Road roughness level recognition method and system considering vehicle speed
Through the neural network method, road surface response at different vehicle speeds is simulated and features are extracted, which solves the problem that the speed factor in the existing technology is not considered, and the rapid and accurate identification of road surface unevenness levels is achieved, which improves recognition accuracy and real-time recognition capabilities.
Patent Information
- Application Number
- CN202210358079.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-04-06
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2042-04-06
AI Technical Summary
The prior art fails to effectively consider vehicle speed factors in the identification of road surface unevenness levels, resulting in low recognition accuracy. Most existing methods are used in laboratories, making it difficult to achieve real-time identification.
By simulating the road surface response at different speeds, the characteristics of the suspension travel and body acceleration signals are extracted, and the characteristic sample sets within the speed range are established to train the neural network, thereby achieving rapid road surface unevenness levels. Accurate identification.
It improves the accuracy and recognition speed of road unevenness level recognition, no sensors are added, the system is simple and reliable, and is suitable for real-time recognition.
Smart Images

Figure CN114818120B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of neural networks, and in particular to a method and system for identifying road surface roughness levels taking vehicle speed into consideration. Background Art
[0002] Road surface roughness is the main source of excitation for vehicle motion, and has a significant impact on ride comfort, ride safety, vehicle maneuverability, and vehicle dynamic load. Road roughness models have been widely studied and applied to vehicle dynamics and other research. The vehicle suspension system has the function of buffering the impact of road roughness on the vehicle body caused by the unevenness of the road during driving. The premise of improving vehicle ride comfort and driving safety through precise control of active suspension is to quickly and accurately identify the road surface roughness level. Accurately identifying the road surface roughness level is of great significance to improving the control performance of the active suspension system, as well as the practicality and authenticity of vehicle dynamic simulation.
[0003] In the field of road roughness grade recognition, domestic and foreign experts have conducted a lot of theoretical research and experimental verification. The road power spectrum density is usually used to describe the statistical characteristics of road roughness, and road roughness is usually divided into 8 levels from A to H. The existing road grade recognition methods mostly use signals such as vehicle body acceleration, wheel dynamic load, and suspension dynamic travel signals. Few people consider the impact of vehicle speed on the road roughness grade recognition results. The dynamic response of the same road surface at different speeds is obviously different, which significantly affects the road roughness grade recognition results. The few studies that consider speed factors do not explain how to reasonably process speed signals, or directly input them into the neural network together with other signals, and the recognition speed is slow and the accuracy is low.
[0004] For the recognition of road roughness level, the existing technology mainly adopts the following two methods for recognition: 1) Direct measurement method, such as "Road profile measurement using the two degrees of freedom response-type mechanism" published by O Kavianipour of Iran University of Science and Technology. This method is mostly used for experimental roads, and the road roughness level of fixed experimental roads is obtained through instruments. However, instrument debugging is usually complicated and is not suitable for vehicles that need to obtain road roughness levels in real time. 2) Using vehicle dynamic response, the road roughness information is obtained through transfer functions. For example, the paper "Research on Road Surface Recognition Based on Vehicle Vibration Acceleration Response" by Mo Mingtao of Beijing Institute of Technology proposes to use wheel vibration speed response to perform signal analysis from correlation, frequency, and time domain, and then propose a road surface recognition algorithm. However, this method does not consider the key factor of speed, which affects the recognition accuracy.
[0005] Therefore, there is a need for a solution that can quickly and accurately identify the road roughness level based on speed. Summary of the invention
[0006] In order to overcome the defects of the prior art, the present invention considers the influence of speed on the recognition of road surface roughness level without improving the existing equipment, and uses a neural network to recognize the road surface roughness level.
[0007] The present invention proposes a method for identifying road roughness level taking vehicle speed into consideration, which comprises the following steps:
[0008] Step 1: According to the road surface spatial power spectrum, the road surface roughness is divided into N levels, and N random road surfaces are simulated respectively. The N random road surfaces correspond to N levels of road surface roughness respectively. The simulation time of the N random road surfaces is set to t seconds, and the sampling period is t1 seconds;
[0009] Step 2: Initialize the vehicle speed segment data and loop count j value;
[0010] Set the vehicle speed range to 0-V max The test vehicle is divided into n speed levels, namely V1~V2, V2~V3, V3~V4, ..., V i ~V i+1 , …, V n ~V n+1 , i = 1, 2, 3, ..., n; where V max is the maximum vehicle speed, V1=0, V n+1 =V max ;
[0011] The loop count j value is initialized to 1, i.e. j=1;
[0012] Step 3: Take V j ~V j+1 Simulating any speed W within a range on the N random roads to obtain suspension dynamic travel and vehicle body acceleration signal data;
[0013] During the simulation, a system dynamics model is first established based on the existing vehicle dynamics. Then, the system dynamics model is used to simulate on N random roads using the speed W. Each random road is simulated to obtain the time domain signal of the suspension travel and the time domain signal of the vertical acceleration of the vehicle body. The speed W can be adjusted during the simulation process. j ~V j+1 Change randomly within the range;
[0014] Step 4: The time domain signal of the suspension travel obtained on the same random road surface is recorded as the first time domain signal A1, and the time domain signal of the vehicle body vertical acceleration is recorded as the second time domain signal A2. The suspension travel data is subjected to empirical mode decomposition (EMD) to obtain the intrinsic modal component IMF 11 、IMF 12 、…、IMF 1h 、…、IMF 1m , m is a positive integer, h = 1, 2, 3, ..., m,; the empirical mode decomposition (EMD) of the vehicle body vertical acceleration data, the intrinsic modal component IMF 21 、IMF 22 、…、IMF 2h 、…、IMF 2m ;
[0015] Step 5: Extract features of time domain signals and connotation modal components;
[0016] For the first time domain signal A1, the second time domain signal A2, and the intrinsic modal component IMF 11 、IMF 12 、…、IMF 1h 、…、IMF 1m 、IMF 21 、IMF 22 、…、IMF 2q 、…、IMF 2m , the 2(m+1) signals are subjected to feature extraction every t2 seconds, and p features are extracted simultaneously each time, where p is a positive integer;
[0017] Step 6: Get the number of feature samples I in the time domain, each sample contains L sample data;
[0018] Step 7: j=j+1, j is a positive integer, if j≤n, return to step 4, otherwise, go to step 9;
[0019] Through loops, simulations are performed on N random roads within different speed ranges. Since there are n levels of vehicle speed, n groups of feature sample groups containing I feature samples are obtained to prepare for offline training of the neural network.
