A Fault Signal Preprocessing Method for Common Rail Injectors Based on IWOA-SVR-EEMD

Abstract

The purpose of the present invention is to provide a common rail injector fault signal preprocessing method based on IWOA-SVR-EEMD. After collecting the high-pressure oil pipe pressure signal, the signal is first extended using support vector regression optimized by the improved whale optimization algorithm; then, the extended signal is subjected to collective empirical mode decomposition to obtain several intrinsic mode components; finally, the Pearson product-moment correlation coefficient of each intrinsic mode component is calculated, and the components with high correlation with the original signal are retained for component synthesis; after the preprocessing is completed, a denoised high-pressure oil pipe pressure signal is obtained. Compared with the original collective empirical mode decomposition method, the present invention has a better effect on suppressing the endpoint effect problem; compared with traditional signal denoising methods, the signal-to-noise ratio of the denoised signal is higher.

Description

Technical Field

[0001] The present invention relates to a method for processing a diesel engine fault, in particular to a method for processing a fuel injector fault. Background Art

[0002] Common rail injectors are key components of diesel engine fuel systems, and their health status affects the overall performance of diesel engines. Therefore, research on fault diagnosis methods for common rail injectors is a current research hotspot. Affected by the working environment of the injector, the pressure signal collected from the high-pressure oil pipe inevitably contains a lot of noise, which may even mask the fault information and cause misdiagnosis. Therefore, preprocessing the fault signal of the common rail injector is of great significance for subsequent fault diagnosis work.

[0003] Traditional preprocessing methods for non-stationary signals often employ wavelet threshold denoising or empirical mode decomposition (EMD). However, the wavelet basis and threshold must be manually determined, significantly impacting the denoising effect. EMD is an adaptive time-frequency analysis method that decomposes complex signals into a finite sum of intrinsic mode components. However, EMD itself has several drawbacks, such as mode mixing, endpoint effects, and stopping conditions. Ensemble Empirical Mode Decomposition (EEMD) is an improvement on EMD and can suppress mode mixing. However, endpoint effects still exist. Endpoint effects refer to the phenomenon in which the envelope of a spline function diverges at its endpoints due to the loss of the constraint of the extreme points. Here, a signal continuation method based on support vector regression (SVR) is introduced to extend the signal sequence and suppress endpoint effects. Since the performance of the support vector regression (SVR) prediction model is largely influenced by the penalty factor c and the kernel function parameter g, optimally selecting c and g is crucial during model development. Swarm intelligence algorithms are widely used in parameter optimization. The Whale Optimization Algorithm (WOA) has the advantages of being simple and easy to understand and requiring fewer parameters to be manually adjusted. However, it has certain shortcomings in terms of convergence accuracy and escaping from local optimality. Summary of the Invention

[0004] The object of the present invention is to provide a common rail injector fault signal preprocessing method based on IWOA-SVR-EEMD that can achieve reasonable denoising of the common rail injector fault signal.

[0005] The object of the present invention is achieved like this:

[0006] The present invention provides a common rail injector fault signal preprocessing method based on IWOA-SVR-EEMD, which is characterized by:

[0007] (1) The pressure fluctuation signal of the high-pressure oil pipeline is collected and the signal sequence is used as the training sample of support vector regression. The penalty factor c and kernel parameter g of the support vector machine are optimized by using the training sample and the improved whale optimization algorithm to construct the support vector machine model;

[0008] (2) The constructed support vector regression model is used to perform endpoint extension on the training sample signal: For a given pressure signal sequence s(1), s(2)...s(N), where N is the number of sampling points of the original pressure signal sequence, the number of sample training l is first determined to generate a training set L = {(x1, y1), (x2, y2), ..., (x l ,y l )},in:

[0009] x i =[s(i) s(i+1) ... s(N-l+i-1)] T

[0010] y i =s(N-l+i),1≤i≤l

[0011] Predict the first endpoint sequence value, and use the support vector regression model to obtain the first predicted value outside the boundary s(N+1), that is,

[0012]

[0013] Where: x l+1 =[s(l+1) s(l+2) ... s(N)] T Then iterate step by step to obtain the predicted signal sequence value;

[0014] (3) Use the method of ensemble empirical mode decomposition to decompose the extended signal sequence and obtain the intrinsic mode component imf;

[0015] (4) Calculate the Pearson product-moment correlation coefficient of each eigenmode component using the following formula:

[0016]

[0017] Where: i On behalf of the IMF i The Pearson product-moment correlation coefficient with the original signal x(t), σ(x) represents the variance of the original signal sequence, and cov(·) represents the covariance of the original signal sequence and each component;

[0018] (5) A threshold is selected based on the correlation between each component and the original signal, and the intrinsic mode components above the threshold are reconstructed. The denoised high-pressure oil pipe pressure signal is output, and the preprocessing is completed.

