Disclosure of Invention
In order to overcome the defects of the prior art, the invention aims to provide a vehicle shock absorber defect diagnosis method, system, device and medium, so as to solve the problems that a plurality of sensors are required for fault detection in the prior art, the detection of the shock absorber in an unmanned vehicle is not comprehensive enough, and the detection result is not accurate enough.
In order to achieve the purpose, the invention adopts the following technical scheme:
in a first aspect, the present invention provides a vehicle shock absorber defect diagnosis method, comprising:
s1: collecting various data of the vehicle and preprocessing the data;
s2: inputting the preprocessed data into a deep learning framework, and performing convolution on the data through a convolution neural network introduced into an SE module to obtain a feature map; and extracting and classifying the features through the pooling layer and the full-connection layer, performing fault classification diagnosis through the ReLU, and outputting a fault classification result.
Further, before data collection, the defect behavior of the air door is simulated by changing the current of the valve; performing test runs at different wheel speeds on different surfaces; the current supply to individual and all dampers was varied, simulating both individual defects and multiple defects.
Further, the various items of data of the vehicle operation include yaw rate, lateral longitudinal acceleration, steering angle and wheel speed.
Further, the yaw rate and the steering angle are obtained through accessing bus communication to obtain data of an anti-lock brake system and a vehicle body electronic stability system, and the yaw rate and the steering angle measured by a yaw rate sensor are collected.
Further, the pretreatment specifically includes: cutting and filtering collected data, dividing a measurement file into a plurality of sequences, fixing the length of the sequences, keeping the number of data points the same, standardizing, and using a discrete Fourier transform signal processing method:
wherein X (k) represents the data after DFT transform, N represents the number of points of fourier transform, k represents the kth spectrum of fourier transform, X (N) is the sampled analog signal, X (N) is a complex or real signal, the imaginary part is 0, and the formula is developed as:
processing by short-time Fourier transform:
R ij =θ(ε-||X i -X j ||)
for time series signal u k (k =1, 2.. Multidot., n), determining the sampling time interval as Δ t, determining the embedding dimension m and the delay time τ, and further reconstructing the event sequence, wherein the reconstructed dynamic system is as follows:
x i =[u i ,u i+τ ,...,u i+(m-1)τ ]
i point x in reconstructed phase space i And j point x j The distance of (c):
S ij =||X i -X j ||
wherein R is ij For recursive values, the delay factor τ, the embedding dimension m, and the threshold ε.
Further, the preprocessed data are input into a deep learning framework, and the data are convoluted through a convolution neural network introduced into an SE module to obtain a feature map; extracting and classifying the features through the pooling layer and the full connection layer, performing fault classification diagnosis through the ReLU, and outputting a fault classification result, wherein the steps comprise:
inputting the processed data into a convolutional neural network, firstly extracting features by using convolution filters with different sizes and parallel of deep separable convolutions, and introducing an SE module into the convolutional neural network; convolution filters of different sizes are used in several branches, and depth separable convolutions are used in one branch; the depth separable convolution firstly carries out channel-by-channel convolution on the channel depth respectively, splicing the output, and then carrying out channel convolution by using a unit convolution kernel to obtain a characteristic diagram; inputting the characteristic diagram into the pooling layer to select the characteristics extracted from the convolution layer, extracting the graphic characteristics, inputting the graphic characteristics into the full-connection layer, and converting all characteristic matrixes of the pooling layer into one-dimensional characteristic large vectors by the full-connection layer; fault classification diagnosis is carried out through the ReLU; and finally, outputting a fault classification result.
Further, the fault classification diagnosis includes whether the four air doors at the front left, the front right, the rear left and the rear right are defective.
In a second aspect, the present invention provides a vehicle shock absorber defect diagnosis system comprising:
the data collection and pretreatment module is used for collecting and pretreating various data of vehicle operation;
the feature extraction and fault classification output module is used for inputting the preprocessed data into the deep learning framework and convolving the data by introducing the convolutional neural network of the SE module to obtain a feature map; and extracting and classifying the features through the pooling layer and the full-connection layer, performing fault classification diagnosis through the ReLU, and outputting a fault classification result.
In a third aspect, the present invention provides a computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements a vehicle shock absorber defect diagnosis method as described in any one of the above when executing the computer program.
