Windscreen wiper fault prediction method based on random forest method
Through the wiper fault prediction method based on the random forest method, the wiper digital twin model and cross-verification technology are used to solve the problem of difficulty in predicting wiper system failures in the existing technology in advance, achieving high-accurate fault prediction and improving the operating efficiency of the vehicle.
Patent Information
- Application Number
- CN202411643245.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-18
- Publication Date
- 2025-02-28
Smart Images
Figure CN119538114A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a fault prediction method, and in particular to a wiper fault prediction method based on a random forest method. Background Art
[0002] With the popularity of electric vehicles, their reliability and safety have become a hot topic of research. As an important part of electric vehicles, the normal operation of the wiper system is crucial to driving safety.
[0003] At present, windshield wiper fault prediction mainly relies on traditional fault diagnosis methods, such as monitoring the operating status and current changes of the wiper through sensors. However, these methods can only detect faults after they occur and cannot predict faults in advance, which may cause problems for the vehicle at critical moments. The existing technology lacks an effective fault prediction model.
[0004] Existing wiper systems usually only record the operating status of the wiper and simple fault codes, and lack comprehensive monitoring of the wiper working environment, such as the impact of environmental factors such as temperature, humidity, and wind speed on the wiper. These environmental factors have an important impact on the life and performance of the wiper, but due to technical limitations, these data are difficult to be effectively collected and analyzed, and there are certain limitations in data collection and processing technology. Summary of the invention
[0005] The purpose of the present invention is to overcome the defects of the prior art and provide a wiper fault prediction method based on the random forest method. The vibration noise of the electric vehicle wiper system is predicted by the random forest algorithm, so as to improve the reliability of the system and reduce the maintenance cost, which helps to reduce the failure rate and improve the operating efficiency of the vehicle.
[0006] The object of the present invention is achieved by: a wiper fault prediction method based on random forest method, comprising the following steps:
[0007] 1) Use the wiper digital twin model to collect wiper operation data and clean the historical data;
[0008] 2) Then the random forest method is used to build a fault prediction model, and the performance of the model is verified by cross-validation method;
[0009] 3) Analyze and verify the results of the fault prediction model constructed using the random forest method.
[0010] As a further limitation of the present invention, the use of a wiper digital twin model to collect wiper operation data in step 1) specifically includes: establishing a wiper three-dimensional model in Adams, and obtaining wiper operation data through simulation testing, wherein the operation data includes acceleration, angular velocity, angular acceleration, angular kinetic energy, and angular momentum; and taking acceleration as the target and using other data as features to predict the acceleration curve of the wiper, thereby comparing the wiper prediction curve with the normal state to see if there are any abnormalities.
[0011] As a further limitation of the present invention, the cleaning of historical data in step 1) specifically includes: after collecting the data, processing the data and deleting outliers; using pandas to read the data, and then generating a violin plot for each feature to view its distribution; defining feature columns `features` and target columns `target` for model training; the feature columns include velocity, angular velocity, angular acceleration, angular momentum and angular kinetic energy, and the target column is acceleration.
[0012] As a further limitation of the present invention, the construction of the fault prediction model using the random forest method in step 2) specifically includes:
[0013] The data set is divided into a training set and a test set. Samples are extracted from the original data set using the bootstrap sampling method to form multiple sub-datasets. For a sample, among m samples, the probability of being collected each time is 1 / m, and the probability of not being collected at all is
[0014]
[0015] That is, in each round of random sampling, about 36.8% of the data in the training set was not sampled;
[0016] For each sub-dataset, build a decision tree, and at each node, randomly select a part of the features for splitting; repeat the above steps until the specified number of decision trees are generated.
[0017] As a further limitation of the present invention, the cross-validation method for verifying the performance of the model in step 2) specifically includes: writing a function 'evaluate_model' to evaluate the model performance, the 'evaluate_model' function calculates and outputs the root mean square error RMSE, the determination coefficient R 2 and explained variance fraction EVS.
[0018] As a further limitation of the present invention, the step 3) specifically includes: comparing and analyzing the results of the fault prediction model constructed using the random forest method with the other five models, grouping the true values and predicted values of the six models into point coordinates one by one, and evaluating the accuracy and stability of the model by comparing the distribution of scattered points of different models; and calculating the root mean square error, determination coefficient and explained variance score of the six models respectively to evaluate the prediction accuracy.
[0019] The present invention adopts the above technical solution, and compared with the prior art, the beneficial effects are as follows: the present invention uses the digital twin model of the wiper to collect the historical operation data and fault records of the electric vehicle wiper system, and then uses data preprocessing technology to clean the data, and then uses six models such as random forest to build a fault prediction model, and evaluates the performance of the model through the cross-validation method. Finally, the prediction results of the model are analyzed and verified; the vibration noise of the electric vehicle wiper system is predicted by the random forest algorithm, in order to improve the reliability of the system and reduce the maintenance cost; the fault prediction model based on random forest can effectively predict the potential faults of the electric vehicle wiper system, with high accuracy and generalization ability, and the prediction accuracy is about 0.9989, which is 0.2% higher than the accuracy of several other models; the establishment of this model provides a new technical means for the intelligent maintenance of the electric vehicle wiper system, which helps to reduce the failure rate and improve the operation efficiency of the vehicle. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 This is a diagram of the digital twin model of the wiper of the present invention.
