Time window dynamic adjustment method and system for fault diagnosis of diesel engine supercharger

Abstract

The invention relates to the technical field of internal combustion engine superchargers, and discloses a time window dynamic adjustment method and system for fault diagnosis of a diesel engine supercharger, and the method comprises the steps: extracting time sequence data of key characteristic parameters of the diesel engine supercharger through mechanism analysis when the fault diagnosis of the diesel engine supercharger is carried out based on deep learning; after preprocessing, inputting into a fault diagnosis model; and adaptively adjusting a time window of the fault diagnosis model based on the principal component analysis result, so that the time window is adjusted according to the fluctuation change of the feature data. According to the method, the time window of the turbocharger in the actual working condition data can be dynamically adjusted, the concept of time domain-feature domain double closed-loop control is utilized, principal component analysis is improved from a traditional dimension reduction tool to a dynamic system regulator, a quantitative mapping model of the window length and the working condition data is established, and on the premise that feature information is guaranteed, the real-time performance of the turbocharger is improved. And the data processing amount of the window can be effectively reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of internal combustion engine supercharger, and particularly relates to a time window dynamic adjustment method and system for fault diagnosis of a diesel engine supercharger. BACKGROUND

[0002] With the continuous development of science and technology, the fault diagnosis method of the diesel engine supercharger has also undergone evolution from traditional methods to modern intelligent methods. The traditional fault diagnosis method of the turbocharger is mainly based on rules and statistics. The modern intelligent method mainly comes from the wide application of deep learning. The rule-based fault diagnosis method is an early diagnosis method, which realizes the diagnosis function by defining the faults that may occur in the turbocharger, the causes and manifestations of the faults. The defined rules often come from expert knowledge, experience accumulation and analysis of historical data, and the fault symptoms, fault causes and fault thresholds are expressed in a certain logical relationship. Due to the strong explainability and easy implementation, the fault diagnosis of the turbocharger in the diesel engine industry usually adopts this method for fault elimination and alarm. However, for complex and variable working conditions and application environments, the acquisition of rules and thresholds is a time-consuming and laborious accumulation process, which usually causes significant economic losses, and has obvious disadvantages in application range and generalization.

[0003] The statistical fault diagnosis method needs to study and analyze a large amount of historical data, and establish the correlation between effective fault features and feature data. There are many monitoring data of the turbocharger, and the essential feature of high-dimensional data statistics is efficient extraction and probabilistic abnormality measurement. The principal component analysis (PCA) commonly used in the prior art can reduce the dimension of high-dimensional fault data, extract the most important features in the data, find the potential rules in the data, and realize the classification and identification of fault types.

[0004] When using the deep learning method for fault diagnosis, the selection of the time window has a very important influence on the feature extraction effect. The time window refers to the continuous time steps selected when processing time series data. A suitable moving time window can accurately capture the feature information of the fault data. If the time window is too small, it cannot contain enough fault information, resulting in insufficient extracted information. When the time window is set too large, the calculation complexity is increased, and noise information is inevitably introduced, which affects the fault diagnosis effect. Therefore, in the process of fault diagnosis of the turbocharger, the size of the time window needs to be corrected according to the application working condition and the diagnosis effect, which plays a key role in balancing the diagnosis effect and the computing power. SUMMARY

[0005] In order to solve the above problems, the present application provides a turbocharger fault diagnosis time window dynamic adjustment method and system, which can dynamically adjust the time window of the turbocharger in the actual working condition data, and establish a quantitative mapping model of the window length and the working condition data.

[0006] The technical scheme adopted by the present application is as follows: A turbocharger fault diagnosis time window dynamic adjustment method, comprising: When performing turbocharger fault diagnosis based on deep learning, the time series data of the key feature parameters of the turbocharger is extracted through mechanism analysis, and after preprocessing, it is input into the fault diagnosis model. Based on the principal component analysis result, the time window of the fault diagnosis model is adaptively adjusted, so that the time window is adjusted according to the fluctuation of the feature data.

