CN113378740A

CN113378740A - Method for identifying cylinder pressure of diesel engine by utilizing orthogonal signal

Info

Publication number: CN113378740A
Application number: CN202110683386.0A
Authority: CN
Inventors: 王双朋; 赵慧敏; 梅检民; 常春; 沈虹; 肖静
Original assignee: Army Military Transportation University
Current assignee: Army Military Transportation University
Priority date: 2021-06-21
Filing date: 2021-06-21
Publication date: 2021-09-10

Abstract

The invention discloses a method for identifying cylinder pressure of a diesel engine by utilizing orthogonal signals. According to the invention, the vibration signals on the upper side of the cylinder cover and the vibration signals on the left side of the cylinder cover of the diesel engine are segmented and resampled at equal angles according to the working cycle to form a group of orthogonal signals, and the orthogonal signals are input into the optimized network, so that the high-precision identification of the cylinder pressure can be realized. The invention utilizes the characteristic of strong feature extraction capability of the convolutional neural network to resample the acquired data, adjusts the network structure and optimizes parameter selection according to the characteristic of the input signal to form a one-dimensional convolutional neural network, and selects the orthogonal signal as the input, thereby improving the cylinder pressure identification precision.

Description

Method for identifying cylinder pressure of diesel engine by utilizing orthogonal signal

Technical Field

The invention belongs to the monitoring and diagnosis of the state of an internal combustion engine, and particularly relates to a method for detecting the combustion quality by analyzing and reconstructing the in-cylinder pressure of the internal combustion engine through a vibration signal.

Background

The working principle of the diesel engine is that chemical energy is converted into mechanical energy, and the core place of energy conversion is a cylinder. It is very valuable to observe the operating state of the engine in real time by studying the cylinder pressure. At present, a pressure sensor is arranged on a large engine and can observe and measure cylinder pressure in real time, but a pressure sensor is not arranged on a small engine such as a vehicle diesel engine. The pressure sensor is arranged on a small engine, and the high-precision measurement of the cylinder pressure can be realized, but the problems of difficult installation, high cost, easy damage to the engine and the like exist. In order to realize the detection of the cylinder pressure without disintegration, the cylinder pressure signal can be fitted by acquiring a cylinder cover vibration signal. At present, methods for realizing cylinder pressure identification by analyzing vibration signals mainly comprise two types, namely reverse filtering and neural networks. Compared with inverse filtering, the cylinder pressure identification precision of the neural network is higher.

Many documents report examples of cylinder pressure identification using cylinder head signals and neural networks. The integrated neural network can be used for accurately identifying the in-cylinder pressure of the diesel engine under different working conditions, and the generalization and the precision of the integrated neural network are greatly improved. The maximum entropy spectrum density is used as a characteristic of a vibration signal, and the in-cylinder pressure is identified by utilizing a multi-hidden layer BP neural network optimized by a genetic algorithm. However, since the extracted feature amount is small, the cylinder pressure identification accuracy is not high. And the LTSA is also used for reducing the dimension of the high-dimensional feature set and inputting the reduced-dimension feature set into the LSSVM model to realize the cylinder pressure identification, and the method has the advantages of strong generalization capability, high identification precision and the like. These methods can effectively identify the cylinder pressure, but all require a large amount of feature extraction work, and the identification accuracy needs to be further improved.

Disclosure of Invention

The present invention is to overcome the above-mentioned technical deficiencies, and to provide a method for identifying a cylinder pressure of a diesel engine by using orthogonal signals, which can realize the purpose of identifying a cylinder pressure curve with high accuracy without disassembling. Not only can reduce a large amount of feature extraction work, but also can the cylinder pressure curve of different operating modes of discernment of high accuracy.

