A data-driven method for automobile component signal estimation

Abstract

A method for signal estimation of automobile parts based on data drive comprises the following steps: step 1: training an offline data model, including a main model family and a secondary model family; step 2: completing online signal estimation according to the secondary model family, comparing the signal estimation result, and selecting a data model in the secondary model family; step 3: screening out the corresponding data model in the main model family according to the data model in the secondary model family in step 2; step 4: completing online signal estimation through the data model in the main model family screened in step 3; the present invention sets the main-secondary model, judges the working condition through the secondary model, enables the main model to accurately predict the unknown output, saves cost, and reduces the maintenance difficulty.

Description

Technical Field

[0001] The invention relates to the field of automobile sensors, and in particular to a method for estimating automobile component signals based on data driving. Background Art

[0002] At present, the popularity of automobiles is very high, and almost every household owns at least one car. As the number of users gradually increases, more and more attention is paid to automobile safety issues. Therefore, a series of tests will be carried out before the automobile is put into production, including vehicle component health monitoring. In the existing technology, in order to accurately obtain the operating status and component status of the tested vehicle, 40 to 50 sensors need to be installed on the body and chassis, and the fatigue analysis of vehicle components is performed through the information collected by the sensors. Among these sensors, there are many expensive sensor types, especially sensors for measuring wheel center force and wheel center torque. Since a large number of samples need to be collected, hundreds or even hundreds of cars need to be tested. If sensors are arranged for each car, it will cost a lot of money; if the sensors are installed and used repeatedly, it will be very cumbersome.

[0003] On the other hand, for a tested vehicle, arranging 40 to 50 sensors requires a lot of effort from the operator, and it also requires a lot of effort and resources for maintenance. In order to solve the above problems, a method is needed to establish a data model and use existing parameters to predict other parameters of the vehicle, so as to reduce the cost and complexity of installation and maintenance. Summary of the invention

[0004] The purpose of the present invention is to solve the deficiencies of the prior art and provide a method for estimating automobile component signals based on data-driven, which can predict the status of other components through the collected information, with accurate prediction and effective cost reduction.

[0005] A method for estimating automobile component signals based on data drive comprises the following steps:

[0006] Step 1: Train offline data models, including the main model family and the secondary model family;

[0007] Step 2: According to the sub-model family, complete the online signal estimation, compare the signal estimation results, and select the data model in the sub-model family;

[0008] Step 3: According to the data models in the secondary model family in step 2, filter out the data models in the corresponding main model family;

[0009] Step 4: Complete the online signal estimation through the data model in the main model family selected in step 3.

[0010] Furthermore, the training of the offline data model in step 1 includes the following steps:

[0011] Step 1.1: Install sensors on the vehicle under test and confirm the pseudo input sensor group and output sensor group;

[0012] Step 1.2: Collect responses of the pseudo input sensor group and the output sensor group of the tested vehicle under different working conditions;

[0013] Step 1.3: Establish a data model between the input sensor group response and the output sensor group response based on a data-driven approach.

[0014] Furthermore, the sensors in step 1.1 include force sensors, acceleration sensors, displacement sensors and strain gauges, and the types of sensors of the pseudo input sensor group and the output sensor group are specified and confirmed by the operator.

[0015] Furthermore, the different operating conditions in step 1.2 include different road conditions and different operating behaviors.

[0016] Furthermore, the step 1.3 establishes a data model between the pseudo input sensor group response and the output sensor group response, which specifically includes the following steps:

[0017] Step 1.3.1: Analyze the relationship between the pseudo input sensor group and the output sensor group, and establish the main model family;

[0018] Step 1.3.2: Divide the pseudo input sensor group into two parts, namely, a secondary pseudo input sensor group and a secondary pseudo output sensor group;

[0019] Step 1.3.3: Analyze the relationship between the secondary pseudo input sensor group and the secondary output sensor group, and establish a secondary model family;