[0020] Step 8: Input the n feature sample groups into n neural networks for training. Each neural network corresponds to a first-level vehicle speed range to obtain the road roughness level, and at the same time obtain n trained neural networks;
[0021] The first neural network classification, the second neural network classification, ..., the i-th neural network classification, ..., the n-th neural network classification, correspond to the vehicle speeds V1~V2, V2~V3, V3~V4, ..., V i ~V i+1 , …, Vn ~V n+1 ;
[0022] Step 9: Obtain the vehicle speed signal v, the suspension travel signal d, and the vehicle body vertical acceleration signal a through the vehicle sensor. According to the speed range where the vehicle speed signal v is located, extract the feature values of the signals d and a through the steps 4 and 5, and then directly execute step 8 to enter the neural network classification of the corresponding speed range, that is:
[0023] If V i ≤v <V i+1 , after the features of signals d and a are extracted in steps 4 and 5, they enter the i-th neural network classification;
[0024] The suspension travel signal d and the vehicle body vertical acceleration signal a correspond to the first time domain signal A1 and the second time domain signal A2 respectively;
[0025] Step 10: The i-th neural network gives the road roughness grade.
[0026] Preferably, in step 5, the feature samples extracted from the time domain signal and the connotation modal component are specifically:
[0027] Extract the first time domain signal A1 and obtain feature sample a through feature extraction 11 、a 12 , …a 1p ; Sequentially for the intrinsic modal component IMF 11 、IMF 12 、…、IMF 1h 、…、IMF 1m Use the selected features to extract features and obtain feature samples, IMF 1h After feature extraction, feature sample a is obtained (h+1)1 、a (h+1)2 , …a (h+1)p ; For example: IMF 11 After feature extraction, feature sample a is obtained 21 、a 22 , …a 2p ;IMF 1m After feature extraction, feature sample a is obtained (m+1)1 、a (m+1)2 , …a (m+1)p ; The time domain signal A2, IMF 21 、IMF 22 、…、IMF 2q 、…、IMF 2m Similarly, after the above feature extraction process, the second time domain signal A2 is feature extracted to obtain feature samples b 11 , b 12 ,…b 1p; Sequentially for the intrinsic modal component IMF 21 、IMF 22 、…、IMF 2h 、…、IMF 2m , extract features to get feature samples, IMF 2q After feature extraction, feature sample b is obtained (h+1)1 、b (h+1)2 ,…b (h+1)p .
[0028] Preferably, in step 6:
[0029] The feature sample extracts the feature value of 2(m+1) signals every t2 seconds, and the number of features is selected as p, so the number of feature samples I = 2×(m+1)×p;
[0030] Because t is simulated on the N random roads, the simulation time of the N random roads is set to t seconds, the sampling period is t1 seconds, and the feature value extraction is performed on 2(m+1) signals every t2 seconds, so each sample contains Sample data.
[0031] Preferably, the step 1 specifically comprises:
[0032] The road surface roughness is divided into 8 levels, and ISO Level A to ISO Level H are used to correspond to road surface roughness levels 1 to 8. Eight random road surfaces are simulated, and the 8 random road surfaces correspond to the 8 levels of road surface roughness. The simulation time of the 8 random road surfaces is set to t = 1000 seconds, and the sampling period is t1 = 0.001 seconds.
[0033] Preferably, in step 5, feature extraction is performed every t2 seconds, where t2=0.1 seconds.
[0034] Preferably, in step 9, the vehicle speed signal v is obtained through a speed sensor, the suspension dynamic travel signal d is obtained through a displacement sensor, and the vehicle body vertical acceleration signal a is obtained through an acceleration sensor.
[0035] Preferably, the step 2 specifically includes: dividing the test vehicles with a speed of 0-180 km / h into 18 speed grades.