[0007] Further, the time series data of the key feature parameters of the turbocharger is extracted through mechanism analysis, and after preprocessing, it is input into the fault diagnosis model, comprising: Data monitoring: collecting the operation monitoring data of the turbocharger to obtain the time series data of the key feature parameters; Data feature fusion: combining the main characteristic parameters of the compressor and the turbine, the time series data of the key feature parameters is converted to obtain the feature fusion result representing the performance of the turbocharger; Data feature layer cleaning: cleaning the noise, missing values and outliers in the feature fusion result to remove redundant and abnormal data; Data feature layer normalization: normalizing the feature data after cleaning to eliminate the dimension and data range difference.

[0008] Further, in the data feature fusion, the abstract features in the operation monitoring data of the turbocharger are extracted by using the feature level fusion technology, and the original data redundancy is eliminated.

[0009] Further, in the data feature layer cleaning, the noise, missing values and outliers in the feature fusion result are cleaned, including: deleting the entire data record containing missing values for the missing values and outliers with an actual proportion less than a preset proportion; for the missing values with a working condition data amount less than a preset data amount, the missing values are filled with mean or median.

[0010] Further, in the data feature layer normalization, the feature data after cleaning is normalized, including: using the Z-Score method to map the feature data after cleaning to a preset range, and retaining the data distribution shape and original distribution characteristics.

[0011] Further, the time window of the fault diagnosis model is adaptively adjusted based on the principal component analysis result, so that the time window is adjusted according to the fluctuation of the characteristic data, and the time window is adaptively adjusted, including: Principal component analysis: initializing the length and step of the time window, performing principal component analysis on the normalized characteristic data in time, calculating the covariance matrix and decomposing the eigenvalue to obtain the principal component, and determining the number of retained principal components; Window adaptive adjustment: taking the cumulative variance contribution rate and its real-time standard deviation as the adjustment basis, combining with the preset threshold range, dynamically adjusting the length and step of the time window, and establishing a quantitative mapping model of the length of the time window and the working condition data; Threshold setting and iterative optimization: traversing the data set to calculate each time window parameter, setting the corresponding threshold based on the cumulative variance contribution rate and the standard deviation under normal and fault working conditions, and iteratively optimizing the size and step of the time window, so that the window data maximally reflects the key information of the characteristic parameters.

[0012] Further, in the window adaptive adjustment, the cumulative variance contribution rate and its real-time standard deviation are taken as the adjustment basis, the preset threshold range is combined, and the length and step of the time window are dynamically adjusted, including: a fixed initial time window is used during normal operation; when the cumulative variance contribution rate exceeds the threshold, the time window is gradually expanded, and after each expansion, the principal component analysis is re-executed and the cumulative variance contribution rate is calculated, until the increase amplitude of the cumulative variance contribution rate is less than the threshold, the expansion is stopped, and the current time window is determined as the optimal time window.

[0013] Further, in the window adaptive adjustment, the standard deviation of the cumulative variance contribution rate is also considered: if the standard deviation of the cumulative variance contribution rate exceeds the set threshold during the expansion of the time window, the size of the time window is adjusted and the principal component analysis is re-executed until the standard deviation of the cumulative variance contribution rate decreases to an acceptable range.

[0014] Further, the principal component analysis further includes data distribution verification: through Klomogorov-Smirnov test, it is verified whether the average value and standard deviation of the cumulative variance contribution rate after principal component analysis approximately obey normal distribution, and the test indicators include KS statistics D and survival function P value; when the survival function P value is greater than the preset significance level, it is determined that the data approximately obeys normal distribution.

[0015] A time window dynamic adjustment system for diesel engine supercharger fault diagnosis, comprising: A data acquisition and processing module configured to extract time series data of key characteristic parameters of a diesel engine supercharger through mechanism analysis when performing diesel engine supercharger fault diagnosis based on deep learning, and input the preprocessed data into a fault diagnosis model; The time window adjustment module is configured to adaptively adjust the time window of the fault diagnosis model based on the principal component analysis result, so that the time window is adjusted according to the fluctuation change of the characteristic data.