[0019] The present invention may also include:

[0020] 1. Improve the whale optimization algorithm in step (1). The specific steps are as follows:

[0021] (a) The vector consisting of the support vector regression penalty factor c and the kernel parameter g is used as an individual whale in the whale population. The initial scale of the whale population is set, the position of the whale population is initialized, and the fitness value of the whale population is calculated. The individual with the smallest fitness value is used as the global optimal position vector of the whale population.

[0022] (b) Whale group iterative optimization: Calculate the entropy of the whale group and calculate the entropy difference. The formula for calculating the entropy of the whale group is as follows:

[0023]

[0024] Where: S(t) is the entropy value of the t-th iteration; N is the total number of whale populations; p(x ti ) is the weight of the i-th whale in the t-th iteration of the whale group, and

[0025]

[0026] Where: g ti is the fitness value of the i-th whale in the t-th iteration;

[0027] The entropy difference calculation formula is:

[0028] d(t)=S(t)-S(t-1)

[0029] (c) Calculate the inertia weight and update the whale position. The formulas for calculating the inertia weight and updating the whale position are as follows:

[0030] The inertia weight calculation formula is:

[0031]

[0032] Where: ω(t) is the inertia weight value of the whale group at the tth iteration; ω min 、ω max are the minimum and maximum values of the inertia weight, respectively, and T is the maximum number of iterations;

[0033] Update the individual whale positions as follows:

[0034]

[0035] Where: t is the current number of iterations; X t is the current position vector; is the current global optimal position vector; D is the distance between the optimal individual position and the current individual position, which is defined as follows:

[0036]

[0037] The coefficient vectors A and C are defined as follows:

[0038] A=2a·rand1-a

[0039] C=2·rand2

[0040] Among them: rand1 and rand2 are random numbers uniformly distributed in the range [0, 1]; a is the convergence factor, which decreases linearly from 2 to 0 with the number of iterations t, that is,

[0041] a=2-2t / t max

[0042] where t max is the maximum number of iterations;

[0043] (d) If the entropy difference d(t) is less than the threshold 1×10 -10 Or when the maximum number of iterations is reached, the iteration ends and the optimal position vector of the whale group is output.

[0044] The advantages of the present invention are: it combines the advantages of the IWOA, SVR, and EEMD algorithms to propose a signal denoising method based on IWOA-SVR-EEMD. The SVR signal continuation method is used to suppress the endpoint effect of the EEMD algorithm. The IWOA-SVR parameter optimization process can avoid the traditional manual trial and error process. The IWOA algorithm not only has the advantages of the WOA algorithm's simple and easy-to-understand principle and the small number of parameters required for manual adjustment and setting, but also has a faster convergence speed and higher convergence accuracy than the WOA algorithm, and is more likely to escape the local optimum. Compared with traditional signal denoising methods, the signal obtained by the present invention has a higher signal-to-noise ratio. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a flow chart of the present invention;

[0046] Figure 2 is the original signal waveform;

[0047] Figure 3 This is the principle diagram of the IWOA-SVR algorithm;

[0048] Figure 4 is the IMF signal diagram after ensemble empirical mode decomposition;

[0049] Figure 5 is the Pearson product-moment correlation coefficient of each IMF;

[0050] Figure 6 Comparison chart before and after signal denoising. DETAILED DESCRIPTION

[0051] The present invention will be described in more detail below with reference to the accompanying drawings:

[0052] Combine Figure 1-6 The present invention provides a common rail injector fault signal preprocessing method based on IWOA-SVR-EEMD, comprising the following steps:

[0053] Step 1: Collect the high-pressure oil pipeline pressure fluctuation signal and use the signal sequence as a training sample for support vector regression. Using the training sample and the improved whale optimization algorithm, the penalty factor c and kernel parameter g of the support vector machine are optimized to construct the support vector machine model.

