Research Article  Open Access
Jiachi Yao, Yang Xiang, Sichong Qian, Shuai Wang, "Noise Source Separation of an Internal Combustion Engine Based on a SingleChannel Algorithm", Shock and Vibration, vol. 2019, Article ID 1283263, 19 pages, 2019. https://doi.org/10.1155/2019/1283263
Noise Source Separation of an Internal Combustion Engine Based on a SingleChannel Algorithm
Abstract
The separation and identification technology of noise sources is the focus and hot spot in the field of internal combustion engine noise research. Combustion noise and piston slap noise are the main noise sources of an internal combustion engine. However, both combustion noise and piston slap noise occur almost at the top dead center. They mix in the time domain and frequency domain. It is difficult to accurately and effectively separate them. A singlechannel algorithm which combines timevarying filteringbased empirical mode decomposition (TVFEMD) and robust independent component analysis (RobustICA) methods is proposed to separate them. Firstly, the TVFEMD method is utilized to decompose the singlechannel noise signal into several intrinsic mode functions (IMFs). Then, the RobustICA method is applied to extract the independent components. Finally, related prior knowledge and timefrequency analysis are employed to identify noise sources. Furthermore, the spectral filtering method and the calculation method of piston slap noise based on the dynamic model are further carried out to verify separation results. The simulation and experimental research results show the effectiveness of the proposed method.
1. Introduction
The internal combustion engine has been widely used in a variety of transportation vehicles, such as ships and automobiles [1]. However, when the internal combustion engine is working, it generates huge noise in the surrounding environment. Noise can disturb people’s daily life and even endanger people’s health. Therefore, noise problem has become a growing concern of society. When people are exposed to noise for a long time, it can cause insomnia, cardiovascular disease, and even death [2, 3]. The global legislative committee is developing strict regulations to make ships, cars, and other transportation vehicles quieter [4].
As the main power source and noise source of ships and automobiles, the internal combustion engine has a significant impact on the total noise level [5]. Therefore, it is urgent to study the corresponding method to reduce the noise level of internal combustion engines. The primary task of reducing the noise of internal combustion engines is to study and analyze the characteristics of each noise source of internal combustion engines and then to develop a corresponding targeted and efficient noise reduction scheme. Thus, separating various noise sources of internal combustion engines has become a hot topic in the field of internal combustion engine noise research.
The noise source of the internal combustion engine is mainly composed of combustion noise, piston slap noise, gear noise, valve knock noise, fuel pump noise, and so forth [6]. Among these noise sources, combustion noise and piston slap noise account for 80% of the total noise [7]. They are the main noise sources of the internal combustion engine. By reducing the combustion noise and piston slap noise, the total noise will be significantly reduced. Therefore, it is necessary to study the combustion noise and piston slap noise.
However, the combustion noise and piston slap noise almost occur near the top dead center, and they severely alias in the timefrequency domain. Thus, it is very difficult to separate them. Currently, some scientists use the multichannel algorithm to separate noise sources, such as blind source separation method [8], independent component analysis method [9], independent component and wavelet analysis method [10], improved spectrofilter method [11], hierarchy diagnosis coherent power spectrum analysis method [12], and acoustic holography and sound intensity method [13].
However, in the practical engineering applications, due to the sensor cost and installation conditions, generally only fewer sensors can be used. The multichannel algorithm is limited in practical engineering applications. Therefore, many scientists begin to study the singlechannel algorithm to separate noise sources. For example, Du et al. [14] used the EMD and independent component analysis methods to separate the combustion excitation signal and piston slap signal. Bi et al. [15] employed the EEMDRobustICA method to separate the combustion noise, piston slap noise, and exhaust noise of gasoline engines. Zhang et al. [16] utilized the EEMD, coherent power spectrum analysis, and improved analytic hierarchy process method to identify the noise sources of internal combustion engines. Zheng et al. [17] employed the EEMD and generalized S transform methods to separate the combustion noise, piston slap noise, gear noise, and so forth.
The current singlechannel algorithm is mainly based on the EMD method and EEMD method to separate the noise sources of internal combustion engines. Besides, there are also some other methods such as the VMDbased method [18] and Gammatonebased method [19]. But the VMDbased method needs to set complex parameters, and the separation result is greatly affected by the setting parameters. The Gammatonebased method is mainly used to separate mixed speech signals. As for the EMD method, it has the modemixing and end effect problems [20, 21]. Although the EEMD method can overcome the modemixing problem, the selection of the parameters such as added noise amplitude and ensemble number is difficult to determine, and they have a great influence on the effect of signal decomposition [22]. In addition, the computational cost of the EEMD method is high [23].
Recently, the novel timevarying filteringbased empirical mode decomposition (TVFEMD) which was proposed by Li et al. could effectively solve the separation problem and the intermittence problem, and it has a better decomposition performance than the EMD method [24]. The TVFEMD method is a datadriven adaptive decomposition method, and it has been applied in bearing fault diagnosis [25, 26], wind speed forecasting [27], modal identification [28], etc. In this paper, the TVFEMD method is explored and applied to the noise source separation of an internal combustion engine. Because the TVFEMD method has the excellent signal decomposition ability and the RobustICA method has the outstanding ability to extract independent components from the mixed components, the singlechannel algorithm which combines TVFEMD and RobustICA methods is proposed to separate the internal combustion engine noise sources. Firstly, the singlechannel noise signal is decomposed into several IMFs by the TVFEMD method. Then, the RobustICA method is carried out to extract the independent components. Finally, related prior knowledge and timefrequency analysis are utilized to identify noise sources. The main contribution of this paper is to propose a singlechannel algorithm based on the TVFEMD and RobustICA methods to separate the noise sources of internal combustion engines. Moreover, the independent combustion noise calculation method (the spectral filtering method) and the piston slap noise calculation method (the calculation method of piston slap noise based on the dynamic model) are integrated together to verify the separation results of the combustion noise and piston slap noise. The separation method and calculation flow of noise sources of the internal combustion engine are obtained.
