High-smoothness automobile crane suspension design method and device based on deep learning

Abstract

The invention discloses a high-smoothness automobile crane suspension design method and device based on deep learning, and relates to the technical field of crane suspension design, and the method comprises the steps: respectively obtaining a first training data set for each pavement spectrum; training a pre-constructed road surface vibration curve prediction model by using the first training data sets corresponding to the different road surface spectrums to obtain trained road surface vibration curve prediction models corresponding to the different road surface spectrums; the trained road surface vibration curve prediction model is utilized, the minimum smoothness index serves as an optimization target, and optimal suspension structure parameters corresponding to all the road surface spectrums are solved; constructing a second training data set according to the optimal suspension structure parameters corresponding to the road surface spectrums; training a pre-constructed optimal suspension structure parameter prediction model by using the second training data set; and carrying out suspension structure design by utilizing the trained optimal suspension structure parameter prediction model. According to the method, rapid optimization design of the high-smoothness crane suspension structure can be achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of crane suspension design, and in particular to a method and device for designing a high-smoothness automobile crane suspension based on deep learning. Background Art

[0002] The performance of a truck crane's suspension system during operation and driving directly impacts the vehicle's ride comfort, stability, and handling. Traditional crane suspension design methods rely on extensive experimental and empirical data for optimization, resulting in long design cycles and poor adaptability. This is particularly problematic when vehicle operating conditions involve complex vibration spectra, making it difficult to efficiently determine optimal design parameters. Furthermore, with the increasing diversity of vehicle loads and road conditions, the demand for more refined and intelligent suspension design is increasing.

[0003] Therefore, there is an urgent need for a truck crane suspension design method that can quickly respond to spectrum input and optimize design parameters. Summary of the Invention

[0004] In order to solve some or all of the technical problems existing in the above-mentioned prior art, the present invention provides a high-smoothness automobile crane suspension design method and device based on deep learning.

[0005] The technical solutions of the present invention are as follows:

[0006] First, a deep learning-based design method for a high-smoothness truck crane suspension is provided, including:

[0007] For each road surface spectrum, a first training data set is obtained, where the first training data set includes a plurality of first training data, each of which includes suspension structure parameters and corresponding cab acceleration vibration curves.

[0008] Using the pre-processed first training data sets corresponding to different road surface spectra to train pre-built road vibration curve prediction models, respectively, to obtain trained road vibration curve prediction models corresponding to different road surface spectra, wherein the input of the road vibration curve prediction model is the suspension structure parameters, and the output of the road vibration curve prediction model is the cab acceleration vibration curve;

[0009] Using the trained road vibration curve prediction model corresponding to each road surface spectrum, the optimal suspension structural parameters corresponding to each road surface spectrum are solved with minimizing the smoothness index as the optimization goal. Smoothness is defined as the smoothness of the cab acceleration vibration curve, and the smoothness index is defined as the root mean square value of the acceleration value of the cab acceleration vibration curve.

[0010] Constructing a second training data set based on the optimal suspension structure parameters corresponding to each road surface spectrum, wherein the second training data set includes a plurality of second training data, and the second training data includes road surface spectrum characteristic parameters and their corresponding optimal suspension structure parameters;

[0011] Using the preprocessed second training data set to train a pre-built optimal suspension structure parameter prediction model, the input of the optimal suspension structure parameter prediction model is the road surface spectrum characteristic parameter, and the output of the optimal suspension structure parameter prediction model is the suspension structure parameter;

[0012] The trained optimal suspension structure parameter prediction model is used to design the suspension structure.

[0013] In some optional implementations, the first training data set is obtained by:

[0014] For a road surface spectrum, multiple groups of suspension structure parameters are set;

[0015] According to the set multiple groups of suspension structure parameters, a suspension model corresponding to each group of suspension structure parameters is constructed;

[0016] Under the current road surface spectrum, dynamic simulation or testing is performed on the multiple suspension models constructed respectively, and the cab acceleration vibration curve corresponding to each suspension model is collected to obtain multiple first training data including suspension structure parameters and their corresponding cab acceleration vibration curves, and obtain the first training data set corresponding to the current road surface spectrum.

[0017] In some optional embodiments, the suspension structural parameters include: the stiffness of the suspension, the sprung mass, the unsprung mass, the type of leaf spring, the material of the leaf spring, the number of leaf springs, the thickness of the leaf spring, the prestress of the leaf spring, the clamping distance of the leaf spring, the length of the leaf spring, the full load of the leaf spring, the damping force of the shock absorber, the stroke of the shock absorber and the axle wheelbase.

[0018] In some optional embodiments, preprocessing the first training data set includes: normalizing the suspension structure parameters in the first training data;

[0019] The preprocessing of the second training data set includes: normalizing the road surface spectrum characteristic parameters in the second training data.