[0016] The present application has the advantages of: 1. The present application can dynamically adjust the time window of the turbocharger under actual working condition data, uses the concept of "time domain-feature domain double closed loop control", improves the principal component analysis from the traditional dimension reduction tool to a dynamic system regulator, establishes a quantitative mapping model of window length and working condition data, and effectively reduces the data processing amount of the window under the premise of guaranteeing the characteristic information.

[0017] 2. The present application extracts the time series data of the key characteristic parameters of the diesel engine turbocharger based on mechanism analysis as the model input, and constructs a time series modeling framework of dynamic cooperation of mechanism analysis and data driven. Through the feature screening process of mechanism constraint, the adaptive focusing of fault sensitive features is realized, which can effectively reduce the redundant parameter input and improve the model running efficiency. The high interpretability extraction method of multi-physical field coupling characteristics is studied, which solves the working condition generalization problem caused by the lack of physical constraints in traditional data driven methods, so that the stability of the model under variable working conditions is greatly improved, and the diagnosis accuracy of compound faults is improved. BRIEF DESCRIPTION OF DRAWINGS

[0018] Figure 1 is the time window dynamic adjustment method flow chart of the diesel engine supercharger fault diagnosis of embodiment 1 of the present application. DETAILED DESCRIPTION

[0019] In order to have a more clear understanding of the technical features, purposes and effects of the present application, the specific embodiments of the present application will be described. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application, that is, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0020] Embodiment 1 As shown in Figure 1 , the present embodiment provides a time window dynamic adjustment method for diesel engine supercharger fault diagnosis, comprising: Data acquisition and processing stage: when the diesel engine supercharger fault diagnosis is based on deep learning, the time series data of the key characteristic parameters of the diesel engine supercharger is extracted through mechanism analysis, and is input to the fault diagnosis model after preprocessing; Time window adjustment stage: based on the principal component analysis result, the time window of the fault diagnosis model is adaptively adjusted, so that the time window is adjusted according to the fluctuation change of the characteristic data.

[0021] Preferably, the data acquisition and processing stage comprises the following steps: Data monitoring: Collecting the operation monitoring data of the diesel engine supercharger to obtain the time series data of key characteristic parameters; Data feature fusion: Combining the main characteristic parameters of the compressor and turbine, the time series data of the key characteristic parameters are converted to obtain the feature fusion results representing the performance of the diesel engine supercharger; Data feature layer cleaning: Cleaning the noise, missing values and outliers in the feature fusion results to remove redundant and abnormal data; Data feature layer normalization: Normalizing the feature data after cleaning to eliminate the differences in dimensions and data ranges.

[0022] Specifically, in the data monitoring step, the common monitoring parameters of the turbocharger in the medium-speed diesel engine are shown in Table 1.

[0023] Table 1-Common monitoring data of turbocharger in medium-speed diesel engine

[0024] Since the change of external application environment will inevitably lead to large changes in the monitoring parameters of the turbocharger, such as air monitoring at the inlet, and multi-focus raw signal processing faces problems such as high noise interference and large calculation load, there may be a weak correlation with the physical mechanism of the turbocharger. If directly used as data input of the model, it will inevitably cause large errors and low accuracy. Therefore, the method extracts multi-physical field coupling characteristic parameters through mechanism analysis as model input, which can realize the dual driving transition of data driving and knowledge driving.

[0025] In the data feature fusion step, first, the parameters in Table 1 are combined with the main characteristic parameters of the compressor and turbine to convert the indicators that best represent the performance of the supercharger, i.e. the results of data feature fusion are used as model input, as shown in Table 2.

[0026] Table 2-Performance characteristic parameters of diesel engine turbocharger

[0027] Generally, multi-modal fusion technology is generally divided into raw data level fusion technology, feature level fusion technology and decision level fusion technology. Among them, feature level fusion can extract abstract features of each modality, eliminate redundancy of raw data, has high-dimensional feature decoupling function, and has strong feature extraction ability for heterogeneous data.