[0020] In step 1.3.1, the excitation of the road surface to the vehicle is first set as u, and the response information collected by the input sensor is used as the pseudo input y I , the response information collected by the output sensor is used as output y O ; Establish a I to y O The main model family G includes K data models, namely G1, G2, ..., G K , where y I,1 ,y I,2 ,…,y I,K and O,1 ,y O,2 ,…,y O,K are the pseudo input and output under K different working conditions respectively;

[0021] In step 1.3.2, the pseudo input sensor group is divided into two parts, namely, y I1 and I2 , where y I1 and I2 The type of sensor is specified by the operator or randomly assigned;

[0022] In step 1.3.3, K different working conditions are established from y I1 to y I2 The sub-model family H is H1, H2, ..., H K , where y I1,1 ,y I1,2 ,…,y I1,K and I2,1 ,y I2,2 ,…,y I2,K are the pseudo input and output under K different working conditions respectively;

[0023] The K operating conditions of the main model family correspond one to one with the K operating conditions of the secondary model family.

[0024] Furthermore, we first collect new pseudo input as y I ',y I The type of sensor in ' and the pseudo input y in building the main model family I The sensor type is consistent; based on the sub-model family H obtained offline, a classifier is constructed to determine the working condition, where the operation of the classifier includes the following steps:

[0025] Step 2.1: Input pseudo I 'Split into y I1 ' and y I2 '; where y I1 'Sensor type and y I1 The sensor type is the same, y I2 'Sensor type and y I2 The sensor types are consistent;

[0026] Step 2.2: Set y I1 'As input to the sub-model family H, we get the estimated value set

[0027] Step 2.3: Compare y I2 ' and the estimated value group similarity, filter out the closest estimate K*∈(1,K).

[0028] Furthermore, the data model selected in the main model family is G K* ; Both the main model family and the secondary model family adopt a multi-input and multi-output finite impulse response model structure.

[0029] Furthermore, to complete the similarity comparison in step 2.3, it is first necessary to verify y I2 'The distribution law of , the specific verification process is as follows:

[0030]

[0031] where B0,…,B n represents the model parameters of a random model in the sub-model family, represents the estimated value of the data model, ε(t) represents the residual at time t; Simplifying formula (1) to obtain

[0032]

[0033] Where Φ(t) represents the regression matrix, θ is the model parameter vector; assuming that ε(t) obeys a Gaussian distribution with a mean of zero and a variance of Σ, is the following Gaussian distribution:

[0034]

[0035] in and is the estimated value of the corresponding parameter, which is obtained by the maximum likelihood estimation method during the data model training process.

[0036] Furthermore, in step 2.3, the closest estimator is selected. By designing a Bayesian classifier, based on the FIR model structure, the posterior probability is calculated as follows:

[0037]

[0038] Where p(H j ) represents the prior probability assigned to the jth data model in the sub-model family H; express The probability density at time t; The expression is:

[0039]

[0040] in, are the model parameters and variance of the j-th data model, and s is the output dimension;

[0041] By obtaining the maximum posterior probability, we select the appropriate data model in the sub-model family H, as shown below:

[0042]

[0043] Where K* represents the operating condition number corresponding to the selected data model; argmax and argmin represent a function respectively.

[0044] Furthermore, the prior probability assigned to each data model in the sub-model family H is

[0045] The beneficial effects of the present invention are:

[0046] The present invention sets a primary-secondary model and uses the secondary model to judge the working condition, so that the primary model can accurately predict the unknown output;

[0047] By setting up the primary-secondary model, prediction is completed, and the response of unknown sensors can be accurately predicted based on the response of existing sensors, saving costs and reducing maintenance difficulty. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a pseudo input-output analysis diagram of the main model of the first embodiment of the present invention;

[0049] Figure 2 is the main model family under K working conditions of the first embodiment of the present invention;

[0050] Figure 3 It is a pseudo input-output analysis diagram of the main model and the secondary model of the first embodiment of the present invention;

[0051] Figure 4 is a sub-model family under K working conditions of the first embodiment of the present invention;

[0052] Figure 5 This is an online estimation flow chart of the first embodiment of the present invention;