2. Related Methods
2.1. TVFEMD Method
The timevarying filteringbased empirical mode decomposition (TVFEMD) can adaptively decompose the nonstationary signals. Compared with the EMD method, the TVFEMD method solved the separation problem and the intermittence problem [24]. Hence, the TVFEMD method has better separation ability than the EMD method, especially in terms of nonstationary signals. The calculation steps of the TVFEMD method are as follows: Step 1: for the signal , use the Hilbert transform to calculate the instantaneous amplitude and the instantaneous frequency . Step 2: locate the local maxima and the local minima . Step 3: interpolate the set of points and to obtain and . Then, and . Step 4: interpolate the and to obtain and . Then, Step 5: calculate the local cutoff frequency through . Step 6: adjust to solve the intermittence problem. Step 7: then, . The Bspline approximation is applied on , and the extreme timings of is taken the knots. The approximate result can be obtained as . Step 8: if the stopping criterion , then an IMF of is obtained. Otherwise, let and continue to perform Step 1 to Step 7.
In the TVFEMD method, the bandwidth and Bspline order are the key parameters which need to be set in advance. According to [24], considering the computational cost and algorithm performance, these two parameters are adopted the default values, and .
Considering that the TVFEMD method has to be used in the nonstationary noise signal of the internal combustion engine, three nonstationary simulated signals are selected to illustrate the decomposition performance of the TVFEMD method. The simulated signals are shown in equation (2). The sampling frequency is 512 Hz. The timedomain waveform of the simulated signal is shown in Figure 1:
Then, the EMD method and TVFEMD method are utilized to decompose the mixed signal. The calculation results are shown in Figures 2 and 3.
As shown in Figure 2, it can be seen that the calculation results by the EMD method have large difference with the original simulation signal, especially shown in the red circle. From Figure 3, the calculation results by the TVFEMD method are similar with the original simulation signal, IMF1 corresponds to S1, IMF2 corresponds to S2, and IMF3 corresponds to S3.
In order to quantitatively compare the separation performance of the two methods, the following index was employed to calculate the matching degree between the separated signal and the original signal [29]:where . The quantitative index on the performance of EMD and TVFEMD methods is shown in Table 1.

According to equation (3), the larger the index is, the better the separation effect is. From Table 1, the TVFEMD method provided the best recovery S3 by the largest index 18.9132 dB, which was consistent with the results in Figure 3. As for S1 and S2, the index is also larger than the EMD method. The average index of the calculated results by the TFMEMD method is 16.9058 dB. But the EMD only scored 3.2405 dB. From the above analysis, it is inferred that the TVFEMD method presents a better decomposition performance than the EMD method.
2.2. RobustICA Method
The separation process of the ICA method is to find a transformation matrix of the observation mixed signal to make the output as independent as possible [30]:where W is the unmixing matrix and N is the additional noise.
The commonly used ICA method has the FastICA [31] and JADE [32]. However, the FastICA and JADE methods have some problems such as low computational accuracy and insufficient robustness. Later, Zarzoso and Comon [33–35] proposed the robust independent component analysis (RobustICA) method. Compared with FastICA and JADE, the RobustICA method is more robust and can better extract independent components from mixed signals.
The RobustICA method separates the independent source signals based on the kurtosis. The calculation function of kurtosis is defined as follows:where represents the mathematical expectations.
The calculation steps of the RobustICA method are as follows:(1)Calculate the coefficients of polynomial of optimal step size:(2)Extract the root of polynomial of optimal step size .(3)Select the root of polynomial to make the absolute value of the objective function in the search direction the largest:(4)Update .(5)Normalize .
In order to illustrate the performance, the sawtooth signal, square signal, and sinusoidal attenuation signal are selected, and they are shown in Figure 4.
A randomly generated mixing matrix by the rand(·) function in MATLAB software is used to mix the original simulation signals. Then the FastICA, JADE, and RobustICA methods are utilized to extract the independent components from the mixed signals. The calculation results are shown in Figure 5.
(a)
(b)
(c)
It can be seen from Figure 5 that the calculation results by RobustICA are very similar to the original simulated signals. However, the calculation results by FastICA and JADE have a certain difference with the original simulated signals, especially in the circled part. Due to the amplitude distortion in the ICA method, considering that the amplitude of the selected simulation signal is in the range of −1 to +1, the extracted independent components are normalized. The calculated quantitative indicators are shown in Table 2.

It can be seen in Table 2 that the RobustICA provided the best recovery x2 by the largest index 265.7350 dB, which was consistent with the results in Figure 5. Although the FastICA obtained the best value of 29.5894 dB for x1, it did not extract the x2 signal very well. The second largest index value for x1 is 24.2738 dB achieved by the JADE, but JADE does not work well for extracting x2 signals. As for x3, FastICA scored −6.3843 dB which is lower than RobustICA and JADE. On the whole, the average score of RobustICA is 103.3583 dB which is much larger than that of FastICA and JADE. From the above analysis, it infers that the RobustICA method has the better performance to extract independent components than the FastICA and JADE methods.
