Embedded engine state monitoring and evaluating method and related device

Abstract

The invention discloses an embedded engine state monitoring evaluation method and a related device, and relates to the technical field of state evaluation, and the method comprises the steps: carrying out the high-pass filtering processing of obtained to-be-detected engine data, carrying out the time domain feature extraction and time-frequency domain feature extraction of the data after the high-pass filtering processing, and obtaining a time domain feature extraction result; obtaining a plurality of time domain features and a plurality of time-frequency domain features; performing normalization processing on the plurality of time domain features and the time-frequency domain features, and inputting the normalized data into a state monitoring evaluation model to obtain a fault state type of the to-be-detected engine; the state monitoring evaluation model is a model constructed based on a machine learning algorithm; the fault state types of the to-be-detected engine comprise normal working condition operation, intake valve clearance abnormity, oil supply advance angle advance and oil supply advance angle lag. According to the method, the time domain features and the time-frequency domain features are used as input of the state monitoring and evaluation model, and the abnormal state analysis and evaluation of the equipment power device can be accurately realized.

Description

Technical Field

[0001] The present application relates to the technical field of state assessment, and in particular to an embedded engine state monitoring and assessment method and related devices. Background Art

[0002] The most critical factor in evaluation modeling based on machine learning is data quality. The actual installation data collected in previous projects are mostly unlabeled data. For the currently widely used supervised learning, it is difficult to meet the model training requirements. The data obtained by the bench fault injection experiment cannot be considered high-quality data because it is not a real fault or not collected from the actual installation.

[0003] Therefore, how to design a solution that can accurately analyze and evaluate the abnormal state of equipment power units has become a technical problem that urgently needs to be solved in this field. Summary of the invention

[0004] The purpose of this application is to provide an embedded engine status monitoring and evaluation method and related devices, which can accurately realize the abnormal status analysis and evaluation of the equipment power unit.

[0005] To achieve the above objectives, this application provides the following solutions:

[0006] In a first aspect, the present application provides an embedded engine state monitoring and evaluation method, the embedded engine state monitoring and evaluation method comprising:

[0007] The engine data to be detected are obtained; the engine data to be detected are vibration data, instantaneous speed data and fuel pressure data under various states.

[0008] The engine data to be detected is subjected to high-pass filtering to obtain data after first high-pass filtering.

[0009] Perform time domain feature extraction on the data processed by the first high-pass filter to obtain a plurality of first time domain features.

[0010] Perform time-frequency domain feature extraction on the data processed by the first high-pass filter to obtain a plurality of first time-frequency domain features.

[0011] Normalization processing is performed on a plurality of the first time domain features and a plurality of the first time-frequency domain features respectively to obtain first normalized data.

[0012] The data after the first normalization processing is input into the state monitoring and evaluation model to obtain the fault state type of the engine to be detected; the state monitoring and evaluation model is a model constructed based on a machine learning algorithm; the fault state types of the engine to be detected include: normal operating conditions, abnormal intake valve clearance, advanced fuel supply advance angle and delayed fuel supply advance angle.

[0013] In a second aspect, the present application provides an embedded engine state monitoring and evaluation system, the embedded engine state monitoring and evaluation system comprising:

[0014] The data acquisition module is used to acquire the engine data to be detected; the engine data to be detected is vibration data, instantaneous speed data and fuel pressure data under various states.

[0015] The high-pass filter processing module is used to perform high-pass filter processing on the engine data to be detected to obtain data after the first high-pass filter processing.

[0016] The time domain feature extraction module is used to extract the time domain features of the data after the first high-pass filtering process to obtain a plurality of first time domain features.

[0017] The time-frequency domain feature extraction module is used to extract the time-frequency domain features of the data after the first high-pass filtering process to obtain a plurality of first time-frequency domain features.

[0018] A normalization processing module, used to perform normalization processing on a plurality of the first time domain features and a plurality of the first time-frequency domain features respectively, to obtain first normalized data;

[0019] A fault state type determination module is used to input the first normalized data into a state monitoring and evaluation model to obtain the fault state type of the engine to be detected; the state monitoring and evaluation model is a model built based on a machine learning algorithm; the fault state types of the engine to be detected include: normal operating conditions, abnormal intake valve clearance, advanced fuel supply advance angle and delayed fuel supply advance angle.

[0020] In a third aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-described embedded engine condition monitoring and assessment methods.

[0021] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described embedded engine condition monitoring and evaluation methods.

[0022] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned embedded engine condition monitoring and evaluation methods.

[0023] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0024] The present application provides an embedded engine state monitoring and evaluation method and related devices. The present application takes the power unit as the research object, performs high-pass filtering on the acquired engine data to be detected, and respectively performs time domain feature extraction and time-frequency domain feature extraction on the data after high-pass filtering to obtain a number of time domain features and a number of time-frequency domain features; after normalizing a number of time domain features and time-frequency domain features, the normalized data is input into a state monitoring and evaluation model to obtain the fault state type of the engine to be detected; the state monitoring and evaluation model is a model constructed based on a machine learning algorithm; the fault state types of the engine to be detected include: normal operating conditions, abnormal intake valve clearance, advanced fuel supply advance angle, and delayed fuel supply advance angle. The present application can accurately realize the abnormal state analysis and evaluation of the equipment power unit through the combination of time domain features and time-frequency domain features. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0026] Figure 1 A flowchart of an embedded engine status monitoring and evaluation method provided in one embodiment of the present application.

[0027] Figure 2 A data transmission flow chart provided for an embodiment of the present application.

[0028] Figure 3 A schematic diagram of the connections between modules of a data acquisition device provided in one embodiment of the present application.

[0029] Figure 4 A functional block diagram of a control module provided in one embodiment of the present application.

[0030] Figure 5 This is a functional block diagram of a collection module provided in one embodiment of the present application.

[0031] Figure 6 This is a functional block diagram of a conditioning module provided in one embodiment of the present application.

[0032] Figure 7 This is a functional block diagram of a charge-type vibration sensor conditioning circuit provided in one embodiment of the present application.

[0033] Figure 8 This is a functional block diagram of an analog voltage conditioning circuit provided in one embodiment of the present application.

[0034] Fig. 9 A functional block diagram of a power module provided in one embodiment of the present application.

[0035] Fig.10 A backplane signal interconnection block diagram provided for one embodiment of the present application.

[0036] Fig.11 A schematic diagram of a time domain waveform of a vibration sensor provided in an embodiment of the present application.

[0037] Fig.12 A schematic diagram of a sensor layout provided in accordance with an embodiment of the present application.

[0038] Fig.13 A schematic diagram of vibration signal timing provided in one embodiment of the present application.

[0039] Fig.14 A timing diagram of a speed signal provided in an embodiment of the present application.

[0040] Fig.15 The IMF component diagrams of various orders extracted from the normal working condition vibration signal provided in an embodiment of the present application.

[0041] Fig.16 This is a spectrum diagram of each IMF component of a normal operating vibration signal provided in an embodiment of the present application.