[0028] Due to the original data often contains noise, missing values and outliers during the acquisition and monitoring process, there are great differences in the distribution of different working conditions and different characteristic data, which cannot be used as input for calculation, affecting the performance and accuracy of the model. The detection and processing of missing values and outliers is an important part of data cleaning, which may cause deviation in statistical analysis results and affect the training and prediction performance of the model. Therefore, for the small proportion of missing values and outliers in the time series data of the turbocharger monitoring, the embodiment adopts the method of directly deleting the whole data record containing missing values; in some working condition data, the missing values are filled by using the mean, median and other methods.

[0029] In the normalization processing step of the data feature layer, the time series data of the key characteristic parameters of the turbocharger is normalized by the Z-Score method, that is, the process of mapping the characteristic data to a specific range, the purpose is to eliminate the influence of the dimension and data range of different characteristics on the model, while retaining the distribution shape and original distribution characteristics of the data, which can effectively improve the stability and efficiency of model training.

[0030] Among them, Z-Score standardization is a method of converting data to a distribution with a mean of 0 and a standard deviation of 1, the formula is expressed as:

[0031] In the formula, X is the original data, μ is the mean of the data, and σ is the standard deviation of the data.

[0032] Preferably, the time window adjustment stage in the embodiment includes the following steps: Principal component analysis: initialize the length and step of the time window, perform principal component analysis on the normalized characteristic data in time, calculate the covariance matrix and decompose the eigenvalues to obtain the principal components, and determine the number of retained principal components; Window adaptive adjustment: taking the cumulative variance contribution rate and its real-time standard deviation as the adjustment basis, combining with the preset threshold range, dynamically adjusting the length and step of the time window, establishing a quantitative mapping model of the length of the time window and the working condition data; Threshold setting and iterative optimization: traverse the data set to calculate the parameters of each time window, set the corresponding threshold based on the cumulative variance contribution rate and standard deviation under normal and fault working conditions, and iteratively optimize the size and step of the time window, so that the window data reflects the key information of the characteristic parameters to the greatest extent.

[0033] It should be noted that as a classic data dimensionality reduction and feature extraction method, the traditional principal component analysis (PCA) technique is often used for static analysis of the entire data set, which cannot capture the dynamic change process of the data in the time dimension, and a reasonable moving window size has a very important influence on the result of fault diagnosis. If the window is too large, the data contains too much redundant information, affecting the efficiency; if the window is too small, it cannot capture the complete fault characteristics. The embodiment proposes a method for self-adaptive adjustment of parameter time window based on the result feedback of principal component analysis, that is, the moving time window of the model is adjusted according to the fluctuation change of the characteristic data, the overall trend is observed from the data perspective, and the robustness of the model is enhanced.

[0034] In turbocharger engineering applications, the method of principal component analysis of characteristic data is used to dynamically adjust the moving window of time series. In order to make the input data of the window contain the most important information, the cumulative variance contribution rate is used as the main basis for window adjustment. Specifically, first, initialize the window length, perform principal component analysis (PCA) on the data in the moving window, then calculate the covariance matrix and decompose the eigenvalue to obtain the principal component, next analyze the variance proportion explained by each principal component, determine that retaining a certain number of principal components can capture the most important information in the data, and find the number of principal components that make the principal component contribution rate reach more than 85%. After fixing the number of principal components, set the cumulative variance contribution rate and its standard deviation and set the threshold range, dynamically adjust the window length and step size under different window lengths by calculating the cumulative contribution rate and its sliding standard deviation, and ensure that the window contains enough information to represent the main change of the data. If the principal component analysis result of the data shows that more information is needed to explain the same proportion of variance, the window should be appropriately increased, and vice versa.

[0035] During normal operation, a relatively small and fixed initial window size is set due to its relatively low and stable cumulative variance contribution rate. When it is detected that the data may have a fault or be in a variable operating condition, that is, when the cumulative variance contribution rate exceeds the threshold, the window size begins to be dynamically adjusted. Due to the high cumulative variance contribution rate and large fluctuation of fault data, the window is gradually expanded. Each time the window is expanded, principal component analysis is performed again to calculate the cumulative variance contribution rate. When the increase in the cumulative variance contribution rate is less than the threshold, the window expansion is stopped, and the moving window size at this time is considered to be the best.