2.3. The SingleChannel Algorithm Based on TVFEMD and RobustICA Methods
Because the TVFEMD method has the excellent performance for decomposing signals and the RobustICA method is a good way to extract independent components from the mixed signals, the TVFEMD and RobustICA methods are combined together to separate noise sources. The specific calculation steps are as follows: Step 1: in the marine engine laboratory which has a similar structural arrangement to an actual ship, the internal combustion engine noise test is carried out to obtain the singlechannel noise signals. Step 2: the TVFEMD method is utilized to decompose the singlechannel noise signal into several IMFs. By the scale selection method, the appropriate IMFs are selected and used for further calculation. Step 3: because these selected IMFs are not always independent of each other, the IMFs and the singlechannel noise signal are combined together to form a new signal group. Then, the RobustICA method is further carried out to extract the independent components from the new signal group. Step 4: according to the prior knowledge of the internal combustion engine and the timefrequency analysis, the combustion noise and piston slap noise are initially identified from the independent components. Step 5: the spectral filtering method and the calculation method of piston slap noise based on the dynamic model are further carried out to identify and verify the separated results. Finally, the combustion noise and piston slap noise can be accurately obtained.
In order to illustrate the performance of the TVFEMDRobustICA method, the simulated signal is adopted in the above equation (2). First, the mixed signal is separated by the TVFEMD method, and the separation result is shown in Figure 3. Then, the RobustICA method is further utilized to extract independent components. Moreover, the EMDRobustICA method is also used to separate the mixed signal. The calculation results are presented in Figure 6.
(a)
(b)
From Figure 6, it can be seen that the TVFEMDRobustICA method has achieved good results. But the EMDRobustICA method does not work well, especially for IC1 and IC3, and they have great difference with the original simulated S1 and S2. For further quantitative analysis, the calculation results of the quantitative indicators are shown in Table 3.

From Table 3, the TVFEMDRobustICA provided the best separated signal S1 by the largest index 15.0491 dB, and the second largest index value is 13.1466 dB for S2. However, the EMDRobustICA only scored 3.0835 dB and −0.5192 dB for S1 and S2, respectively. As for S3, the EMDRobustICA scored −4.2874 dB, and it is slightly larger than the TVFEMDRobustICA which scored −5.3101 dB. In total, the average index of TVFEMDRobustICA is 7.6285 dB, and the EMDRobustICA is only −0.5744 dB. Thus, it infers that the TVFEMDRobustICA method has the better separation effect than the EMDRobustICA method, and it is utilized to separate the noise sources of the internal combustion engine.
3. Internal Combustion Engine Test
3.1. Test Platform
The internal combustion engine test was carried out in the marine engine laboratory which has a similar structural arrangement to the actual ship, and it is shown in Figure 7.
In the marine engine laboratory, there is a MAN B&W 6L16/24type diesel engine. The main technical parameters are shown in Table 4.

When the internal combustion engine is running, the six cylinders will generate lots of noise. It is very difficult to directly separate the noise sources generated by the six cylinders. In fact, each cylinder of the internal combustion engine produces the similar noise source. Therefore, it is only necessary to study the noise source generated by one specified cylinder. Hence, it is necessary to isolate the interference noise generated by other five cylinders.
3.2. Lead Coverage Method
In this experiment, the lead coverage method is adopted to wrap five cylinders (No. 1 cylinder, No. 2 cylinder, No. 3 cylinder, No. 5 cylinder, and No. 6 cylinder) of the internal combustion engine, and only No. 4 cylinder is not wrapped. In the process of the lead coverage method, firstly, the 10 mm thick flame retardant sound insulation cotton is wrapped outside the internal combustion engine. Secondly, the 1.5 mm thick lead plate is wrapped outside the internal combustion engine. Finally, the 10 mm thick flame retardant sound insulation cotton is again wrapped outside the internal combustion engine. The threelayer coverage method is adopted to isolate the interference noise. The lead coverage method is shown in Figure 8.
(a)
(b)
(c)
(d)
Through the lead coverage method, the interference noise generated by other five cylinders can effectively be isolated. It is beneficial to the noise source separation of the designated cylinder.
3.3. Measuring System and Measuring Points
Through the lead coverage method, the interference noise produced by other five cylinders can be isolated as much as possible. Then as for No. 4 cylinder, the microphones are utilized to measure the noise signals at the cylinder head top, the main slap side, and the vice slap side. The specific location of noise measurement points are shown in Figure 9.
(a)
(b)
(c)
The internal combustion engine measurement system consists of sensor equipment (sound pressure sensor, cylinder pressure sensor, charge amplifier, and top dead center sensor), NI cDAQ 9172 data acquisition chassis, NI 9234 data acquisition card, computer, etc., and it is shown in Figure 10.
The type of the cylinder pressure sensor is 6013CA, in which the range is 250 bar and the sensitivity is 21 pC/bar. The type of the singlechannel charge amplifier is 5018A1000. The PCB130F20 microphones are calibrated before measurement. The noise data measured in the test are saved on the computer through LabVIEW integrated data acquisition system. The sampling frequency is 25600 Hz.