[0042] Fig.17 This is a diagram of the relative ratios of the energies of the IMF components of a vibration signal under normal operating conditions provided in one embodiment of the present application.

[0043] Fig.18 This is a diagram of the wavelet decomposition results of a normal operating vibration signal provided in an embodiment of the present application.

[0044] Fig.19 This is a spectrum diagram of each node of the wavelet decomposition of a vibration signal under normal operating conditions provided in an embodiment of the present application.

[0045] Fig. 20 A diagram showing the relative ratio of energies of each node of a normal operating condition vibration signal decomposed by wavelet decomposition according to an embodiment of the present application.

[0046] Fig.21 A neural network fault diagnosis flowchart provided in one embodiment of the present application.

[0047] Fig. 22 A flowchart of SVM fault diagnosis provided in one embodiment of the present application.

[0048] Fig.23 This is a schematic diagram of the research (development) process of this application.

[0049] Fig.24 This is a schematic diagram of the overall architecture of the embedded engine condition monitoring and evaluation system of this application.

[0050] Fig.25 A schematic diagram of functional modules of an embedded engine status monitoring and evaluation system provided in one embodiment of the present application. DETAILED DESCRIPTION

[0051] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0052] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0053] This application aims to address the problems of single function of equipment cloud point devices, lack of self-assessment and self-diagnosis functions, and low intelligence level. Taking the power unit as the research object, this application conducts status assessment of the equipment power unit by tracking and collecting the extreme performance and data of the fault state of the equipment power unit. By deploying an embedded intelligent assessment and diagnosis system on the terminal, an abnormal status assessment model for power unit power seriously reduced, engine abnormal noise, misfire, severe cylinder wear, etc. is formed. The abnormal status assessment model is used to realize abnormal status analysis and assessment of the equipment power unit, providing equipment status assessment means for the reliability of the equipment.

[0054] In an exemplary embodiment, Figure 1 As shown, an embedded engine condition monitoring and evaluation method is provided, which includes the following steps.

[0055] S1: Acquire the engine data to be detected; the engine data to be detected includes vibration data, instantaneous speed data and fuel pressure data under various states.

[0056] S2: Performing high-pass filtering on the engine data to be detected to obtain data after first high-pass filtering.

[0057] S3: extracting time domain features from the data processed by the first high-pass filter to obtain a plurality of first time domain features.

[0058] S4: extracting time-frequency domain features from the data processed by the first high-pass filter to obtain a plurality of first time-frequency domain features.

[0059] S5: performing normalization processing on a plurality of the first time domain features and a plurality of the first time-frequency domain features respectively to obtain first normalized data.

[0060] S6: Input the first normalized data into a state monitoring and evaluation model to obtain the fault state type of the engine to be detected; the state monitoring and evaluation model is a model constructed based on a machine learning algorithm; the fault state types of the engine to be detected include: normal operating conditions, abnormal intake valve clearance, advanced fuel supply advance angle and delayed fuel supply advance angle.

[0061] By implementing the above steps S1 to S6, the fault state type of the engine to be detected can be accurately identified through the state monitoring evaluation model.

[0062] In another exemplary embodiment of the present application, the determination of the condition monitoring evaluation model specifically includes:

[0063] A1: Acquire historical engine data; the historical engine data is vibration data under various states.

[0064] A2: Perform high-pass filtering on the historical engine data to obtain data after second high-pass filtering.

[0065] A3: Perform time domain feature extraction on the data processed by the second high-pass filter to obtain a plurality of second time domain features.

[0066] A4: Perform time-frequency domain feature extraction on the data processed by the second high-pass filter to obtain a plurality of second time-frequency domain features.

[0067] A5: Normalize the plurality of second time domain features and the plurality of second time-frequency domain features respectively to obtain second normalized data.

[0068] A6: Based on the data after the second normalization process, a machine learning algorithm is trained to obtain a condition monitoring and evaluation model.

[0069] When the model is a BP neural network, the machine learning algorithm is trained according to the data after the second normalization process to obtain a condition monitoring and evaluation model, which specifically includes:

[0070] a1: According to the data after the second normalization processing, the BP neural network is trained using a back propagation algorithm.

[0071] a2: Determine whether the termination condition is met and obtain the first determination result.

[0072] a3: If the first judgment result is no, then return to "using the back propagation algorithm to train the BP neural network".

[0073] a4: If the first judgment result is yes, then the optimal weight and threshold of the BP neural network are output to obtain a trained BP neural network model; the trained BP neural network model is the first state monitoring and evaluation model.

[0074] When the model is an SVM network, the machine learning algorithm is trained according to the data after the second normalization process to obtain a condition monitoring and evaluation model, which specifically includes:

[0075] b1: According to the data after the second normalization processing, the SMO algorithm is used to train the SVM network.

[0076] b2: Determine whether the termination condition is met and obtain a second determination result.

[0077] b3: If the second judgment result is no, return to "using the SMO algorithm to train the SVM network".

[0078] b4: If the second judgment result is no, the optimal parameters of the SVM network are output to obtain a trained SVM model; the trained SVM model is the second state monitoring and evaluation model.

[0079] In an exemplary embodiment, in step S4, it specifically includes:

[0080] S41: performing three-layer wavelet packet decomposition on the data after the first high-pass filtering process to obtain a plurality of first wavelet features.

[0081] S42: Using the first wavelet feature as a first time-frequency domain feature.

[0082] Specifically, in step S6, it specifically includes:

[0083] The data after the first normalization processing is input into the state monitoring and evaluation model, and the fault state type of the engine to be detected is output using an anti-normalization algorithm.

[0084] In an exemplary embodiment, the embedded engine monitoring terminal prototype (data acquisition device) can realize the collection and monitoring of the engine working status, and send the monitoring data to the host computer (sensor signal acquisition, data processing and result reporting). The host computer combines the fault diagnosis and status evaluation algorithm to complete the status analysis and evaluation of the engine. The data transmission flow chart is as follows Figure 2 shown.

[0085] The engine abnormal state monitoring system includes: a data acquisition device and a state evaluation system.

[0086] like Figure 3 As shown, the data acquisition device includes: a control module, an acquisition module, a conditioning module, a power module, a backplane and a sensor system.

[0087] The control module is the core module of the data acquisition device, using model HM250X_CPU, and its main functions include:

[0088] a) Data preprocessing and analysis.

[0089] b) Data storage and upload.

[0090] c) Control the acquisition module to complete data acquisition.

[0091] d) Complete the system's external communication function.

[0092] e) System initialization, self-test, resource allocation, etc.

[0093] The control module is divided according to the functions to be implemented, including: CPU core circuit, CAN interface circuit, debug interface circuit, bus and signal driving circuit and reset control unit, etc.

[0094] The CAN interface realizes external communication function and is connected to the system display and control computer. A 120Ω termination resistor is designed on the board. The control module function block diagram is as follows Figure 4 shown.

