CN113761649A

CN113761649A - Intelligent automobile tire eccentric wear prediction method based on one-dimensional convolutional neural network

Info

Publication number: CN113761649A
Application number: CN202110897163.4A
Authority: CN
Inventors: 周海超; 关浩东; 李慧云; 王国林
Original assignee: Jiangsu University
Current assignee: Jiangsu University
Priority date: 2021-08-05
Filing date: 2021-08-05
Publication date: 2021-12-07
Anticipated expiration: 2041-08-05
Also published as: CN113761649B

Abstract

The invention discloses an intelligent automobile tire eccentric wear prediction method based on a one-dimensional convolutional neural network, which comprises four steps of acquiring basic data, constructing a training set X and a verification set C, establishing a tire eccentric wear prediction model and predicting tire eccentric wear; and acquiring vibration signal data of the tire in real time, intercepting and normalizing the acquired vibration signal to obtain a test sample, and inputting the test sample into a trained tire eccentric wear prediction model to obtain an eccentric wear prediction result of the automobile tire. Has the advantages that: according to the method, the automobile tire eccentric wear is predicted according to the collected vibration signal data of the tire, the established automobile tire eccentric wear prediction model can automatically extract the characteristics in the data, the tire eccentric wear state can be effectively detected in real time under the condition that parking detection is not needed, the labor is saved, the economic loss is reduced, and the tire state monitoring and safe cruising driving capability of the automobile are improved.

Description

Intelligent automobile tire eccentric wear prediction method based on one-dimensional convolutional neural network

Technical Field

The invention relates to a method for predicting automobile tire eccentric wear, in particular to an intelligent automobile tire eccentric wear prediction method based on a one-dimensional convolutional neural network, and belongs to the technical field of intelligent automobile tire wear prediction.

Background

The tire is used as the only part of the automobile contacting with the road surface, transmits the force and moment required in the running process of the automobile, monitors the tire state information in real time, provides information input and response feedback for a vehicle power control system, and has important significance for realizing automobile intellectualization and networking. The intelligent tire can directly monitor various state parameters of the tire by using sensors such as optical sensors, strain sensors, acceleration sensors and the like, such as tire pressure, six component force of the tire, tire/road adhesion characteristics and the like, and provides information reference for an intelligent automobile power control system. But the abrasion phenomenon often occurs in the long-time and long-distance driving process of the automobile. The eccentric wear of the automobile tire is the unilateral wear of the tire caused by external factors such as four-wheel positioning parameters, tire installation, unbalanced cargo loading and the like in the use process of the tire. The wear of the patterns of the single tire is uneven, wherein the patterns on one side are rapidly reduced, and the wear on the other side is not obvious; or wavy abrasion, block abrasion and the like appear before and after the patterns. The eccentric wear of the tire can cause abnormal vibration of a vehicle, affect the handling performance of an automobile and aggravate the normal wear of the tire, so that the service life of the tire is directly affected; the intelligent automobile tire monitoring system has a large interference effect on monitoring information of tire states, and further has adverse effects on an entire automobile control system of the intelligent automobile. Therefore, the eccentric wear problem of the automobile tire needs to be found in time, so that the tire state information is effectively improved, the stability and the accuracy of the intelligent automobile power control are improved, and a driver and passengers are actively prompted to eliminate the source of the eccentric wear problem of the tire.

In the prior art, the eccentric wear and the normal state of the tire are usually judged by a manual visual detection method or a computer visual detection method, which not only consumes time and labor for detection personnel, but also cannot detect whether the tire is in the eccentric wear state in real time in the driving process of an automobile and prompt a driver.

With the application and development of a simulation model and an intelligent algorithm, real-time data of the rolling tire can be effectively acquired and processed, so that the intelligent automobile tire is more intelligent due to the acquired large amount of tire data. Deep learning methods have a powerful self-learning capability from feature extraction to pattern recognition. The convolutional neural network is a typical neural network in deep learning, has a characteristic learning ability, and can perform translation invariant classification on input information according to a hierarchical structure, so that the convolutional neural network is also called as a translation invariant artificial neural network and is effectively applied to the engineering fields of bearings, motors and the like. The convolutional neural network has strong feature extraction and pattern recognition capabilities, is applied to the eccentric wear prediction of the intelligent automobile tire, realizes real-time monitoring and prejudgment, and has great significance.

Disclosure of Invention

The purpose of the invention is as follows: the invention provides an intelligent automobile tire eccentric wear prediction method based on a one-dimensional convolutional neural network, which aims to accurately and really monitor and predict the tire eccentric wear phenomenon, can effectively detect the tire eccentric wear state in real time under the condition of no need of parking detection, saves manpower, reduces economic loss and improves the tire state monitoring and safe cruising driving capability of an intelligent automobile.