The rated speed of MAN B&W 6L16/24type diesel engine is 1000 rpm. Thus, the test condition is selected as 1000 rpm and noload condition. Through the internal combustion engine test, the cylinder pressure signal p and each side noise signals (the cylinder head top y1, the main slap side y2 and the vice slap side y3) of a working cycle of the internal combustion engine can be obtained, and it is shown in Figure 11.
4. Separation and Identification of Noise Sources
Through the internal combustion engine test, each side noise signals can be obtained. Firstly, the TVFEMD method is applied to decompose the cylinder head top noise signal y1 into several IMFs. The decomposition results are shown in Figure 12.
Through the TVFEMD method, nine IMFs can be obtained. The scale selection method is utilized to select the IMFs which have the high coherence with noise source of the internal combustion engine for further calculation. The calculation expression of the correlation coefficient is shown in the following equation [15]:where and are, respectively, the mean of signal x and signal y.
Through calculation, the correlation coefficient between each IMF and noise signal is shown in Table 5.

Considering that the internal combustion engine has many noise sources, in order to retain more components for the next calculation, the components whose correlation coefficient is less than 0.1 are removed. The IMF3∼IMF9 are retained for subsequent calculation. However, these IMFs are not always independent of each other. Therefore, it is needed to further extract the independent component from these IMFs. These IMFs and the original cylinder head top noise signals are combined together to form a new signal group. Then, the RobustICA method is further employed to extract the independent component from the new signal group. The calculation results are shown in Figure 13.
From Figure 13, each independent component may be correspond to the internal combustion engine noise source, such as combustion noise, piston slap noise, air valve knock noise, fuel injection pump noise, gear meshing noise, and so forth. The main goal of this paper is to study combustion noise and piston slap noise. As for the other noise sources, it needs to be further researched.
According to the prior knowledge of the internal combustion engine, the IC1 and IC2 may be the combustion noise and the piston slap noise. The timedomain waveform, spectrum, and timefrequency diagram of IC1 and IC2 are shown in Figure 14.
(a)
(b)
From Figure 14(a), firstly, from the perspective of the time domain, the timedomain waveform amplitude of IC1 varies greatly at 370°CA. According to the relevant prior knowledge of the internal combustion engine, the firing order of the internal combustion engine is 124653. In the test, the firing angle of No. 4 cylinder is about 370°CA. Thus, the change time of IC1 amplitude is consistent with the firing time of No. 4 cylinder. Secondly, from the perspective of the frequency domain, the frequency component of IC1 is concentrated around 3,000 Hz. When combustion occurs in No. 4 cylinder, highfrequency oscillation occurs due to combustion, resulting in a higher frequency component. In addition, as can be seen from the timefrequency diagram of IC1, the frequency component energy at about 370°CA and 3,000 Hz is the highest. It coincides with the moment when No. 4 cylinder burns and the highfrequency oscillation is caused by combustion. Thus, it can be considered that IC1 is the combustion noise.
From Figure 14(b), from the perspective of the time domain, the timedomain waveform amplitude of IC2 varies greatly at 370°CA. It coincides with the moment when the piston of No. 4 cylinder hits the cylinder wall. From the perspective of the frequency domain, the frequency component of IC2 is concentrated around 400 Hz. Usually the frequency of piston slap noise is low. From the timefrequency diagram, at round 370°CA and 400 Hz, the frequency component energy is large. Thus, it can be considered that the IC2 is the piston slap noise.
Similarly, as for the main slap side noise signal y2 and the vice slap side noise signal y3, the calculation results are shown in Figures 15 and 16, respectively.
(a)
(b)
(a)
(b)
From Figure 15(a), the moment of amplitude change of IC1 and the ignition time of No. 4 cylinder are both at 370°CA. The frequency of IC1 is concentrated around 3,000 Hz. In the timefrequency diagram, the frequency component energy is large at round 370°CA and 3,000 Hz. However, there is still a certain frequency component at around 5,000 Hz. The noise measurement position is at main slap side, and there is an injection pump next to the main slap side. The 5,000 Hz frequency component may be caused by the injection pump. The main component of IC1 is the combustion noise. From Figure 15(b), the frequency component of IC2 is mainly concentrated around 400 Hz. In the timefrequency diagram, the frequency component energy is large at around 370°CA and 400 Hz. But the frequency component energy also appears at other locations. The piston has a certain contact area with the inner wall of the cylinder liner. This may be caused by repeated collisions between the piston and the cylinder liner. Thus, it can be considered that the IC2 component is mainly the piston slap noise.
From Figure 16(a), the moment of amplitude change of IC1 and the ignition time of No. 4 cylinder are also both at 370°CA. The frequency component of IC1 is also concentrated around 3,000 Hz. When No. 4 cylinder burns at around 370°CA, it produces highfrequency oscillation components. Thus, it can be considered that the IC1 is mainly the combustion noise. From Figure 16(b), the frequency component of IC2 is mainly concentrated around 400 Hz. In the timefrequency diagram, the frequency component energy is large at around 370°CA and 400 Hz. But it also has some other frequency components around 1,000 Hz. Near the vice slap side, there are some other parts such as exhaust pipe, and the frequency component may be caused by them. Therefore, the main component of IC2 is the piston slap noise.
5. Evaluation and Discussion
In order to further evaluate the separation effect of the separated combustion noise and piston slap noise, the spectral filtering method [11] and the calculation method of piston slap noise based on the dynamic model [36–39] are further utilized to calculate the independent combustion noise and piston slap noise, respectively.