[0036] The present invention also discloses a road surface roughness level recognition system considering vehicle speed, comprising the following modules: a simulated road surface generation module, an initialization module, a signal simulation module, an empirical mode decomposition module, a feature acquisition module, a training data module, a neural network module and a vehicle sensor module; wherein the simulated road surface generation module and the initialization module both send data to the signal simulation module, the signal simulation module generates a time domain signal and sends it to the empirical mode decomposition module, the empirical mode decomposition module then sends the time domain signal and the connotation mode component to the feature acquisition module, the feature acquisition module extracts features and sends them to the training data module or the neural network module according to the source of the time domain signal, the training data module saves training samples and trains the neural network module, the vehicle sensor module collects real-time signals of the vehicle and sends them to the empirical mode decomposition module, and each module is described in detail below:
[0037] The simulated road surface generation module is used to simulate N random roads of different levels according to the road surface spatial power spectrum. The simulation time of the N random roads is set to t seconds, and the sampling period is t1 seconds;
[0038] An initialization module, used for dividing the vehicle speed into n continuous non-intersecting vehicle speed segments, and initializing a loop count j value;
[0039] The signal simulation module is used to simulate any speed within each speed segment on the N random roads according to the system dynamics model to obtain the time domain signal of the suspension dynamic travel and the time domain signal of the vehicle body vertical acceleration; the system dynamics model is established according to the existing vehicle dynamics, and the speed can be changed arbitrarily within the segment range during the simulation process;
[0040] The empirical mode decomposition module is used to perform empirical mode decomposition on the time domain signal of the suspension dynamic stroke and the time domain signal of the vehicle body vertical acceleration to obtain m corresponding intrinsic modal components respectively; the time domain signal of the suspension dynamic stroke obtained on the same random road surface is recorded as the first time domain signal A1, and the time domain signal of the vehicle body vertical acceleration is recorded as the second time domain signal A2, and the suspension dynamic stroke data is empirically decomposed (EMD) to obtain the intrinsic modal component IMF 11 、IMF 12 、…、IMF 1h 、…、IMF 1m , m is a positive integer, h = 1, 2, 3, ..., m,; the empirical mode decomposition (EMD) of the vehicle body vertical acceleration data, the intrinsic modal component IMF 21 、IMF 22 、…、IMF 2h 、…、IMF 2m ;
[0041] A feature acquisition module is used to extract features from the time domain signal of the suspension dynamic travel, the time domain signal of the vehicle body vertical acceleration, and their intrinsic modal components; the 2(m+1) signals are subjected to feature value extraction every t2 seconds, and the feature values include mean, variance, root amplitude, root mean square value, peak value, skewness or kurtosis, and p feature values are selected from the feature values, where p is a positive integer;
[0042] The training data module is used to save the feature samples extracted in the feature acquisition module. A total of I = 2 × (m + 1) × p feature samples are obtained. Each sample contains Sample data, put the feature samples in the same speed range into the same feature sample group, there are n feature sample groups in total;
[0043] A neural network module is used to give a road roughness grade, the neural network module includes a first neural network classification, a second neural network classification, ..., an i-th neural network classification, ..., and an n-th neural network classification, which correspond to n vehicle speed segments respectively, and the n neural network classifications are trained using feature sample groups of corresponding vehicle speed segments respectively;
[0044] The vehicle sensor module includes a speed sensor, a displacement sensor and a velocity sensor, which are used to obtain a vehicle speed signal v, a suspension travel signal d and a vehicle body vertical acceleration signal a respectively; the suspension travel signal d and the vehicle body vertical acceleration signal a respectively correspond to a first time domain signal A1 and a second time domain signal A2;
[0045] Preferably, after the training of the neural network module is completed using the simulated road surface generation module, initialization module, signal simulation module, empirical mode decomposition module, feature acquisition module, training data module and judgment module, only the neural network module, empirical mode decomposition module and feature acquisition module are installed on the vehicle body, and the vehicle sensor module uses the sensors in the existing vehicle.
[0046] Compared with the prior art, the present invention has the following beneficial effects:
[0047] (1) The present invention utilizes sensors in the existing active suspension system to identify the road surface without adding any sensors, and the system is simple and reliable.
[0048] (2) The vehicle speed, a factor that significantly affects the recognition of road roughness level, is taken into account, thereby improving the recognition accuracy.
[0049] (3) The recognition method of the present invention has high accuracy and works reliably. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] Figure 1 A flow chart of the steps of a method for identifying road surface roughness level taking vehicle speed into consideration according to the present invention;
[0051] Figure 2 A schematic diagram of a module of a road surface roughness level recognition system taking vehicle speed into consideration according to the present invention;
[0052] Figure 3 A schematic diagram of the position of sensors in the vehicle sensor module of the present invention;
[0053] Figure 4 The figure is a schematic diagram of an embodiment of a method for identifying road roughness level taking vehicle speed into consideration according to the present invention. DETAILED DESCRIPTION
[0054] In order to better understand the technical solution of the present invention, the specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings need not be drawn to scale unless otherwise specified.
[0055] The present invention discloses a method for identifying road surface roughness level taking vehicle speed into consideration, such as Figure 1 As shown, the specific steps are as follows:
[0056] Step 1: According to the road surface spatial power spectrum, the road surface roughness is usually divided into 8 levels, represented by ISO Level N, where N is a character from A to H, and A to H correspond to road surface roughness levels 1 to 8 respectively. The present invention simulates 8 random road surfaces respectively, and the 8 random road surfaces correspond to 8 levels of road surface roughness respectively. The simulation time of the 8 random road surfaces is set to t seconds, and the sampling period is t1 seconds (t1<<t).