In order to achieve the purpose, the invention adopts the following technical scheme: firstly, collecting vibration signals on the upper side and the left side of a cylinder cover, synchronously collecting in-cylinder pressure signals, forming a group of orthogonal signals by the upper side and the left side of the cylinder cover together with the in-cylinder pressure signals as input, training a network and optimizing the network, and finally inputting a test signal to identify the cylinder pressure and comparing the cylinder pressure with the real cylinder pressure. The method comprises the following specific steps:

(1) mounting vibration sensors on the upper side and the left side of the cylinder cover, and punching a hole in the front side of the 6 th cylinder to mount a pressure sensor;

(2) collecting different rotating speed signals under normal working conditions, wherein the rotating speeds are respectively 800s/min, 1000s/min, 1200s/min, 1400s/min, 1600s/min and 1800 s/min;

(3) segmenting the acquired signals according to a working cycle, wherein each segment comprises a complete working cycle of air inlet, compression, work application and air exhaust, and forming orthogonal signals by the signals on the upper side of the cylinder cover and the signals on the left side;

(4) performing equal-angle resampling on the signals, performing selective analysis on multiple sampling points, and finally selecting 120 resampling points;

(5) grouping 600 collected groups of data, wherein 550 groups of training sets and 50 groups of training sets are test sets;

(6) training the network by taking the orthogonal signal and the in-cylinder pressure signal as input, and further adjusting a network algorithm and different parameters to optimize the network;

(7) inputting the test concentrated orthogonal signal into the optimized network to obtain an identified cylinder pressure curve, and comparing and analyzing the identified cylinder pressure curve with an actual cylinder pressure curve to obtain the validity of the identified cylinder pressure;

the vibration sensors are PCB M603C01 vibration acceleration sensors and are arranged on the upper side and the left side of the 6 th cylinder cover in a magnetic type installation mode; the pressure sensor is a Kistler6052 type sensor and is arranged at a punching hole at the front side of the No. 6 cylinder in a fixed mounting mode.

The invention has the following beneficial effects:

in the prior art, cylinder pressure identification mostly needs feature extraction, and in addition, identification precision needs to be improved; the invention utilizes the characteristic of strong feature extraction capability of the convolutional neural network, omits the manual feature extraction work, optimizes the network aiming at the characteristics of the vibration signal, and finally selects the orthogonal signal as the input for cylinder pressure identification, thereby greatly improving the identification precision and realizing the non-disintegrated cylinder pressure curve identification.

The convolutional neural network has the advantage of weight sharing, so that the convolutional neural network can learn corresponding features from huge data sets, and a complex feature extraction process is effectively avoided. The invention relates to a novel diesel engine cylinder pressure identification method based on orthogonal signals and a one-dimensional convolutional neural network. Optimizing the network from the aspects of structure and parameter selection; the upper side and the left side of the cylinder cover and the orthogonal signals are respectively used as input signals, and the one-dimensional convolutional neural network can be obtained through comparative analysis, so that the vibration signal characteristics can be effectively extracted, and the cylinder pressure identification is realized; the orthogonal signal is used as an input signal, so that the identification precision of the cylinder pressure curve can be greatly improved.

Drawings

FIG. 1 is a view of a sensor mounting location;

FIG. 2 is a flow chart of the actual vibration signal acquisition;

FIG. 3 is a graph of cylinder pressure versus vibration signals;

FIG. 4 is a diagram of a one-dimensional convolution model;

FIG. 5 is a top dead center ± 60 ° cylinder internal pressure plot;

FIG. 6 is a workflow diagram;

FIG. 7 is a graph of cylinder pressure identification for different algorithms;

FIG. 8 is a different signal RMSE;

fig. 9 is a graph of cylinder pressure identification for different signals.

Detailed Description

The following is an implementation of identifying cylinder pressure using an orthogonal signal and a one-dimensional convolutional neural network, which is described in detail in the accompanying drawings.

Firstly, collecting related data, collecting vibration signals on the upper side and the left side of a cylinder cover to form orthogonal signals, synchronously collecting in-cylinder pressure signals, and dividing the collected data into a training set 550 group and a testing set 50 group.

(1) A PCB M603C01 vibration acceleration sensor is respectively arranged on the upper side and the left side of a cylinder cover of the diesel engine to obtain vibration signals, and a Kistler6052 type pressure sensor is arranged on the front side of a 6 th cylinder of the diesel engine in a punching way. The sensor is installed as shown in figure 1, a vibration acceleration sensor 2 and a pressure sensor 3.

(2) The collected data is transmitted to a computer through a data acquisition card to form a digital signal. As shown in fig. 2.