[0036] In order to ensure the stability of the input data of the window, the standard deviation of the cumulative variance contribution rate is considered when adjusting the window size. When the standard deviation of the cumulative variance contribution rate exceeds the set threshold during window expansion, it indicates that the stability of the data in the current window is poor. At this time, the window size should be appropriately adjusted, and principal component analysis should be performed again until the standard deviation is reduced to an acceptable range.

[0037] The calculation of turbocharger acquisition parameters as characteristic values as model input is a feature selection about mechanism constraint analysis, which is ultimately based on the calculation of time series data. The distribution of turbocharger time series data set is determined according to the application condition, and more subjective operation leads to the fact that time series data cannot meet the normal distribution rule. After the feature data set is normalized and principal component analysis, the average value and standard deviation of the cumulative variance contribution rate of the fixed principal component number under different windows are obtained, and it is found through verification that they are approximately subject to normal distribution. The verification method is through Klomogorov-smirnov, which detects the goodness of fit test of the sample in any known distribution, that is, whether it meets the normal distribution, and the indicators of the verification algorithm are KS statistics D and survival function P value, as follows:

[0038]

[0039]

[0040] In the formula, F n ( x ) is the sample empirical distribution function, F( x ) is the cumulative distribution function of the normal distribution, n is the sample size, the larger the value of D, the more significant the deviation of the sample from the theoretical distribution. If the P value is greater than the significance level , the null hypothesis (the null hypothesis that the data deviates from the normal distribution) is not rejected.

[0041] Next, the average value and standard deviation of the cumulative variance contribution rate under different windows and steps are calculated in the window self-learning process. According to the 3σ rule, the normal fluctuation range of the cumulative variance contribution rate is (k value can be selected as 1, 2, 3), (k ) contains about 68% of the data information, (k ) contains about 95% of the data, and 99.7% of the data falls within the range of k ), and data outside the range is judged as abnormal data. However, the number of principal components has been fixed here, so the distribution of the cumulative variance contribution rate will show a skewed structure. In order to find the adaptive threshold of the normal data set and the fault data set, the following method is used to set the threshold: a. Threshold setting of cumulative variance contribution rate CVR First, calculate the average value and the standard deviation of the average value of the cumulative variance contribution rate of the normal data under different windows and different steps, and the threshold of the normal data is set as (k is 1, 2, 3) is set as the upper and lower threshold of the cumulative variance contribution rate CVR, ensuring that the window contains the main characteristics of the normal state; similarly, the average value of the cumulative variance contribution rate is calculated using fault data and standard deviation , (k is 1, 2, 3) is set as the upper and lower threshold of the fault state CVR, capturing the high specificity characteristics in the fault state.

[0042] b. Standard deviation STD threshold setting of cumulative variance contribution rate The standard deviation STD of the variance contribution rate is the fluctuation of the cumulative variance contribution rate of each moving window under the condition of fixed window and step, that is, the standard deviation STD of the cumulative variance contribution rate. Similarly, the threshold setting process of the average value of CVR, the STD threshold setting is to filter and find the unstable window in the fault state, and the greater the value of STD indicates that the fluctuation of the cumulative variance contribution rate is greater. By calculating the standard deviation of the cumulative variance contribution rate of different windows and different steps in the data set, the average value of the standard deviation at this time is calculated and standard deviation , (k is 1, 2, 3), ensuring that the data fluctuation in the window is within the normal range, otherwise the window is invalid.

[0043] c. Adaptive logic rule of data moving window size By comparing the various index parameters obtained by principal component analysis of the normal data set and the fault data set, the change rule of each feature parameter under different states, different window sizes and different steps is found, and the window adjustment threshold suitable for the performance of the turbocharger is found.

[0044] d. Threshold iteration optimization process First, the data set is traversed in the full window, the main parameters of each window are calculated, the moving window size and step are initialized and set, the window is adjusted according to the adaptive window setting logic, and the appropriate window size and step are found, so that the time series window data can reflect the important information of each feature parameter to the greatest extent.