5.1. Spectral Filtering Method
The MAN B&W 6L16/24type internal combustion engine has six cylinders. Suppose the cylinder pressure of these six cylinders is . The combustion noise generated by these six cylinders is . The transfer function between cylinder pressure and combustion noise is . The combustion noise calculation model is shown in Figure 17.
From Figure 17, the combustion noise can be calculated as follows:
The total combustion noise of the internal combustion engine is the sum of combustion noise generated by the six cylinders:
However, in the test, the five cylinders of the internal combustion engine were wrapped by the lead coverage method, and only No. 4 cylinder was exposed. Therefore, this combustion noise calculation model becomes a singlecylinder combustion noise calculation model, and it is shown in Figure 18.
Assume that the cylinder pressure and combustion noise of No. 4 cylinder are and . The radiation noise signal of No. 4 cylinder is . The other interference noise signal is :
In the singlecylinder combustion noise calculation model, the cylinder pressure and the radiation noise signal can directly be obtained by the test. Firstly, the spectral filtering function is needed to be calculated:where is the crossspectrum of cylinder pressure and radiation noise signal . is the autospectrum of cylinder pressure and radiation noise signal .
After calculating the spectral filtering function , the combustion noise estimation signal can be obtained through convolution the spectral filtering function with the cylinder pressure signal :where represents the convolution operation.
The error between the actual combustion noise signal and the combustion noise estimation signal is calculated as follows:
The combustion noise signal can be calculated by the spectral filtering method. However, there is an error between the calculated combustion noise and the actual combustion noise. In this paper, the calculated combustion noise by the spectral filtering method is mainly utilized to evaluate the separated combustion noise. Although there is an error between the calculated combustion noise and the actual combustion noise, the calculated combustion noise has the main characteristics of the actual combustion noise. Thus, the calculated combustion noise can be used to evaluate the separated combustion noise. As for the error, it needs to be further researched.
The combustion noise estimation signal can be calculated through the spectral filtering method. Then, the calculated combustion noise estimation signal is compared with the separated combustion noise. The calculated results are shown in Figure 19.
(a)
(b)
(c)
From Figure 19, at the cylinder head top, the separated combustion noise is consistent with the calculated combustion noise by the spectral filtering method. But it has many other interference components in the separated combustion noise. This is because the cylinder head top receives interference noise from both main slap side and vice slap side. But its main component is combustion noise. At main slap side, the separated combustion noise is basically consistent with the calculated combustion noise by the spectral filtering method, but there is a small difference in the vicinity of 2,000 Hz. At vice slap side, the separated combustion noise is also basically consistent with the calculated combustion noise, but there is a small difference in the vicinity of 5,000 Hz. In general, the frequency characteristics of the separated combustion noise are consistent with the calculated combustion noise by the spectral filtering method. Thus, it can be considered that the combustion noise can accurately be obtained by using the proposed singlechannel algorithm. The frequency component of combustion noise is concentrated around 3,000 Hz.
5.2. Calculation Method of Piston Slap Noise Based on Dynamic Model
The piston hits the cylinder wall to generate piston slap noise. When the internal combustion engine is working, the movement of the piston in the cylinder liner is very complicated, including the main (axial reciprocating) motion and the secondary (lateral and tilting) motion. The dynamic model of the piston in the cylinder liner is shown in Figure 20.
The schematic diagram of the force of the piston in the cylinder liner is shown in Figure 21.
In Figure 21, is the thrust of the cylinder pressure generated in the incylinder combustion on the top surface of the piston. and are, respectively, the piston inertia force and the piston pin inertia force generated by the reciprocating movement of the piston. , , and are, respectively, the inertial force of the piston, the inertial force of the piston pin, and the inertial moment of the piston produced by the secondorder motion of the piston. and are, respectively, the lubricating force and lubricating torque acting on the piston skirt. and are, respectively, the viscous friction of the lubricating oil and the frictional moment generated by the piston skirt. is the force of the connecting rod. is the distance between the center of the piston pin and the centerline of the piston. and are, respectively, the the distance between the center of mass of the piston pin and the upper part of the piston skirt and the distance between the center of mass of the piston and the upper part of the piston skirt. is the angle between the connecting rod, and the axis of the piston.
According to the force of the piston in the cylinder liner, the dynamic equation of piston movement in the cylinder liner can be obtained:
In fact, there is a lubricant film between the piston and the cylinder liner. Therefore, it is also necessary to consider the effect of lubricant film on the movement of the piston in the cylinder liner. The Reynolds equation for the effect of lubricant film is defined as follows:where is the circumferential direction of the piston liner, is the axial direction along the liner, is the reciprocating speed of the piston, is the viscosity of lubricating oil, is the time, is the lubricating oil film pressure, is the thickness of the lubricant film, and is the lubricating oil density.
Through formulas (15) and (16), the lubricant film pressure and the piston secondorder motion displacement, velocity, and acceleration can be obtained.
The piston collides with the inner wall of the cylinder liner to generate piston slap noise, and it will attenuate through the body structure. Therefore, it is necessary to calculate the attenuation factor of sound propagation of the body structure:where is the surface acoustic power of the body and is the piston and liner inner wall collision sound power.
The formula for calculating the body surface acoustic power is defined as follows:where is the surface radiation efficiency, is the air density, is the velocity of sound waves in air, is the surface area of noise radiation, and is the surface vibration velocity.
The formula for calculating piston and liner inner wall collision sound power is as follows:where is the impact impedance between piston and liner, . is the impact speed.
