Fault detection method of new energy vehicle engine based on wavelet transform and support vector machine

Abstract

In order to solve the problem that existing methods are affected by fundamental frequency and harmonic frequency oscillation of new energy vehicle engine, a new energy vehicle engine fault detection method based on wavelet transform and support vector machine is proposed. Firstly, a detection model of abnormal noise signal of automobile engine fault is established, and the time-frequency parameters of basis function are adjusted adaptively. Then, the mechanical excitation component and the battery excitation component in the engine surface radiation noise are separated, and the new energy vehicle engine fault signal is decomposed by feature decomposition and multi-scale separation. Finally, wavelet transform combined with support vector machine algorithm was used to extract fault features of new energy vehicles, fuzzy clustering was carried out, and time-frequency analysis of fault signals was carried out in fractional Fourier domain to realize fault detection of new energy vehicle engines. The test results show that the method has high fault identification level and good clustering of fault feature points for new energy vehicle engine fault detection, and can effectively improve the ability of engine fault detection with good application effect.

Keywords

Wavelet transform support vector machine new energy vehicle engine failure fault detection

1. Introduction

In combination with the engine stability monitoring method of new energy vehicles, a fault detection and control model of new energy vehicles is established to improve the performance of the drive motor and control system [1]. By optimizing the design of the engine and motor control system of new energy vehicles, combined with the fault detection method, the occurrence rate of faults is reduced. By analyzing the fault stability of new energy vehicles, continuously support the scientific and technological innovation of new energy vehicles, study the fault detection methods of new energy vehicle engines, and achieve fault feature analysis in combination with time-frequency analysis algorithm, blind source separation algorithm and signal estimation algorithm to improve the stability of vehicle engines [2].

The fault detection of new energy vehicle engine is based on the analysis of abnormal noise characteristics of the vehicle engine, combined with the analysis of characteristics of steady and transient operating conditions, and uses the diagnostic process monitoring and analysis methods of abnormal noise to achieve the analysis of vehicle fault characteristics. TRIVEDI M et al. proposed a deep learning and blockchain-based electric vehicle fault detection framework, which identifies tire pressure, temperature and battery failure [3]. At the same time, we use the 5G wireless network with high scalability and reliability and the Interplanetary File System (IPFS) protocol to conduct data transactions. The framework uses convolutional neural network (CNN) and long short memory (LSTM) models to deal with air tire pressure failure, temperature anomaly and battery failure detection, and predict the existence of failure data, so as to ensure a safer journey for users. The framework has the advantages of high safety, economic efficiency and reliability. The deep learning model used in this method requires a lot of computational resources and time to train, which leads to a long response time for its detection and poor detection effect. Abboush Mohammerd et al. used a real-time fault injection framework based on hil to generate fault data without changing the original system model [4]. The model structure is constructed by the combination of convolutional neural network (CNN) and Long and short memory (LSTM), but this method requires more advanced hardware equipment, higher cost, and its detection effect still has a large room for improvement, which leads to the failure of this method to be widely applied. The reliability of the machine is improved and defects in adhesives and sliding attachments are prevented [5]. The experiments are aimed at vibration signals and motor currents. Six fault conditions of linear bearing are established. The analysis of vehicle fault characteristics is realized by means of spectrum analysis, data analysis, peak factor analysis and variance analysis. This method is mainly aimed at the detection and diagnosis of linear bearing faults, which may not be suitable for other types of fault detection and diagnosis, and the application effect is not good.

To solve these problems, this paper proposes a new energy vehicle engine fault detection method based on wavelet transform and support vector machine. Firstly, the detection model of abnormal noise signal of automobile engine fault is established according to various mechanical noises and abnormal noises, and the mechanical excitation component and battery excitation component in the radiated noise of engine surface are separated [6]. Then, the attenuation curve, independent component analysis and empirical mode decomposition methods are used to realize the feature decomposition and multi-scale separation of the fault signal of new energy automobile engine. The wavelet transform is used to extract fault features, and the support vector machine algorithm is used to carry out fuzzy clustering processing on the extracted fault features of new energy. Finally, the simulation test shows the superior performance of the method in improving the engine fault detection capability of new energy vehicles.

