Intelligent tire key state estimation method based on multi-sensor fusion

Abstract

The invention belongs to the field of tire dynamic characteristic application, and particularly relates to an intelligent tire key state estimation method based on multi-sensor fusion. According to the method, a PVDF sensor and an MEMS three-axis acceleration sensor are arranged on the inner wall of a tire, and an intelligent tire mechanical property test is carried out; according to encoder data and sensor signal characteristics, extracting a sensor signal of the intelligent tire grounding print area, and processing the sensor signal; establishing three different neural network models by taking different sensor signal characteristics as input; training neural network models by using sensor test data, and respectively solving optimal fusion weights of estimated values of the two neural network models according to training set data by using a particle swarm algorithm; and preliminarily estimating the key state of the intelligent tire by using a neural network model, and fusing two preliminary estimated values with the minimum difference value by using the optimal weight to obtain a final estimated value of the key state of the intelligent tire. The method plays an important role in improving the key state estimation precision of the intelligent tire.

Description

Technical Field

[0001] The present invention belongs to the field of application of tire dynamic characteristics, and particularly relates to a method for estimating key states of intelligent tires based on multi-sensor fusion. Background Art

[0002] With the development of automotive electrification and intelligence, intelligent electric vehicles have gradually become an important development direction in the automotive field today. Their vehicle state estimation and active safety control capabilities pose higher requirements for the state perception of the vehicle chassis. As a component that contacts and continuously acts on the road surface during vehicle driving, accurate estimation of the key states of tires is of great significance for improving the accuracy of vehicle chassis state perception and enhancing vehicle active safety.

[0003] Intelligent tire technology embeds sensors, energy supply devices, and microprocessors inside the tire. The sensors generate induction signals under the interaction between the tire and the road surface, and the microprocessor processes the induction signals according to the estimation algorithm to directly estimate the mechanical characteristics, motion states, and road condition information of the tire. However, there are dilemmas of serious coupling and integration errors in indirectly estimating the key states of tires based on vehicle dynamics models. Therefore, in order to improve the estimation accuracy of the key states of intelligent tires, a method for estimating key states of intelligent tires based on multi-sensor fusion has been developed.

[0004] The invention patent with the patent number ZL201910065646.0 discloses a method for identifying vertical wheel forces based on multi-sensor information fusion. This method can only realize the real-time identification of vertical wheel forces of tires under normal vehicle driving by using the tire pressure, rotational speed, and vertical deformation information of tire movement. However, this patent mainly uses the PVDF sensor signals and MEMS triaxial acceleration sensor signals of tire movement to realize the estimation of multiple key states of intelligent tires, including tire vertical force, lateral force, longitudinal force, sideslip angle, and slip ratio. Summary of the Invention

[0005] To solve the above technical problems, the present invention provides a method for estimating key states of intelligent tires based on multi-sensor fusion, which is characterized in that it uses multiple PVDF (Polyvinylidene Fluoride) sensor signals and MEMS (Micro-Electro-Mechanical System) triaxial acceleration sensor signals to realize the estimation of key states of intelligent tires, thereby improving the estimation accuracy of key states of intelligent tires.

[0006] To achieve the above object, the technical solution adopted by the present invention is as follows: Step 1: Arrange PVDF sensors and MEMS triaxial acceleration sensors on the inner wall of the tire. Install the tire on the tire six-component force test machine, set the test conditions according to the test requirements, and conduct intelligent tire mechanical property tests under different working conditions to obtain PVDF sensor test data, MEMS triaxial acceleration sensor test data, and intelligent tire key state test data; Step 2: According to the encoder data and sensor signal characteristics, extract the PVDF sensor signal and MEMS triaxial acceleration sensor signal in the contact patch area of the intelligent tire respectively. Perform different processing on the MEMS triaxial acceleration sensor signal to obtain the time-domain characteristics of the PVDF sensor signal, the time-domain characteristics and frequency-domain characteristics of the MEMS triaxial acceleration sensor signal in the contact patch area of the intelligent tire; Step 3: Taking the PVDF sensor signal characteristics and MEMS triaxial acceleration sensor signal characteristics in the contact patch area of the intelligent tire as inputs and the intelligent tire key state as the output, establish neural network model 1. Taking the PVDF sensor signal characteristics in the contact patch area of the intelligent tire as the input and the intelligent tire key state as the output, establish neural network model 2. Taking the MEMS triaxial acceleration sensor signal characteristics in the contact patch area of the intelligent tire as the input and the intelligent tire key state as the output, establish neural network model 3; Step 4: Use the processed PVDF sensor test data and MEMS triaxial acceleration sensor test data to train the three neural network models, and adopt the particle swarm optimization algorithm to solve the optimal fusion weights of the estimated values of the two neural network models according to the training set data respectively; Step 5: Use the trained three neural network models to preliminarily estimate the intelligent tire key state, and fuse the two preliminary estimated values of the intelligent tire key state with the smallest difference using the optimal weights to obtain the final estimated value of the intelligent tire key state.