The technical scheme is as follows: an intelligent automobile tire eccentric wear prediction method based on a one-dimensional convolutional neural network comprises the following steps:

s1: acquiring basic data: respectively acquiring vibration signal data of a tire with partial wear and a tire without partial wear in the running process;

s2: constructing a training set X and a verification set C: firstly, preprocessing the vibration signal data in the S1, and then intercepting the vibration signal data to establish sample data corresponding to two tires; then normalizing the sample data; labeling the sample; finally, respectively dividing the marked samples into a training set X and a verification set C;

s3: building a tire eccentric wear prediction model: establishing a one-dimensional convolution neural network initial model, training the one-dimensional convolution neural network initial model by using a training set X in S2, constructing a tire eccentric wear prediction model, and checking the diagnosis performance of the tire eccentric wear prediction model by using a verification set C;

s4: and (3) tyre eccentric wear prediction: and acquiring vibration signal data of the tire in real time, intercepting and normalizing the acquired vibration signal in S2 to obtain a test sample, and inputting the test sample into the tire partial wear prediction model trained in S3 to obtain a partial wear prediction result of the automobile tire.

According to the method, the intelligent automobile tire eccentric wear prediction is carried out according to the collected vibration signal data of the tire, the established automobile tire eccentric wear prediction model can automatically extract the characteristics in the data, the tire eccentric wear state can be effectively detected in real time under the condition that parking detection is not needed, the labor is saved, the economic loss is reduced, and the tire state monitoring and safe cruising driving capability of the intelligent automobile are improved.

Preferably, in order to accurately acquire the data of the vibration signal, the vibration signal data in step S1 is the vibration signal data of the relationship between the radial acceleration at the center of the rim where the tire with or without partial wear has been mounted and the time. The relation between the radial acceleration and the time at the center of the rim is used as vibration signal data in the running process of the tire, so that the data can be collected and analyzed more conveniently.

Preferably, the vibration signal data in step S1 is obtained by modeling and simulating a tire with or without partial wear on a finite element analysis software by a "data twinning" method, and obtaining n signal data points of radial vibration at the center position of the rim through transient dynamics analysis.

Preferably, the n signal data points of the radial vibration of the central position of the wheel rim are determined according to the frequency of the sampling points.

Preferably, in step S2, the same preprocessing is performed on two sets of vibration signal data during the running process of the tire with partial wear and the tire without partial wear, the vibration signal data are intercepted to establish sample data corresponding to the two tires, and the method for intercepting the sample includes: and selecting L data points of the vibration signal data as the sample length, wherein the step length is S, and the vibration signal of each type comprises n data points, so that each type can obtain (n-L)/S +1 groups of samples, and the two groups of samples have 2 x ((n-L)/S +1) groups of samples.

Preferably, the method for normalizing the sample data in step S2 includes: in all the above-mentioned vibration signal data, the positive value is divided by the maximum value among the positive values, and the negative value is divided by the absolute value of the minimum value among the negative values, thereby normalizing the vibration signal data in all the above-mentioned samples to the range of (-1, 1).

Preferably, the method for labeling the sample in step S2 is as follows: the two types of samples are labeled with labels, and one-hot coding is carried out by utilizing a tensierflow frame, and the eccentric wear state is marked as [ 10 ] and the health state is marked as [ 01 ].

Preferably, the method for dividing the labeled samples into the training set X and the verification set C in step S2 includes: 80-90% of the number of 2X ((n-L)/S +1) samples were selected as training set X, and the remaining 10-20% were selected as validation set C.

Preferably, the method for establishing the tire eccentric wear prediction model in step S3 includes:

inputting the training set X obtained in the step S2 into the established one-dimensional convolutional neural network initial model, starting to train the model, wherein an Adam optimizer is adopted in the model training process, the initial learning rate is set to be 0.001, the initial momentum is set to be 0.9, and the iteration step is set to be 500; continuously optimizing the model weight through a BP algorithm; during the training, a termination criterion is set: if the loss function value is less than 1e-4 or the accuracy rate reaches 100%, terminating the training in advance;

after the training set trains the model, the verification set C is input into the trained model, a model diagnosis result is output and compared with a label corresponding to the verification set C, and the accuracy of the model is obtained by comparing the error data volume in the verification set C with the total verification set data volume;

and finally, storing the model to obtain the trained tire eccentric wear prediction model.

Preferably, the one-dimensional convolutional neural network initial model comprises: the system comprises two one-dimensional convolution layers, two one-dimensional pooling layers, a full-connection layer and a softmax classifier; the one-dimensional convolution layer and the one-dimensional pooling layer are arranged in a staggered mode.