[0095] The acquisition module realizes analog switch control, analog-to-digital conversion, data caching and uploading functions. The functional block diagram of the acquisition module is as follows: Figure 5 shown.

[0096] The acquisition module supports 16 channels of analog signal synchronous acquisition. The onboard FPGA implements ADC chip acquisition control, bus driver, internal data cache, register implementation, analog switch control, etc. The model is HM250X_DAQ.

[0097] On the conditioning module, various physical quantity signals or electrical signals are conditioned by conversion circuits, filtering circuits, and amplification circuits and transformed into analog signals that meet certain amplitude and bandwidth requirements, and finally connected to the input end of the acquisition board through the backplane. Model: HM250X_SCD.

[0098] like Figure 6 As shown, the analog switch 2 circuit is designed to realize the self-check function of the conditioning channel, and the analog switch 1 circuit is designed to support the function of multiple conditioning modules. Other functional circuits need to be specifically designed according to the type of signal to be conditioned.

[0099] a) Charge type vibration sensor conditioning circuit.

[0100] The charge type vibration sensor has a very small output charge, a very high internal impedance, and poor anti-electromagnetic interference capability. Signal conditioning is necessary to convert the analog signal from the sensor into a digital signal. The conditioning circuit consists of a charge amplifier, a high-pass filter, a low-pass filter, an output amplifier, and a regulated power supply. The high-resistance charge signal output by the sensor is converted into a low-resistance voltage signal by the charge conversion circuit, and then the interference is filtered out by the high-pass and low-pass filters before being sent to the amplifier for amplification and output. The regulated power supply provides a suitable and stable DC regulated power supply for the components and chips in the circuit. The functional block diagram of the charge type vibration sensor conditioning circuit is shown below: Figure 7 shown.

[0101] Since the internal impedance of the piezoelectric sensor is extremely high and the output charge signal is very weak, the charge amplifier is required to have extremely high input impedance and extremely low bias current. Otherwise, when the bias current of the amplifier is close to the signal current, the signal may be overwhelmed by the bias current and normal amplification cannot be achieved.

[0102] b) Analog voltage conditioning circuit.

[0103] The analog voltage conditioning circuit consists of a voltage divider resistor, an amplifier, and a low-pass filter. The principle block diagram is shown in the figure. Figure 8 shown.

[0104] The power module converts the external input 24VDC power into +3.3V, +5V and ±12V power through the isolated power supply. The power module performs necessary over-current and over-voltage protection and power filtering on each group of power output to provide stable power supply for each module. The model is HM250X_PWR. The functional block diagram of the power module is as follows Fig. 9 shown.

[0105] The backplane provides power for each functional module and provides signal interconnection channels. The signal interconnection block diagram of each module on the backplane is as follows: Fig.10 shown.

[0106] Fig.10 The backplane signal interconnection relationship is displayed. The currently designed system supports 1 control module, 1 acquisition module, 2 conditioning modules, and 1 power module. If the signal line passes above the slot, it means that the group of signal lines is electrically connected to the slot. If the signal line passes below the slot, it means that the group of signal lines is not electrically connected to the slot.

[0107] Signal topology and function description:

[0108] GPIO: connects to all slots except the power module slot, and is used for triggering, control, status transmission and other functions;

[0109] Digital bus: Communication bus between control module and acquisition module, used for parameter configuration, status query and data transmission;

[0110] AI: Conditioning output analog signal;

[0111] Power supply: External +24V DC power supply is input, and the power module converts it into +5V, +3.3V, ±12V power supply to power each module.

[0112] Sensor system: mainly uses vibration sensors and fuel pressure sensors to collect signals about the working status of the diesel engine.

[0113] 1. Selection and installation of vibration sensors.

[0114] Diesel engines have many vibration excitation sources and complex transmission paths. System failures are both "longitudinal" and "lateral". Different results may be obtained when measuring at different measuring points. The main transmission path of combustion excitation is through the piston, connecting rod, crankshaft, and main bearing to the engine body. However, this transmission path greatly reduces the response signal, so the measuring point is generally not selected on this transmission path. In addition, the gas force caused by combustion will directly act on the cylinder head, causing the cylinder head to produce a vibration response. At the same time, the intake and exhaust throttling impact and the seating impact directly act on the cylinder head system. Therefore, it is more ideal to select the measuring point on the cylinder head close to the combustion chamber. In addition, the gas force acts on the cylinder liner and then on the cylinder body. Therefore, the measuring point can also be set on the transmission path. The piston knocking excitation directly acts on the cylinder liner and is transmitted to the cylinder body, and finally spreads throughout the engine body. Therefore, the measuring point can also be arranged on the diesel engine body.

[0115] 1.1 Vibration sensor installation.

[0116] In actual engineering, there are three main methods for installing vibration sensors, magnetic attraction, gluing and screw fixing. Based on reliability and durability, screw fixing is the first choice, followed by magnetic attraction + gluing. Experience shows that as long as the sensor is installed near the cylinder head, satisfactory results can be achieved whether it is the side cylinder body or the top cylinder head. Therefore, we first check whether there are threaded holes near the cylinder head, and try to use screw fixing whenever possible.

[0117] 1.2 Selection of vibration sensor.

[0118] There are two main options for vibration sensors: 1) charge type vibration sensor; 2) ICP sensor. This project chooses charge type sensor mainly based on the following considerations: low price, suitable for large-scale applications; good environmental adaptability, suitable for the harsh environment of the power cabin; passive and will not drift.

[0119] 1.3 Vibration sensor layout.

[0120] The vibration sensor time domain wave has a large attenuation on the propagation of vibration signals due to the large cylinder of the selected engine, such as Fig.11As shown, only the signals of this cylinder and the adjacent cylinders can be collected. Therefore, the layout of the vibration sensor in this project is as shown in the figure. There is no threaded hole on the cylinder block near the cylinder head, so only the gluing method can be selected. Since the top of the cylinder head is completely covered by the cylinder head cover, only the square vertical sensitive direction sensor can be installed on the side of the cylinder block.

[0121] 2. Fuel pressure sensor installation.

[0122] The fuel pressure sensor is installed on the fuel pipeline of a cylinder to collect the injection pressure signal. The data of this signal can be used to evaluate the working status of the fuel pump. It can also be used as a top dead center position signal of a cylinder to provide a benchmark for vibration and speed data analysis.

[0123] Sensor layout: Fig.12 As shown in the figure, the oil pressure sensors are installed on the oil pipes of the left 1 and right 1 cylinders respectively. Under normal circumstances, the oil pressure data of the left 1 cylinder is used as the basis for data analysis. When the left 1 data fails, the right 1 data is used. The installation positions of the vibration sensor and the oil pressure sensor are shown in the figure. Fig.12 shown.

[0124] The data collection is as follows:

[0125] 1) Vibration data collection.