[0020] In some optional embodiments, the road vibration curve prediction model adopts a multi-layer feedforward neural network, the activation function of the multi-layer feedforward neural network adopts a Leaky ReLU activation function, and a Dropout layer is set between the hidden layers of the multi-layer feedforward neural network.

[0021] In some optional implementations, the root mean square value of the acceleration value of the cab acceleration vibration curve is expressed as:

[0022]

[0023] Where RMS represents the root mean square value of the acceleration value of the cab acceleration vibration curve, T represents the time series length, t represents the time variable, and y(t) represents the cab acceleration value at time t.

[0024] In some optional implementations, the optimal suspension structure parameters corresponding to the road surface spectrum are solved in the following manner:

[0025] Step 301, setting initial values ​​of suspension structure parameters for a road surface spectrum;

[0026] Step 302: Input the suspension structure parameters into a trained road vibration curve prediction model corresponding to the current road surface spectrum to obtain a cab acceleration vibration curve output by the road vibration curve prediction model;

[0027] In step 303, the smoothness index is calculated based on the cab acceleration vibration curve output by the road vibration curve prediction model, and it is determined whether the smoothness index reaches the minimum value. If so, the current suspension structure parameters are used as the optimal suspension structure parameters corresponding to the current road surface spectrum. If not, the suspension structure parameters are used as optimization variables, and the suspension structure parameters are iteratively updated using the gradient descent method, and the process returns to step 302.

[0028] In some optional implementations, the road surface spectrum characteristic parameters include: road surface frequency spectrum and vibration intensity.

[0029] In some optional implementations, the optimal suspension structure parameter prediction model uses a multi-layer perceptron, and the activation function of the multi-layer perceptron uses a Leaky ReLU activation function.

[0030] Secondly, a high-smoothness truck crane suspension design device based on deep learning is also provided, including:

[0031] a first training data set acquisition unit, configured to acquire a first training data set corresponding to each road surface spectrum, wherein the first training data set includes a plurality of first training data, each of which includes suspension structure parameters and corresponding cab acceleration vibration curves;

[0032] a road vibration curve prediction model generating unit, configured to train a pre-established road vibration curve prediction model using a first training data set corresponding to different road surface spectra after preprocessing, to obtain trained road vibration curve prediction models corresponding to different road surface spectra, wherein the input of the road vibration curve prediction model is suspension structure parameters, and the output of the road vibration curve prediction model is a cab acceleration vibration curve;

[0033] An optimal suspension structural parameter solving unit is used to solve the optimal suspension structural parameters for each road surface spectrum using a trained road vibration curve prediction model corresponding to each road surface spectrum, with minimizing a smoothness index as the optimization objective. Smoothness is defined as the smoothness of the cab acceleration vibration curve, and the smoothness index is defined as the root mean square value of the acceleration value of the cab acceleration vibration curve;

[0034] a second training data set acquisition unit, configured to construct a second training data set based on the optimal suspension structure parameters corresponding to each road surface spectrum, wherein the second training data set includes a plurality of second training data, and the second training data includes road surface spectrum characteristic parameters and their corresponding optimal suspension structure parameters;

[0035] an optimal suspension structure parameter prediction model generating unit, configured to train a pre-constructed optimal suspension structure parameter prediction model using the pre-processed second training data set, wherein the input of the optimal suspension structure parameter prediction model is the road surface spectrum characteristic parameter, and the output of the optimal suspension structure parameter prediction model is the suspension structure parameter;

[0036] The suspension structure design unit is used to perform suspension structure design using the trained optimal suspension structure parameter prediction model.

[0037] The main advantages of the technical solution of the present invention are as follows:

[0038] The deep learning-based high-smoothness automobile crane suspension design method and device of the present invention constructs and trains a road vibration curve prediction model to fit the mapping relationship between the suspension structure parameters and the cab acceleration vibration curve, and then uses the road vibration curve prediction model to determine the optimal suspension structure parameters corresponding to different road surface spectra. The optimal suspension structure parameter prediction model is trained using different road surface spectrum characteristic parameters and their corresponding optimal suspension structure parameters. The optimal suspension structure parameter prediction model can be used to quickly predict the optimal suspension structure parameters under any road surface spectrum conditions, and the accuracy and reliability of the predicted optimal suspension structure parameters can be guaranteed, thereby realizing the rapid optimization design of the high-smoothness automobile crane suspension structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings, which are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application. In the drawings:

[0040] Figure 1 A flowchart of a high-ride comfort automobile crane suspension design method based on deep learning provided by an embodiment of the application;

[0041] Figure 2 A structural schematic diagram of a high-ride comfort automobile crane suspension design device based on deep learning provided by an embodiment of the application. DETAILED DESCRIPTION

[0042] To make the objectives, technical solutions and advantages of the application clearer, the technical solutions of the application will be described below in detail with reference to the embodiments of the application and the corresponding drawings. Obviously, the described embodiments are only some of the embodiments of the application, but not all the embodiments of the application. Based on the embodiments of the application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the application.