2. New energy vehicle engine fault signal processing

2.1 Fault signal extraction

In order to realize the fault detection of new energy vehicle engine based on wavelet transform and support vector machine, the fault information collection model of new energy engine was first constructed [7]. When the automobile engine is in the transient condition, the detection model of the abnormal noise signal of the automobile engine fault is established according to various mechanical noises and abnormal noises. The abnormal noise is sensitive to the parameters of the variable camshaft phase system (VCT), and the abnormal noise components in the cylinder head surface vibration signal are extracted. Assume that the instantaneous speed sample set of crankshaft and camshaft contains $n$ samples, including diagnostic samples of transient mechanical abnormal sound of motor $x_{i}$ , the characteristic vector of $i = 1, 2, \dots, n$ is:

$𝒙_{i} = {(x_{i 1}, x_{i 2}, \dots, x_{i s})}^{T}$ (1)

The sensor array detection method is used to extract multi-dimensional data noise under steady and transient conditions [8]. The Wiener filter based noise component method is used for adaptive optimization, and the data set is divided into $2^{n}$ subset, based on the combustion noise separation method of multiple regression, sets the fault data of automobile engine $X$ is divided into category $c$ , where $1 < c < n$ . Fault dynamic separation is carried out based on Fourier transform (ShortTimeFourier Transform) [9]. According to the time characteristics of non-stationary signals, the fuzzy clustering center matrix of fault detection for automobile engines is obtained as follows:

$𝑽 = {v_{i j} | i = 1, 2, \dots, c, j = 1, 2, \dots, s}$ (2)

Among them, $v_{i j}$ is the cluster center $i$ vectors of No. $j$ concept of window function is introduced to expand the one-dimensional time-domain signal to the two-dimensional time-frequency plane [10]. According to the feature similarity in the clustering process, the spectrum division matrix of the signal cut within each window width is expressed as:

$𝑼 = {μ_{i k} | i = 1, 2, \dots, c, k = 1, 2, \dots, n}$ (3)

The method of changing the truncation position of window function is used to improve the time-frequency resolution of STFT for engine fault, construct a synthetic simulation signal to display the time-frequency characteristics of STFT, truncate the signal non integrally, so that there is a certain proportion of overlap between adjacent window functions, and obtain the clustering objective function matrix of KNN for the engine fault signal of new energy vehicles (define the clustering objective function):

$𝑱_{m} (U, V) = \sum_{k = 1}^{n} \sum_{i = 1}^{c} μ_{i k}^{m} {(d_{i})}^{2}$ (4)

Among them, $d_{i}$ is the variable window width feature vector in frequency domain. The shorter window width is used to analyze the time-frequency energy distribution of the signal $p_{0}$ through adaptive optimization, the time-frequency energy distribution characteristic window of new energy vehicle engine fault is:

$\sum_{n = 1}^{N} p_{K, n} = \frac{p_{0} + ρ - 1}{𝑱_{m} (U, V)}$ (5)

Among them, $ρ$ is the number of optimization times. Average of variable window width in time domain $\bar{K}$ can be expressed as:

$\bar{K} = \sum_{k = 1}^{K} \sum_{n = 1}^{N} k p_{k, n}$ (6)

In the time domain variable window width, calculate the time varying and adaptive optimal window of engine fault, compare the subclasses of the first category in the second layer, if $D_{2} ⩽ D_{0}$ , then the convergence condition is satisfied, and the recall feature quantity of the distributed clustering of engine fault signals in the time domain variable window width is obtained as follows:

$W = \frac{\bar{K}}{γ}$ (7)

In the subclass of class k in class 1, S-transform is used to gradually transform the time-frequency diagram from low frequency to high frequency to obtain the S-transform time-frequency diagram of engine fault signal, as shown in Fig. 1.

Figure 1.

S-transform time-frequency diagram of engine fault signal.

Although the concept of variable window width is introduced into S-transform in Fig. 1, the window width slows down with the increase of frequency. According to the signal acquisition results, the time-frequency characteristics of the engine fault signal are adaptively scheduled. By separating the mechanical excitation component from the battery excitation component in the radiated noise from the engine surface, and then using the attenuation curve, independent component analysis and empirical mode decomposition methods, the feature information in the engine fault signal can be more accurately extracted and transformed into feature vectors that can be used for classification and diagnosis, so as to achieve accurate detection and diagnosis of engine faults [11].