[0054] Step 2: Use the constructed support vector regression model to perform endpoint extension on the training sample signal. For a given pressure signal sequence s(1), s(2)...s(N) (N is the number of sampling points of the original pressure signal sequence), first determine the number of sample training l, and generate a training set L = {(x1, y1), (x2, y2), ..., (x l ,y l )},in:

[0055] x i =[s(i) s(i+1) ... s(N-l+i-1)] T

[0056] y i =s(N-l+i),1≤i≤l

[0057] Predict the first endpoint sequence value, and use the support vector regression model to obtain the first predicted value outside the boundary s(N+1), that is,

[0058]

[0059] Where: x l+1 =[s(l+1) s(l+2) ... s(N)] T Then iterate step by step to obtain the predicted signal sequence value;

[0060] Step 3: Use the method of ensemble empirical mode decomposition to decompose the extended signal sequence and obtain the intrinsic mode component imf;

[0061] Step 4: Calculate the Pearson product-moment correlation coefficient of each eigenmode component using the following formula:

[0062]

[0063] Where ρ i On behalf of the IMF i The Pearson product-moment correlation coefficient with the original signal x(t), σ(x) represents the variance of the original signal sequence, and cov(·) represents the covariance of the original signal sequence and each component;

[0064] Step 5: Select a threshold based on the correlation between each component and the original signal, reconstruct the intrinsic mode components above the threshold, and output the denoised high-pressure oil pipe pressure signal. The preprocessing is completed.

[0065] The improved whale optimization algorithm described in step 1 is as follows:

[0066] The vector consisting of the penalty factor c of support vector regression and the kernel parameter g is used as a whale individual in the whale population. The initial scale of the whale group is set, the position of the whale group is initialized, and the fitness value of the whale group is calculated. The individual with the smallest fitness value is used as the global optimal position vector of the whale group.

[0067] Whale group iterative optimization. Calculate the entropy value of the whale group and calculate the entropy difference. The formula for calculating the entropy value of the whale group is as follows:

[0068]

[0069] Where: S(t) is the entropy value of the t-th iteration; N is the total number of whale populations; p(x ti ) is the weight of the i-th whale in the t-th iteration of the whale group, and

[0070]

[0071] Where: g t i is the fitness value of the i-th whale in the t-th iteration.

[0072] The entropy difference calculation formula is:

[0073] d(t)=S(t)-S(t-1)

[0074] Calculate the inertia weight and update the whale position. The inertia weight calculation formula and the whale position update formula are as follows.

[0075] The inertia weight calculation formula is:

[0076]

[0077] Where: ω(t) is the inertia weight value of the whale group at the tth iteration; ω min 、ω max are the minimum and maximum values of the inertia weight, respectively. Here we take ω min =0.3,ωmax =0.9; T is the maximum number of iterations.

[0078] Update the individual whale positions as follows:

[0079]

[0080] Where: t is the current iteration number; X t is the current position vector; is the current global optimal position vector; D is the distance between the optimal individual position and the current individual position, which is defined as follows:

[0081]

[0082] The coefficient vectors A and C are defined as follows:

[0083] A=2a·rand1-a

[0084] C=2·rand2

[0085] Among them: rand1 and rand2 are random numbers uniformly distributed in the range [0, 1]; a is the convergence factor, which decreases linearly from 2 to 0 with the number of iterations t, that is,

[0086] a=2-2t / t max

[0087] where t max is the maximum number of iterations.

[0088] If the entropy difference d(t) is less than the threshold 1×10 -10 Or when the maximum number of iterations is reached, the iteration ends and the optimal position vector of the whale group is output.

[0089] Example:

[0090] Step 1: The high-pressure oil pipe pressure signal is collected through the pressure sensor. The injector fault type is needle valve wear fault, the rail pressure is set to 70MPa, the injector injection frequency is 2Hz, and the sampling frequency is 25kHz. The sampling point is N=4096, and the original signal waveform is as follows Figure 2 As shown;

[0091] Step 2: Use the collected oil pressure signal as a training sample and the improved whale optimization algorithm to optimize the penalty factor c and kernel parameter g of the support vector machine to build a support vector machine model.

[0092] Reference Figure 3 , the IWOA-SVR method, the specific steps are as follows:

[0093] The vector consisting of the penalty factor c of support vector regression and the kernel parameter g is used as a whale individual in the whale population. The initial scale of the whale group is set, the position of the whale group is initialized, and the fitness value of the whale group is calculated. The individual with the smallest fitness value is used as the global optimal position vector of the whale group.

[0094] Whale group iterative optimization. Calculate the entropy value of the whale group and calculate the entropy difference. The formula for calculating the entropy value of the whale group is as follows:

[0095]

[0096] Where: S(t) is the entropy value of the t-th iteration; N is the total number of whale populations; p(x ti ) is the weight of the i-th whale in the t-th iteration of the whale group, and

[0097]

[0098] Where: g ti is the fitness value of the i-th whale in the t-th iteration.