[0057] Step 2: Initialize the vehicle speed segment data and loop count j value.
[0058] Set the vehicle speed range to 0-V max The test vehicle is divided into n speed levels, namely V1~V2, V2~V3, V3~V4, ..., V i ~V i+1 , …, V n ~V n+1 , i = 1, 2, 3, ..., n; where V max is the maximum vehicle speed, V1=0, V n+1 =V max .
[0059] The loop count j value is initialized to 1, i.e. j=1;
[0060] Step 3: Take V j ~V j+1The suspension dynamic travel and vehicle body acceleration signal data are obtained by simulating any speed W within the range on 8 random roads.
[0061] During the simulation, a system dynamics model is first established based on the existing vehicle dynamics. Then, the system dynamics model is used to simulate on 8 random roads using the speed W. Each random road is simulated to obtain the time domain signal of the suspension travel and the time domain signal of the vertical acceleration of the vehicle body. The speed W can be adjusted during the simulation process. j ~V j+1 The range can be changed at will.
[0062] Step 4: The time domain signal of the suspension travel obtained on the same random road surface is recorded as the first time domain signal A1, and the time domain signal of the vehicle body vertical acceleration is recorded as the second time domain signal A2. The suspension travel data is subjected to empirical mode decomposition (EMD) to obtain the intrinsic modal component IMF 11 、IMF 12 、…、IMF 1h 、…、IMF 1m , m is a positive integer, h = 1, 2, 3, ..., m,; the empirical mode decomposition (EMD) of the vehicle body vertical acceleration data, the intrinsic modal component IMF 21 、IMF 22 、…、IMF 2h 、…、IMF 2m .
[0063] Step 5: Extract features of time domain signals and intrinsic modal components.
[0064] For the first time domain signal A1, the second time domain signal A2, and the intrinsic modal component IMF 11 、IMF 12 、…、IMF 1h 、…、IMF 1m 、IMF 21 、IMF 22 、…、IMF 2q 、…、IMF 2m , a total of 2 (m + 1) signals are extracted every t2 seconds (t1 << t2 << t). Features include mean, variance, root amplitude, root mean square value, peak, skewness, kurtosis, etc., and p features are selected from the features, where p is a positive integer. The first time domain signal A1 is extracted to obtain feature sample a 11 、a 12 , …a 1p ; Sequentially for the intrinsic modal component IMF 11 、IMF 12 、…、IMF 1h 、…、IMF 1mUse the selected features to extract features and obtain feature samples, IMF 1h After feature extraction, feature sample a is obtained (h+1)1 、a (h+1)2 , …a (h+1)p ; For example: IMF 11 After feature extraction, feature sample a is obtained 21 、a 22 , …a 2p ;IMF 1m After feature extraction, feature sample a is obtained (m+1)1 、a (m+1)2 , …a (m+1)p . Time domain signal A2, IMF 21 、IMF 22 、…、IMF 2q 、…、IMF 2m Similarly, after the above feature extraction process, the second time domain signal A2 is feature extracted to obtain feature sample b 11 、b 12 ,…b 1p ; Sequentially for the intrinsic modal component IMF 21 、IMF 22 、…、IMF 2h 、…、IMF 2m , extract features to get feature samples, IMF 2h After feature extraction, feature sample b is obtained (h+1)1 、b (h+1)2 ,…b (h+1)p ; For example: from the IMF 21 After feature extraction, feature sample b is obtained 21 、b 22 ,…b 2p ; From IMF 2m After feature extraction, feature sample b is obtained (m+1)1 、b (m+1)2 ,…b (m+1)p .
[0065] Step 6: Obtain the number of feature samples I in the time domain, each sample contains L sample data.
[0066] The feature sample extracts the feature value of 2(m+1) signals every t2 seconds, and the number of features is selected as p, so the number of feature samples I = 2×(m+1)×p;
[0067] Because t is simulated on 8 random roads, the simulation time of the 8 random roads is set to t seconds, the sampling period is t1 seconds, and the feature value extraction is performed on 2(m+1) signals every t2 seconds, so each sample contains Sample data.
[0068] Step 7: j=j+1, j is a positive integer. If j≤n, return to step 4; otherwise, execute step 9.
[0069] By cyclically performing simulations within different speed ranges on 8 random road surfaces, n groups of feature samples containing I feature samples are obtained to prepare for offline training of the neural network.
[0070] Step 8: Input the n feature sample groups into n neural networks for training. Each neural network corresponds to a first-level vehicle speed range to obtain the road roughness level, and at the same time obtain n trained neural networks.
[0071] The first neural network classification, the second neural network classification, ..., the i-th neural network classification, ..., the n-th neural network classification, correspond to the vehicle speeds V1~V2, V2~V3, V3~V4, ..., V i ~V i+1 , …, V n ~V n+1 .