[0043] The technical solutions provided by the embodiments of the application will be described in detail below with reference to the drawings.

[0044] Reference Figure 1 In a first aspect, the embodiments of the application provide a high-ride comfort automobile crane suspension design method based on deep learning, which comprises the following steps:

[0045] Step 1: For each road spectrum, a first training data set is obtained respectively.

[0046] In the embodiments of the application, each road spectrum has a corresponding first training data set, and the first training data set comprises a plurality of first training data. The first training data comprises a suspension structure parameter and a corresponding cab acceleration vibration curve.

[0047] In the embodiments of the application, the types of road spectrum are set according to actual requirements.

[0048] Step 2: The pre-constructed road vibration curve prediction model is trained by using the first training data set corresponding to the different road spectrum after preprocessing, to obtain a trained road vibration curve prediction model corresponding to the different road spectrum.

[0049] In the embodiments of the application, the road vibration curve prediction model is constructed based on a neural network.

[0050] In the embodiment of the present invention, the input of the road surface vibration curve prediction model is the suspension structure parameters, and the output of the road surface vibration curve prediction model is the cab acceleration vibration curve.

[0051] In the embodiment of the present invention, each road surface spectrum has a corresponding trained road surface vibration curve prediction model.

[0052] Step 3: Using the trained road vibration curve prediction model corresponding to each road surface spectrum, the optimal suspension structural parameters corresponding to each road surface spectrum are solved with minimizing the smoothness index as the optimization goal;

[0053] In the embodiment of the present invention, the smoothness is defined as: the smoothness of the cab acceleration vibration curve; the smoothness index is defined as: the root mean square value of the acceleration value of the cab acceleration vibration curve.

[0054] Among them, the smaller the smoothness index is, the smoother the cab acceleration vibration curve is and the higher the smoothness is.

[0055] Step 4: construct a second training data set based on the optimal suspension structure parameters corresponding to each road surface spectrum;

[0056] In an embodiment of the present invention, the second training data set includes a plurality of second training data, and the second training data includes road surface spectrum characteristic parameters and their corresponding optimal suspension structure parameters.

[0057] In the embodiment of the present invention, a second training data set is constructed based on the optimal suspension structure parameters corresponding to each road surface spectrum obtained in step 3. The optimal suspension structure parameters corresponding to one road surface spectrum can determine one second training data set.

[0058] Step 5: using the pre-processed second training data set to train the pre-built optimal suspension structure parameter prediction model;

[0059] In an embodiment of the present invention, the optimal suspension structure parameter prediction model is constructed based on a neural network.

[0060] In the embodiment of the present invention, the input of the optimal suspension structure parameter prediction model is the road surface spectrum characteristic parameter, and the output of the optimal suspension structure parameter prediction model is the suspension structure parameter.

[0061] Step 6: Use the trained optimal suspension structure parameter prediction model to design the suspension structure.

[0062] In an embodiment of the present invention, the specified road surface spectrum characteristic parameters are input into a trained optimal suspension structure parameter prediction model to obtain suspension structure parameters corresponding to the specified road surface spectrum characteristic parameters output by the optimal suspension structure parameter prediction model. The suspension structure parameters output by the optimal suspension structure parameter prediction model are the optimal suspension structure parameters corresponding to the specified road surface spectrum characteristic parameters, and the suspension structure design is performed based on the obtained optimal suspension structure parameters.

[0063] The high-smoothness automobile crane suspension design method based on deep learning provided by an embodiment of the present invention constructs and trains a road vibration curve prediction model to fit the mapping relationship between suspension structure parameters and cab acceleration vibration curves, and then uses the road vibration curve prediction model to determine the optimal suspension structure parameters corresponding to different road surface spectra. The optimal suspension structure parameter prediction model is trained using different road surface spectrum characteristic parameters and their corresponding optimal suspension structure parameters. In this way, the optimal suspension structure parameter prediction model can be used to quickly predict the optimal suspension structure parameters under any road surface spectrum conditions, and the accuracy and reliability of the predicted optimal suspension structure parameters can be guaranteed, thereby realizing the rapid optimization design of the high-smoothness automobile crane suspension structure.