[0099] The entropy difference calculation formula is:

[0100] d(t)=S(t)-S(t-1)

[0101] Calculate the inertia weight and update the whale position. The inertia weight calculation formula and the whale position update formula are as follows.

[0102] The inertia weight calculation formula is:

[0103]

[0104] Where: ω(t) is the inertia weight value of the whale group at the tth iteration; ω min 、ω max are the minimum and maximum values of the inertia weight, respectively. Here we take ω min =0.3,ω max =0.9; T is the maximum number of iterations.

[0105] Update the individual whale positions as follows:

[0106]

[0107] Where: t is the current iteration number; X t is the current position vector; is the current global optimal position vector; D is the distance between the optimal individual position and the current individual position, which is defined as follows:

[0108]

[0109] The coefficient vectors A and C are defined as follows:

[0110] A=2a·rand1-a

[0111] C=2·rand2

[0112] Among them: rand1 and rand2 are random numbers uniformly distributed in the range [0, 1]; a is the convergence factor, which decreases linearly from 2 to 0 with the number of iterations t, that is,

[0113] a=2-2t / t max

[0114] where t max is the maximum number of iterations.

[0115] If the entropy difference d(t) is less than the threshold 1×10 -10 Or when the maximum number of iterations is reached, the iteration ends and the optimal position vector of the whale group is output.

[0116] Step 3: Use the constructed support vector regression model to perform endpoint extension on the pressure signal sequence. Take the extension of the right endpoint as an example. Construct a training set based on the original signal sequence. For a given pressure signal sequence s(1), s(2)...s(4096), first determine the number of sample training l=4096 to generate a training set; predict the first endpoint sequence value, and use the support vector regression model to obtain the first predicted value outside the boundary s(4097), that is,

[0117]

[0118] Then, we iterate step by step and predict 500 data points. Similarly, we extend the data to the left and obtain a signal sequence with 5096 data points.

[0119] Step 4: Use the method of ensemble empirical mode decomposition to decompose the extended signal sequence and obtain the intrinsic mode component imf;

[0120] Step 5: Calculate the Pearson product-moment correlation coefficient of each eigenmode component using the following formula:

[0121]

[0122] Where ω i On behalf of the IMF i The Pearson product-moment correlation coefficient with the original signal x(t), σ(x) represents the variance of the original signal sequence, and cov(·) represents the covariance of the original signal sequence and each component;

[0123] Step 4: Use the endpoint effect evaluation index and orthogonality index to prove that the support vector regression extension can effectively suppress the endpoint effect. Use the optimization time to prove that the IWOA algorithm converges faster. The endpoint effect evaluation index θ is defined as:

[0124]

[0125] Where RMS original is the root mean square value of the original signal, RMS i is the RMS value of the ith IMF. If the endpoint effect has a more serious impact on the internal contamination of the signal, the larger the θ value will be, so the smaller the θ value, the better. The orthogonality index OI is defined as:

[0126]

[0127] Where n imf Indicates the number of IMF components generated by decomposition, N indicates the length of the IMF component; x k and r k Represent the kth value of the original signal and the residual component respectively; the smaller the OI value, the better the algorithm decomposition performance. imf =12, N=4096;

[0128] The three indicators compare the endpoint effect of the original signal and the signal after extension by three different methods, namely WOA-SVR, mirror extension and extreme value extension, as shown in Table 1. It can be seen from Table 1 that IWOA-SVR extension can effectively suppress the endpoint effect;

[0129] Endpoint effect evaluation index Orthogonality Index time IWOA-SVR-EEMD <![CDATA[8.1520×10 -5 ]]> <![CDATA[5.88×10 -4 ]]> 162.489 WOA-SVR-EEMD <![CDATA[2.7×10 -3 ]]> <![CDATA[6.67×10 -4 ]]> 1352.53 Mirror extension <![CDATA[2.9×10 -3 ]]> <![CDATA[8.05×10 -4 ]]> \ No extension <![CDATA[3.5×10 -3 ]]> <![CDATA[8.42×10 -4 ]]> \

[0130] Table 1

[0131] Step 5: Use the method of ensemble empirical mode decomposition to decompose the extended signal sequence and obtain the intrinsic mode component imf. The decomposition result is as follows: Figure 4 As shown;

[0132] Step 6: Calculate the Pearson product-moment correlation coefficient of each eigenmode component using the following formula:

[0133]

[0134] Where ρ i On behalf of the IMF i The Pearson product-moment correlation coefficient with the original signal x(t), σ(x) represents the variance of the original signal sequence, and cov(·) represents the covariance of the original signal sequence and each component;

[0135] The calculated Pearson product-moment correlation coefficients of each component and the original signal are as follows: Figure 5 As shown,

[0136] Step 7: Select the threshold a according to the correlation between each component and the original signal. In this example, a is 0.1. The intrinsic mode components above the threshold are reconstructed and the denoised high-pressure oil pipe pressure signal is output. The preprocessing is completed. Figure 6 shown.