[0007] Preferably, in Step 1, a PVDF sensor and a MEMS triaxial acceleration sensor are longitudinally arranged at the middle position on the inner wall of the tire, and a PVDF sensor is longitudinally arranged at each of the two side positions.

[0008] Preferably, in Step 2, the PVDF sensor signal for one full rotation of the tire is extracted according to the encoder data, and the PVDF sensor signal in the contact patch area of the intelligent tire is extracted according to the extreme value characteristics of the sensor signal to obtain its time-domain characteristics.

[0009] Preferably, in step two, the MEMS triaxial acceleration sensor signals for one tire revolution are extracted from the encoder data, and the MEMS triaxial acceleration sensor signals in the intelligent tire contact patch area are extracted according to the extreme value characteristics of the sensor signals. Digital filtering is performed on the vertical acceleration signal and the longitudinal acceleration signal to obtain their time-domain characteristics, and time-frequency conversion is performed on the lateral acceleration signal to obtain its frequency-domain characteristics.

[0010] Preferably, in step three, neural network model 1 and neural network model 2 adopt the BP neural network model (BackPropagation Neural Network), including an input layer, two hidden layers, and an output layer, and the mean square error function is used as the loss function.

[0011] Preferably, in step three, neural network model 3 adopts the CNN neural network model (Convolutional Neural Network), including an input layer, a convolutional layer, a normalization layer, an activation layer, a pooling layer, a fully connected layer, and an output layer, and the mean square error function is used as the loss function.

[0012] Preferably, the key states of the intelligent tire in step three include tire vertical force, lateral force, longitudinal force, sideslip angle, and slip ratio.

[0013] Preferably, the key states of the intelligent tire in step five include tire vertical force, lateral force, longitudinal force, sideslip angle, and slip ratio.

[0014] Advantages of the present invention: 1. The method disclosed in the present invention realizes the direct estimation of the key states of the intelligent tire by using the in-tire sensor signals, overcoming the integration error problem existing in the indirect estimation of the key states of the tire based on the vehicle dynamics model.

[0015] 2. The method disclosed in the present invention arranges multiple PVDF sensors and MEMS triaxial acceleration sensors inside the tire, and realizes the estimation of the key states of the intelligent tire by collecting multi-dimensional tire motion signals, improving the estimation accuracy of the key states of the intelligent tire.

[0016] 3. The method disclosed in the present invention realizes the estimation of the key states of the intelligent tire by fusing the information of multiple in-tire sensors, and has high robustness. Description of the drawings

[0017] Figure 1 It is a flow chart of the method for estimating the key states of an intelligent tire based on multi-sensor fusion.

[0018] Figure 2 It is the PVDF sensor signal collected in the intelligent tire mechanical property test in the embodiment.

[0019] Figure 3 The filtered MEMS triaxial acceleration sensor signals collected during the mechanical property tests of the intelligent tire in the embodiments.

[0020] Figure 4 A comparison example between the estimated value of the vertical force of the intelligent tire and the true value of the tire test under the pure cornering condition in the embodiments.

[0021] Figure 5 A comparison example between the estimated value of the lateral force of the intelligent tire and the true value of the tire test under the pure cornering condition in the embodiments.

[0022] Figure 6 A comparison example between the estimated value of the slip angle of the intelligent tire and the true value of the tire test under the pure cornering condition in the embodiments.

[0023] Figure 7 A comparison example between the estimated value of the vertical force of the intelligent tire and the true value of the tire test under the pure longitudinal slip condition in the embodiments.

[0024] Figure 8 A comparison example between the estimated value of the longitudinal force of the intelligent tire and the true value of the tire test under the pure longitudinal slip condition in the embodiments.

[0025] Figure 9 A comparison example between the estimated value of the slip ratio of the intelligent tire and the true value of the tire test under the pure longitudinal slip condition in the embodiments. Detailed implementation manners

[0026] Please refer to Figures 1 to 9 , which is an embodiment of the present invention.

[0027] Taking a 245 / 45 R19 specification radial tire of a passenger car as an example, a PVDF sensor and a MEMS triaxial acceleration sensor are longitudinally arranged at the middle position of the inner wall of the tire, and a PVDF sensor is longitudinally arranged at each of the two side positions. The tire is installed on an MTS Flat-Trac high-speed tire test bench. The signal sampling frequency of the PVDF sensor is set to 695 Hz, and the signal sampling frequency of the MEMS triaxial acceleration sensor is set to 3475 Hz. The mechanical property tests of the intelligent tire are carried out under the pure cornering condition and the pure longitudinal slip condition to obtain the test data of the PVDF sensor, the test data of the MEMS triaxial acceleration sensor, and the test data of the key states of the intelligent tire.