[0072] Step 9: Obtain the vehicle speed signal v, the suspension travel signal d, and the vehicle body vertical acceleration signal a through the vehicle sensor. According to the vehicle speed range where the vehicle speed signal v is located, extract the feature values of the signals d and a through steps 4 and 5, and then directly execute step 8 to enter the corresponding neural network classification; wherein the suspension travel signal d and the vehicle body vertical acceleration signal a correspond to the first time domain signal A1 and the second time domain signal A2, respectively.
[0073] The vehicle speed signal v is obtained through the speed sensor 3 , the suspension dynamic travel signal d is obtained through the displacement sensor 1 , and the vehicle body vertical acceleration signal a is obtained through the acceleration sensor 2 .
[0074] If V i ≤v <V i+1 , signals d and a are extracted through features in step 5 and then enter the i-th neural network classification.
[0075] Step 10: The i-th neural network gives the road roughness grade.
[0076] In the method of the present invention, steps 1 to 8 belong to offline recognition, and steps 9 to 10 belong to online recognition.
[0077] In order to implement the method, the present invention also proposes a road surface roughness level recognition system taking vehicle speed into consideration, such as Figure 2As shown, it includes the following modules: In order to implement this method, the present invention also proposes a road surface roughness level recognition system taking vehicle speed into consideration, including the following modules: simulated road surface generation module 1, initialization module 2, signal simulation module 3, empirical mode decomposition module 4, feature acquisition module 5, training data module 6, neural network module 7 and vehicle sensor module 8; wherein, the simulated road surface generation module 1 and the initialization module 2 both send data to the signal simulation module 3, the signal simulation module 3 generates a time domain signal and sends it to the empirical mode decomposition module 4, the empirical mode decomposition module 4 then sends the time domain signal and the connotation modal component to the feature acquisition module 5, the feature acquisition module 5 extracts features and sends them to the training data module 6 or the neural network module 7 according to the source of the time domain signal, the training data module 6 saves the training samples and trains the neural network module 7, and the vehicle sensor module 8 collects the real-time signal of the vehicle and sends it to the empirical mode decomposition module 4. The following is a detailed description of each module:
[0078] The simulated road surface generation module 1 is used to simulate 8 different levels of random road surfaces according to the road surface spatial power spectrum. The simulation time of the 8 random road surfaces is set to t seconds, and the sampling period is t1 seconds.
[0079] Initialization module 2 is used to divide the vehicle speed into n continuous non-intersecting vehicle speed segments and initialize the cycle count j value.
[0080] Signal simulation module 3 is used to simulate any speed within each speed segment on 8 random roads according to the system dynamics model to obtain the time domain signal of the suspension dynamic travel and the time domain signal of the vehicle body vertical acceleration. The system dynamics model is established based on existing vehicle dynamics, and the speed can be changed arbitrarily within the segment range during the simulation process.
[0081] The empirical mode decomposition module 4 is used to perform empirical mode decomposition on the time domain signal of the suspension dynamic travel and the time domain signal of the vehicle body vertical acceleration to obtain m corresponding intrinsic modal components. The time domain signal of the suspension dynamic travel obtained on the same random road surface is recorded as the first time domain signal A1, and the time domain signal of the vehicle body vertical acceleration is recorded as the second time domain signal A2. The empirical mode decomposition (EMD) of the suspension dynamic travel data is performed to obtain the intrinsic modal component IMF 11 、IMF 12 、…、IMF 1h 、…、IMF 1m , m is a positive integer, h = 1, 2, 3, ..., m,; the empirical mode decomposition (EMD) of the vehicle body vertical acceleration data, the intrinsic modal component IMF 21 、IMF 22 、…、IMF 2h 、…、IMF 2m .
[0082] The feature acquisition module 5 is used to extract features from the time domain signal of the suspension travel, the time domain signal of the vehicle body vertical acceleration, and their intrinsic modal components. The 2(m+1) signals are subjected to feature value extraction every t2 seconds. The feature values include mean, variance, root amplitude, root mean square value, peak value, skewness, kurtosis, etc., and p feature values are selected from the feature values, where p is a positive integer.
[0083] The training data module 6 is used to store the feature samples extracted in the feature acquisition module. A total of I = 2 × (m + 1) × p feature samples are obtained. Each sample contains Sample data are collected and feature samples in the same speed range are placed in the same feature sample group. There are n feature sample groups in total.
[0084] The neural network module 7 is used to give a road roughness grade. The neural network module includes a first neural network classification, a second neural network classification, ..., an i-th neural network classification..., and an n-th neural network classification, which correspond to n vehicle speed segments respectively. The n neural network classifications are trained using feature sample groups of the corresponding vehicle speed segments respectively.
[0085] The vehicle sensor module 8 includes a speed sensor 83, a displacement sensor 81 and a speed sensor 82, which are used to obtain a vehicle speed signal v, a suspension travel signal d and a vehicle body vertical acceleration signal a, respectively. This module is a collection of sensors in existing vehicles and does not need to be installed additionally. The suspension travel signal d and the vehicle body vertical acceleration signal a correspond to the first time domain signal A1 and the second time domain signal A2, respectively.
[0086] In actual reality, all modules can be installed on the vehicle body, or the simulated road surface generation module, initialization module, signal simulation module, empirical mode decomposition module, feature acquisition module, training data module and judgment module can be used. After the training of the neural network module is completed, only the neural network module, empirical mode decomposition module and feature acquisition module are installed on the vehicle body, and the vehicle sensor module uses the sensors in the existing vehicle to realize the identification of the road surface grade.