[0064] Furthermore, in an embodiment of the present invention, the first training data set is obtained by:

[0065] For a road surface spectrum, multiple groups of suspension structure parameters are set;

[0066] According to the set multiple groups of suspension structure parameters, a suspension model corresponding to each group of suspension structure parameters is constructed;

[0067] Under the current road surface spectrum, dynamic simulation or testing is performed on the multiple suspension models constructed respectively, and the cab acceleration vibration curve corresponding to each suspension model is collected to obtain multiple first training data including suspension structure parameters and their corresponding cab acceleration vibration curves, and obtain the first training data set corresponding to the current road surface spectrum.

[0068] In the embodiment of the present invention, the above method can quickly obtain the first training data set corresponding to each road surface spectrum, and ensure the accuracy of the obtained first training data set.

[0069] In the embodiment of the present invention, the types and values ​​of the suspension structure parameters are specifically set according to actual conditions.

[0070] Specifically, for the leaf spring suspension structure, the suspension structure parameters include: the stiffness of the suspension, the sprung mass, the unsprung mass, the type of leaf spring, the material of the leaf spring, the number of leaf springs, the thickness of the leaf spring, the prestress of the leaf spring, the clamping distance of the leaf spring, the length of the leaf spring, the full load of the leaf spring, the damping force of the shock absorber, the stroke of the shock absorber and the axle wheelbase.

[0071] Further, in the embodiment of the present application, in step 2, the preprocessing includes: performing normalization processing on the suspension structure parameters in the first training data.

[0072] In the embodiment of the present application, by performing normalization processing on the suspension structure parameters, the dimensional differences of different parameters can be eliminated, the parameter scales are unified, it is ensured that each parameter in the suspension structure parameters input into the model has a consistent data range, the influences of each parameter are balanced, and the model generalization capability is improved.

[0073] Further, in the embodiment of the present application, the road surface vibration curve prediction model adopts a multi-layer feedforward neural network (FNN), the activation function of the multi-layer feedforward neural network adopts a Leaky ReLU activation function, and a Dropout layer is arranged between the hidden layers of the multi-layer feedforward neural network.

[0074] In the embodiment of the present application, by constructing the road surface vibration curve prediction model by using the above structure, the complex nonlinear relationship between the suspension structure parameters and the acceleration vibration response can be fully captured by the road surface vibration curve prediction model; by using the Leaky ReLU activation function, the training effect of the model can be improved; and by arranging the Dropout layer, the model can be prevented from excessively relying on the training data, overfitting can be prevented, and the model generalization capability can be improved.

[0075] In the embodiment of the present application, the multi-layer feedforward neural network includes an input layer, an output layer, and a plurality of hidden layers arranged between the input layer and the output layer, and the output of each hidden layer of the multi-layer feedforward neural network can be expressed as:

[0076] h l =f(w l h l-1 +b l );

[0077] wherein h l represents the output of the lth layer of the multi-layer feedforward neural network, h l-1 represents the output of the (l-1)th layer of the multi-layer feedforward neural network, w l represents the weight matrix of the lth layer of the multi-layer feedforward neural network, b l represents the bias coefficient of the lth layer of the multi-layer feedforward neural network, and f represents the activation function of the multi-layer feedforward neural network.

[0078] wherein the activation function adopts a Leaky ReLU activation function.

[0079] Furthermore, setting the dropout rate of the Dropout layer to p, the output of each layer after the Dropout layer processing can be expressed as:

[0080]

[0081] in, Represents the output of the lth layer after being processed by the Dropout layer.

[0082] In the embodiment of the present invention, the Dropout layer enhances the generalization ability of the network by randomly discarding some neurons.

[0083] Furthermore, in an embodiment of the present invention, in step 2, the road surface vibration curve prediction model is trained in the following manner:

[0084] The preprocessed suspension structure parameters are used as the input of the road vibration curve prediction model, and the cab acceleration vibration curve corresponding to the suspension structure parameters is used as the output of the road vibration curve prediction model to train the road vibration curve prediction model.

[0085] In an embodiment of the present invention, the road vibration curve prediction model is trained using the preprocessed suspension structural parameters as input and the cab acceleration vibration curve corresponding to the suspension structural parameters as output, further comprising the following steps:

[0086] Step 201: input the pre-processed suspension structural parameters into a road vibration curve prediction model to obtain a cab acceleration vibration curve prediction result output by the road vibration curve prediction model;

[0087] Step 202 , calculating a first loss function based on a cab acceleration vibration curve corresponding to the input suspension structural parameters and a cab acceleration vibration curve prediction result output by a road surface vibration curve prediction model corresponding to the input suspension structural parameters;

[0088] Step 203, determine whether the preset training stop condition is met, if so, use the current road vibration curve prediction model as the road vibration curve prediction model to complete the training, if not, use the first loss function to update the parameters of the road vibration curve prediction model, and return to step 201.