[0045] In this embodiment, the time series data of the key feature parameters of the diesel turbocharger is used as the model input, and the window adaptive stretching mechanism is performed based on the principal component analysis result feedback of the feature fusion parameters. Using a lightweight Transformer framework, the core multi-head attention layer focuses on the cross-parameter transient interaction, captures the global dependence relationship of the data and locates the key fault sensitive features, and adapts to the small sample training of the turbocharger fault. Bidirectional LSTM captures the degradation trend of forward and backward long-term dependencies, and realizes the dimensional expansion technology research from time dependence and local sensitive interaction.

[0046] Embodiment 2 The embodiment provides a time window dynamic adjustment system for diesel engine supercharger fault diagnosis, comprising: The data acquisition and processing module is configured to, when diesel engine supercharger fault diagnosis is performed based on deep learning, extract time series data of key characteristic parameters of the diesel engine supercharger through mechanism analysis, and input the preprocessed time series data to the fault diagnosis model. The time window adjustment module is configured to adaptively adjust the time window of the fault diagnosis model based on the principal component analysis result, so that the time window is adjusted according to the fluctuation of the characteristic data.

[0047] Embodiment 3 The embodiment is based on the embodiment 1, and comprises: The embodiment provides a computer device, comprising a memory and a processor, the memory stores a computer program, and the processor implements the time window dynamic adjustment method for diesel engine supercharger fault diagnosis of the embodiment 1 when executing the computer program. The computer program can be in the form of source code, object code, executable file or some intermediate form.

[0048] Embodiment 4 The embodiment is based on the embodiment 1, and comprises: The embodiment provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the time window dynamic adjustment method for diesel engine supercharger fault diagnosis of the embodiment 1. The computer program can be in the form of source code, object code, executable file or some intermediate form. The storage medium includes any entity or device capable of carrying computer program code, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium. It should be noted that the content included in the storage medium can be appropriately added or reduced, for example, the storage medium does not include the electrical carrier signal and the telecommunication signal.

[0049] The above only describes the preferred embodiments of the present application, and it should be understood that the present application is not limited to the forms disclosed herein, and should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be modified by the above-mentioned teaching or related art or knowledge within the scope of the concept described herein. The modifications and changes made by those skilled in the art without departing from the spirit and scope of the present application shall be within the scope of protection of the appended claims of the present application.

[0050] It is apparent that, for the method embodiments described previously, the steps of the methods have been described as being arranged in a particular order. However, it is to be appreciated that this is merely one example, and that the steps of the methods can be arranged in other orders or performed contemporaneously. Furthermore, it is to be appreciated that the embodiments described in the specification are merely preferred embodiments, and that the steps of the methods need not be performed in the order described.

Claims

1. A method for dynamically adjusting the time window for diesel engine turbocharger fault diagnosis, characterized in that, include: When using deep learning for fault diagnosis of diesel engine turbochargers, time series data of key characteristic parameters of diesel engine turbochargers are extracted through mechanism analysis, preprocessed, and then input into the fault diagnosis model. The time window of the fault diagnosis model is adaptively adjusted based on the principal component analysis results, so that the time window is adjusted according to the fluctuations of the feature data.

2. The method for dynamically adjusting the time window for diesel engine turbocharger fault diagnosis according to claim 1, characterized in that, The time-series data of key characteristic parameters of the diesel engine turbocharger extracted through mechanistic analysis, after preprocessing, are input into the fault diagnosis model, including: Data monitoring: Collect operational monitoring data of the diesel engine turbocharger and obtain time series data of key characteristic parameters; Data feature fusion: Combining the main characteristic parameters of the compressor and turbine, the time series data of key characteristic parameters are converted to obtain the feature fusion results characterizing the performance of the diesel engine turbocharger; Data feature layer cleaning: Cleaning and processing of noise, missing values ​​and outliers in the feature fusion results to remove redundant and abnormal data; Normalization of data feature layer: Normalize the cleaned feature data to eliminate differences in units and data range.