[0053] Figure 6 This is a working principle diagram of the classifier according to the first embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0055] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and thus the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0056] Embodiment 1:

[0057] A method for estimating automobile component signals based on data drive comprises the following steps:

[0058] Step 1: Train offline data models, including the main model family and the secondary model family;

[0059] Step 2: According to the sub-model family, complete the online signal estimation, compare the signal estimation results, and select the data model in the sub-model family;

[0060] Step 3: According to the data models in the secondary model family in step 2, filter out the data models in the corresponding main model family;

[0061] Step 4: Complete the online signal estimation through the data model in the main model family selected in step 3.

[0062] The training of the offline data model in step 1 includes the following steps:

[0063] Step 1.1: Install sensors on the vehicle under test and confirm the pseudo input sensor group and output sensor group;

[0064] Step 1.2: Collect responses of the pseudo input sensor group and the output sensor group of the tested vehicle under different working conditions;

[0065] Step 1.3: Establish a data model between the input sensor group response and the output sensor group response based on a data-driven approach.

[0066] like Figure 1 As shown, the sensors in step 1.1 include force sensors, acceleration sensors, displacement sensors, strain gauges, etc., and the types of sensors of the pseudo input sensor group and the output sensor group are specified and confirmed by the operator. Because in fact, the pseudo input sensor group and the output sensor group are both output responses to the vehicle status, but in this embodiment, the pseudo input sensor group establishes a data model to the output sensor group, so it is called a pseudo input sensor group.

[0067] The different working conditions in step 1.2 include different road conditions and different operating behaviors, wherein the road conditions include Belgian block roads, smooth stone roads, asphalt roads, etc.; the operating behaviors include acceleration, turning, braking, etc.

[0068] The step 1.3 establishes a data model between the pseudo input sensor group response and the output sensor group response, specifically comprising the following steps:

[0069] Step 1.3.1: Analyze the relationship between the pseudo input sensor group and the output sensor group, and establish the main model family;

[0070] Step 1.3.2: Divide the pseudo input sensor group into two parts, namely, a secondary pseudo input sensor group and a secondary pseudo output sensor group;

[0071] Step 1.3.3: Analyze the relationship between the secondary pseudo input sensor group and the secondary output sensor group, and establish a secondary model family.

[0072] like Figure 2 As shown, in step 1.3.1, the excitation of the road surface to the vehicle is first set as u, and the response information collected by the input sensor is used as the pseudo input y I , the response information collected by the output sensor is used as output y O , after training, a data model is established to reflect the relationship between the pseudo input sensor group and the output sensor group. Due to different working conditions, multiple data models can be established. In this embodiment, K different working conditions are established from y I to y O The main model family G includes K data models, namely G1, G2, ..., G K , where y I,1 ,y I,2 ,…,y I,K and O,1 ,y O,2 ,…,y O,K are the pseudo input and output under K different working conditions respectively.

[0073] like Figure 3 As shown, in step 1.3.2, the pseudo input sensor group is divided into two parts, namely y I1 and I2 , where y I1 and I2 The type of sensor is specified by the operator or randomly assigned.

[0074] like Figure 4 As shown, in step 1.3.3, K different working conditions are established from y I1 to y I2 The sub-model family H is H1, H2, ..., H K , where y I1,1 ,y I1,2 ,…,y I1,K and I2,1 ,y I2 ,2,…,yI2,K are the pseudo inputs and outputs under K different working conditions. It should be noted that the K working conditions of the main model family correspond one to one with the K working conditions of the secondary model family.

[0075] like Figure 5 , 6 As shown, the online signal estimation in step 2 requires the collection of new pseudo input y I ', it should be noted that y I The type of sensor in ' and the pseudo input y in building the main model family I The sensor type is consistent. In order to obtain the unknown output y O ', firstly, based on the sub-model family H obtained offline, a classifier is constructed to judge the working condition, wherein the operation of the classifier includes the following steps:

[0076] Step 2.1: Input pseudo I 'Split into y I1 ' and y I2 '; where y I1 'Sensor type and y I1 The sensor type is the same, y I2 'Sensor type and y I2 The sensor types are consistent;

[0077] Step 2.2: Set y I1 'As input to the sub-model family H, we get the estimated value set

[0078] Step 2.3: Compare y I2 ' and the estimated value group similarity, filter out the closest estimate K*∈(1,K).