[0089] In the embodiment of the present invention, before training begins, the parameters of each layer of the road surface vibration curve prediction model are all initialized parameters. During the training process, the parameters of each layer of the road surface vibration curve prediction model are continuously updated.

[0090] In this embodiment of the present invention, the first loss function adopts the mean square error (MSE) loss function, which is specifically expressed as:

[0091]

[0092] Among them, L1 represents the first loss function value, T represents the length of the time series, and t represents the time variable. It represents the predicted value of the cab acceleration at time t, and y(t) represents the cab acceleration value at time t.

[0093] Among them, the predicted value of cab acceleration The cab acceleration value y(t) is directly obtained from the prediction result of the cab acceleration vibration curve.

[0094] In an embodiment of the present invention, the training stop condition is specifically set according to the actual situation, for example, the number of training iterations reaches a set number or the optimization index reaches a set threshold, wherein the optimization index can adopt the above-mentioned first loss function value.

[0095] In an embodiment of the present invention, a gradient descent method is used to iteratively update the parameters of the road surface vibration curve prediction model.

[0096] Furthermore, in the embodiment of the present invention, in step 3, the root mean square value of the acceleration value of the cab acceleration vibration curve is expressed as:

[0097]

[0098] Where RMS represents the root mean square value of the acceleration value of the cab acceleration vibration curve, T represents the time series length, t represents the time variable, y(t) represents the cab acceleration value at time t, and the cab acceleration value y(t) is directly obtained from the cab acceleration vibration curve.

[0099] Furthermore, in the embodiment of the present invention, in step 3, the optimal suspension structure parameters corresponding to the road surface spectrum are solved in the following manner:

[0100] Step 301, setting initial values ​​of suspension structure parameters for a road surface spectrum;

[0101] Step 302: Input the suspension structure parameters into a trained road vibration curve prediction model corresponding to the current road surface spectrum to obtain a cab acceleration vibration curve output by the road vibration curve prediction model;

[0102] In step 303, the smoothness index is calculated based on the cab acceleration vibration curve output by the road vibration curve prediction model, and it is determined whether the smoothness index reaches the minimum value. If so, the current suspension structure parameters are used as the optimal suspension structure parameters corresponding to the current road surface spectrum. If not, the suspension structure parameters are used as optimization variables, and the suspension structure parameters are iteratively updated using the gradient descent method, and the process returns to step 302.

[0103] In the embodiment of the present invention, for each road surface spectrum, the optimal suspension structure parameters are solved respectively in the above-mentioned manner, thereby determining the optimal suspension structure parameters corresponding to each road surface spectrum.

[0104] In an embodiment of the present invention, iteratively updating the suspension structure parameters by the gradient descent method can be expressed as:

[0105]

[0106] Among them, X t+1 represents the suspension structure parameters at the t+1th iteration, X t represents the suspension structure parameter at the tth iteration, η represents the learning rate, It represents the gradient of the smoothness index with respect to the suspension structure parameters. The learning rate is set according to actual needs. The gradient of the smoothness index with respect to the suspension structure parameters represents the rate of change of the smoothness index relative to the suspension structure parameters under the current suspension structure parameters.

[0107] In the embodiment of the present invention, by solving the optimal suspension structure parameters corresponding to each road surface spectrum in the above manner, the optimal suspension structure parameters corresponding to each road surface spectrum can be quickly solved and the accuracy of the solved optimal suspension structure parameters can be guaranteed.

[0108] Furthermore, in the embodiment of the present invention, the road surface spectrum characteristic parameters are specifically set according to actual conditions.

[0109] Specifically, in the embodiment of the present invention, the road surface spectrum characteristic parameters include: road surface frequency spectrum and vibration intensity.

[0110] Furthermore, in the embodiment of the present invention, in step 5, the preprocessing includes: performing normalization processing on the road surface spectrum characteristic parameters in the second training data.

[0111] In the embodiment of the present invention, by normalizing the pavement spectrum characteristic parameters, the dimensional differences of different parameters can be eliminated, the parameter scales can be unified, and the pavement spectrum characteristic parameters of the input model can be ensured to have a consistent data range. The influence of each parameter is balanced, and the generalization ability of the model is improved.

[0112] Furthermore, in an embodiment of the present invention, the optimal suspension structure parameter prediction model adopts a multilayer perceptron (MLP), and the activation function of the multilayer perceptron adopts a Leaky ReLU activation function.

[0113] In an embodiment of the present invention, by adopting the above-mentioned structure to construct an optimal suspension structure parameter prediction model, the optimal suspension structure parameter prediction model can fully capture the complex relationship between the road surface spectrum characteristic parameters and the optimal suspension structure parameters; by adopting the Leaky ReLU activation function, the training effect of the model can be improved.