2.2 Decomposition of fault signal

According to the time-frequency energy concentration degree of low frequency and high frequency parts, the KNN clustering model of engine fault signal is established, so that both high frequency and low frequency have better frequency resolution [12]. According to the time-frequency energy distribution and optimization idea, the fault correlation analysis is carried out in the direction of the frequency axis, and the directivity characteristics of engine fault characteristics are obtained as follows:

$Φ_{k} = \frac{𝑊𝐶𝑜𝑣 (𝑿, 𝒀)}{\sqrt{D (𝑿)} \sqrt{D (𝒀)}}$ (8)

Among them, $𝐶𝑜𝑣 (𝑿, 𝒀)$ indicates the bandwidth number of steady state signal and linear frequency modulation signal, $D (𝑿)$ and $D (𝒀)$ respectively represent the average energy. The wavelet scale decomposition method is used to decompose the characteristics of the steady signal in the time-frequency diagram of automobile engine fault [13]. The characteristics of the energy distribution of the steady signal in the steady state interval are:

$s (t) = \underset{u (n)}{\underset{⏟}{\sum_{k = 1}^{N} p_{k} \sin (ω_{k} n + Φ_{k})}} + ζ (n)$ (9)

Among them, $ζ (n)$ is time-frequency distribution and energy amplitude, $ω_{k}$ is the recursive feature on the central frequency line of the signal. The whole signal rotates at the same angle on the time-frequency plane, and the central frequency equalization scheduling model of the three components is expressed as:

$U_{n} = h [s (t) (t_{0} + n Δ t)] + ω_{n}$ (10)

Where, $h (.)$ is the time window function of engine fault sampling for new energy vehicles, $ω_{n}$ is the frequency modulation measurement error with nonlinear change. The sample training set of fundamental frequency multiplied by order is $X = {x_{1}, x_{2}, \dots, x_{n}}$ , $n$ is the number of the slope distribution set $X$ of the center frequency curve of the first component in the fault signal of the automobile engine. According to the rotation degree corresponding to the first component, the feature extraction result of multi parameter fusion is obtained as follows:

$y (t) = \frac{U_{n}}{π} * K \int \frac{x (τ)}{t - τ} 𝑑 τ$ (11)

Among them, $K$ represents the matching coefficient of engine fault characteristics of new energy vehicles, $x (τ)$ represents the discriminant statistics of engine failure prediction for new energy vehicles, $*$ represents convolution. Since the length and position of the window function are fixed, the nonlinear chirplet transform (NLCT) method is used to distribute the fault of the automobile engine with big data $x (n)$ which is decomposed into several IMF components, and any training sample is $𝑿_{k} = [x_{k 1}, x_{k 2}, \dots, x_{k m}, \dots, x_{k M}]$ . When the central frequency line of the signal remains horizontal, the fault trend prediction output is:

$𝒀_{k} = [y_{k 1}, y_{k 2}, \dots, y_{k j}, \dots, y_{k J}] (k = 1, 2, \dots, N)$ (12)

Based on this data, an intelligent expert system is established, and the subsequent STFT transformation is carried out. The order of engine fault allocation is $w (t)$ , the engine failure delay parameter of new energy vehicle is $a_{i} (t)$ , filtered color noise $n (t)$ can be expressed by the following formula.

$n (t) = - \sum_{i = 1}^{p} a_{i} (t) n (t - i) w (t)$ (13)

IMF signals contain signals of different scales. Wavelet scale decomposition is used to obtain the decomposition results of fault signals as follows:

$P (t, f) = n (t) s_{0} (u_{0} + \frac{τ_{0}}{2}) * (u - \frac{τ}{2}) α (τ, v)$ (14)

The bandwidth of the narrowband signal is determined by the penalty factor $α$ . The fault signal of the new energy vehicle engine is divided into some local characteristic statistics to obtain the clustering control parameters according to the fault of the new energy vehicle engine $s_{0}$ , $τ_{0}$ . Utilize $s_{0}$ , $τ_{0}$ mean square error estimate of ${\hat{s}}_{0 M L}$ , ${\hat{τ}}_{0 M L}$ breakdown of engine fault characteristics of new energy vehicles, namely:

$l (r) = \int r (t) \sqrt{{\hat{s}}_{0 M L}} f^{*} [{\hat{s}}_{0 M L} (t - {\hat{τ}}_{0 M L})] 𝑑 t = max_{a, b} | W_{f} r (a, b) | a < λ_{1} < b$ (15)

