Intelligent tire wear life estimation method and device based on BP neural network

[0065] In order to more clearly illustrate the technical solutions of the present invention, the specific embodiments of the present invention will be described below with reference to the drawings. It will be apparent that the drawings in the following description are merely some embodiments of the invention, and those of ordinary skill in the art will also obtain other drawings according to these figures without paying creative labor. Other embodiments.

[0066] First embodiment of the invention, such as figure 1 As shown, a smart tire wear life estimation method based on a BP neural network, including:

[0067] S10 acquires data sets containing tire pressure, speed, load, and tire riped 2 to 6 steps, and randomly split data sets as training sets, verification sets and test sets;

[0068] S20 creates a BP neural network model, the input of the network model is data in the data set, and the output is tire grinding.

[0069] S30 based on training set, verification set and pre-set mean square error on BP neural network model training, determine the structure and network parameters of BP neural network model;

[0070] S40 will train the data in the test concentration, and the BP neural network model is trained to estimate the wear life of the tires belonging to the data.

[0071] In the present embodiment, the tire wear life is estimated, and the basic modal formula is (1):

[0072]

[0073] Where f is the modal frequency, k is the rigidity of the tire, and m is the quality of the tire.

[0074] When the tire pressure, speed, load, and tire grinding changes, the corresponding tires will change the rigidity and quality, thereby affecting the modal frequency of the tire, that is, the tire pressure, vehicle speed, load, tire wear and tire mode There is a certain relationship between the frequency, satisfying the basic relationship between neural network output, output parameters, based on this, in this embodiment, the most easily identified radial modular frequency of the finite element tire is selected as the reference object. In the selected process, by Figure 2 ~ 5 ( figure 2 Relationship with the relationship between the tire mode frequency and modal step under different air pressure; image 3 Relationship between the radial mode frequency and modal step; Figure 4 Relationship between the radial mode frequency and modal step; Figure 5 It can be seen for the relationship between the radial modal frequency and modal step of different wear content), and the radial modal frequency of the tire has increased with the increase of inflatable pressure and tire grinding. figure 2 Medium pressure gradually rose from 0.18 MPa to 0.24 MPa, Figure 5 The medium tire wear is gradually rising from no wear 6mm. The overall trend is generally upward, and the load and speed increase ( image 3 Medium speed is gradually rising from 0 to 60km / h, Figure 4 The load is gradually increased from 2000N to 5000 N). However, from image 3 It can be seen that the radial first-order mode frequency is not sensitive to the speed of the vehicle, so in order to improve the prediction accuracy of the BP neural network model, the data of 2 to 6 steps rising mode frequency is selected as the input of the BP neural network model. The output is the amount of tire grinding.

[0075] After the input and output of the BP neural network model is selected, the BP neural network model is created, and the collected data set is randomly split into training set, verification set and test set, where the training set is for BP neural network models. Training, the verification set is used in the network training process to prevent the network from falling into the fitting, and the test set is used to test the performance of the network completed. For training sets, verification sets, and test sets, it can be set according to the actual situation, such as in an example, training set, verification set and test sets are 0.7, 0.2 and 0.1, respectively.

[0076] For the created BP neural network model, considering the amount of data collected and the number of input parameters is not large, in order to avoid complex, the hidden layer number is set to 1, and TAN SIG is used as the activation function. In order to determine the number of hidden layers in the hidden layer, in the creation of the BP neural network model, the hidden layer node is determined based on the number of hidden layers of the formula (2), the number of input layer nodes and output layer nodes. The scope of the number, then established the corresponding BP neural network model based on the range of hidden layers nodes.

[0077]

[0078] Wherein, p is the number of hidden layers, n is the number of input layer nodes (determined by the actual input number), and Q is a constant between the output layer node (according to the actual demand), a constant between 1 to 10. The number of input parameters and output parameters in this embodiment (n = 8, q = 1) can initially determine the range of implicit layer nodes in the [4, 13] section, and establish hidden The number of layered nodes is incremented from 4 to the BP neural network model of 13.