[0126] according to Fig.12 The vibration sensor layout shown is installed in the vibration sensor. According to the test requirements, under the condition of no-load and idling of the diesel engine, four faults are set for the diesel engine: abnormal intake valve clearance, abnormal exhaust valve clearance, advance fuel supply advance angle fault, lag fuel supply advance angle fault and fuel cut-off fault, and the vibration signals of the diesel engine cylinder head under different states are collected. The experiment shows that the vibration response generated by the opening and closing of the intake valve is similar, and its energy is mainly concentrated in 6-8kHz; the vibration energy generated by combustion is mainly concentrated in 1-2kHz; the vibration energy generated in the later stage of combustion is mainly concentrated in 0.5-1kHz. This shows that the valve seat response is a high-frequency vibration signal, and the combustion and explosion response is a medium-to-low-frequency vibration signal. When a diesel engine fails, the energy in the corresponding frequency band or the frequency band range corresponding to the spectrum peak will change, so that diagnostic features can be provided. Since the fault frequency of the diesel engine cylinder head vibration signal is analyzed to be within 10kHz, according to the sampling theorem, the sampling frequency domain of the acceleration sensor is determined to be 25kHz.

[0127] 2) Speed ​​data collection.

[0128] A signal is output from the original engine eddy current instantaneous speed sensor to the monitoring terminal in parallel (experiments have shown that it has no effect on the ECU). The instantaneous speed sensor is installed directly above the engine flywheel tooth surface. When the flywheel rotates, each tooth passes through the eddy current sensor in turn. The change in the distance between the tooth surface and the probe is converted into a change in the output voltage amplitude, generating a signal that approximates a sine wave. The period of this approximate sine wave signal is the time it takes for each tooth to rotate through the magnetoelectric sensor. The high-speed data acquisition is used to collect and count the sensor output, so that the time each tooth rotates can be measured and converted into instantaneous speed. To ensure the test accuracy, a 200kHZ sampling rate is used.

[0129] 3) Oil pressure data collection.

[0130] The oil pressure sensor uses a piezoelectric clamping pressure sensor installed on cylinder 1 or left cylinder 1. It outputs a pulse every time cylinder 1 injects oil, and uses a 25kHZ sampling rate. In subsequent data analysis, the oil pressure data is used as the top dead center of cylinder 1, and the faulty cylinder is located based on this. Fig.13 and Fig.14 As shown, the time domain waveform of each cylinder's work and the speed fluctuation curve can be clearly seen.

[0131] Condition Assessment System: Uses fault diagnosis and condition assessment algorithms based on collected data to perform condition analysis and assessment on the engine.

[0132] The main research work carried out was on data mining technology, embedded autonomous assessment and diagnosis model (modeling technology) and its standardization, mainly including: 1) collecting vibration data, instantaneous speed data and fuel pressure data under various engine states through experiments; extracting and selecting features from the original data to find out the features that are most sensitive to health status or faults; 2) using health assessment and fault diagnosis algorithms to perform status assessment.

[0133] Step 1: Use time domain, frequency domain and time-frequency domain feature extraction algorithms to extract features from the collected fault data or various health status data. By analyzing the results, select the features that are most sensitive to the fault state or health state to form a feature system.

[0134] 1.1 Time domain analysis

[0135] When the diesel engine is running normally, 5 groups of samples are randomly selected for time domain feature extraction, and the time domain characteristic parameters of the cylinder head vibration signal are shown in Table 1. The indicators of each time domain characteristic parameter fluctuate within a relatively small range.

[0136] Table 1 Time domain characteristic parameters of cylinder head vibration signal under normal working conditions

[0137]

[0138] By comparing the four operating conditions of normal operation, abnormal intake valve clearance, advanced fuel injection advance angle and delayed fuel injection advance angle, it can be found that there are certain differences between the characteristic indicators in the time domain, so the differences between them can be used to identify different fault categories. Therefore, these time domain characteristic parameters can be used as the input of the neural network to diagnose the fault.

[0139] Table 2 Vibration time domain characteristics

[0140] Car name Working conditions RMS Kurtosis Factor Pulse Factor Margin Factor Kurtosis factor Skewness Factor 515 Normal working condition 0.193 134.857 23986.375 25.967 141.841 -0.701 515 Normal working condition 0.202 123.136 25732.755 24.813 199.827 -0.266 515 Normal working condition 0.216 107.469 20557.236 23.180 197.802 0.385 515 Normal working condition 0.186 145.280 22621.986 26.952 166.558 -0.248 515 Normal working condition 0.213 109.749 24072.933 23.425 141.227 -0.098 515 1 cylinder oil cut 0.166 181.178 17205.437 30.098 226.701 -0.865 515 1 cylinder oil cut 0.194 132.186 31984.315 25.708 244.828 -0.444 515 2 cylinder oil cut 0.175 163.573 16748.625 28.598 288.822 -0.452 515 2 cylinder oil cut 0.166 181.311 17486.387 30.109 243.415 -0.689 515 3 cylinder oil cut 0.211 112.512 48232.718 23.718 232.380 1.141 518 Normal working condition 0.049 598.226 2872.609 29.569 71.854 -0.102 518 Normal working condition 0.052 522.076 2773.695 26.954 103.119 0.523 518 Normal working condition 0.051 412.421 2151.305 21.014 70.005 0.334 518 Normal working condition 0.050 619.284 3092.452 31.046 88.479 0.510 518 Normal working condition 0.052 563.487 3025.918 29.308 102.697 0.908 518 1 cylinder oil cut 0.049 558.382 4845.654 27.421 132.253 -0.094 518 1 cylinder oil cut 0.049 558.382 4845.654 27.421 132.253 -0.094 518 2 cylinder oil cut 0.067 282.963 5934.607 18.972 85.271 -0.391 518 2 cylinder oil cut 0.067 282.963 5934.607 18.972 85.271 -0.391 518 3 cylinder oil cut 0.071 356.429 9140.408 25.462 96.308 1.391 Stand Cylinder pull(200k) 0.114 110.615 3910.482 12.593 26.545 -0.072 Stand Cylinder pull(300k) 0.147 97.804 4587.900 14.339 31.224 -0.204 Stand Cylinder pull(500k) 0.144 112.791 3520.927 16.278 32.748 0.045

[0141] Table 2 is a comparison of the time domain characteristics of the engines of the two vehicles under different working conditions in the actual vehicle experiment. Through the comparison of the above time domain signal characteristics, it can be seen that the dimensionless characteristics: waveform index, peak index, pulse index, margin index, kurtosis index and skewness index will have large differences when the engine is in different health states, which can be used as the basis for health status assessment. When the valve clearance is abnormal, the characteristics of the peak index, pulse index, margin index, kurtosis index and other reaction impact characteristics will also be different, which can be used as the basis for fault diagnosis. The maximum value, root mean square value, etc. will not only increase with the development of the fault state, but also change due to changes in working conditions, so they are not suitable for use as the basis for health assessment and fault diagnosis.

[0142] 1.2 Extract IMF component features based on empirical mode decomposition.