Where, $a = 1 / s$ , $b = τ$ , $λ_{1}$ is the detection threshold. Multi scale wavelet decomposition is used to decompose a single actual component into several BLIMF. Calculate the fault feature decomposition of the automobile engine in the optimal classification plane, and obtain the fault feature distribution sequence between the data analysis layer and the data processing layer $x_{1}, x_{2}, \dots, x_{n}, \dots$ , set the total points as $N$ the results of feature decomposition and multi-scale separation of engine fault signals of new energy vehicles are obtained. It can better extract the fault features in the signal and reduce the noise interference in the signal, so as to achieve effective detection and diagnosis of engine faults.

3. Realization of automobile engine fault detection

3.1 Fault signal feature extraction

The wavelet transform is used to extract the fault features separated from the fault signal, and then the support vector machine algorithm is used to carry out fuzzy clustering processing on the extracted fault features of new energy vehicles, so as to realize the fault detection of new energy vehicle engines [14].

The vibration data sequence of new energy vehicle engine fault will be obtained ${x_{i}}$ at sampling interval $j τ$ the output autocorrelation function of new energy vehicle engine fault detection is:

$R (j τ) = \frac{l (r)}{N} \sum_{i = 0}^{N - 1} x_{i} x_{i + j τ}$ (16)

Calculation of prediction characteristic function of new energy vehicle engine under certain fault conditions $X (l, n_{i})$ , calculate the within class dispersion matrix of the fault category distribution of the new energy vehicle engine, and obtain the optimal solution of the fault characteristic parameters in the non divisible frequency band set within four consecutive different bandwidths. After using the signal components obtained by the VMD decomposition strategy from coarse to fine, the fault detection signal is completely equal to the input signal by linear superposition. Idle speed is the most common steady-state operating condition of the vehicle. When the vehicle is in idle speed, the maximum gradient difference of fault trend prediction is obtained:

${𝐴𝑉𝐺}_{X} = \frac{R (j τ)}{m \times n} \sum_{x = 1}^{n} \sum_{y = 1}^{m} | G_{X} (x, y) |$ (17)

Where, $m$ and $n$ are the autocorrelation coefficients of the direct noise from the engine or the body structure change caused by the engine vibration respectively [15]. Perform time-frequency analysis and sound source location for abnormal knocking noise under idle speed, infer the sound generation mechanism of abnormal knocking noise at idle speed $Δ t$ as the scale $Δ u$ is the time interval, and the scale convolution function of the fault signal obtained is:

$Δ u | Δ t = 2 π | S | \frac{\sin α}{M} {𝐴𝑉𝐺}_{X}$ (18)

Analyze the information of rotating speed, voltage and other signals to get $| S |$ and $M$ with homomorphic attenuation, the value is generally 1. The expression of time-frequency characteristics of passenger cars under idling conditions is:

(1) When $α = 2 D . π$ Hour

$X_{α} (m) = x (m)$ (19)

(2) When $α = (2 D + 1) . π$ Hour

$X_{α} (m) = x (- m)$ (20)

Among the above, $D$ can be found that when the engine of one sample vehicle with obvious abnormal noise is replaced with the engine of another sample vehicle with unknown abnormal noise in the ABA replacement test (that is, the engine is exchanged first and then restored), the abnormal noise changes with the movement of the engine. When $M = N$ , when meeting the condition of $α \neq D . π$ , the scale equalization matrix expression of the signal is obtained:

$X = F_{α} \cdot x$ (21)

Where:

$X = {[X_{α} (0), X_{α} (1), \dots, X_{α} (N - 1)]}^{T}$ (22) $x = {[x_{α} (0), x_{α} (1), \dots, x_{α} (N - 1)]}^{T}$ (23)

Among them, $F_{α}$ is $N \times N$ dimension matrix, support vector machine learning method is adopted to further check the source of abnormal noise, and the time-frequency characteristics of fault signal can be expressed as:

$x = F_{- α} . X$ (24)

Infer and verify abnormal sound according to various digital signals collected in the experiment $d$ the data receiving model of fault characteristic quantity is:

$x_{m} (t) = \sum_{i = 1}^{I} \frac{s_{i} (t) e}{d}$ (25)