[0079] During training, the set loop algorithm uses training sets and verification sets into the BP neural network model, and calculates the number of preset times for the BP neural network model containing different hidden layers nodes (such as 50, 100 Supreme and even more), and set the store automatically store the minimum mean square error MSE for each hidden layer node loop iterative calculation training set, the training method is the TrainLM training method, and the calculation formula of MSE is (3)), The BP neural network model corresponding to the current minimum mean square error MSE is automatically saved via the Save function. After all BP neural network model iterate, compare the mean square error of each iteration, select the number of nodes corresponding to the number of nodes as the best hidden layer node, and to determine the structure of the BP neural network model and Network parameters, then use the BP neural network model to estimate the data in the test concentration in tire wear life.

[0080]

[0081] Where Y i For the training set I input data, the true value of the tire mill, Y ' i The expected value of the corresponding input data is predicted for the BP neural network module, and n is the number of training sets.

[0082] In the prediction of the test set in the use of the training BP neural network model, the prediction accuracy is measured with the absolute error ε of the tire wear prediction value and the actual value, as in the formula (4):

[0083] ε = x-a (4)

[0084] Where X is the tire wear prediction value, and A is the tire wear.

[0085] Since the acquisition data numerical gap is too large, in order to improve the accuracy of the estimated algorithm, in other embodiments, the data in the training set, verification set and test concentration needs to be normalized and used in order to disrupted with the Randperm function. Increasing the robustness of the estimation algorithm, the value of the data after normalization is between (0, 1), in line with the standard normal distribution, avoiding 0 in subsequent calculations; after the prediction is completed, Normalization. Specifically, normalization and anti-entering, as equation (5) ~ (7):

[0086]

[0087]

[0088] Y predict = (Y predict,Nor +1) Y max -Y predict,Nor Y min (7)

[0089] Where X and Y represent the input value and output value of the training data, X nor Sum nor Data input value and output value after normalization processing, respectively, X min X max Indicates the minimum value and maximum value of the input value in the training data, y min Sum max Indicates the minimum and maximum value of the output value, y predict,Nor Represents normalized BP neural network model forecast results, y predict Prediction of BP Neural Network Model for Inverse Culture.

[0090] Another embodiment of the present invention, a smart tire wear life estimation device 100 based on a BP neural network, such as Figure 12 The data acquisition module 110 is configured to obtain a data set containing the tire pressure, vehicle speed, load, and tire ridier 2 to 6 steps, and randomly split the data set as a training set, verification set And test set; neural network creation module 120, used to create BP neural network models, input of network models is data in the data set, output as tire wear, neural network training module 130, based on training set, verification set and pre-set The mean square error is trained to the BP neural network model, determine the structure and network parameters of the BP neural network model; the tire wear life estimation module 140 is used to train the data input in the test concentration, the BP neural network model is used, and the data belonging to the data The wear life of the tire is estimated.

[0091]In the present embodiment, the tire wear life is prevised based on the modal analysis related theory. When the tire pressure, vehicle speed, load and tire grinding have changed, the corresponding tires will change the rigidity and quality, respectively, thereby affecting the tire. There is a certain relationship between the modal frequency, that is, the tire pressure, vehicle speed, load, tire wear and tire mode frequency, satisfying the basic relationship between the neural network output, output parameters, based on this, in this embodiment, the finite element tire is selected The most easy-to-identify radial modal rise in modal frequencies acts as a reference object. In the selected process, by Figure 2 ~ 5 ( figure 2 Relationship with the relationship between the tire mode frequency and modal step under different air pressure; image 3 Relationship between the radial mode frequency and modal step; Figure 4 Relationship between the radial mode frequency and modal step; Figure 5 It can be seen for the relationship between the radial modal frequency and modal step of different wear content), and the radial modal frequency of the tire has increased with the increase of inflatable pressure and tire grinding. figure 2 Medium pressure gradually rose from 0.18 MPa to 0.24 MPa, Figure 5 The medium tire wear is gradually rising from no wear 6mm. The overall trend is generally upward, and the load and speed increase ( image 3 Medium speed is gradually rising from 0 to 60km / h, Figure 4 The load is gradually increased from 2000N to 5000 N). However, from image 3 It can be seen that the radial first-order mode frequency is not sensitive to the speed of the vehicle, so in order to improve the prediction accuracy of the BP neural network model, the data of 2 to 6 steps rising mode frequency is selected as the input of the BP neural network model. The output is the amount of tire grinding.