[0143] When the diesel engine is working normally, a group of samples are randomly selected to perform empirical mode decomposition on the cylinder head vibration signal, and the first 6-order IMF components are extracted. The decomposed waveform is as follows: Fig.15 As shown, then make the corresponding spectrum diagram according to the time domain signal of each IMF component, such as Fig.16 As shown, finally, the energy of each component is calculated according to the time domain signal of each node and standardized, as shown in Fig.17 As shown, it can be clearly seen that the relative energy size of each node, the energy of component 1 is relatively larger than the energy of other components, and the energy of nodes 5 and 6 is relatively smaller than the energy of other nodes.

[0144] By comparing the relative energies of the IMF components of each order under the four operating conditions of normal operation, abnormal intake valve clearance, advanced fuel supply advance angle and delayed fuel supply advance angle, it can be seen that the energy of component 1 is relatively large compared with the energies of other components, and the differences between the operating conditions are not large. The energies of nodes 2, 3 and 4 in each operating condition are quite different, and the energies of nodes 5 and 6 are relatively small compared with the energies of other nodes, and the differences between the operating conditions are not large. Therefore, the energy characteristics of nodes 2, 3 and 4 can be incorporated into the characteristic system of health status assessment and fault diagnosis.

[0145] 1.3 Wavelet analysis.

[0146] Normal working condition: When the diesel engine is working normally, a group of samples are randomly selected and the cylinder head vibration signal is decomposed by three layers of wavelet packets using wavelet analysis technology. The decomposed waveform is as follows: Fig.18 As shown, then make the corresponding spectrum diagram according to the time domain signal of each node, such as Fig.19 As shown, finally, the energy of each node is calculated according to the time domain signal of each node and normalized, as shown in Fig. 20 As shown, it can be clearly seen that the relative energy size of each node, the energy of nodes 1 and 2 is relatively larger than the energy of other nodes, and the energy of nodes 3 and 4 is relatively smaller than the energy of other nodes.

[0147] By comparing the wavelet relative energy of four operating conditions, namely, normal operation, abnormal intake valve clearance, advanced fuel injection advance angle and delayed fuel injection advance angle, it can be seen that the characteristic indicators of each node in the wavelet packet decomposition are somewhat different from each other, and the energy of different nodes in each operating condition varies greatly. Therefore, all eight nodes can be included in the characteristic system of health status assessment and fault diagnosis.

[0148] Step 2: Perform status assessment using health assessment and fault diagnosis algorithms.

[0149] 2.1BP neural network diagnosis.

[0150] 1) Obtain training sample data.

[0151] The experimental data of this embodiment are all from the time domain data collected by the vibration sensor of cylinder 1. The sampling rate is set to 20.48kHz, and the length of each sample is 1 second of continuous sampling. The number of samples for each working condition is: 500 groups of samples for normal engine working condition, 500 groups of samples for working condition with 4 degrees of advanced fuel advance angle, 500 groups of samples for working condition with 4 degrees of delayed fuel advance angle, and 500 groups of samples for working condition with excessive clearance of intake valve of cylinder 1. Among them, 400 groups of samples of each working condition are randomly selected as training samples, with a total of 1600 groups of training samples, and then the remaining 100 groups of samples of each working condition are used as test samples, with a total of 400 groups of test samples.

[0152] 2) Perform high-pass filtering on the training samples to remove power frequency interference below 50 Hz.

[0153] 3) The time domain feature extraction algorithm is applied to extract the time domain features of each sample, including five time domain features: peak index, pulse index, margin index, kurtosis index, and skewness index. Then the time-frequency domain features of the vibration signal are extracted. A three-layer wavelet packet decomposition algorithm is used to extract a total of eight wavelet features. EMD is used to extract three features of each order IMF component.

[0154] 4) Then, a total of 13 features of time domain and wavelet features are normalized, and a total of 8 features of time domain and EMD features are normalized.

[0155] 5) Use the back propagation algorithm to train the neural network model and obtain the final neural network diagnostic model.

[0156] The BP neural network model is trained by applying the back propagation algorithm, and the optimal weights and thresholds of the network are output after reaching the termination condition, which is the final neural network diagnosis model.

[0157] 6) Apply the neural network model to diagnose and identify test data samples of various working conditions.

[0158] The neural network diagnosis process is as follows: Fig.21 As shown in Table 3, the results of the BP neural network training and diagnosis experiments using 5 time domain features are shown in Table 3, the results of the BP neural network training and diagnosis experiments using 3 empirical mode decomposition IMF component features are shown in Table 4, the results of the BP neural network training and diagnosis experiments using 8 wavelet packet decomposition features are shown in Table 5, the results of the BP neural network training and diagnosis experiments using 8 features including 5 time domain features and 3 empirical mode decomposition features are shown in Table 6, the results of the BP neural network training and diagnosis experiments using 13 features including 5 time domain features and 8 wavelet packet decomposition features are shown in Table 7, and the accuracy of diagnosis using different feature parameters is shown in Table 8.

[0159] Table 3 Fault diagnosis results of BP neural network based on time domain characteristics

[0160] Normal working condition Fuel supply advance angle Fuel supply advance angle lag Intake valve clearance is too large Normal working condition 67 23 10 Fuel supply advance angle advance 4 degrees 17 72 8 2 Fuel supply advance angle lag 4 degrees 22 9 66 3 Intake valve clearance is too large 15 9 7 69

[0161] It can be seen from Table 3 that only time domain features are used to train the network. In normal working conditions, only 67 groups of 100 samples are correctly identified; 72 groups of samples are correctly identified for the advance fuel advance angle fault; only 66 groups of samples are correctly identified for the lag fuel advance angle fault, of which 22 groups of samples are misdiagnosed as normal working conditions; only 69 groups of samples are correctly identified for the excessive intake valve clearance fault, of which 15 groups of samples are diagnosed as normal working conditions; 274 groups of 400 samples are correctly diagnosed, and the overall accuracy is 68.5%.

[0162] Table 4 Fault diagnosis results of BP neural network based on EMD features

[0163] Normal working condition Fuel supply advance angle Fuel supply advance angle lag Intake valve clearance is too large Normal working condition 56 30 14 Fuel supply advance angle advance 4 degrees 3 62 35 Fuel supply advance angle lag 4 degrees 7 55 38 Intake valve clearance is too large 29 71

[0164] It can be seen from Table 4 that only EMD features are used to train the network. In normal working conditions, only 56 samples out of 100 are correctly identified; 62 samples are correctly identified for the advance fuel angle fault, of which 35 samples are misdiagnosed as the lag fuel angle fault; only 55 samples are correctly identified for the lag fuel angle fault, of which 38 samples are misdiagnosed as the excessive valve clearance fault; only 71 samples are correctly identified for the excessive intake valve clearance fault, of which 29 samples are diagnosed as the lag fuel angle fault; 244 samples are correctly diagnosed out of 400 samples, and the overall accuracy is 61.0%.