Among them, $s_{i} (t)$ is vibration data sensed by the engine’s vibration sensor, $x_{m} (t)$ is an array element $m$ the received thermal sensing data sequence. Build the impulse model for fault detection of mechanical structure system of cylinder head assembly:

$h (t) = x_{m} (t) e θ_{i} (t)$ (26)

In the above formula, $θ_{i} (t)$ indicates that the engine is running at idle speed. In the cylinder head assembly of the engine, the valve mechanism is mainly responsible for controlling the opening and closing of the engine inlet and exhaust valves. The distribution value of the support vector machine is:

$c (τ, t) = \frac{a_{n} (t) τ_{n} (t)}{h (t)}$ (27)

Among them, $a_{n} (t)$ is the sampling frequency of the n-th channel signal acquisition system, $τ_{n} (t)$ is the time delay of the nth data channel. Four substructure replacement schemes of cylinder head assembly are adopted for experimental research, and the structural model of engine fault wavelet transform feature extraction is obtained as shown in Fig. 2.

Figure 2.

Structural model of engine fault feature extraction.

3.2 Fault detection

According to the above feature extraction results, time-frequency analysis of the fault signal is carried out in the fractional Fourier domain to extract time-frequency features, which can more comprehensively describe the time-domain and frequency-domain features of the signal, so as to effectively extract and diagnose the fault features in the signal [16].

Assumed input signal $x (t)$ , whose fractional Fourier transform is defined as:

$X_{p} (u) = c (τ, t) \int_{- \infty}^{\infty} K_{p} (t, u) x (t) 𝑑 t$ (28)

Where, $p$ is the characteristic component of camshaft angle time correlation, and the phase offset of the signal is $α = p π / 2$ . $F^{α} [\cdot]$ represented as transformation operator formal mark, $K_{p} (t, u)$ is the accounting sub of abnormal sound vibration signal in angle domain, and it is the fractional Fourier transform kernel for time-frequency conversion of fault signal $K_{p} (t, u)$ , can be described as:

$K_{p} (t, u) = {\begin{matrix} \sqrt{\frac{1 - j \cot α}{2 π}} \exp (j \frac{t^{2} + u^{2}}{2} \cot α - t u \csc α) & α \neq n π \\ \exp (t - u), & α = 2 n π \\ \exp (t + u), & α = (2 n \pm 1) π \end{matrix}$ (29)

Where, $n$ is an integer, $n \in Z$ . The time series is introduced to resample the original vibration signal, and the kernel function of the intake valve lift phase is obtained $K_{p} (t, u)$ meet the following properties:

$K_{- α} (t, u) = K_{α}^{*} (t, u)$ (30) $\int_{- \infty}^{\infty} K_{α} (t, u) K_{α}^{*} (t, u^{'}) 𝑑 t = \exp (u - u^{'})$ (31)

Figure 3.

Implementation flow of algorithm.

The fault signal is processed by fractional Fourier transform $x (t)$ normal orthogonal decomposition is carried out in the fractional Fourier domain, and then support vector machine algorithm is used to realize the fault feature clustering of new energy vehicle engines [17]. The implementation flow of the improved algorithm is shown in Fig. 3.

4. Experimental test

In the experiment, the characteristic components of idle abnormal noise were extracted on the engine test bench in a semi anechoic chamber, and the engine was running at 700 rpm. Cut a window on the engine timing cover, and install a BAUMER speed encoder with a resolution of 720 PPR (pulses per revolution) on the inlet camshaft (reference shaft) through the window. Place three Kulite pressure sensors (XCEL-100 type) on the wall of the OCV valve, which are respectively connected to cavities A and B of the camshaft adjuster and their oil supply pipe P. Install two PCB vibration acceleration sensors on the intake side of the cylinder head: one is between cylinders 1 and 2 (number cy12), and the other is between cylinders 3 and 4 (number cy34). A BSWA microphone is placed on the exhaust side of the engine.

In the experiment, all physical signals are simultaneously collected by the Hyde acoustic data acquisition system at a sampling frequency of 48 kHz, and the steps are as follows:

Set data acquisition parameters: By configuring the Hyde Sound data acquisition system, set the sampling frequency to 48 kHz, and determine other parameters that need to be set, such as the number of channels, gain, etc.

Connect the sensor/signal source: Connect the sensor or signal source that needs to be collected with the Hyde Sound data acquisition system to ensure the correct access of the signal source.