3. The method for dynamically adjusting the time window for diesel engine turbocharger fault diagnosis according to claim 2, characterized in that, In the data feature fusion, feature-level fusion technology is used to extract abstract features from the operation monitoring data of the diesel engine turbocharger and eliminate redundancy in the original data.

4. The method for dynamically adjusting the time window for diesel engine turbocharger fault diagnosis according to claim 2, characterized in that, In the cleaning process of the data feature layer, noise, missing values ​​and outliers in the feature fusion results are cleaned, including: for missing values ​​and outliers whose actual proportion is less than the preset proportion, deleting the entire data record containing the missing value; for missing values ​​whose working condition data volume is less than the preset data volume, filling with the mean or median.

5. The method for dynamically adjusting the time window for diesel engine turbocharger fault diagnosis according to claim 2, characterized in that, In the normalization process of the data feature layer, the cleaned feature data is normalized, including: using the Z-Score method to map the cleaned feature data to a preset range, preserving the data distribution shape and original distribution features.

6. The method for dynamically adjusting the time window for diesel engine turbocharger fault diagnosis according to claim 1, characterized in that, The adaptive adjustment of the time window of the fault diagnosis model based on the principal component analysis results, which adjusts the time window according to the fluctuations in the feature data, includes: Principal Component Analysis: Initialize the length and step size of the time window, perform principal component analysis on the normalized feature data within the time period, calculate the covariance matrix and decompose the feature values ​​to obtain the principal components, and determine the number of principal components to retain. Window adaptive adjustment: Based on the cumulative variance contribution rate and its real-time standard deviation, and combined with the preset threshold range, the length and step size of the time window are dynamically adjusted to establish a quantitative mapping model between the time window length and the working condition data. Threshold setting and iterative optimization: Traverse the dataset to calculate the parameters of each time window. Based on the cumulative variance contribution rate and standard deviation under normal and fault conditions, set the corresponding thresholds and iteratively optimize the size and step size of the time window to make the window data reflect the key information of the feature parameters to the greatest extent.

7. The method for dynamically adjusting the time window for diesel engine turbocharger fault diagnosis according to claim 6, characterized in that, In the adaptive window adjustment, the cumulative variance contribution rate and its real-time standard deviation are used as the adjustment basis. Combined with the preset threshold range, the length and step size of the time window are dynamically adjusted, including: a fixed initial time window is used during normal operation; when the cumulative variance contribution rate exceeds the threshold, the time window is gradually expanded. After each expansion, principal component analysis is re-executed and the cumulative variance contribution rate is calculated until the increase in the cumulative variance contribution rate is less than the threshold, at which point the expansion is stopped and the current time window is determined as the optimal time window.

8. The method for dynamically adjusting the time window for diesel engine turbocharger fault diagnosis according to claim 7, characterized in that, In the adaptive window adjustment, the standard deviation of the cumulative variance contribution rate is also considered simultaneously: if the standard deviation of the cumulative variance contribution rate exceeds the set threshold during the expansion of the time window, the size of the time window is adjusted and the principal component analysis is re-executed until the standard deviation of the cumulative variance contribution rate drops to an acceptable range.

9. The method for dynamically adjusting the time window for diesel engine turbocharger fault diagnosis according to claim 6, characterized in that, The principal component analysis is followed by data distribution verification: the Klomogorov-Smirnov test is used to verify whether the average and standard deviation of the cumulative variance contribution rate after principal component analysis approximately follow a normal distribution. The test indicators include the KS statistic D and the survival function P value. When the survival function P value is greater than the preset significance level, it is determined that the data approximately follow a normal distribution.

10. A dynamic adjustment system for the time window of a diesel engine turbocharger fault diagnosis system, characterized in that, include: The data acquisition and processing module is configured to extract time series data of key characteristic parameters of the diesel engine turbocharger through mechanism analysis when performing fault diagnosis of diesel engine turbochargers based on deep learning, and input the preprocessed data into the fault diagnosis model. The time window adjustment module is configured to adaptively adjust the time window of the fault diagnosis model based on the principal component analysis results, so that the time window is adjusted according to the fluctuations of the feature data.

Images