[0165] Table 5 Fault diagnosis results of BP neural network based on wavelet features

[0166] Normal working condition Fuel supply advance angle Fuel supply advance angle lag Intake valve clearance is too large Normal working condition 86 7 7 Fuel supply advance angle advance 4 degrees 10 82 6 2 Fuel supply advance angle lag 4 degrees 12 5 83 Intake valve clearance is too large 9 6 85

[0167] It can be seen from Table 5 that only wavelet features are used to train the network. In normal working conditions, only 86 groups of samples are correctly identified out of 100 groups of samples; 82 groups of samples are correctly identified for the advance fuel injection advance angle fault, of which 10 groups of samples are misdiagnosed as normal working conditions; 83 groups of samples are correctly identified for the lag fuel injection advance angle fault, of which 12 groups of samples are misdiagnosed as normal working conditions; 85 groups of samples are correctly identified for the excessive intake valve clearance fault, of which 9 groups of samples are diagnosed as normal working conditions; 336 groups of samples are correctly diagnosed out of 400 groups of samples, and the overall accuracy is 84.0%.

[0168] Table 6 Fault diagnosis results of BP neural network based on time domain features and EMD features

[0169] Normal working condition Fuel supply advance angle Fuel supply advance angle lag Intake valve clearance is too large Normal working condition 88 10 2 Fuel supply advance angle advance 4 degrees 11 82 7 Fuel supply advance angle lag 4 degrees 9 6 84 1 Intake valve clearance is too large 2 7 91

[0170] It can be seen from Table 6 that the network is trained using a total of 8 features, including time domain features and EMD features. In normal working conditions, only 88 samples out of 100 groups are correctly identified; 82 groups of samples are correctly identified for the advance fuel injection angle fault, of which 11 groups are misdiagnosed as normal working conditions; 84 groups of samples are correctly identified for the lag fuel injection angle fault, of which 9 groups are misdiagnosed as normal working conditions; 91 groups of samples are correctly identified for the excessive intake valve clearance fault, of which 7 groups are diagnosed as the lag fuel injection angle fault; 345 groups of 400 samples are correctly diagnosed, and the overall accuracy is 86.25%.

[0171] Table 7 Fault diagnosis results of BP neural network based on time domain features and wavelet features

[0172] Normal working condition Fuel supply advance angle Fuel supply advance angle lag Intake valve clearance is too large Normal working condition 97 3 Fuel supply advance angle advance 4 degrees 5 95 Fuel supply advance angle lag 4 degrees 3 4 93 Intake valve clearance is too large 3 1 96

[0173] It can be seen from Table 7 that the network is trained using 13 features, including time domain features and wavelet features. In normal working conditions, only 97 groups of 100 samples are correctly identified; 95 groups of samples are correctly identified for the advance fuel supply angle fault, of which 5 groups of samples are misdiagnosed as normal working conditions; 93 groups of samples are correctly identified for the lag fuel supply angle fault, of which 4 groups of samples are misdiagnosed as the advance fuel supply angle fault; 96 groups of samples are correctly identified for the excessive intake valve clearance fault, of which 3 groups of samples are diagnosed as normal working conditions; 381 groups of 400 samples are correctly diagnosed, and the overall accuracy is 95.25%.

[0174] Table 8 Diagnostic accuracy of the model trained using different feature parameters

[0175]

[0176] It can be seen from Table 8 that when only 5 features of the time domain are used to train the BP neural network, its diagnostic recognition accuracy is 68.5%; when only 3 features of the EMD are used to train the BP neural network, its diagnostic recognition accuracy is 61.0%; when only 8 features of wavelet packet decomposition are used to train the BP neural network, its diagnostic recognition accuracy is 84.0%; when a total of 8 features, including 5 features of the time domain and 3 features of the EMD, are used to train the BP neural network, its diagnostic recognition accuracy is 86.25%; when 5 features of the time domain are used to train the BP neural network, its diagnostic recognition accuracy is 84.0%. The BP neural network is trained with a total of 13 features, including 8 features of wavelet packet decomposition, and its diagnostic recognition accuracy is 95.25%; based on the above analysis, it can be seen that when only time domain features and EMD features are used to train the neural network model, its diagnostic recognition accuracy is low, while when only wavelet features are used to train the neural network model, the accuracy is relatively higher than the first two models. After combining the time domain and EMD features, its recognition accuracy is significantly improved. After combining the time domain features and wavelet features, the diagnostic recognition rate of the model is the highest, reaching 95.25%.

[0177] 2.2SVM model diagnosis.

[0178] 1) Obtain training sample data

[0179] The experimental data of this embodiment are all from the time domain data collected by the vibration sensor of cylinder 1. The sampling rate is set to 20.48kHz, and the length of each sample is 1 second of continuous sampling. The number of samples for each working condition is: 500 groups of samples for normal engine working condition, 500 groups of samples for working condition with 4 degrees of advanced fuel advance angle, 500 groups of samples for working condition with 4 degrees of delayed fuel advance angle, and 500 groups of samples for working condition with excessive clearance of intake valve of cylinder 1. Among them, 400 groups of samples of each working condition are randomly selected as training samples, with a total of 1600 groups of training samples, and then the remaining 100 groups of samples of each working condition are used as test samples, with a total of 400 groups of test samples.

[0180] 2) Perform high-pass filtering on the training samples to remove power frequency interference below 50 Hz.

[0181] 3) The time domain feature extraction algorithm is applied to extract the time domain features of each sample, including five time domain features: root mean square value, waveform index, peak index, pulse index, margin index, kurtosis index, and skewness index. Then the time-frequency domain features of the vibration signal are extracted. A three-layer wavelet packet decomposition algorithm is used to extract a total of eight wavelet features. EMD is used to extract three features of each order IMF component.

[0182] 4) Apply the normalization algorithm to normalize the training data samples.

[0183] A total of 13 features, including time domain and wavelet features, are normalized, and a total of 8 features, including time domain and EMD features, are normalized.

[0184] 5) Apply the SMO algorithm to optimize the SVM network and obtain the final SVM diagnosis model.

[0185] The back propagation algorithm is then used to train the SVM network model. After the termination condition is reached, the optimal parameters of the network are output as the final SVM diagnostic model.

[0186] 6) Apply the SVM diagnostic model to diagnose and identify the test data samples of various working conditions.

[0187] The neural network diagnosis process is as follows: Fig. 22 As shown in Table 9, the SVM network training and diagnosis experimental results using 5 time domain features are shown in Table 9, the SVM network training and diagnosis experimental results using 3 empirical mode decomposition IMF component features are shown in Table 10, the SVM network training and diagnosis experimental results using 8 wavelet packet decomposition features are shown in Table 3, the SVM network training and diagnosis experimental results using 8 features including 5 time domain features and 3 empirical mode decomposition features are shown in Table 11, the SVM network training and diagnosis experimental results using 13 features including 5 time domain features and 8 wavelet packet decomposition features are shown in Table 12, and the diagnostic accuracy using different feature parameters is shown in Table 13.