[0092] After the input and output of the BP neural network model is selected, the BP neural network model is created, and the collected data set is randomly split into training set, verification set and test set, where the training set is for BP neural network models. Training, the verification set is used in the network training process to prevent the network from falling into the fitting, and the test set is used to test the performance of the network completed. For training sets, verification sets, and test sets, it can be set according to the actual situation, such as in an example, training set, verification set and test sets are 0.7, 0.2 and 0.1, respectively.

[0093] For the created BP neural network model, considering the amount of data collected and the number of input parameters is not large, in order to avoid complex, the hidden layer number is set to 1, and TAN SIG is used as the activation function. In order to determine the number of hidden layers in the hidden layer, in the creating BP neural network model, the hidden layer node scope determining unit first based on the number of hidden layers of the formula (2), the number of input layer nodes and output layer nodes The relationship between the hidden layer node is the range of the number of hidden layers, and the neural network creation unit establishes the corresponding BP neural network model according to the range of hidden layers. Based on the number of input parameters and output parameters in this embodiment, the range of implicit layer nodes can be initially determined within the [4,13] section, and this sequentially establishes the number of implicit layer nodes from 4 to increment. 13 BP neural network model.

[0094] During training, the loop algorithm set in the training unit will take training set and verification set into the BP neural network model, and calculate the number of preset times for the BP neural network model containing different hidden layer nodes (such as 50 Second, 100 or even more), and set the store automatically store the minimum mean square error MSE for each hidden layer node recirculation iterative training set (training method is the TrainLM training method, the calculation formula of MSE is (3) ), Automatically saving the BP neural network model corresponding to the current minimum mean square error MSE through the SAVE function. After all BP neural network model iterates, the network determination unit compares the mean square error of each iteration, and the number of nodes corresponding to the number of norms, the number of nodes corresponding to the best hidden layer node, and determine the BP neural network model The structure and network parameters, then use the BP neural network model to perform the data of the tire wear life. In predicts the test sets in the use of the well-trained BP neural network model, the prediction accuracy is measured by the absolute error ε of the tire wear prediction value and the actual value, as in the formula (4).

[0095] Since the acquisition data numerical gap is too large, in order to improve the accuracy of the estimated algorithm, in other embodiments, the data in the training set, verification set and test concentration needs to be normalized and used in order to disrupted with the Randperm function. Increasing the robustness of the estimation algorithm, the value of the data after normalization is between (0, 1), in line with the standard normal distribution, avoiding 0 in subsequent calculations; after the prediction is completed, Normalization. Specifically, normalization and inverselation are as embodied (5) to (7).

[0096] In one example, based on finite element modal analysis, combined with control variable method is extracted in a tire finite element simulation environment, 324 sets of different tire pressure, load, vehicle speed, and radial direction 2 to 6 steps. Simulation data between state frequencies acts as a data set to reduce the cost of testing and capital costs. Part of the data is as Image 6 As shown, each group of data includes wear, tire pressure, load, vehicle speed, radial 2-order tire modal frequency, radial 3-order tire mode frequency, radial 4-order tire mode frequency and radial 5th order tire Modal frequency 8 input parameters.

[0097] like Figure 13 As shown, after loading 324 sets of data samples, the DivideParam function is divided into training sets, verification sets and test sets, accounting for 0.7, 0.2 and 0.1, where the training set is used to train BP nerves. Network model, the verification set is used to prevent the network from being fed into the process, and the test set is used to test the performance of the network completed. The data in the training set, verification set and test concentration is then subsequently processed separately.

[0098] After creating a BP neural network model, the accuracy of the parameters of the setting model is 0.0004, the learning rate is 0.1, the maximum iterative number is 10,000 step, the hidden layer activation function is TAN SIG, the expression is The activation function of the output layer is Purelin, the expression is purelin (n) = n, the training method uses TrainLM adaptive LRBP gradient decreasing training function, and uses the [NET, TR] function to train, which is easy to obtain a detailed result of network training later. parameter.