[0087] In order to better understand the present invention, the specific implementation steps of a method for identifying road roughness level considering vehicle speed are illustrated as follows:
[0088] Step 1: According to the road surface spatial power spectrum, the road surface roughness is usually divided into 8 levels from A to H, and 8 random road surfaces of ISO Level A, ISO Level B, ISO Level C..., and ISO Level H are simulated. The simulation time for each road surface roughness level is set to t = 1000 seconds, and the sampling period is t1 = 0.001 seconds.
[0089] Step 2: As attached Figure 3 As shown, this example selects a vehicle with a speed table of 0 to 180 km / h for detailed description, as shown in the attached Figure 4 As shown, the test vehicle with a speed of 0-180km / h is divided into 18 speed grades, namely 0-10km / h, 10-20km / h, 20-30km / h, ..., 170-180km / h.
[0090] Initialize the loop count j to j=1.
[0091] Step 3: Take any value within the j-th speed range and simulate on A-H road surfaces to obtain the suspension dynamic travel and body acceleration corresponding to the 8 speed levels.
[0092] Step 4: As attached Figure 3 First, the time domain signal of the suspension travel is recorded as the first time domain signal A1, and the time domain signal of the vehicle body acceleration is recorded as the second time domain signal A2. The suspension travel data is empirically decomposed (EMD) to obtain the intrinsic modal component IMF 11 、IMF 12 、IMF 13 、…、IMF 15 ; The empirical mode decomposition (EMD) of the vehicle body vertical acceleration data and the intrinsic modal component IMF 21 、IMF 22 、IMF 23 、…、IMF 25 .
[0093] Step 5: Extract features from the time domain signal and the modal component. Extract feature values from the 12 signals every t2 = 0.1 seconds. In this example, three feature values, root mean square, root square amplitude and root mean square value, are selected. The first time domain signal A1 is extracted to obtain feature sample a 11 、a 12 、a 13 ;IMF 11 The characteristic sample a is obtained by eigenvalue extraction 21 、a 22 、a 23 ;IMF 12 The characteristic sample a is obtained by eigenvalue extraction 31 、a 32 、a 33 ;…;IMF 15 The characteristic sample a is obtained by eigenvalue extraction 61 、a 62 、a 63 ; Time domain signal A2, IMF 21 、IMF 22 、IMF23 、IMF 24 、IMF 25 The same feature value extraction process is carried out as above.
[0094] Step 6: A total of 36 feature samples are obtained, each containing 800,000 sample data.
[0095] Step 7: j=j+1, j is a positive integer, and it is determined that j≤18. If yes, return to step 4, otherwise, execute step 9.
[0096] This step is repeated 18 times in a cycle, in each speed range of 10-20km / h, 20-30km / h, ..., 170-180km / h, and so on, to obtain 18 groups of feature sample groups containing 36 feature samples, each of which contains 800,000 sample data. Figure 3 As shown, the above data processing is prepared for offline training of the neural network.
[0097] Step 8: Input the 18 feature sample groups into 18 neural networks for training to obtain the road roughness grade. At the same time, 18 trained neural networks are obtained, which are recorded as the first neural network classification, the second neural network classification, ..., the 18th neural network classification, corresponding to the vehicle speeds V1~V2, V2~V3, ..., V 18 ~V 19 .
[0098] Step 9: As attached Figure 4 As shown, the dual displacement sensor (1) obtains the suspension dynamic travel signal d, the acceleration sensor (2) obtains the vehicle body vertical acceleration signal a, and the speed sensor (3) obtains the vehicle speed signal v. Different neural networks are entered according to the vehicle speed signal v, specifically:
[0099] If 0≤v<10km / s, signals d and a are subjected to feature extraction in steps 4 and 5 and then enter the first neural network classification;
[0100] If 0≤v<20km / s, signals d and a are subjected to feature extraction in steps 4 and 5 and then enter the second neural network classification;
[0101] If 0≤v<30km / s, signals d and a are subjected to feature extraction in steps 4 and 5 and then enter the third neural network classification;
[0102] …
[0103] If 0≤v<170km / s, signals d and a are subjected to feature extraction in steps 4 and 5 and then enter the 17th neural network classification;
[0104] If 0≤v<180km / s, signals d and a are subjected to feature extraction in steps 4 and 5 and then enter the 18th neural network classification.
[0105] Step 10: Output the road surface roughness level according to step 9.