[0028] The test conditions for the pure cornering condition are set as follows: (1) The tire pressure is set to 250 kPa; (2) The tire side inclination angle is set to 0°; (3) The test speeds are respectively set to 30 km / h, 60 km / h, and 90 km / h; (4)The vertical loads of the tire are set to 2000 N, 4000 N, and 6000 N respectively; (5)For the tire sideslip angle, the quasi-steady-state sweep method is adopted. It sweeps from the starting point of 0° to 10° at a rate of 3° / s, then sweeps to -10° at the same rate, and finally returns to 0°.

[0029] The test conditions for the pure longitudinal slip condition are set as follows: (1)The tire pressure is set to 250 kPa; (2)The tire sideslip angle is set to 0°; (3)The test speeds are set to 30 km / h, 60 km / h, and 90 km / h respectively; (4)The vertical loads of the tire are set to 500 N and 2000 N respectively; (5)For the tire slip ratio, the quasi-steady-state sweep method is adopted. It sweeps from the starting point of 0% to 30% at a rate of 7.5% / s, then sweeps to -30% at the same rate, and finally returns to 0%.

[0030] Extract the PVDF sensor signals for one tire rotation according to the encoder data. Extract the PVDF sensor signals in the grounding contact area of the intelligent tire according to the extreme value characteristics of the PVDF sensor signals, and obtain the time-domain characteristics of the PVDF sensor signals in the grounding contact areas of the three intelligent tires.

[0031] Extract the MEMS triaxial acceleration sensor signals for one tire rotation according to the encoder data. Extract the MEMS triaxial acceleration sensor signals in the grounding contact area of the intelligent tire according to the extreme value characteristics of the sensor signals. Perform digital filtering on the vertical acceleration signal and the longitudinal acceleration signal, with the cut-off frequency set to 400 Hz, to obtain the time-domain characteristics of the vertical acceleration signal and the longitudinal acceleration signal in the grounding contact area of the intelligent tire. Perform time-frequency conversion on the lateral acceleration signal to obtain the frequency-domain characteristics of the lateral acceleration signal in the grounding contact area of the intelligent tire.

[0032] Establish a BP neural network model 1. The neural network model includes an input layer, two hidden layers, and an output layer. The input layer consists of the time-domain characteristics of three PVDF sensor signals, the time-domain characteristics of the vertical acceleration signal of the MEMS triaxial acceleration sensor, the time-domain characteristics of the longitudinal acceleration signal, and the frequency-domain characteristics of the lateral acceleration signal. The first hidden layer has 240 neurons, and the activation function is the Sigmoid function. The second hidden layer has 120 neurons, and the activation function is the Sigmoid function. In the pure sideslip condition, the output layer is the vertical force, lateral force, and sideslip angle of the intelligent tire. In the pure longitudinal slip condition, the output layer is the vertical force, longitudinal force, and slip ratio of the intelligent tire. The loss function uses the mean square error function.

[0033] Build a BP neural network model 2, which includes an input layer, two hidden layers, and an output layer. The input layer consists of the time-domain features of three PVDF sensor signals. The first hidden layer has 160 neurons with the Sigmoid function as the activation function, and the second hidden layer has 80 neurons with the Sigmoid function as the activation function. In the pure cornering condition, the output layer is the vertical force, lateral force, and slip angle of the intelligent tire. In the pure longitudinal slip condition, the output layer is the vertical force, longitudinal force, and slip ratio of the intelligent tire. The loss function uses the mean square error function.

[0034] Build a CNN neural network model 3, which includes an input layer, a convolutional layer, a normalization layer, an activation layer, a pooling layer, a fully connected layer, and an output layer. The input layer consists of the time-domain features of the vertical acceleration signal, longitudinal acceleration signal, and the frequency-domain features of the lateral acceleration signal of the MEMS triaxial acceleration sensor. The ReLU function is used in the activation function layer, and the max pooling layer is used in the pooling layer. In the pure cornering condition, the output layer is the vertical force, lateral force, and slip angle of the intelligent tire. In the pure longitudinal slip condition, the output layer is the vertical force, longitudinal force, and slip ratio of the intelligent tire. The loss function uses the mean square error function.

[0035] Use the processed PVDF sensor test data and MEMS triaxial acceleration sensor test data to train the three neural network models, and adopt the particle swarm optimization algorithm to solve the optimal fusion weights of the estimated values of the two neural network models according to the training set data respectively.

[0036] Use the trained neural network models to preliminarily estimate the key states of the intelligent tire, and fuse the two preliminary estimated values of the key states of the intelligent tire with the smallest difference using the optimal weights to obtain the final estimated values of the key states of the intelligent tire. The key states of the intelligent tire include vertical force, lateral force, longitudinal force, slip angle, and slip ratio.