[0099] In training for BP neural network models including hidden nodes in the [4,13] interval, the set loop algorithm will bring training sets and verification sets into the network, which hidden layers of nodes from 4 Up to 13, and repeatedly calculated 100 times at each node, set the store to automatically store the minimum mean square error MSE of each hidden layer node to calculate the training set, and automatically save the current minimum MSE through the SAVE function. The corresponding BP neural network model, the storage algorithm should be saved in the reservoir, and the minimum MSE corresponding to nodes 4 to 13 should be saved. like Figure 7 The MSE is shown as the hidden layer node number change curve, which can be seen from the figure, the best implicit layer node of the BP neural network model in this example is 11, the corresponding BP neural network model structure Figure 8 As shown, including wear, tire pressure, load, vehicle speed, radial 2-order tire modal frequency, radial 3-order tire mode frequency, radial 4-order tire mode frequency and radial 5-order tire mode frequency 8 An input parameter, a tire grinding capacity 1 output parameter, 1 hidden layer containing 11 hidden layers nodes. Figure 9 The relationship between the MSE and iterative steps (corresponding icons in the corresponding illustration) is 11 o'clock, that is, the training of the BP neural network model is stopped in step 45. The root mean square error MSE is 7.0026. E-4, lower than the set target value 4E5 (the maximum verification step of the verification set in the training process reaches the maximum number of setup steps, so the training of the network is terminated to prevent the production of the prefraction).

[0100] Prediction of test sets in the use of training BP neural network models, predictive results Figure 10 As shown, it can be seen that the error value in different test samples (samples in the test set) is small, and the average of the tire wear of the neural network is 0.0874 mm. Figure 11 The percentage of the error of the test set is predicted for the BP neural network model, and the percentage of prediction error is within ± 10%, with an average percentage of an average error, and the prediction accuracy is high.

[0101] At this time, the incentive function of each node of the hidden layer is as in the formula (8):

[0102]

[0103] Among them, O j For the incentives of the jth hidden node, j = 1, 2, ..., 11; TAN SIG is the transfer function of the implicit layer; W jl For the weight of the hidden layer, J20 to the weight of the output layer, the weight of the first neuron; X i For the i = 1, 2, ..., 8; θ j In order to impose the J20 of the hidden layer;

[0104] BP neural network model prediction tire wear amount M as in formula (9):

[0105]

[0106] Among them, the Purelin function is the incentive function of the output layer; W ij We value for the i-th hidden node to the sequence of the output layer; θ m The mth neuron threshold of the output layer.

[0107] After the BP neural network model, the data sample in the test concentration is predicted, and the prediction data is reversed, and the prediction accuracy is evaluated according to the formula (4) to determine the final tire wear life estimation. Neural Networks.

[0108] At this time, it is assumed that the thermometer thickness of the new tire is d. new MM, then the remaining wear life of the tire is use It can be formula (10) to achieve prediction of tire service life:

[0109]

[0110] In practical applications, the predicted tire wear life can be displayed on the automotive dashboard as the tire pressure. It can also be displayed in other ways, and the time to remind the driver's tire wear and life, solve the driver often Ignore the problem of tire status, thereby improving the safety of the car.

[0111] Those skilled in the art will appreciate that in order to describe convenient and concise, only the partitions of the various program modules are described, and the above-described functionality can be done by different program modules as needed. The internal structure of the device is divided into a different program unit or module to complete all or part of the above described above. The various program modules in the embodiment can be integrated into one processing unit, but each unit can be generated separately, or two or more units can be integrated in one processing unit, and the integrated unit can be implemented in the form of hardware. It can also be implemented in the form of a software program unit. In addition, the specific names of each program module are also intended to distinguish each other, and is not intended to limit the protection range of the present application.

[0112] Figure 13It is a schematic structural diagram of the terminal device provided in one embodiment of the present invention, as shown, the terminal device 200 includes a processor 220, a memory 210, and a computer program 211 stored in memory 210 and can run on processor 220. For example, a smart tire wear life estimate based on a BP neural network is estimated. When the processor 220 performs a computer program 211, the process is implemented in the step of the intelligent tire wear life estimation method embodiment based on the BP neural network, or when the processor 220 performs a computer program 211 to achieve the above-described BP neural network-based smart tire wear life. The function of each module in the device embodiment is estimated.

