the rotating component to be evaluated.

14. A health assessment device for a rotating component of a vehicle's running gear, characterized in that, include: The fault feature acquisition module is used to acquire the target sensitive fault features of the rotating component to be evaluated in the vehicle running gear within the current time period; wherein, the target sensitive fault features are features that are associated with the degradation of the rotating component to be evaluated; The level probability mapping module is used to process the spliced ​​target sensitive fault features using the target state level probability mapping model to obtain the target probability of the rotating component to be evaluated under each degradation state level; wherein, the target state level probability mapping model is a deep neural network model built based on a deep belief network. The health index acquisition module is used to quantify the degradation state of the rotating component to be evaluated based on each of the target probabilities, so as to obtain the health index of the rotating component to be evaluated.

15. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the health assessment method for rotating components of a vehicle running gear as described in any one of claims 1 to 13.

16. A computer-readable storage medium, characterized in that, Used to store a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the health assessment method for the rotating components of a vehicle running gear as described in any one of claims 1 to 13.

Images