[0188] Table 9 Fault diagnosis results of SVM network based on time domain features

[0189] Normal working condition Fuel supply advance angle Fuel supply advance angle lag Intake valve clearance is too large Normal working condition 92 8 Fuel supply advance angle advance 4 degrees 6 86 8 Fuel supply advance angle lag 4 degrees 66 14 20 Intake valve clearance is too large 3 29 68

[0190] It can be seen from Table 9 that only time domain features are used to train the network. In normal working conditions, 92 out of 100 groups of samples are correctly identified; 86 groups of samples are correctly identified for the advance fuel angle fault; only 20 groups of samples are correctly identified for the lag fuel angle fault, of which 66 groups of samples are misdiagnosed as normal working conditions; only 68 groups of samples are correctly identified for the excessive intake valve clearance fault, of which 29 groups of samples are diagnosed as the advance fuel angle fault; 66 groups of 400 groups of samples are correctly diagnosed, and the overall accuracy is 66.5%.

[0191] Table 10 Fault diagnosis results of SVM network based on EMD features

[0192] Normal working condition Fuel supply advance angle Fuel supply advance angle lag Intake valve clearance is too large Normal working condition 91 8 1 Fuel supply advance angle advance 4 degrees 50 46 2 2 Fuel supply advance angle lag 4 degrees 15 77 8 Intake valve clearance is too large 4 19 14 63

[0193] It can be seen from Table 10 that only EMD features are used to train the network. In normal working conditions, 91 out of 100 groups of samples are correctly identified; 46 groups of samples are correctly identified for the advance fuel angle fault, of which 50 groups of samples are misdiagnosed as normal working conditions; only 77 groups of samples are correctly identified for the lag fuel angle fault, of which 15 groups of samples are misdiagnosed as normal working conditions; only 63 groups of samples are correctly identified for the excessive intake valve clearance fault, of which 19 groups of samples are diagnosed as the advance fuel angle fault; 277 groups of 400 groups of samples are correctly diagnosed, and the overall accuracy is 69.25%.

[0194] Table 11 Fault diagnosis results of SVM network based on wavelet features

[0195] Normal working condition Fuel supply advance angle Fuel supply advance angle lag Intake valve clearance is too large Normal working condition 100 Fuel supply advance angle advance 4 degrees 61 39 Fuel supply advance angle lag 4 degrees 9 91 Intake valve clearance is too large 3 2 5 90

[0196] It can be seen from Table 11 that only wavelet features are used to train the network. Only 100 groups of samples in normal working conditions are correctly identified; 39 groups of samples are correctly identified for the advance fuel injection angle fault, of which 61 groups of samples are misdiagnosed as normal working conditions; 91 groups of samples are correctly identified for the lag fuel injection angle fault, of which 9 groups of samples are misdiagnosed as normal working conditions; 90 groups of samples are correctly identified for the excessive intake valve clearance fault, of which 5 groups of samples are diagnosed as the lag fuel injection angle fault; 320 groups of 400 groups of samples are correctly diagnosed, and the overall accuracy is 80.0%.

[0197] Table 12 Fault diagnosis results of SVM network based on time domain features and EMD features

[0198] Normal working condition Fuel supply advance angle Fuel supply advance angle lag Intake valve clearance is too large Normal working condition 97 3 Fuel supply advance angle advance 4 degrees 39 56 1 4 Fuel supply advance angle lag 4 degrees 12 3 82 3 Intake valve clearance is too large 3 97

[0199] It can be seen from Table 12 that the network is trained using 13 features, including time domain features and EMD features. In normal working conditions, only 97 samples out of 100 groups are correctly identified; 56 samples are correctly identified for the advance fuel angle fault, of which 39 samples are misdiagnosed as normal working conditions; 82 samples are correctly identified for the lag fuel angle fault, of which 12 samples are misdiagnosed as normal working conditions; 97 samples are correctly identified for the excessive intake valve clearance fault, of which 3 samples are diagnosed as the advance fuel angle fault; 332 samples are correctly diagnosed out of 400 groups, with an overall accuracy of 83.0%.

[0200] Table 13 Fault diagnosis results of SVM network based on time domain features and wavelet features

[0201] Normal working condition Fuel supply advance angle Fuel supply advance angle lag Intake valve clearance is too large Normal working condition 100 Fuel supply advance angle advance 4 degrees 11 86 3 Fuel supply advance angle lag 4 degrees 9 6 85 Intake valve clearance is too large 5 3 92

[0202] It can be seen from Table 13 that the network is trained using 15 features, including time domain features and wavelet features. Only 100 groups of samples are correctly identified out of 100 groups of samples in normal working conditions; 86 groups of samples are correctly identified for the advance fuel advance angle fault, of which 11 groups of samples are misdiagnosed as normal working conditions; 85 groups of samples are correctly identified for the lag fuel advance angle fault, of which 9 groups of samples are misdiagnosed as normal working conditions; 92 groups of samples are correctly identified for the excessive intake valve clearance fault, of which 5 groups of samples are diagnosed as normal working conditions; 363 groups of samples are correctly diagnosed out of 400 groups of samples, with an overall accuracy of 90.75%.

[0203] Table 14 Diagnostic accuracy of the model trained using different feature parameters

[0204]

[0205] It can be seen from Table 14 that when only 5 features of the time domain are used to train the SVM network, its diagnostic recognition accuracy is 66.5%; when only 3 features of the EMD are used to train the SVM network, its diagnostic recognition accuracy is 69.25%; when only 8 features of wavelet packet decomposition are used to train the SVM network, its diagnostic recognition accuracy is 80.0%; when a total of 8 features, including 5 features of the time domain and 3 features of the EMD, are used to train the SVM network, its diagnostic recognition accuracy is 83.0%; when 5 features of the time domain are used to train the SVM network, its diagnostic recognition accuracy is 84.9%. The SVM network was trained with a total of 13 features, including 8 features of wavelet packet decomposition, and its diagnostic recognition accuracy was 90.75%; based on the above analysis, it can be seen that when only time domain features and EMD features are used to train the network model, its diagnostic recognition accuracy is low, while when only wavelet features are used to train the neural network model, the accuracy is relatively higher than the first two models. After combining the time domain and EMD features, its recognition accuracy has been significantly improved. After combining the time domain features and wavelet features, the diagnostic recognition rate of the model is the highest, reaching 90.75%.

[0206] This application uses machine learning-based assessment modeling to acquire labeled fault samples and establish a standardized condition assessment database architecture. This application uses pre-overhaul data collection and tracks the overhaul process. It uses actual fault labeling data to establish a standardized condition assessment database architecture, so it should be able to meet the requirements of model training. A universal, open, and standardized condition assessment operating platform has been implemented, and the algorithm modules can be distributed and cross-platform configured; this application is a universal, open, and standardized condition assessment operating environment and engine assessment system development method, including a universal embedded condition assessment monitoring terminal, a universal condition assessment operating environment software, an engine feature extraction software module, a health assessment / fault diagnosis software module, etc. These hardware, software, and development methods can be applied to the health management system of tracked vehicles. The schematic diagram of the research (development) process is shown in the figure. Fig.23 shown.