According to formulas (17)–(19), the following formula can be obtained:
Assume that the energy generated at the inner wall of the cylinder liner is not changed when it is transmitted to the body surface through the body structure:where is the density of the body structure, is the thickness of the body structure, is the surface area of the body structure, is the vibration speed of the body structure, is the density of the liner structure, is the thickness of the liner structure, is the surface area of the liner structure, and is the vibration speed of the liner structure.
By combining formula (20) and formula (21), the attenuation factor of sound propagation of the body structure can be calculated:
Now, the attenuation factor of sound propagation of the body structure and the sound power of the piston and cylinder liner at the point of impact are obtained. The sound power level at each node on the body structure surface is defined as follows:where is the reference sound power, and it is usually .
Then, the sound pressure level needs to be calculated:where is the distance between the calculation area location and the structural surface of the internal combustion engine body.
Finally, the piston slap noise of the internal combustion engine can be calculated:where is the piston slap noise on thrust side and is the piston slap noise on antithrust side.
The calculated piston slap noise based on the dynamic model is utilized to evaluate the separated piston slap noise. The calculated results are shown in Figure 22.
(a)
(b)
(c)
From Figure 22, at cylinder head top, the separated piston slap noise is consistent with the calculated piston slap noise. There is a difference at about 1,500 Hz. At main slap side, the separated piston slap noise is basically consistent with the calculated piston slap noise. At vice slap side, the separated piston slap noise is also basically consistent with the calculated piston slap noise. But it has some other frequency components around 1,000 Hz. This may be caused by some other parts, such as exhaust pipe, and it needs to be further researched. On the whole, for the real measured noise signals of internal combustion noise, it can be considered that the separated piston slap noise is generally consistent with the calculated piston slap noise. Thus, piston slap noise is accurately separated by using the proposed singlechannel algorithm.
5.3. Error Analysis
From Figures 19 and 22, it can be seen that the separated combustion noise and piston slap noise is basically consistent with the calculated independent combustion noise and piston slap noise. But there are still some errors. The main reasons for the error are as follows: (1) The internal combustion engine has complex structure and numerous parts, which will produce a lot of noise. Besides combustion noise and piston slap noise, there are also fuel injection pump noise, valve mechanism knocking noise, etc. The combustion noise and piston slap noise separated by using the singlechannel algorithm will contain a small amount of other noise components. Therefore, it is necessary to further study the highprecision separation algorithm. (2) The independent combustion noise calculation method (the spectral filtering method) and the piston slap noise calculation method (the calculation method of piston slap noise based on the dynamic model) are mainly based on mathematical algorithms and models to calculate independent combustion noise and piston slap noise. The calculated results can reflect the main characteristics of combustion noise and piston slap noise. But the calculated combustion noise and piston slap noise are slightly different from the actual combustion noise and piston slap noise of an internal combustion engine. In order to better evaluate the accuracy of combustion noise and piston slap noise, it is necessary to further build independent combustion noise and piston slap noise simulation test bench to obtain independent combustion noise and piston slap noise. It needs to be further researched.
In engineering, the sound pressure level (SPL) is usually used to represent the noise situation of the internal combustion engine. The calculation formula of sound pressure level is as follows [7]:where .
The calculated sound pressure level of the combustion noise and piston slap noise is shown in Figure 23.
(a)
(b)
(c)
(d)
It can be seen from Figure 23 that the sound pressure level of separated combustion noise and piston slap noise is consistent with the calculated combustion noise and piston slap noise. In order to quantitatively analyze the error, the following error calculation formula is adopted [7]:where abs means to take the absolute value.
The calculation results are shown in Table 6.

From Table 6, it can be seen that the sound pressure level errors between the calculated combustion noise and each side (cylinder head top, main slap side, and vice slap side) combustion noise are 4.14 dB(A), 3.86 dB(A), 3.81 dB(A), respectively. The sound pressure level error at the cylinder head top is the largest. This is because the cylinder head top receives interference noise from main slap side and vice slap side. Since the main slap side has a fuel injection pump, the sound pressure level error is also large. The fuel injection pump noise has a greater impact on main slap side piston slap noise, and the sound pressure level error between the calculated piston slap noise and main slap side piston slap noise is 4.24 dB(A). On the whole, compared with main slap side and vice slap side, the average error of cylinder head top is 3.67 dB(A). The main slap side has the largest error, and the average is up to 4.05 dB(A). From the perspective of engineering applications, the error of sound pressure level within 3 dB indicates the calculation results are very good. In general, the error of sound pressure level within 5 dB is also acceptable. In the next step, it is needed to study highprecision separation algorithms and build the independent noise source test bench for more indepth research.
6. Conclusions
(1)In order to obtain the radiation noise signal of an internal combustion engine in the marine engine laboratory, the lead coverage method using a threelayer covering is utilized to isolate interference noise from other cylinders. The interference noise is effectively reduced from the noise source level, and the quality and reliability of the radiation noise signal measured by the test are greatly improved.(2)The singlechannel algorithm based on the TVFEMD and RobustICA methods is proposed to separate the noise sources. Moreover, the spectral filtering method and the calculation method of piston slap noise based on the dynamic model are further proposed to calculate independent combustion noise and piston slap noise. The research results show the proposed method can effectively separate noise sources. The separation method and calculation flow of internal combustion engine noise sources are obtained.
Data Availability
The MATfile data used to support the findings of this study are available from the corresponding author upon request.
Conflicts of Interest
The authors declare that they have no conflicts of interest.