(3) The sampling frequency is selected to be 65536Hz, 800, 1000, 1200, 1400, 1600 and 1800s/min are respectively collected, and the collected signal time domain waveform is shown in figure 3.

(4) The convolutional neural network is optimized according to the characteristics of the vibration signal, and the optimized 1-DCNN structure is shown in FIG. 4.

(5) And intercepting the pressure signal in the cylinder, and intercepting a signal of +/-60 degrees of a top dead center, as shown in figure 5.

(6) Firstly, training the network by utilizing orthogonal signals, then inputting the orthogonal signals of the test set into the network, and comparing and analyzing the identified cylinder pressure and the actual cylinder pressure. The specific work flow diagram is shown in fig. 6.

(7) The network is optimized in terms of algorithm and parameter selection, and cylinder pressure identification curves of different algorithms are shown in FIG. 7.

(8) Selecting proper algorithm, respectively taking the upper side of the cylinder cover, the left side of the cylinder cover and the orthogonal signals as input, and comparing training conditions of different signals, as shown in fig. 8.

(9) Three signals are input into the optimized network, and the cylinder pressure identification condition is shown in fig. 9. By taking RMSE as an evaluation index, the cylinder pressure identification precision of the orthogonal signal can be higher through analysis.

TABLE 1 root mean square error of different vibration signals

As described above, the result of training by randomly selecting data mostly satisfies the minimum RMSE of the orthogonal signal, and the fitting effect of the orthogonal signal is better from the view of the fitting graph. The method shows that the cylinder pressure curve can be effectively identified by using the 1-DCNN and the orthogonal signal, and then the curve is analyzed, so that the non-disintegration detection of the combustion condition of the diesel engine is realized.

Claims

1. A method for identifying cylinder pressure of a diesel engine by utilizing orthogonal signals is characterized by comprising the following steps: segmenting the vibration signals on the upper side and the left side of the cylinder cover of the diesel engine according to a working cycle, and resampling at equal angles to form a group of orthogonal signals; according to the characteristics of the one-dimensional vibration signal, a 1-DCNN network model is adjusted and optimized, collected data are divided into a training set and a testing set, the testing set is input into a trained network, and cylinder pressure is accurately obtained, and the method specifically comprises the following steps:

(1) respectively installing vibration sensors on the upper side and the left side of a cylinder cover of the 6 th cylinder, and punching a hole on the front side of the 6 th cylinder to install a pressure sensor;

(2) synchronously acquiring vibration signals and in-cylinder pressure signals on the upper side and the left side of a cylinder cover, and acquiring normal working condition data from idle speed to medium and high speed of 800s/min, 1000s/min, 1200s/min, 1400s/min, 1600s/min and 1800s/min respectively;

(3) segmenting the acquired data according to the working cycle, wherein each segment of data comprises a complete working cycle of air intake, compression, work application and air exhaust;

(4) dividing 600 groups of data into 550 groups of training sets and 50 groups of testing sets, combining vibration signals on the upper side and the left side of the cylinder cover into a group of orthogonal signals as input, and taking in-cylinder pressure signals as output;

(5) the network is structurally adjusted by adjusting the number of hidden layers to be in accordance with the characteristics of one-dimensional signals, and the network is optimized from the aspects of algorithm and parameter selection by constructing Root-Mean-Square error (RMSE) evaluation indexes;

(6) the orthogonal signal is input into the optimized network, the RMSE value of the identified cylinder pressure curve and the actual cylinder pressure curve is smaller than that of a single-path signal, and the cylinder pressure can be effectively identified by the method.

2. The method for identifying the cylinder pressure of the diesel engine by using the orthogonal signal as set forth in claim 1, wherein: the vibration sensor is a PCB M603C01 vibration acceleration sensor, and the vibration sensor is arranged on the upper side of the cylinder cover and on the left side of the cylinder cover.

3. The method for identifying the cylinder pressure of the diesel engine by using the orthogonal signal as set forth in claim 1, wherein: the cylinder pressure sensor is a Kistler6052 type sensor, and the installation mode is that the front side of the No. 6 cylinder is fixedly installed in a punching mode.