[0137] This example collects a high-pressure oil pipe pressure signal at a sampling frequency of 25kHz. A simulated injector needle valve wear fault is applied to the fault signal. The fault signal undergoes IWOA-SVR signal extension, ensemble empirical mode decomposition, component screening, and component synthesis to obtain a denoised pressure signal. The signal-to-noise ratio (SNR), root mean square difference (RMS), and RMS difference percentage (RMS percentage) of the proposed signal denoising method and the original EEMD decomposition denoised signal are calculated. Table 2 shows that the proposed method outperforms traditional denoising methods in terms of SNR, RMS difference, and RMS difference percentage.

Claims

1. A common rail injector fault signal preprocessing method based on the whale optimization algorithm, support vector regression algorithm and ensemble empirical mode decomposition algorithm, characterized by: (1) The pressure fluctuation signal of the high-pressure oil pipeline is collected and the signal sequence is used as the training sample of support vector regression. The penalty factor c and kernel parameter g of support vector regression are optimized by using the training sample and the improved whale optimization algorithm to construct a support vector regression model; (a) The vector consisting of the support vector regression penalty factor c and the kernel parameter g is used as an individual whale in the whale population. The initial scale of the whale population is set, the position of the whale population is initialized, and the fitness value of the whale population is calculated. The individual with the smallest fitness value is used as the global optimal position vector of the whale population. (b) Whale group iterative optimization: Calculate the entropy of the whale group and calculate the entropy difference. The formula for calculating the entropy of the whale group is as follows: Where S(t) is the entropy value of the t-th iteration; N is the total number of whale populations; p(x ti ) is the weight of the i-th whale in the t-th iteration of the whale group, g ti is the fitness value of the i-th whale in the t-th iteration; The entropy difference calculation formula is: d(t)=S(t)-S(t-1) (c) Calculate the inertia weight and update the whale position. The formulas for calculating the inertia weight and updating the whale position are as follows: The inertia weight calculation formula is: Where ω(t) is the inertia weight value of the whale group at the tth iteration; ω min 、ω max are the minimum and maximum values of the inertia weight, respectively, and T is the maximum number of iterations; Update the individual whale positions as follows: Among them, t is the current iteration number; X t is the current position vector; is the current global optimal position vector; D is the distance between the optimal individual position and the current individual position, which is defined as follows: The coefficient vectors A and C are defined as follows: A=2a·rand1-a C=2·rand2 Where: rand1 and rand2 are random numbers uniformly distributed in the range [0, 1]; a is the convergence factor, which decreases linearly from 2 to 0 with the number of iterations t, a = 2-2t / T; (d) If the entropy difference d(t) is less than the threshold 1×10 -10 Or when the maximum number of iterations is reached, the iteration ends and the optimal position vector of the whale group is output; (2) The constructed support vector regression model is used to perform endpoint extension on the training sample signal: For a given pressure signal sequence s(1), s(2)...s(N), where N is the number of sampling points of the original pressure signal sequence, the number of sample training l is first determined to generate a training set L = {(x1, y1), (x2, y2), ..., (x l ,y l )},in: x i =[s(i)s(i+1)...s(N-l+i-1)] T y i =s(N-l+i),1≤i≤l Predict the first endpoint sequence value and use the support vector regression model to obtain the first predicted value s(N+1) outside the boundary, that is, Among them, x l+1 =[s(l+1)s(l+2)...s(N)] T Then iterate step by step to obtain the predicted signal sequence value; (3) Use the method of ensemble empirical mode decomposition to decompose the extended signal sequence and obtain the intrinsic mode component imf i ; (4) Calculate the Pearson product-moment correlation coefficient of each eigenmode component using the following formula: Among them, ρ i On behalf of the IMF i The Pearson product-moment correlation coefficient with the original signal x(t), σ(x) represents the variance of the original signal sequence, and cov(·) represents the covariance of the original signal sequence and each component; (5) A threshold is selected based on the correlation between each component and the original signal, and the intrinsic mode components above the threshold are reconstructed. The denoised high-pressure oil pipe pressure signal is output, and the preprocessing is completed.

Images