Acknowledgments
This work was supported by the National Natural Science Foundation of China (51279148) and the China Scholarship Fund (201806950033).
References
 S. Verhelst, J. W. Turner, L. Sileghem, and J. Vancoillie, “Methanol as a fuel for internal combustion engines,” Progress in Energy and Combustion Science, vol. 70, pp. 43–88, 2019. View at: Publisher Site  Google Scholar
 D. Vienneau, H. Héritier, M. Foraster et al., “Façades, floors and mapsinfluence of exposure measurement error on the association between transportation noise and myocardial infarction,” Environment International, vol. 123, pp. 399–406, 2019. View at: Publisher Site  Google Scholar
 D. Singh, N. Kumari, and P. Sharma, “A review of adverse effects of road traffic noise on human health,” Fluctuation and Noise Letters, vol. 17, no. 1, pp. 1–10, 2018. View at: Publisher Site  Google Scholar
 J. K. Brueckner and R. Girvin, “Airport noise regulation, airline service quality, and social welfare,” Transportation Research Part B: Methodological, vol. 42, no. 1, pp. 19–37, 2008. View at: Publisher Site  Google Scholar
 S. Delvecchio, P. Bonfiglio, and F. Pompoli, “Vibroacoustic condition monitoring of internal combustion engines: a critical review of existing techniques,” Mechanical Systems and Signal Processing, vol. 99, pp. 661–683, 2018. View at: Publisher Site  Google Scholar
 A. Acri, E. Nijman, M. Klanner, G. Offner, and R. Corradi, “On the influence of cyclic variability on surface noise contribution analysis of internal combustion engines,” Applied Acoustics, vol. 132, pp. 97–108, 2018. View at: Publisher Site  Google Scholar
 E. G. Giakoumis, D. C. Rakopoulos, and C. D. Rakopoulos, “Combustion noise radiation during dynamic diesel engine operation including effects of various biofuel blends: a review,” Renewable and Sustainable Energy Reviews, vol. 54, pp. 1099–1113, 2016. View at: Publisher Site  Google Scholar
 C. Servière, J.L. Lacoume, and M. El Badaoui, “Separation of combustion noise and pistonslap in diesel enginepart II: separation of combustion noise and pistonslap using blind source separation methods,” Mechanical Systems and Signal Processing, vol. 19, no. 6, pp. 1218–1229, 2005. View at: Publisher Site  Google Scholar
 W. Li, F. Gu, A. D. Ball, A. Y. T. Leung, and C. E. Phipps, “A study of the noise from diesel engines using the independent component analysis,” Mechanical Systems and Signal Processing, vol. 15, no. 6, pp. 1165–1184, 2001. View at: Publisher Site  Google Scholar
 X. Wang, F. Bi, C. Liu et al., “Blind source separation and identification of internal combustion engine noise based on independent component and wavelet analysis,” in Proceedings of the IEEE 2011 International Conference on Electrical and Control Engineering (ICECE), pp. 113–116, Yichang, China, September 2011. View at: Google Scholar
 L. Pruvost, Q. Leclère, and E. Parizet, “Diesel engine combustion and mechanical noise separation using an improved spectrofilter,” Mechanical Systems and Signal Processing, vol. 23, no. 7, pp. 2072–2087, 2009. View at: Publisher Site  Google Scholar
 G. Shu and X. Liang, “Identification of complex diesel engine noise sources based on coherent power spectrum analysis,” Mechanical Systems and Signal Processing, vol. 21, no. 1, pp. 405–416, 2007. View at: Publisher Site  Google Scholar
 J. Xi, Z. Feng, G. Wang, and F. Wang, “Vibration and noise source identification methods for a diesel engine,” Journal of Mechanical Science and Technology, vol. 29, no. 1, pp. 181–189, 2015. View at: Publisher Site  Google Scholar
 X. Du, Z. Li, F. Bi, J. Zhang, X. Wang, and K. Shao, “Source separation of diesel engine vibration based on the empirical mode decomposition and independent component analysis,” Chinese Journal of Mechanical Engineering, vol. 25, no. 3, pp. 557–563, 2012. View at: Publisher Site  Google Scholar
 F. Bi, L. Li, J. Zhang, and T. Ma, “Source identification of gasoline engine noise based on continuous wavelet transform and EEMDrobustICA,” Applied Acoustics, vol. 100, pp. 34–42, 2015. View at: Publisher Site  Google Scholar
 J. Zhang, J. Wang, J. Lin et al., “Diesel engine noise source identification based on EEMD, coherent power spectrum analysis and improved AHP,” Measurement Science and Technology, vol. 26, no. 9, pp. 1–16, 2015. View at: Publisher Site  Google Scholar
 X. Zheng, Z. Y. Hao, Y. Jin et al., “Application of EEMD and GST method in noise characteristics analysis of internal combustion engine,” Chinese Internal Combustion Engine Engineering, vol. 32, no. 5, pp. 68–73, 2011. View at: Google Scholar
 J. Yao, Y. Xiang, S. Qian, S. Wang, and S. Wu, “Noise source identification of diesel engine based on variational mode decomposition and robust independent component analysis,” Applied Acoustics, vol. 116, pp. 184–194, 2017. View at: Publisher Site  Google Scholar
 J. Yao, Y. Xiang, S. Qian, and S. Wang, “Radiation noise separation of internal combustion engine based on gammatonerobustICA method,” Shock and Vibration, vol. 2017, Article ID 7565041, 14 pages, 2017. View at: Publisher Site  Google Scholar