Start data acquisition: Start the data acquisition process by operating the control interface or command of the Hyde Sound data acquisition system.

Signal acquisition: The data acquisition system begins to collect physical signals, and continuously obtains sampling values from the sensor or signal source according to the preset sampling frequency. These sampled values are usually expressed in numerical form and stored in memory or on disk.

Data processing: The collected physical signal can be further processed, such as filtering, amplification, denoising and other operations, in order to improve the signal quality and adapt to the needs of subsequent analysis.

Data preservation: Store the processed data in a suitable format, such as text files, databases, etc., to facilitate subsequent data analysis and use. The specific channel number, measured signal and corresponding sensor position of the data acquisition front-end are shown in Fig. 4.

Figure 4.

Layout of automobile engine fault information collection test bench and sensors.

According to the above simulation test environment parameter settings, the vehicle engine fault signal is collected, as shown in Fig. 5.

Figure 5.

Vibration and noise results.

According to the analysis of Fig. 5, the collected vibration and noise signals of the automobile engine are analyzed for fault characteristics, and there is a peak value in the A-weighted sound pressure level spectrum of the exhaust side noise within 400 $\sim$ 600 Hz. Accordingly, the vibration signals of cy12 and cy34 also have peaks in this frequency band, and the vibration amplitude of cy12 is large, so the vibration excitation source corresponding to the abnormal sound should theoretically be closer to the No. 1 cylinder side of the cylinder head. However, it is still impossible to determine the source of the abnormal sound from the time-domain waveform and Fourier spectrum. It is necessary to further extract the vibro acoustic component containing the abnormal sound. Based on the above analysis results, the design method is adopted to realize the fault detection. The NLCT time-frequency diagram detection results are shown in Fig. 4.

Figure 6.

NLCT time-frequency diagram of three components of fault detection.

Since the frequency band of idle abnormal noise is 400 $\sim$ 600 Hz, in order to shorten the calculation time of the algorithm, the sound signal and the vibration signals of cy12 and cy34 are resampled to 8192 Hz. Then, the DVMD based algorithm is used to decompose the vibration and noise signals to extract the narrow-band components containing the characteristics of abnormal sound. Finally, after the DVMD algorithm decomposition, the noise signal is decomposed into seven BLIMF components, and the vibration signal of cy12 is decomposed into six BLIMF components. The accuracy and time cost of fault detection by different methods are tested, and the comparison results are shown in Fig. 7 and Table 1.

Table 1

Comparison of fault detection performance

Method	Time cost/s	Output signal to noise ratio/dB
Methods in this paper	1.186	36.169
Methods of literature [3]	4.009	12.056
Methods of literature [4]	8.165	19.996

Figure 7.

Comparison of fault signal detection performance.

According to the analysis in Fig. 7, the detection accuracy of the proposed method is between 90% and 98%; the [3] method is between 75% and 90%; and the [4] method is between 60% and 90%. Since this method is combined with the SVM algorithm for fuzzy clustering of the extracted new energy failure features, specifically, SVM can classify and diagnose using the extracted feature vectors. Fuzzy clustering can divide data points into multiple fuzzy categories, which can better handle the uncertainty and ambiguity between data and thus more accurately describe the characteristics of the data. By combining SVM and fuzzy clustering, the fault signal of the new energy vehicle engine can be effectively processed and accurately classified, so as to improve the detection performance of the fault signal.

It can be seen from the analysis in Table 1 that the fault detection time of the design method is less, and the output signal has a higher signal-to-noise ratio, which can more accurately detect the existence of the fault signal, thus improving the performance of fault detection. This method can better capture useful information in the engine fault signal by separating the mechanical excitation component from the battery excitation component. Then, attenuation curve, independent component analysis and empirical mode decomposition are used to extract the characteristic information of the signal to realize the rapid detection and diagnosis of engine faults.

The proposed method is used to extract fault features from fault signals, and the extraction results are shown in Fig. 8.

Figure 8.

Comparison of fault feature extraction results.

It can be seen from Fig. 8 that due to the large amount of noise in the diagnosis process of the engine fault of new energy vehicles, the fault signal is unstable, and there are a large number of false peaks, which lead to the difficulty of fault diagnosis and easy error; The denoising effect of this method is obvious, the ridges in the figure are clearly visible, and the fault features have obvious aggregation effect, which shows that the proposed method has a high fault signal extraction ability and fault recognition rate.