[0207] Based on the same inventive concept, the embodiment of the present application also provides an embedded engine state monitoring and evaluation system for implementing the embedded engine state monitoring and evaluation method involved above. The implementation scheme for solving the problem provided by the system is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more embedded engine state monitoring and evaluation system embodiments provided below can refer to the limitations of the embedded engine state monitoring and evaluation method above, and will not be repeated here.

[0208] In an exemplary embodiment, Fig.24 and Fig.25 As shown, an embedded engine state monitoring and evaluation system is provided, and the embedded engine state monitoring and evaluation system comprises:

[0209] The data acquisition module M1 is used to acquire the engine data to be detected; the engine data to be detected is vibration data, instantaneous speed data and fuel pressure data under various states.

[0210] The high-pass filter processing module M2 is used to perform high-pass filter processing on the engine data to be detected to obtain data after the first high-pass filter processing.

[0211] The time domain feature extraction module M3 is used to extract time domain features from the data processed by the first high-pass filter to obtain a plurality of first time domain features.

[0212] The time-frequency domain feature extraction module M4 is used to extract time-frequency domain features from the data processed by the first high-pass filter to obtain a plurality of first time-frequency domain features.

[0213] The normalization processing module M5 is used to perform normalization processing on a plurality of the first time domain features and a plurality of the first time-frequency domain features respectively to obtain first normalized data.

[0214] The fault state type determination module M6 is used to input the first normalized data into the state monitoring and evaluation model to obtain the fault state type of the engine to be detected; the state monitoring and evaluation model is a model constructed based on a machine learning algorithm; the fault state types of the engine to be detected include: normal operating conditions, abnormal intake valve clearance, advanced fuel supply advance angle and delayed fuel supply advance angle.

[0215] In an exemplary embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the above-mentioned method embodiments are implemented when the processor executes the computer program.

[0216] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the above-mentioned method embodiments are implemented.

[0217] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the above-mentioned method embodiments are implemented.

[0218] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0219] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. An embedded engine condition monitoring and evaluation method, characterized in that: The embedded engine condition monitoring and evaluation method comprises: Acquire the engine data to be tested; the engine data to be tested includes vibration data, instantaneous speed data and fuel pressure data under various states; Performing high-pass filtering on the engine data to be detected to obtain data after first high-pass filtering; Performing time domain feature extraction on the data processed by the first high-pass filter to obtain a plurality of first time domain features; Performing time-frequency domain feature extraction on the data processed by the first high-pass filter to obtain a plurality of first time-frequency domain features; Normalizing the plurality of first time domain features and the plurality of first time-frequency domain features respectively to obtain first normalized data; The data after the first normalization processing is input into the state monitoring and evaluation model to obtain the fault state type of the engine to be detected; the state monitoring and evaluation model is a model constructed based on a machine learning algorithm; the fault state types of the engine to be detected include: normal operating conditions, abnormal intake valve clearance, advanced fuel supply advance angle and delayed fuel supply advance angle.

2. The embedded engine condition monitoring and evaluation method according to claim 1 is characterized in that: The determination of the condition monitoring and evaluation model specifically includes: Acquire historical engine data; the historical engine data is vibration data under various states; Performing high-pass filtering on the historical engine data to obtain data after second high-pass filtering; Performing time domain feature extraction on the data processed by the second high-pass filter to obtain a plurality of second time domain features; Performing time-frequency domain feature extraction on the data processed by the second high-pass filter to obtain a plurality of second time-frequency domain features; performing normalization processing on a plurality of the second time domain features and a plurality of the second time-frequency domain features respectively to obtain second normalized data; Based on the data after the second normalization process, a machine learning algorithm is trained to obtain a condition monitoring and evaluation model.

3. The embedded engine condition monitoring and evaluation method according to claim 2 is characterized in that: According to the data after the second normalization process, the machine learning algorithm is trained to obtain a condition monitoring and evaluation model, which specifically includes: According to the data after the second normalization process, a back propagation algorithm is used to train the BP neural network; Determine whether the termination condition is met and obtain a first determination result; If the first judgment result is no, then return to "using the back propagation algorithm to train the BP neural network"; If the first judgment result is yes, the optimal weight and threshold of the BP neural network are output to obtain a trained BP neural network model; the trained BP neural network model is the first state monitoring and evaluation model.

4. The embedded engine condition monitoring and evaluation method according to claim 2, characterized in that: According to the data after the second normalization process, the machine learning algorithm is trained to obtain a condition monitoring and evaluation model, which specifically includes: According to the data after the second normalization process, the SVM network is trained using the SMO algorithm; Determine whether the termination condition is met, and obtain a second determination result; If the second judgment result is no, then return to "using the SMO algorithm to train the SVM network"; If the second judgment result is no, the optimal parameters of the SVM network are output to obtain a trained SVM model; the trained SVM model is the second state monitoring and evaluation model.

5. The embedded engine condition monitoring and evaluation method according to claim 1, characterized in that: Performing time-frequency domain feature extraction on the data processed by the first high-pass filter to obtain a plurality of first time-frequency domain features, specifically including: Performing three-layer wavelet packet decomposition on the data processed by the first high-pass filter to obtain a plurality of first wavelet features; The first wavelet feature is used as the first time-frequency domain feature.

6. The embedded engine condition monitoring and evaluation method according to claim 1, characterized in that: Inputting the first normalized data into a state monitoring and evaluation model to obtain the fault state type of the engine to be detected specifically includes: The data after the first normalization processing is input into the state monitoring and evaluation model, and the fault state type of the engine to be detected is output using an anti-normalization algorithm.

7. An embedded engine condition monitoring and evaluation system, characterized in that: The embedded engine condition monitoring and evaluation system comprises: A data acquisition module is used to acquire the engine data to be detected; the engine data to be detected is vibration data, instantaneous speed data and fuel pressure data in various states; A high-pass filter processing module, used for performing high-pass filter processing on the engine data to be detected to obtain data after the first high-pass filter processing; A time domain feature extraction module, used to extract time domain features from the data processed by the first high-pass filter to obtain a plurality of first time domain features; A time-frequency domain feature extraction module, used to extract time-frequency domain features from the data processed by the first high-pass filter to obtain a plurality of first time-frequency domain features; A normalization processing module, used to perform normalization processing on a plurality of the first time domain features and a plurality of the first time-frequency domain features respectively, to obtain first normalized data; A fault state type determination module is used to input the first normalized data into a state monitoring and evaluation model to obtain the fault state type of the engine to be detected; the state monitoring and evaluation model is a model built based on a machine learning algorithm; the fault state types of the engine to be detected include: normal operating conditions, abnormal intake valve clearance, advanced fuel supply advance angle and delayed fuel supply advance angle.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the embedded engine condition monitoring and evaluation method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the embedded engine condition monitoring and evaluation method described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the embedded engine condition monitoring and evaluation method according to any one of claims 1 to 6 is implemented.

Images