Has the advantages that: according to the method, the automobile tire eccentric wear is predicted according to the collected vibration signal data of the tire, the established automobile tire eccentric wear prediction model can automatically extract the characteristics in the data, the tire eccentric wear state can be effectively detected in real time under the condition that parking detection is not needed, the labor is saved, the economic loss is reduced, and the tire state monitoring and safe cruising driving capability of the automobile are improved.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.

FIG. 1 is a flow chart of a method of predicting tire bias wear in accordance with the present invention;

FIG. 2 is a graph showing the vibration signals of the tire of the present invention in comparison with the vibration signals of the tire in an eccentric wear state and an unbiased wear state;

FIG. 3 is a schematic diagram of sample generation of vibration signal data according to the present invention;

FIG. 4 is a schematic structural diagram of a one-dimensional convolutional neural network model constructed by the present invention;

FIG. 5 is a flow chart of the training of the one-dimensional convolutional neural network model constructed by the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

As shown in fig. 1, an intelligent automobile tire eccentric wear prediction method based on a one-dimensional convolutional neural network includes the following steps:

s1: acquiring basic data: respectively acquiring vibration signal data of a tire with partial wear and a tire without partial wear in the running process; the vibration signal data is vibration signal data of a relation between radial acceleration at a rim center of a tire in which partial wear has occurred or a tire in which partial wear has not occurred and time.

As shown in FIG. 2, the present invention is applied to vibration signal data of radial acceleration versus time at the center of a rim of a tire in which partial wear has occurred or a tire in which partial wear has not occurred.

Modeling simulation is respectively carried out on the tire with eccentric wear or the tire without eccentric wear on ABAQUS software by a data twin method, and n signal data points of radial vibration of the center position of the rim are obtained by transient dynamics analysis.

As shown in fig. 3, two types of data samples of the tire with or without partial wear are respectively captured, the rim center radial vibration signal data length L is selected as the sample length, the step length is S, the rim center radial vibration signal of each type includes n data points, and then each type can obtain (n-L)/S +1 groups of samples, and the two types of samples have a total of 2 ((n-L)/S +1) groups.

In all the above-mentioned vibration signal data, the positive value is divided by the maximum value among the positive values, and the negative value is divided by the absolute value of the minimum value among the negative values, thereby normalizing the vibration signal data in all the above-mentioned samples to the range of (-1, 1).

The two types of samples are labeled with labels, and one-hot coding is carried out by utilizing a tensierflow frame, and the eccentric wear state is marked as [ 10 ] and the health state is marked as [ 01 ].

80-90% of the number of 2X ((n-L)/S +1) samples were selected as training set X, and the remaining 10-20% were selected as validation set C.

As shown in fig. 4, upon completion of the data preprocessing, the specific steps in step S3 are:

the established one-dimensional convolutional neural network (1D-CNN) model comprises the following steps: two one-dimensional convolutional layers, two one-dimensional pooling layers, a full link layer and a softmax classifier.

The convolution kernel of the first convolution layer has the size of 1 × 20, the number of 32, the step size of 1, the activation function selects the "tanh" function, the sample size after convolution for one sample is 1 × 1000 × 32, then the pooling operation is carried out, the size of the pooling layer is 1 × 3, the step size is 2, and the size of the output characteristic data is 1 × 500 × 32.

The convolution kernel of the second convolution layer has the size of 1 × 3, the number of 64, the step size of 1, the activation function selects the "tanh" function, the sample size after convolution of the previous layer of single sample data is 1 × 500 × 64, then the pooling operation is carried out, the size of the pooling layer is 1 × 3, the step size is 2, and the size of the output characteristic data is 1 × 250 × 64.

The convolution operation formula of the first two layers is as follows:

=f(

+

) i=1,2, . . . , K

in the formula (I), the compound is shown in the specification,

for the convolutional layer output, f is the nonlinear activation function, X is the sample data,

representing the convolution operation of the convolution kernel and the sample data for the weight value of each convolution kernel,

for each channel offset, K is the number of channels after output.

The pooling operation formula of the first two layers is

=max(

) r=1,2, . . . , S

In the formula (I), the compound is shown in the specification,

is the output of the pooling layer and,

is composed of

The element in (b), S, is the size of the pooled layer output.

And a third layer: and fully connecting the layers, inputting data output by the second layer, activating a function to select a ReLu function, performing dropout processing to avoid overfitting, setting the probability to be 0.5, outputting characteristic data with the output of 1 x 200, and setting the number of neurons to be 200.

A fourth layer: and in the output layer, a Softmax classifier is adopted to carry out secondary classification on the characteristic data output by the third layer, so that the eccentric wear prediction of the intelligent automobile tire is realized.

In the above model, the convolution operation is performed by multiplying and summing the data points and the corresponding positions of the convolution kernel, and the purpose of the convolution operation is to extract the features in the vibration signal.