[0106] Finally, it should be noted that the above-described embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some or all of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for identifying road roughness level taking vehicle speed into consideration, characterized in that: It includes the following steps: Step 1: According to the road surface spatial power spectrum, the road surface roughness is divided into N levels, and N random road surfaces are simulated respectively. The N random road surfaces correspond to N levels of road surface roughness respectively. The simulation time of the N random road surfaces is set to t seconds, and the sampling period is t1 seconds; Step 2: Initialize the vehicle speed segment data and loop count j value; Set the vehicle speed range to 0-V max The test vehicle is divided into n speed levels, namely V1~V2, V2~V3, V3~V4, ..., V i ~V i+1 , …, V n ~V n+1 , i = 1, 2, 3, ..., n; where V max is the maximum vehicle speed, V1=0, The loop count j value is initialized to 1, i.e. j=1; Step 3: Take V j ~V j+1 Simulating any speed W within a range on the N random roads to obtain suspension dynamic travel and vehicle body acceleration signal data; During the simulation, a system dynamics model is first established based on the existing vehicle dynamics. Then, the system dynamics model is used to simulate on N random roads using the speed W. Each random road is simulated to obtain the time domain signal of the suspension travel and the time domain signal of the vertical acceleration of the vehicle body. The speed W can be adjusted during the simulation process. j ~V j+1 Change randomly within the range; Step 4: The time domain signal of the suspension travel obtained on the same random road surface is recorded as the first time domain signal A1, and the time domain signal of the vehicle body vertical acceleration is recorded as the second time domain signal A2. The suspension travel data is subjected to empirical mode decomposition (EMD) to obtain the intrinsic modal component IMF 11 、IMF 12 、…、IMF 1h 、…、IMF 1m , m is a positive integer, h = 1, 2, 3, ..., m, the empirical mode decomposition (EMD) of the vehicle body vertical acceleration data, the intrinsic modal component IMF 21 、IMF 22 、…、IMF 2h 、…、IMF 2m ; Step 5: Extract features of time domain signals and connotation modal components; For the first time domain signal A1, the second time domain signal A2, and the intrinsic modal component IMF 11 、IMF 12 、…、IMF 1h 、…、IMF 1m 、IMF 21 、IMF 22 、…、IMF 2q 、…、IMF 2m , the 2(m+1) signals are subjected to feature extraction every t2 seconds, and p features are extracted simultaneously each time, where p is a positive integer; Step 6: Get the number of feature samples I in the time domain, each sample contains L sample data; Step 7: j=j+1, j is a positive integer, if j≤n, return to step 4, otherwise, go to step 9; Through loops, simulations are performed on N random roads within different speed ranges. Since there are n levels of vehicle speed, n groups of feature sample groups containing I feature samples are obtained to prepare for offline training of the neural network. Step 8: Input the n feature sample groups into n neural networks for training. Each neural network corresponds to a first-level vehicle speed range to obtain the road roughness level, and at the same time obtain n trained neural networks; The first neural network classification, the second neural network classification, ..., the i-th neural network classification, ..., the n-th neural network classification, correspond to the vehicle speeds V1~V2, V2~V3, V3~V4, ..., V i ~V i+1 , …, V n ~V n+1 ; Step 9: Obtain the vehicle speed signal v, the suspension travel signal d, and the vehicle body vertical acceleration signal a through the vehicle sensor. According to the speed range where the vehicle speed signal v is located, extract the feature values of the signals d and a through the steps 4 and 5, and then directly execute step 8 to enter the neural network classification of the corresponding speed range, that is: If V i ≤v <V i+1 , after the features of signals d and a are extracted in steps 4 and 5, they enter the i-th neural network classification; The suspension dynamic travel signal d and the vehicle body vertical acceleration signal a correspond to the first time domain signal A1 and the second time domain signal A2 respectively; Step 10: The i-th neural network gives the road roughness grade.
2. The method for identifying road roughness level considering vehicle speed according to claim 1, characterized in that: In step 5, the feature samples extracted from the time domain signal and the connotation modal component are specifically: Extract the first time domain signal A1 and obtain feature sample a through feature extraction 11 、a 12 , …a 1p ; Sequentially for the intrinsic modal component IMF 11 、IMF 12 、…、IMF 1h 、…、IMF 1m Use the selected features to extract features and obtain feature samples, IMF 1h After feature extraction, feature sample a is obtained (h+1)1 、a (h+1)2 , …a (h+1)p ; For example: IMF 11 After feature extraction, feature sample a is obtained 21 、a 22 , …a 2p ;IMF 1m After feature extraction, feature sample a is obtained (m+1)1 、a (m+1)2 , …a (m+1)p ; The time domain signal A2, IMF 21 、IMF 22 、…、IMF 2q 、…、IMF 2m Similarly, after the above feature extraction process, the second time domain signal A2 is feature extracted to obtain feature samples b 11 、b 12 ,…b 1p ; Sequentially for the intrinsic modal component IMF 21 、IMF 22 、…、IMF 2h 、…、IMF 2m , extract features to get feature samples, IMF 2q After feature extraction, feature sample b is obtained (h+1)1 、b (h+1)2 ,…b (h+1)p .
3. The method for identifying road roughness level considering vehicle speed according to claim 1, characterized in that: In step 6: The feature sample extracts the feature value of 2(m+1) signals every t2 seconds, and the number of features is selected as p, so the number of feature samples I = 2×(m+1)×p; Because t is simulated on the N random roads, the simulation time of the N random roads is set to t seconds, the sampling period is t1 seconds, and the feature value extraction is performed on 2(m+1) signals every t2 seconds, so each sample contains Sample data.
4. The method for identifying road roughness level considering vehicle speed according to claim 1, characterized in that: The specific steps of step 1 are: The road surface roughness is divided into 8 levels, and ISO LevelA to ISO Level H are used to correspond to road surface roughness levels 1 to 8. Eight random road surfaces are simulated, and the 8 random road surfaces correspond to the 8 levels of road surface roughness. The simulation time of the 8 random road surfaces is set to t = 1000 seconds, and the sampling period is t1 = 0.001 seconds.