 N. E. Huang, Z. Shen, S. R. Long et al., “The empirical mode decomposition and the Hilbert spectrum for nonlinear and nonstationary time series analysis,” Proceedings of the Royal Society of London. Series A: Mathematical, Physical and Engineering Sciences, vol. 454, no. 1971, pp. 903–995, 1998. View at: Publisher Site  Google Scholar
 G. Rilling and P. Flandrin, “One or two frequencies? The empirical mode decomposition answers,” IEEE Transactions on Signal Processing, vol. 56, no. 1, pp. 85–95, 2007. View at: Publisher Site  Google Scholar
 Z. Wu and N. E. Huang, “Ensemble empirical mode decomposition: a noiseassisted data analysis method,” Advances in Adaptive Data Analysis, vol. 1, no. 1, pp. 1–41, 2009. View at: Publisher Site  Google Scholar
 J. Gilles, “Empirical wavelet transform,” IEEE Transactions on Signal Processing, vol. 61, no. 16, pp. 3999–4010, 2013. View at: Publisher Site  Google Scholar
 H. Li, Z. Li, and W. Mo, “A time varying filter approach for empirical mode decomposition,” Signal Processing, vol. 138, pp. 146–158, 2017. View at: Publisher Site  Google Scholar
 Y. Xu, Z. Cai, and K. Ding, “An enhanced bearing fault diagnosis method based on TVFEMD and a highorder energy operator,” Measurement Science and Technology, vol. 29, no. 9, pp. 10–24, 2018. View at: Publisher Site  Google Scholar
 X. Zhang, Z. Liu, Q. Miao, and L. Wang, “An optimized time varying filtering based empirical mode decomposition method with grey wolf optimizer for machinery fault diagnosis,” Journal of Sound and Vibration, vol. 418, pp. 55–78, 2018. View at: Publisher Site  Google Scholar
 W. Zheng, X. Peng, D. Lu et al., “Composite quantile regression extreme learning machine with feature selection for shortterm wind speed forecasting: a new approach,” Energy Conversion and Management, vol. 151, pp. 737–752, 2017. View at: Publisher Site  Google Scholar
 M. Lazhari and A. Sadhu, “Decentralized modal identification of structures using an adaptive empirical mode decomposition method,” Journal of Sound and Vibration, vol. 447, pp. 20–41, 2019. View at: Publisher Site  Google Scholar
 Z. Li, X. Yan, X. Wang, and Z. Peng, “Detection of gear cracks in a complex gearbox of wind turbines using supervised bounded component analysis of vibration signals collected from multichannel sensors,” Journal of Sound and Vibration, vol. 371, pp. 406–433, 2016. View at: Publisher Site  Google Scholar
 S. Gepshtein and Y. Keller, “Iterative spectral independent component analysis,” Signal Processing, vol. 155, pp. 368–376, 2019. View at: Publisher Site  Google Scholar
 A. Hyvarinen, “Fast and robust fixedpoint algorithms for independent component analysis,” IEEE Transactions on Neural Networks, vol. 10, no. 3, pp. 626–634, 1999. View at: Publisher Site  Google Scholar
 J.F. Cardoso, “Highorder contrasts for independent component analysis,” Neural Computation, vol. 11, no. 1, pp. 157–192, 1999. View at: Publisher Site  Google Scholar
 V. Zarzoso and P. Comon, “Comparative speed analysis of fastICA,” in Proceedings of the International Conference on Independent Component Analysis and Signal Separation, pp. 293–300, Springer, London, UK, September 2007. View at: Google Scholar
 V. Zarzoso and P. Comon, “Robust independent component analysis for blind source separation and extraction with application in electrocardiography,” in Proceedings of the 30th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, pp. 3344–3347, Vancouver, Canada, August 2008. View at: Google Scholar
 V. Zarzoso and P. Comon, “Robust independent component analysis by iterative maximization of the kurtosis contrast with algebraic optimal step size,” IEEE Transactions on Neural Networks, vol. 21, no. 2, pp. 248–261, 2010. View at: Publisher Site  Google Scholar
 D. Zhu, H. S. Cheng, T. Arai, and K. Hamai, “A numerical analysis for piston skirts in mixed lubricationpart I: basic modeling,” Journal of Tribology, vol. 114, no. 3, pp. 553–562, 1992. View at: Publisher Site  Google Scholar
 D. Zhu, Y.Z. Hu, H. S. Cheng, T. Arai, and K. Hamai, “A numerical analysis for piston skirts in mixed lubrication: part IIdeformation considerations,” Journal of Tribology, vol. 115, no. 1, pp. 125–133, 1993. View at: Publisher Site  Google Scholar
 N. Dolatabadi, B. Littlefair, M. De la Cruz, S. Theodossiades, S. J. Rothberg, and H. Rahnejat, “A transient tribodynamic approach for the calculation of internal combustion engine piston slap noise,” Journal of Sound and Vibration, vol. 352, pp. 192–209, 2015. View at: Publisher Site  Google Scholar
 N. Dolatabadi, S. Theodossiades, and S. J. Rothberg, “Design optimization study of a nonlinear energy absorber for internal combustion engine pistons,” Journal of Computational and Nonlinear Dynamics, vol. 13, no. 9, Article ID 090910, 2018. View at: Publisher Site  Google Scholar
Copyright
Copyright © 2019 Jiachi Yao et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.