[0021] Figure 2 This is a schematic diagram of the random forest principle of the present invention.
[0022] Figure 3 This is a partial model diagram of the structure tree of the present invention.
[0023] Figure 4 This is a coordinate comparison diagram of the true value and the test value of the present invention.
[0024] Figure 5 It is the absolute error diagram of the present invention and the other five models.
[0025] Figure 6 It is a comparison chart between the true value and the test value of the present invention and the other five models. DETAILED DESCRIPTION
[0026] A wiper fault prediction method based on random forest method comprises the following steps:
[0027] 1) Using a windshield wiper digital twin model (such as Figure 1 (as shown) collects wiper operation data and cleans historical data;
[0028] A three-dimensional model of the wiper is established in Adams, and the operation data of the wiper is obtained through simulation tests. The operation data includes acceleration, angular velocity, angular acceleration, angular kinetic energy, and angular momentum. The acceleration is used as the target, and other data are used as features to predict the acceleration curve of the wiper, so as to compare the predicted wiper curve with the normal state to see if there is any abnormality.
[0029] After collecting the data, process the data and delete outliers; use pandas to read the data, and then generate a violin plot for each feature to view its distribution; define the feature column `features` and the target column `target` for model training; the feature columns include velocity, angular velocity, angular acceleration, angular momentum and angular kinetic energy, and the target column is acceleration.
[0030] 2) Then the random forest method is used to build a fault prediction model, and the performance of the model is verified by cross-validation method;
[0031] like Figure 2 As shown in Figure 1, random forest is an integrated learning model based on the Bagging (Bootstrap Aggregating) strategy. It can effectively handle nonlinear problems and is good at handling a large number of samples and features. In addition, random forest also focuses on reducing variance.
[0032] The data set is divided into a training set and a test set. The test set is the last 350 records in the data set. Samples are extracted from the original data set using the bootstrap sampling method to form multiple sub-datasets. For a sample, among m samples, the probability of being collected each time is 1 / m, and the probability of not being collected at all is
[0033]
[0034] That is, in each round of random sampling, about 36.8% of the data in the training set was not sampled;
[0035] For each sub-dataset, build a decision tree, and at each node, randomly select a portion of features for splitting; repeat the above steps until the specified number of decision trees are generated;
[0036] A function 'evaluate_model' is written to evaluate the model performance. The 'evaluate_model' function calculates and outputs the root mean square error RMSE and the coefficient of determination R 2 and explained variance fraction EVS.
[0037] 3) Analyze and verify the results of the fault prediction model constructed using the random forest method; compare the results of the fault prediction model constructed using the random forest method with the other five models, and combine the true values and predicted values of the six models into point coordinates one by one. By comparing the distribution of scattered points of different models, this straight line represents the situation where the predicted value is equal to the actual value. The greater the deviation, the greater the difference between the model's prediction and the actual value. By comparing the distribution of scattered points of different models, it can be intuitively seen which model's prediction is closer to the actual value, thereby evaluating the accuracy and stability of the model. The root mean square error, determination coefficient and explained variance score of the six models are calculated respectively to evaluate the prediction accuracy.
[0038] To further illustrate the advantages of the method of the present invention, the remaining five models were constructed for comparative analysis. Figure 3 It is one of the tree model branches, and the parameters set are that the number of trees is 1100 and the random state is 42. In order to utilize the time series information, the data of the first three time points are added as new features to obtain model two, namely random forest (original data + data of the first three moments), which means that for each original feature, its value at the first three time points is now available as an additional feature. Delete any rows containing missing values, and update the feature list to include these newly added lagging features. Use the new data set containing time series features to re-divide the training set and test set, and keep the test set as the last 350 records. Use the random forest model for training again, and the parameters are the same as the model of the present invention. In addition, two linear regression models (models three and four) are created, which are trained on the original features and the data containing time series features, respectively. In order to improve the expressiveness of the model, the original features are polynomially transformed, and the order of the polynomial is set to 2. Afterwards, two linear regression models (models five and six) are trained respectively using the converted features, and the coefficients and intercepts of the two models are output.
[0039] In order to intuitively obtain the training results of the present invention and the other five models, the real values and predicted values of the present invention and the other five models are combined into point coordinates one by one, as follows Figure 4 The comparison between the true value and the predicted value is shown in the figure. The model accuracy is shown in the six model accuracies in Table 1. The root mean square error (Root Mean Square Error), coefficient of determination (coefficient of determination) and explained variance (Explained variance) of the six models are calculated respectively.