[0114] In an embodiment of the present invention, the multilayer perceptron includes an input layer, an output layer, and a plurality of hidden layers arranged between the input layer and the output layer.

[0115] Specifically, assuming that the multilayer perceptron includes L hidden layers, the output of the L hidden layers of the multilayer perceptron can be expressed as:

[0116]

[0117] The output of the output layer of the multi-layer perceptron can be expressed as:

[0118] X * =W out H L +B out ;

[0119] Among them, H1 represents the output of the first hidden layer of the multilayer perceptron, H2 represents the output of the second hidden layer of the multilayer perceptron, and H L-1 Represents the output of the L-1 hidden layer of the multilayer perceptron, H L represents the output of the Lth hidden layer of the multilayer perceptron, S represents the vector containing the road spectrum characteristic parameter information received by the input layer, W1 and B1 represent the weight matrix and bias coefficient of the first hidden layer of the multilayer perceptron, W2 and B2 represent the weight matrix and bias coefficient of the second hidden layer of the multilayer perceptron, W L and B L They represent the weight matrix and bias coefficient of the Lth hidden layer of the multilayer perceptron, σ represents the activation function of the multilayer perceptron, X * Represents the output of the multilayer perceptron output layer, W out and B out Represent the weight matrix and bias coefficient of the multilayer perceptron output layer respectively.

[0120] Among them, the activation function adopts the Leaky ReLU activation function.

[0121] In the embodiment of the present invention, the output X of the output layer of the multi-layer perceptron is * That is the predicted optimal suspension structure parameters.

[0122] Furthermore, in an embodiment of the present invention, in step 5, the optimal suspension structure parameter prediction model is trained in the following manner:

[0123] The preprocessed road surface spectrum characteristic parameters are used as the input of the optimal suspension structure parameter prediction model, and the optimal suspension structure parameters corresponding to the road surface spectrum characteristic parameters are used as the output of the optimal suspension structure parameter prediction model to train the optimal suspension structure parameter prediction model.

[0124] In an embodiment of the present invention, the preprocessed road surface spectrum characteristic parameters are used as the input of the optimal suspension structure parameter prediction model, and the optimal suspension structure parameters corresponding to the road surface spectrum characteristic parameters are used as the output of the optimal suspension structure parameter prediction model. Training the optimal suspension structure parameter prediction model further includes the following steps:

[0125] Step 501: inputting the pre-processed road surface spectrum characteristic parameters in the plurality of second training data into the optimal suspension structure parameter prediction model in sequence, and obtaining the optimal suspension structure parameter prediction result corresponding to each road surface spectrum characteristic parameter output by the optimal suspension structure parameter prediction model;

[0126] Step 502 , calculating a second loss function based on the optimal suspension structure parameters corresponding to the input road surface spectrum characteristic parameters and the optimal suspension structure parameter prediction results output by the optimal suspension structure parameter prediction model corresponding to the input road surface spectrum characteristic parameters;

[0127] Step 503, determine whether the preset training stop condition is met, if so, use the current optimal suspension structure parameter prediction model as the optimal suspension structure parameter prediction model to complete the training, if not, use the second loss function to update the parameters of the optimal suspension structure parameter prediction model, and return to step 501.

[0128] In the embodiment of the present invention, before training begins, the parameters of each layer of the optimal suspension structure parameter prediction model are all initialized parameters. During the training process, the parameters of each layer of the optimal suspension structure parameter prediction model are continuously updated.

[0129] In this embodiment of the present invention, the second loss function adopts the mean square error (MSE) loss function, which is specifically expressed as:

[0130]

[0131] Where L2 represents the second loss function value, N represents the number of second training data used for training, and X true,i represents the optimal suspension structure parameters in the i-th training data, It represents the optimal suspension structure parameter prediction result output by the optimal suspension structure parameter prediction model corresponding to the road surface spectrum characteristic parameter in the i-th training data.

[0132] In an embodiment of the present invention, the training stop condition is specifically set according to the actual situation, for example, the number of training iterations reaches a set number or the optimization index reaches a set threshold, wherein the optimization index can adopt the above-mentioned second loss function value.

[0133] In an embodiment of the present invention, a gradient descent method is used to iteratively update the parameters of the optimal suspension structure parameter prediction model.

[0134] Specifically, the parameters of the optimal suspension structure parameter prediction model are updated using the following formula:

[0135]

[0136] Among them, θ t+1 represents the parameters of the optimal suspension structure parameter prediction model at the t+1th iteration, θ t represents the parameters of the optimal suspension structure parameter prediction model at iteration t, η2 represents the learning rate, and θ represents the parameters of the optimal suspension structure parameter prediction model. The learning rate is pre-set to control the speed of model parameter updates. The parameters of the optimal suspension structure parameter prediction model include the weight matrix and bias coefficient.