[0079] In step 2.3, the multi-input multi-output finite impulse response (FIR) is used as the structure of the data model in the sub-model family H. In order to complete the similarity comparison, it is first necessary to verify y I2 'The distribution law of , the specific verification process is as follows:

[0080]

[0081] where B0,…,B n A model parameter representing a data model in a sub-model family, represents the estimated value of the data model, ε(t) represents the residual at time t; Indicates that at time t The value of . Simplifying formula (1) gives

[0082]

[0083] Where Φ(t) represents the regression matrix, which is expressed as represents the tensor product, The expression is I is the unit matrix. θ is the model parameter vector, which is expressed as θ=vec([B0(t)…B n (tn)] T ), where vec(·) represents a vectorized operation. Assuming that ε(t) follows a Gaussian distribution with a mean of zero and a variance of Σ, it is clear that is the following Gaussian distribution:

[0084]

[0085] in and is the estimated value of the corresponding parameter, which is obtained by the maximum likelihood estimation method during the data model training process.

[0086] Based on the FIR model structure, a Bayesian classifier is designed to calculate the posterior probability as shown below:

[0087]

[0088] Where p(H j ) represents the prior probability assigned to the jth data model in the sub-model family H, j∈(1,K), express The probability density at time t; if there are no other interfering factors, then the prior probability of each data model in the sub-model family H is consistent, so The expression is:

[0089]

[0090] in, are the model parameters and variance of the jth data model, and s is the output dimension. By calculating the maximum value of the posterior probability, the appropriate data model in the sub-model family H is selected as follows:

[0091]

[0092] Where K* represents the operating condition number corresponding to the selected data model, argmax and argmin represent a function respectively; the estimated value

[0093] In step 3, according to the above calculation, the working condition corresponding to the maximum posterior probability in the secondary model family is K*, so the data model G with the same working condition is selected in the main model family.K* .

[0094] The data model of the main model family in step 4 is also a multi-input and multi-output finite impulse response (FIR) model structure.

[0095] During the implementation process, the working condition is judged by the secondary model family, and then the data model of the same working condition in the trained main model family is used to complete the estimation of the unknown output to improve the accuracy. Through the above estimation, the response of more sensors can be predicted by building a data model through the response of limited sensors, saving costs and facilitating maintenance.

[0096] The above description is only a specific example of the present invention and does not constitute any limitation to the present invention. It is obvious that for professionals in this field, after understanding the content and principle of the present invention, various modifications and changes in form and details may be made without departing from the principle and structure of the present invention, but these modifications and changes based on the idea of ​​the present invention are still within the scope of protection of the claims of the present invention.