[0113] The terminal device 200 can be a notebook, a handheld computer, a flat-panel computer, a mobile phone and other devices. The terminal device 200 can include, but is not limited to the processor 220, the memory 210. Those skilled in the art will appreciate that Figure 13 It is only an example of the terminal device 200, and does not constitute a defined of the terminal device 200, and may include more or less components, or a combination of certain components, or different components, for example, the terminal device 200 can also include Enter the output device, display device, network access device, bus, etc.

[0114] Processor 220 can be a central processing unit (CPU), as well as other universal processors, digital signal processors, DSPs, dedicated integrated circuits (ASIC), on-site Field-Programmable Gate Array, FPGA or other programmable logic devices, separate doors or transistor logic devices, discrete hardware components, and the like. Universal processor 220 can be a microprocessor or the processor or any conventional processor or the like.

[0115] Memory 210 can be an internal storage unit of the terminal device 200, such as a hard disk or memory of the terminal device 200. The memory 210 may be an external storage device of the terminal device 200, for example, a plug-in hard disk equipped with a terminal device 200, a smart media card, SMC, secure digital (SD) card, flash card Flash Card, etc. Further, the memory 210 may also include both the internal storage unit of the terminal device 200, also including an external storage device. Memory 210 is used to store other programs and data required for computer program 211 and terminal device 200. Memory 210 can also be used to temporarily store data that has been output or will output.

[0116] In the above embodiment, each of the descriptions of each of the embodiments have each other, and there is no detailed description thereof in a detailed description, and a description of other embodiments can be found.

[0117] One of ordinary skill in the art will appreciate that the unit and algorithm steps described herein described herein can be accomplished by electronic hardware, or computer software and electronic hardware. These features are executed by hardware or software, depending on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each particular application, but this implementation should not be considered exceeded the scope of this application.

[0118] In the embodiments provided herein, it should be understood that the disclosed device / terminal apparatus and method can be implemented in other ways. For example, the device / terminal device embodiment described above is merely schematic, for example, the division of the module or unit is only one logical function division, and there may be additional division modifications, such as, multiple units or Components can be combined or can be integrated into another system, or some features can be ignored, or not executed. Another point, the coupling or direct coupling or communication connection of the displayed or discussed may be an indirect coupling or communication connection of the interface, the device, or unit, which may be electrical, mechanical or other forms.

[0119] As the unit illustrated as the separation component may be or may not be physically separated, the components displayed as the unit may be or may not be a physical unit, i.e., in one place, or can also be distributed to a plurality of network elements. The object of the present embodiment can be implemented in accordance with the actual needs to select some or all units.

[0120] Further, each functional unit in the various embodiments of the present application may be integrated into one processing unit, or each unit is generated separately, or two or more units can be integrated into one unit. The above-described integrated units can be implemented in the form of hardware, or may be implemented in the form of a software functional unit.

[0121] The integrated module / unit can be stored in a computer readable storage medium if implemented in the form of a software functional unit and as a stand-alone product sales or in use. Based on this, the present invention implements all or part of the flow in the above embodiment, or may be stored by the computer program 211 to the related hardware, and the computer program 211 can be stored in a computer readable storage medium, the computer program 211 When executed by the processor 220, the steps of the various method embodiments can be realized. The computer program 211 includes: computer program code, computer program code can be a source code form, object code form, executable file, or some intermediate form. Computer readable storage media can include any entity or device capable of carrying computer program 211 code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random Access memory (RAM, RANDOMACCESS MEMORY), electrical carrier signal, telecommunications signal, and software distribution media, etc. It should be noted that the contents of the computer readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice within the jurisdiction, such as: in some jurisdictions, according to legislation and patent practice, computer readable media Does not include electrical carrier signals and telecommunications signals.

[0122] It should be noted that the above embodiments can be freely combined as needed. The above is merely the preferred embodiment of the present invention, and it should be noted that several improvements and moisters can also be made without departing from the principles of the present invention, and these improvements and moisters should also be regarded. It is a range of protection of the present invention.