The maximum value pooling is selected in the pooling operation, namely, a maximum value is selected between every two adjacent data values in the data of the upper layer based on the size of the pooling layer, the step length of the pooling layer is slightly lower than the size, so that the pooling operation parts are overlapped, the comprehensiveness of output signals is improved, a large amount of data is simplified, and overfitting is avoided to a certain extent.

dropout operation: the probability of the neuron is set to be 0.5, namely the probability of the neuron on the upper layer transmitting data to the neuron on the lower layer is set to be 0.5, so that the data on the lower layer does not depend on the data on the upper layer too much, the coupling relation between the neurons is reduced, and the overfitting phenomenon is greatly reduced.

The purpose of the Softmax classifier is to convert the values of the neurons of the output layer into a range between 0 and 1, while the sum of these neurons is 1. I.e., when the value of the last layer of neurons is

The formula of Softmax operation is as follows:

S(

)=

j=1,2, . . . , n

wherein S: (

) For the output of the Softmax classifier, n is the number of the last layer of neurons.

The loss function adopts a cross entropy function, and the purpose of the cross entropy function can lead the output probability of the network to tend to the probability of the target. For the classification labels of the present invention to be one-hot type vectors, and the number of output neurons to be greater than 1, the formula of the loss function can be expressed as:

L=-1*

wherein L is the loss function output,

is the value of the ith neuron.

And (4) storing the model obtained by the data training, and bringing the data of the verification set into the model for verification to obtain the accuracy and feasibility of the model.

The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent automobile tire eccentric wear prediction method based on a one-dimensional convolutional neural network is characterized by comprising the following steps:

2. The intelligent automobile tire eccentric wear prediction method based on the one-dimensional convolutional neural network as claimed in claim 1, characterized in that: the vibration signal data in step S1 is vibration signal data of the relationship between the radial acceleration at the rim center of the tire mounted with or without partial wear and time.

3. The intelligent automobile tire eccentric wear prediction method based on the one-dimensional convolutional neural network as claimed in claim 2, characterized in that: the vibration signal data in step S1 is obtained by modeling and simulating a tire that has been subjected to partial wear or a tire that has not been subjected to partial wear on a finite element analysis software by a "data twinning" method, and obtaining n signal data points of radial vibration at the center position of the rim through transient dynamics analysis.

4. The intelligent automobile tire eccentric wear prediction method based on the one-dimensional convolutional neural network as claimed in claim 3, characterized in that: and determining n signal data points of radial vibration of the center position of the wheel rim according to the frequency of the sampling points.

5. An intelligent automobile tire eccentric wear prediction method based on a one-dimensional convolutional neural network as claimed in claim 1, wherein in step S2, the same preprocessing is performed on two sets of vibration signal data during the driving process of the tire with eccentric wear and the tire without eccentric wear, the vibration signal data is intercepted to establish sample data corresponding to the two tires, and the sample intercepting method is as follows: and selecting L data points of the vibration signal data as the sample length, wherein the step length is S, and the vibration signal of each type comprises n data points, so that each type can obtain (n-L)/S +1 groups of samples, and the two groups of samples have 2 x ((n-L)/S +1) groups of samples.

6. The intelligent automobile tire eccentric wear prediction method based on the one-dimensional convolutional neural network as claimed in claim 5, wherein the method for normalizing the sample data in step S2 is as follows: in all the above-mentioned vibration signal data, the positive value is divided by the maximum value among the positive values, and the negative value is divided by the absolute value of the minimum value among the negative values, thereby normalizing the vibration signal data in all the above-mentioned samples to the range of (-1, 1).

7. The intelligent automobile tire eccentric wear prediction method based on the one-dimensional convolutional neural network as claimed in claim 6, wherein the method for labeling the samples in step S2 is as follows: the two types of samples are labeled with labels, and one-hot coding is carried out by utilizing a tensierflow frame, and the eccentric wear state is marked as [ 10 ] and the health state is marked as [ 01 ].

8. The intelligent automobile tire eccentric wear prediction method based on the one-dimensional convolutional neural network as claimed in claim 7, wherein the method for dividing the labeled samples into the training set X and the verification set C in step S2 is as follows: 80-90% of the number of 2X ((n-L)/S +1) samples were selected as training set X, and the remaining 10-20% were selected as validation set C.

9. The intelligent automobile tire eccentric wear prediction method based on the one-dimensional convolutional neural network as claimed in claim 1, wherein the method for establishing the tire eccentric wear prediction model in step S3 is as follows:

10. An intelligent automobile tire eccentric wear prediction method based on a one-dimensional convolutional neural network as claimed in claim 1, wherein the one-dimensional convolutional neural network initial model comprises: the system comprises two one-dimensional convolution layers, two one-dimensional pooling layers, a full-connection layer and a softmax classifier; the one-dimensional convolution layer and the one-dimensional pooling layer are arranged in a staggered mode.