[0040] Root mean square error: The square root of the ratio of the square of the deviation between the observed value and the true value to the number of observations N. The root mean square error is also represented by σ, which reflects the degree to which the measured data deviates from the true value. The smaller σ is, the higher the measurement accuracy is. The calculation formula is as shown in the formula, where N represents the number of observations, X OBS,iCorresponding to the true value, X model,i Corresponding predicted value:
[0041]
[0042] Coefficient of determination: It can be generally understood as using the mean as the error benchmark to see whether the prediction error is greater or less than the mean benchmark error. If: R 2 score=1, the predicted value and the true value in the sample are completely equal, without any error, indicating that the independent variable in the regression analysis explains the dependent variable better; R2 score=0, at this time, the numerator is equal to the denominator, and each predicted value of the sample is equal to the mean; the specific calculation formula is as follows, RMSE means root mean square error, Var is variance;
[0043]
[0044] Explained variance: It is a statistical concept. The core idea is to measure the model's ability to explain data changes or its prediction accuracy. In regression analysis (such as linear regression), explained variance can be used to evaluate the relative quality of the model's predicted values to the actual observed values. The calculation of explained variance is shown in the formula, where y is the actual target variable (i.e., the true value), is the output value predicted by the model, To represent the residual sum of squares divided by the sample size (that is, the square of the standard deviation of the error term), Var(y) is the variance of the target variable y;
[0045]
[0046] Table 1 Accuracy of six models
[0047]
[0048] Model 1 is the random forest model of the present invention (original data), model 2 is random forest (original data + data from the previous 3 moments), model 3 is linear regression (original data), model 4 is linear regression (original data + data from the previous 3 moments), model 5 is polynomial regression (original data), and model 6 is polynomial regression (original data + data from the previous 3 moments).
[0049] To more intuitively display the prediction accuracy of the six models, we plot the absolute error graphs of the six models, subtract the true value from the predicted value and take the absolute value, and get the following: Figure 5 (a) Figure 5 (b) is a comparison chart of three models with relatively high accuracy.
[0050] like Figure 6The comparison chart of the true value and predicted value of each model is shown. Based on the comparison of the above models, the Follow-up Forest algorithm shows higher accuracy, with an accuracy of 0.9989, which is 0.2% higher than that of several other models. This will be a big improvement in the prediction accuracy of engineering failures.
[0051] The present invention is not limited to the above-mentioned embodiments. On the basis of the technical solution disclosed in the present invention, technicians in this field can make some substitutions and deformations to some technical features therein according to the disclosed technical content without creative labor, and these substitutions and deformations are all within the protection scope of the present invention.
Claims
1. A wiper fault prediction method based on random forest method, characterized in that: The following steps are involved: 1) Use the wiper digital twin model to collect wiper operation data and clean the historical data; 2) Then the random forest method is used to build a fault prediction model, and the performance of the model is verified by cross-validation method; 3) Analyze and verify the results of the fault prediction model constructed using the random forest method.
2. The wiper fault prediction method based on random forest method according to claim 1, characterized in that: The use of the wiper digital twin model to collect wiper operation data in step 1) specifically includes: establishing a three-dimensional wiper model in Adams, and obtaining wiper operation data through simulation testing, wherein the operation data includes acceleration, angular velocity, angular acceleration, angular kinetic energy, and angular momentum; and taking acceleration as the target and using other data as features to predict the acceleration curve of the wiper, thereby comparing the wiper prediction curve with the normal state to see if there are any abnormalities.
3. The wiper fault prediction method based on random forest method according to claim 1, characterized in that: The cleaning of historical data described in step 1) specifically includes: after collecting data, processing the data and deleting outliers; using pandas to read the data, and then generating a violin plot for each feature to view its distribution; defining the feature column `features` and the target column `target` for model training; the feature columns include velocity, angular velocity, angular acceleration, angular momentum and angular kinetic energy, and the target column is acceleration.
4. The wiper fault prediction method based on random forest method according to claim 1, characterized in that: The use of the random forest method to construct a fault prediction model in step 2) specifically includes: The data set is divided into a training set and a test set. Samples are extracted from the original data set using the bootstrap sampling method to form multiple sub-datasets. For a sample, among m samples, the probability of being collected each time is 1 / m, and the probability of not being collected at all is That is, in each round of random sampling, about 36.8% of the data in the training set was not sampled; For each sub-dataset, build a decision tree, and at each node, randomly select a part of the features for splitting; repeat the above steps until the specified number of decision trees are generated.
5. The wiper fault prediction method based on random forest method according to claim 1, characterized in that: The performance of the model verified by cross-validation in step 2) specifically includes: writing a function 'evaluate_model' to evaluate the model performance, and the 'evaluate_model' function calculates and outputs the root mean square error RMSE, the coefficient of determination R 2 and explained variance fraction EVS.
6. The wiper fault prediction method based on random forest method according to claim 1, characterized in that: The step 3) specifically includes: comparing and analyzing the results of the fault prediction model constructed using the random forest method with the other five models, combining the true values and predicted values of the six models into point coordinates one by one, and evaluating the accuracy and stability of the model by comparing the distribution of scattered points of different models; and calculating the root mean square error, determination coefficient and explained variance score of the six models respectively to evaluate the prediction accuracy.