In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program is executed by a processor to implement a vehicle shock absorber defect diagnosis method according to any one of the above.
The invention has at least the following beneficial effects:
the method comprises the steps of inputting preprocessed data into a deep learning framework, convolving the data through a convolutional neural network introduced into an SE module to obtain a feature map, and judging which feature maps have the largest contribution to the optimal solution of a classification problem through the SE module; extracting and classifying the features through the pooling layer and the full-link layer, and then performing fault classification diagnosis through the ReLU; the invention can detect the defects of the shock absorber of the chassis of the unmanned vehicle without a plurality of sensors, and can ensure the accuracy of the detection result and improve the safety of the vehicle by using the algorithm of basic data.
Detailed Description
The present invention will be described in detail below with reference to the embodiments with reference to the attached drawings. It should be noted that the embodiments and features of the embodiments may be combined with each other without conflict.
The following detailed description is exemplary in nature and is intended to provide further details of the invention. Unless otherwise defined, all technical terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the invention.
Example 1
A vehicle shock absorber defect diagnostic method comprising:
s1: collecting various data of vehicle operation and preprocessing the data;
various items of data of vehicle operation include yaw rate, lateral longitudinal acceleration, steering angle, and wheel speed.
Data of an anti-lock brake system (ABS) and a vehicle body electronic stability system (ESP) are obtained by accessing bus communication, and a yaw rate and a steering angle measured by a yaw rate sensor are collected.
The damper characteristic is adjusted by the current of the damper valve, and the defect behavior of the damper is simulated by changing the valve current. Different wheel speeds were run on different surfaces. The current supply to individual and all dampers was varied to simulate individual defects and multiple defects.
The Controller Area Network (CAN) signals for all wheel speeds are used, along with the calculated longitudinal and lateral accelerations at the center of gravity of the vehicle, to make a defect diagnosis. All signals are sampled at the same frequency. Due to the particularity of the CAN communication such as bus load fluctuation, signals cannot be supplied in exactly the same time, and therefore, sampling rate conversion (resampling) is performed.
The collected data needs to be pre-processed before being transmitted into the network. In the preprocessing part, collected data are cut and filtered, a measurement file is divided into a plurality of sequences, the length of the sequences is fixed, and the number of data points is the same; and to standardize to prevent certain inputs from being implicitly too heavy or too light. To cope with the problem that time-based data tends to contain a high noise component and has a high dimensionality, a Discrete Fourier Transform (DFT) signal processing method may be used:
wherein X (k) represents data after DFT transform, N represents the number of points of fourier transform, k represents the kth frequency spectrum of fourier transform, X (N) is a sampled analog signal, X (N) in the formula can be a complex signal, and in practice, X (N) is a real signal, that is, the imaginary part is 0, and at this time, the formula can be expanded as follows:
a Fast Fourier Transform (FFT) is implemented with a discrete fourier transform. A discrete fourier transform can transform a signal from the time domain to the frequency domain, and both the time and frequency domains are discrete. The fast fourier transform is a fast algorithm of discrete fourier transform, and can transform a signal into the frequency domain. Some signals are difficult to see what features are in the time domain, but if transformed to the frequency domain, the features are easy to see. This is why many signal analyses employ a fast fourier transform. In addition, the fast fourier transform can extract the spectrum of a signal, which is also often used in the field of spectral analysis. Data samples transformed with the fast fourier transform lose the information of the changes in the time domain, and to overcome this problem a Short Time Fourier Transform (STFT) can be used, in which a time shifted analysis window is used to extract the local frequencies by fourier transform. The glamipan angular field is a two-dimensional representation of the time signal, a time signal is scaled to a fixed interval and then converted to polar coordinates, the values are represented as angle cosines, and time is radius. The recursion graph is an important method for analyzing the periodicity, chaos and non-stationarity of the time sequence, and can reveal the internal structure of the time sequence and provide a priori knowledge about similarity, information quantity and predictability. The recursive graph is particularly suitable for short-time sequence data, and can check the stationarity and the internal similarity of the time sequence. The short-time Fourier transform is adopted for processing and is expressed by the following formula:
R ij =θ(ε-||X i -X j ||)
for time series signal u k (k =1,2,. N), determining the sampling time interval as Δ t, determining a suitable embedding dimension m and a delay time τ through correlation theory calculation, and further reconstructing an event sequence, wherein the reconstructed dynamic system is as follows:
x i =[u i ,u i+τ ,...,u i+(m-1)τ ]
i point x in reconstructed phase space i And j point x j The distance of (c):
S ij =||X i -X j ||
wherein R is ij For recursive values, the delay factor τ, the embedding dimension m, and the threshold ε. A more common embedding dimension selection method is a pseudo-neighborhood method, delay coefficient selection is an average mutual information method, the optimal recursive threshold value is not a good method at present, and 10% of the peak value is generally selected. The results are evaluated using a confusion matrix.