5. The method for identifying road roughness level considering vehicle speed according to claim 1, characterized in that: In step 5, feature extraction is performed every t2 seconds, where t2=0.1 seconds.
6. The method for identifying road roughness level considering vehicle speed according to claim 1, characterized in that: In step 9, the vehicle speed signal v is obtained through a speed sensor, the suspension dynamic travel signal d is obtained through a displacement sensor, and the vehicle body vertical acceleration signal a is obtained through an acceleration sensor.
7. The method for identifying road roughness level considering vehicle speed according to claim 1, characterized in that: The step 2 specifically includes: dividing the test vehicles with a speed of 0-180 km / h into 18 speed grades.
8. A road surface roughness level recognition system taking vehicle speed into consideration for use in the road surface roughness level recognition method taking vehicle speed into consideration as claimed in claim 1, characterized in that: The road surface roughness level recognition system considering vehicle speed includes the following modules: a simulated road surface generation module, an initialization module, a signal simulation module, an empirical mode decomposition module, a feature acquisition module, a training data module, a neural network module and a vehicle sensor module; wherein the simulated road surface generation module and the initialization module both send data to the signal simulation module, the signal simulation module generates a time domain signal and sends it to the empirical mode decomposition module, the empirical mode decomposition module then sends the time domain signal and the connotation modal component to the feature acquisition module, the feature acquisition module extracts features and sends them to the training data module or the neural network module according to the source of the time domain signal, the training data module saves training samples and trains the neural network module, the vehicle sensor module collects the real-time signal of the vehicle and sends it to the empirical mode decomposition module, and each module is described in detail below: The simulated road surface generation module is used to simulate N random roads of different levels according to the road surface spatial power spectrum. The simulation time of the N random roads is set to t seconds, and the sampling period is t1 seconds; An initialization module, used for dividing the vehicle speed into n continuous non-intersecting vehicle speed segments, and initializing a loop count j value; The signal simulation module is used to simulate any speed within each speed segment on the N random roads according to the system dynamics model to obtain the time domain signal of the suspension dynamic travel and the time domain signal of the vehicle body vertical acceleration; the system dynamics model is established according to the existing vehicle dynamics, and the speed can be changed arbitrarily within the segment range during the simulation process; The empirical mode decomposition module is used to perform empirical mode decomposition on the time domain signal of the suspension dynamic stroke and the time domain signal of the vehicle body vertical acceleration to obtain m corresponding intrinsic modal components respectively; the time domain signal of the suspension dynamic stroke obtained on the same random road surface is recorded as the first time domain signal A1, and the time domain signal of the vehicle body vertical acceleration is recorded as the second time domain signal A2, and the suspension dynamic stroke data is empirically decomposed (EMD) to obtain the intrinsic modal component IMF 11 、IMF 12 、…、IMF 1h 、…、IMF 1m , m is a positive integer, h = 1, 2, 3, ..., m, the empirical mode decomposition (EMD) of the vehicle body vertical acceleration data, the intrinsic modal component IMF 21 、IMF 22 、…、IMF 2h 、…、IMF 2m ; A feature acquisition module is used to extract features from the time domain signal of the suspension dynamic travel, the time domain signal of the vehicle body vertical acceleration, and their intrinsic modal components; the 2(m+1) signals are subjected to feature value extraction every t2 seconds, and the feature values include mean, variance, root amplitude, root mean square value, peak value, skewness or kurtosis, and p feature values are selected from the feature values, where p is a positive integer; The training data module is used to save the feature samples extracted in the feature acquisition module. A total of I = 2 × (m + 1) × p feature samples are obtained. Each sample contains Sample data, put the feature samples in the same speed range into the same feature sample group, there are n feature sample groups in total; A neural network module is used to give a road roughness grade, the neural network module includes a first neural network classification, a second neural network classification, ..., an i-th neural network classification, ..., and an n-th neural network classification, which correspond to n vehicle speed segments respectively, and the n neural network classifications are trained using feature sample groups of corresponding vehicle speed segments respectively; The vehicle sensor module includes a speed sensor, a displacement sensor and an acceleration sensor, which are respectively used to obtain a vehicle speed signal v, a suspension dynamic travel signal d and a vehicle body vertical acceleration signal a; the suspension dynamic travel signal d and the vehicle body vertical acceleration signal a correspond to a first time domain signal A1 and a second time domain signal A2 respectively.
9. The road surface roughness level recognition system considering vehicle speed according to claim 8, characterized in that: After the neural network module is trained using the simulated road surface generation module, initialization module, signal simulation module, empirical mode decomposition module, feature acquisition module, training data module and judgment module, only the neural network module, empirical mode decomposition module and feature acquisition module are installed on the vehicle body, and the vehicle sensor module uses the sensors in the existing vehicle.
Citation Information
Patent Citations
Method and system for identifying road surface in a road
CN102519577A
Method for obtaining surface roughness load spectrum of soft road surface
CN105973189A
Road surface grade recognition system and method based on suspension dynamic stroke
CN110001335A
Device for control suspension using neural circuit network of vehicle
KR1020000043025A