Claims

1. A method for estimating automobile component signals based on data-driven, characterized in that: The steps include: Step 1: training an offline data model, including a main model family and a secondary model family. The training of the offline data model in step 1 includes the following steps: Step 1.1: Install sensors on the tested vehicle and confirm the pseudo input sensor group and output sensor group; Step 1.2: Collect responses of the pseudo input sensor group and the output sensor group of the tested vehicle under different working conditions; Step 1.3: Establish a data model between the input sensor group response and the output sensor group response based on a data-driven approach, which specifically includes the following steps: Step 1.3.1: Analyze the relationship between the pseudo input sensor group and the output sensor group, and establish the main model family; Step 1.3.2: Divide the pseudo input sensor group into two parts, namely, a secondary pseudo input sensor group and a secondary pseudo output sensor group; Step 1.3.3: Analyze the relationship between the secondary pseudo input sensor group and the secondary output sensor group, and establish a secondary model family; in the step 1.3.1, first set the road surface excitation to the vehicle as u, and the response information collected by the input sensor as the pseudo input y I , the response information collected by the output sensor is used as output y O ; Establish a I to y O The main model family G includes K data models, namely G1, G2, ..., G K , where y I,1 ,y I,2 ,…,y I,K and O,1 ,y O,2 ,…,y O,K are the pseudo input and output under K different working conditions respectively; In step 1.3.2, the pseudo input sensor group is divided into two parts, namely, y I1 and I2 , where y I1 and I2 The type of sensor is specified by the operator or randomly assigned; In step 1.3.3, K different working conditions are established from y I1 to y I2 The sub-model family H is H1, H2, ..., H K , where y I1,1 ,y I1,2 ,…,y I1,K and I2,1 ,y I2,2 ,…,y I2,K are the pseudo input and output under K different working conditions respectively; The K working conditions of the main model family correspond one to one with the K working conditions of the secondary model family; Step 2: According to the sub-model family, complete the online signal estimation, compare the signal estimation results, and select the data model in the sub-model family; Step 3: According to the data models in the secondary model family in step 2, filter out the data models in the corresponding main model family; Step 4: Complete the online signal estimation through the data model in the main model family selected in step 3.

2. The method for estimating automobile component signals based on data drive according to claim 1, characterized in that: The sensors in step 1.1 include force sensors, acceleration sensors, displacement sensors and strain gauges, and the types of sensors of the pseudo input sensor group and the output sensor group are specified and confirmed by the operator.

3. The method for estimating automobile component signals based on data drive according to claim 2, characterized in that: The different working conditions in step 1.2 include different road conditions and different operating behaviors.

4. The method for estimating automobile component signals based on data drive according to claim 1, characterized in that: In step 2, the online signal estimation is firstly to collect a new pseudo input as y I ',y I The type of sensor in ' and the pseudo input y in building the main model family I The sensor type is consistent; based on the sub-model family H obtained offline, a classifier is constructed to determine the working condition, where the operation of the classifier includes the following steps: Step 2.1: Input pseudo I 'Split into y I1 ' and y I2 '; where y I1 'Sensor type and y I1 The sensor type is the same, y I2 'Sensor type and y I2 The sensor types are consistent; Step 2.2: Set y I1 'As input to the sub-model family H, we get the estimated value set Step 2.3: Compare y I2 ' and the estimated value group similarity, filter out the closest estimate 5. The method for estimating automobile component signals based on data drive according to claim 4, characterized in that: The data model selected in the main model family is G K* ; Both the main model family and the secondary model family adopt a multi-input and multi-output finite impulse response model structure.

6. The method for estimating automobile component signals based on data drive according to claim 5, characterized in that: In step 2.3, the similarity comparison is completed. First, it is necessary to verify y I2 'The distribution law of , the specific verification process is as follows: where B0,…,B n A model parameter representing a data model in a sub-model family, represents the estimated value of the data model, ε(t) represents the residual at time t; Simplifying formula (1) to obtain Where φ(t) represents the regression matrix, θ is the model parameter vector; assuming that ε(t) obeys a Gaussian distribution with a mean of zero and a variance of Σ, is the following Gaussian distribution: in and is the estimated value of the corresponding parameter, which is obtained by the maximum likelihood estimation method during the data model training process.

7. The method for estimating automobile component signals based on data drive according to claim 6, characterized in that: In step 2.3, the closest estimator is selected By designing a Bayesian classifier, based on the FIR model structure, the posterior probability is calculated as follows: Where p(H j ) represents the prior probability assigned to the jth data model in the sub-model family H; express The probability density at time t; The expression is: in, are the model parameters and variance of the j-th data model, and s is the output dimension; By calculating the maximum value of the posterior probability, the appropriate data model in the sub-model family H is selected as follows: Where K* represents the operating condition number corresponding to the selected data model; argmax and argmin represent a function respectively.

8. The method for estimating automobile component signals based on data drive according to claim 7, characterized in that: The prior probability assigned to each data model in the sub-model family H is

Images