[0137] refer to Figure 2 In a second aspect, an embodiment of the present invention provides a high-smoothness truck crane suspension design device based on deep learning, the device comprising:

[0138] a first training data set acquisition unit, configured to acquire a first training data set corresponding to each road surface spectrum, the first training data set including a plurality of first training data, the first training data including suspension structure parameters and corresponding cab acceleration vibration curves;

[0139] a road vibration curve prediction model generation unit, configured to train a pre-established road vibration curve prediction model using a first training data set corresponding to different road surface spectra after preprocessing to obtain trained road vibration curve prediction models corresponding to different road surface spectra, wherein the input of the road vibration curve prediction model is the suspension structure parameters, and the output of the road vibration curve prediction model is the cab acceleration vibration curve;

[0140] An optimal suspension structural parameter solving unit is used to solve the optimal suspension structural parameters for each road surface spectrum using a trained road vibration curve prediction model corresponding to each road surface spectrum, with minimizing a smoothness index as the optimization objective. Smoothness is defined as the smoothness of the cab acceleration vibration curve, and the smoothness index is defined as the root mean square value of the acceleration value of the cab acceleration vibration curve;

[0141] a second training data set acquisition unit, configured to construct a second training data set based on the optimal suspension structure parameters corresponding to each road surface spectrum, wherein the second training data set includes a plurality of second training data, and the second training data includes road surface spectrum characteristic parameters and their corresponding optimal suspension structure parameters;

[0142] an optimal suspension structure parameter prediction model generating unit, configured to train a pre-built optimal suspension structure parameter prediction model using the pre-processed second training data set, wherein the input of the optimal suspension structure parameter prediction model is the road surface spectrum characteristic parameter, and the output of the optimal suspension structure parameter prediction model is the suspension structure parameter;

[0143] The suspension structure design unit is used to perform suspension structure design using the trained optimal suspension structure parameter prediction model.

[0144] Each unit in the high-smoothness automobile crane suspension design device based on deep learning provided by an embodiment of the present invention is a device corresponding to the steps of the above-mentioned method, and can implement all the processes of the high-smoothness automobile crane suspension design method based on deep learning described in any of the above-mentioned embodiments. The specific working principle, function and technical effect achieved by each unit are respectively the same as those corresponding to the high-smoothness automobile crane suspension design method based on deep learning described in the above-mentioned embodiments, and will not be repeated here.

[0145] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In addition, "front", "back", "left", "right", "upper" and "lower" in this document are all referenced to the placement states shown in the accompanying drawings.

[0146] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A high-smoothness truck crane suspension design method based on deep learning, characterized by: include: For each road surface spectrum, a first training data set is obtained, where the first training data set includes a plurality of first training data, each of which includes suspension structure parameters and corresponding cab acceleration vibration curves. Using the pre-processed first training data sets corresponding to different road surface spectra to train pre-built road vibration curve prediction models, respectively, to obtain trained road vibration curve prediction models corresponding to different road surface spectra, wherein the input of the road vibration curve prediction model is the suspension structure parameters, and the output of the road vibration curve prediction model is the cab acceleration vibration curve; Using the trained road vibration curve prediction model corresponding to each road surface spectrum, the optimal suspension structural parameters corresponding to each road surface spectrum are solved with minimizing the smoothness index as the optimization goal. Smoothness is defined as the smoothness of the cab acceleration vibration curve, and the smoothness index is defined as the root mean square value of the acceleration value of the cab acceleration vibration curve. Constructing a second training data set based on the optimal suspension structure parameters corresponding to each road surface spectrum, wherein the second training data set includes a plurality of second training data, and the second training data includes road surface spectrum characteristic parameters and their corresponding optimal suspension structure parameters; Using the preprocessed second training data set to train a pre-built optimal suspension structure parameter prediction model, the input of the optimal suspension structure parameter prediction model is the road surface spectrum characteristic parameter, and the output of the optimal suspension structure parameter prediction model is the suspension structure parameter; The trained optimal suspension structure parameter prediction model is used to design the suspension structure.

2. The high-ride comfort vehicle crane suspension design method based on deep learning according to claim 1 is characterized in that: The first training data set is obtained by: For a road surface spectrum, multiple groups of suspension structure parameters are set; According to the set multiple groups of suspension structure parameters, a suspension model corresponding to each group of suspension structure parameters is constructed; Under the current road surface spectrum, dynamic simulation or testing is performed on the multiple suspension models constructed respectively, and the cab acceleration vibration curve corresponding to each suspension model is collected to obtain multiple first training data including suspension structure parameters and their corresponding cab acceleration vibration curves, and obtain the first training data set corresponding to the current road surface spectrum.