5. Conclusion

In order to help the further development of new energy vehicles and improve the engine fault detection effect of new energy vehicles, this paper proposes a new energy vehicle engine fault detection method based on wavelet transform and support vector machine. According to the result of signal acquisition, the time-frequency characteristic of engine fault signal is adaptive scheduling. Then, according to the signal acquisition results, the time-frequency characteristics of the engine fault signal are adaptive scheduling. Finally, the mechanical excitation component and the battery excitation component in the engine surface radiation noise are separated, and the fault features are extracted by wavelet transform, combined with the support vector machine algorithm, and the extracted fault features of new energy are fuzzy clustering. Four kinds of cylinder head substructure replacement schemes are used to carry out experimental research, and the fault structure model of the engine is extracted by wavelet transform feature to realize fault detection. According to the test results, the application of the design method can improve the fault feature clustering effect, the fault detection accuracy is higher than 90%, and the fault detection time cost is smaller, better than the comparison method, and the application effect is good.

References

Deng

Xie

Tao

Wen

. Identification method of wheel out-of-roundness state of high-speed train based on axle box vibration and dynamic model. Journal of Mechanical Engineering.2023; 59(3): 110-121.

Song

Liang

Wang

, et al. Wheel polygon state recognition method based on improved EEMD-WVD joint time-frequency analysis. Journal of Traffic and Transportation Engineering.2021; 21(6): 259-268.

Trivedi

Kakkar

Gupta

, et al. Blockchain and deep learning-based fault detection framework for electric vehicles. Mathematics.2022; 10(19): 3626.

Abboush

Bamal

Knieke

, et al. Intelligent fault detection and classification based on hybrid deep learning methods for hardware-in-the-loop test of automotive software systems. Sensors.2022; 22(11).

Chommuangpuck

Wanglomklang

Srisertpol

. Fault detection and diagnosis of linear bearing in auto core adhesion mounting machines based on condition monitoring. Systems Science & Control Engineering.2021; 9(1): 290-303.

Mühlroth

Kölbl

Grottke

. Innovation signals: Leveraging machine learning to separate noise from news. Scientometrics.2023; 128(5): 2649-2676.

Pak

Han

. Prediction of characteristic blast-induced vibration frequency during underground excavation by using wavelet transform. Frontiers of Structural and Civil Engineering.2022; 16(8): 1029-1039.

Gauthier

Pruvost

Gamache

, et al. A new pressure sensor array for normal stress measurement in complex fluids. Journal of Rheology.2021; 65(4): 583-594.

Tang

Liu

. Simulation of molecular property prediction method based on improved fourier transform. Computer Simulation.2023; 40(01): 505-509.

10.

Giri

Majumder

. Eigenvalue-based cooperative spectrum sensing using kernel fuzzy c-means clustering. Digital Signal Processing.2021; 111(2): 102996.

11.

Dong

Xun

. Fault diagnosis of aeroengine fan based on generative adversarial network and acoustic features. Aerospace Systems.2022; 5(4): 567-575.

12.

Radović

Ivanović Veselin

Djurović

, et al. A step toward real-time time-frequency analyses with varying time-frequency resolutions: Hardware implementation of an adaptive s-transform. Circuits, Systems, and Signal Processing.2022; 42(2): 853-874.

13.

Gültekin

Çinar

, et al. A novel deep learning approach for intelligent fault diagnosis applications based on time-frequency images. Neural Computing and Applications.2022; 34(6): 4803-4812.

14.

Sushir Rupesh

Wakde Dinkar

. An improved detection of blind image forgery using hybrid deep belief network and adaptive fuzzy clustering. Multimedia Tools and Applications.2022; 81(20): 29177-29205.

15.

Boztas

. Sound source localization for auditory perception of a humanoid robot using deep neural networks. Neural Computing and Applications.2022; 35(9): 6801-6811.

16.

Oyarzun

Aizpuru

BaraiaEtxaburu

. Time-Frequency Analysis of Experimental Measurements for the Determination of EMI Noise Generators in Power Converters. Electronics.2022; 11(23): 3898.

17.

Alimisis

Gennis

Gourdouparis

, et al. A low-power analog integrated implementation of the support vector machine algorithm with on-chip learning tested on a bearing fault application. Sensors (Basel, Switzerland).2023; 23(8): 3978.