The deep learning architecture is shown in table 1.
TABLE 1 deep learning architecture
S2: extracting the characteristics of fault diagnosis through a deep learning framework, and outputting a fault classification result; the deep learning architecture comprises: convolutional neural networks, pooling layers, full connectivity layers, and relus for the SE module were introduced. Inputting the preprocessed data into a deep learning framework, and performing convolution on the data through a convolution neural network introduced into an SE module to obtain a feature map; and extracting and classifying the features through the pooling layer and the full-connection layer, performing fault classification diagnosis through the ReLU, and outputting a fault classification result.
In order to cover the influence of different filter sizes, processed data are input into a convolutional neural network, firstly, features are extracted by using convolution filters with different sizes and the juxtaposition of deep separable convolutions, and an SE module is introduced into the convolutional neural network; the SE module is used for judging which characteristic graphs contribute most to the optimal solution of the classification problem; convolution filters of different sizes are used in several branches and depth separable convolution is used in one branch as shown in fig. 1. The depth separable convolution first performs channel-by-channel convolution (DC) on channels (depths), respectively, and concatenates outputs, followed by channel convolution (PC) using a unit convolution kernel to obtain a feature map. Inputting the characteristic diagram into the pooling layer to select the characteristics extracted from the convolutional layer, extracting the graphic characteristics, then inputting the characteristics into the full-link layer, converting all characteristic matrixes of the pooling layer into one-dimensional characteristic large vectors by the full-link layer, and generally placing the full-link layer at the end of a convolutional neural network structure for classification; then fault classification diagnosis is carried out through a ReLU (modified Linear Unit), wherein the ReLU is an activation function and is generally used in a classification problem; finally, outputting a fault classification result; the fault classification diagnosis comprises the following steps: whether the four air doors at the front left, the front right, the back left and the back right have defects or not.
Compared with the conventional convolution operation, the number of parameters and the operation cost are lower. One convolution kernel of the channel-by-channel convolution is responsible for one channel, one channel is only convoluted by one convolution kernel, and the number of the characteristic image channels generated by the process is identical to the number of the input channels. And carrying out weighted combination on the feature maps in the previous step in the depth direction by point-by-point convolution operation to generate a new feature map. This results in each branch having a different feature map, which can be generated by a common convolution operation with different sized filter masks on the one hand and by extracting features on the other hand. By extracting features from each data channel separately and then combining them. And their subsequent combinations.
The SE module implements scaling of a single signature graph by a Network-in-Network (NIN) method. Internal multi-layer perceptrons (MLPs) allow the network to autonomously learn which feature maps contribute most to the optimal solution to the classification problem. Therefore, an SE block is additionally introduced into the network. The network with the SE blocks is followed by a max pooling layer. The sub-sampling layer improves performance while reducing the number of parameters. dropout is used to reduce overfitting.
Example 2
A vehicle shock absorber defect diagnostic system comprising:
the data collection and pretreatment module is used for collecting and pretreating various data of vehicle operation;
the characteristic extraction and fault classification output module is used for inputting the preprocessed data into the deep learning framework and convolving the data by introducing a convolutional neural network of the SE module to obtain a characteristic diagram; and extracting and classifying the features through the pooling layer and the full-connection layer, performing fault classification diagnosis through the ReLU, and outputting a fault classification result.
Example 3
The invention provides a computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements a vehicle shock absorber defect diagnosis method as described in embodiment 1 when executing the computer program.
Example 4
The present invention provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements a vehicle shock absorber defect diagnosis method as described in embodiment 1.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting the same, and although the present invention is described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: modifications and equivalents may be made to the embodiments of the invention without departing from the spirit and scope of the invention, which is to be covered by the claims.