3. The high-ride comfort vehicle crane suspension design method based on deep learning according to claim 1 is characterized in that: The suspension structural parameters include: suspension stiffness, sprung mass, unsprung mass, type of leaf spring, material of leaf spring, number of leaf springs, thickness of leaf spring, prestress of leaf spring, clamping distance of leaf spring, length of leaf spring, full load of leaf spring, damping force of shock absorber, stroke of shock absorber and axle wheelbase.

4. The high-ride comfort truck crane suspension design method based on deep learning according to claim 1 is characterized in that: Preprocessing the first training data set includes: normalizing the suspension structure parameters in the first training data; The preprocessing of the second training data set includes: normalizing the road surface spectrum characteristic parameters in the second training data.

5. The high-ride comfort vehicle crane suspension design method based on deep learning according to claim 1 is characterized in that: The road vibration curve prediction model adopts a multi-layer feedforward neural network, the activation function of the multi-layer feedforward neural network adopts a Leaky ReLU activation function, and a Dropout layer is set between the hidden layers of the multi-layer feedforward neural network.

6. The high-ride comfort vehicle crane suspension design method based on deep learning according to claim 1 is characterized in that: The root mean square value of the acceleration value of the cab acceleration vibration curve is expressed as: Where RMS represents the root mean square value of the acceleration value of the cab acceleration vibration curve, T represents the time series length, t represents the time variable, and y(t) represents the cab acceleration value at time t.

7. The high-ride comfort vehicle crane suspension design method based on deep learning according to claim 1 is characterized in that: The optimal suspension structure parameters corresponding to the road surface spectrum are solved in the following way: Step 301, setting initial values ​​of suspension structure parameters for a road surface spectrum; Step 302: Input the suspension structure parameters into a trained road vibration curve prediction model corresponding to the current road surface spectrum to obtain a cab acceleration vibration curve output by the road vibration curve prediction model; In step 303, the smoothness index is calculated based on the cab acceleration vibration curve output by the road vibration curve prediction model, and it is determined whether the smoothness index reaches the minimum value. If so, the current suspension structure parameters are used as the optimal suspension structure parameters corresponding to the current road surface spectrum. If not, the suspension structure parameters are used as optimization variables, and the suspension structure parameters are iteratively updated using the gradient descent method, and the process returns to step 302.

8. The high-ride comfort vehicle crane suspension design method based on deep learning according to claim 1 is characterized in that: The road surface spectrum characteristic parameters include: road surface spectrum and vibration intensity.

9. The method for designing a high-ride truck crane suspension based on deep learning according to claim 1, characterized in that: The optimal suspension structure parameter prediction model adopts a multi-layer perceptron, and the activation function of the multi-layer perceptron adopts the LeakyReLU activation function.

10. A high-smoothness vehicle crane suspension design device based on deep learning, characterized in that: include: a first training data set acquisition unit, configured to acquire a first training data set corresponding to each road surface spectrum, wherein the first training data set includes a plurality of first training data, each of which includes suspension structure parameters and corresponding cab acceleration vibration curves; a road vibration curve prediction model generating unit, configured to train a pre-established road vibration curve prediction model using a first training data set corresponding to different road surface spectra after preprocessing, to obtain trained road vibration curve prediction models corresponding to different road surface spectra, wherein the input of the road vibration curve prediction model is suspension structure parameters, and the output of the road vibration curve prediction model is a cab acceleration vibration curve; An optimal suspension structural parameter solving unit is used to solve the optimal suspension structural parameters for each road surface spectrum using a trained road vibration curve prediction model corresponding to each road surface spectrum, with minimizing a smoothness index as the optimization objective. Smoothness is defined as the smoothness of the cab acceleration vibration curve, and the smoothness index is defined as the root mean square value of the acceleration value of the cab acceleration vibration curve; a second training data set acquisition unit, configured to construct a second training data set based on the optimal suspension structure parameters corresponding to each road surface spectrum, wherein the second training data set includes a plurality of second training data, and the second training data includes road surface spectrum characteristic parameters and their corresponding optimal suspension structure parameters; an optimal suspension structure parameter prediction model generating unit, configured to train a pre-constructed optimal suspension structure parameter prediction model using the pre-processed second training data set, wherein the input of the optimal suspension structure parameter prediction model is the road surface spectrum characteristic parameter, and the output of the optimal suspension structure parameter prediction model is the suspension structure parameter; The suspension structure design unit is used to perform suspension structure design using the trained optimal suspension structure parameter prediction model.

Images