[0037] The test results of this embodiment are as follows: Figure 2 PVDF sensor signals collected for the mechanical property test of the intelligent tire; Figure 3 Filtered MEMS triaxial acceleration sensor signals collected for the mechanical property test of the intelligent tire; Figure 4 A comparison example of the estimated value and the true value of the vertical force of the intelligent tire in the pure cornering condition; Figure 5 A comparison example of the estimated value and the true value of the lateral force of the intelligent tire in the pure cornering condition; Figure 6 A comparison example of the estimated value and the true value of the slip angle of the intelligent tire in the pure cornering condition; Figure 7 A comparison example of the estimated value and the true value of the vertical force of the intelligent tire in the pure longitudinal slip condition; Figure 8 A comparison example of the estimated value and the true value of the longitudinal force of the intelligent tire in the pure longitudinal slip condition; Figure 9It is a comparison example of the estimated value of the intelligent tire slip ratio and the true value of the tire test under the pure longitudinal slip condition.

Claims

1. A method for estimating key states of an intelligent tire based on multi-sensor fusion, characterized in that: The following steps are involved: Step 1: Arrange PVDF sensors and MEMS triaxial acceleration sensors on the inner wall of the tire, install the tire on a tire six-component force testing machine, set test conditions according to test requirements, conduct smart tire mechanical property tests under different working conditions, and obtain PVDF sensor test data, MEMS triaxial acceleration sensor test data, and smart tire key state test data; Step 2: According to the encoder data and the sensor signal characteristics, the PVDF sensor signal and the MEMS three-axis acceleration sensor signal of the smart tire contact patch area are extracted respectively, and the MEMS three-axis acceleration sensor signal is processed differently to obtain the PVDF sensor signal time domain characteristics, the MEMS three-axis acceleration sensor signal time domain characteristics and the frequency domain characteristics of the smart tire contact patch area respectively; Step 3, using the PVDF sensor signal characteristics and the MEMS triaxial acceleration sensor signal characteristics of the smart tire contact patch area as inputs and the smart tire key state as outputs, a neural network model 1 is established, using the PVDF sensor signal characteristics of the smart tire contact patch area as inputs and the smart tire key state as outputs, a neural network model 2 is established, and using the MEMS triaxial acceleration sensor signal characteristics of the smart tire contact patch area as inputs and the smart tire key state as outputs, a neural network model 3 is established; Step 4: Use the processed PVDF sensor test data and MEMS triaxial acceleration sensor test data to train three neural network models, and use the particle swarm algorithm to solve the optimal fusion weights of the estimated values ​​of the two neural network models according to the training set data; Step 5: Use the three trained neural network models to preliminarily estimate the key states of the smart tires, and fuse the two preliminary estimated values ​​of the key states of the smart tires with the smallest difference using the optimal weights to obtain the final estimated value of the key states of the smart tires.

2. The method for estimating key states of an intelligent tire based on multi-sensor fusion according to claim 1, characterized in that: In the step 1, a PVDF sensor and a MEMS three-axis acceleration sensor are longitudinally arranged in the middle of the inner wall of the tire, and a PVDF sensor is longitudinally arranged at each of the two side positions.

3. The method for estimating key states of an intelligent tire based on multi-sensor fusion according to claim 1, characterized in that: The second step extracts the PVDF sensor signal of the tire rolling one circle according to the encoder data, and extracts the PVDF sensor signal of the smart tire contact patch area according to the extreme value characteristics of the sensor signal to obtain its time domain characteristics.

4. The method for estimating key states of an intelligent tire based on multi-sensor fusion according to claim 1, characterized in that: The second step extracts the MEMS three-axis acceleration sensor signal of the tire rolling one circle according to the encoder data, extracts the MEMS three-axis acceleration sensor signal of the smart tire contact patch area according to the extreme value characteristics of the sensor signal, performs digital filtering on the vertical acceleration signal and the longitudinal acceleration signal to obtain their time domain characteristics, The lateral acceleration signal is processed by time-frequency conversion to obtain its frequency domain characteristics.

5. The method for estimating key states of an intelligent tire based on multi-sensor fusion according to claim 1, characterized in that: In step three, neural network model 1 and neural network model 2 adopt BP neural network model, and neural network model 3 adopts CNN neural network model.

6. The method for estimating key states of an intelligent tire based on multi-sensor fusion according to claim 1, characterized in that: The key states of the smart tire in step three and step five include tire vertical force, lateral force, longitudinal force, sideslip angle and slip rate.

7. Application of a method for estimating key states of an intelligent tire based on multi-sensor fusion according to any one of claims 1 to 6 in estimating mechanical characteristics and motion states of intelligent tires under pure sideslip conditions and pure longitudinal slip conditions.

Images