Marine engine turbocharger fault prediction method based on digital twinning

Abstract

The invention provides a marine engine turbocharger fault prediction method based on digital twinning. The marine engine turbocharger fault prediction method comprises the steps that operation parameters and vibration signals of a marine turbocharger are acquired; performing signal processing on the vibration signal; establishing a three-dimensional solid model of the supercharger; establishing a fluid excitation reconstruction model and a vibration response model according to the three-dimensional entity model; performing all-condition fault simulation based on the vibration response model to obtain fault data; constructing a full-condition fault data set and training a fault prediction model, and screening sensitive vibration characteristic parameter categories based on various vibration characteristic parameters; according to a data transmission relationship among the fluid excitation reconstruction model, the vibration response model and the fault prediction model, constructing a corresponding data twinborn model; and inputting the operation parameters and the sensitive characteristic parameters corresponding to the real-time vibration signals into a digital twinborn model, and obtaining an output fault type and a fault degree. According to the method, the fault type and degree of the supercharger are accurately predicted, and the reliability and safety of equipment are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault prediction of marine engine turbochargers, and particularly relates to a method for fault prediction of marine engine turbochargers based on digital twin. Background Art

[0002] With the development of Internet of Things, big data and artificial intelligence technologies, the intelligentization of ship shipping has gradually become a new trend. As an indispensable key component in the ship power system, the marine turbocharger significantly improves the power and fuel economy of the ship engine. The performance of the marine turbocharger directly affects the safety and economy of ship navigation, and it is one of the key maintenance equipment of the ship. According to statistics, the failure rate of the turbocharging system ranks first among the failures of key components of the ship engine.

[0003] There are various types of failures that occur in turbochargers. Among them, five types of failures such as water leakage, oil leakage, surge, overheating, abnormal noise and abnormal boost pressure account for more than 90% of the proportion. The occurrence of these failures is often closely related to the imbalance problem of the rotor system and the bearing wear problem. At present, the management of turbochargers on ships mainly judges whether the turbocharging system fails and what kind of failure occurs by monitoring whether the working parameters such as temperature and pressure are abnormal or the time-frequency characteristics of vibration signals, and then conducts targeted repairs and maintenance. However, this management method often only stays at the surface diagnosis and emergency treatment of the fault phenomenon, and rarely pays attention to the deep reasons for the occurrence of these failures, that is, the operating states of the rotor system and the bearings, and fails to identify and solve potential systematic problems fundamentally. It not only affects the stable operation of the ship power system, but also increases the maintenance cost and ship downtime, and has many negative impacts on the normal navigation and economic benefits of the ship. Conducting condition monitoring and fault prediction on two typical failures of turbocharger rotor imbalance and bearing wear can help identify potential problems in advance, optimize the maintenance cycle, reduce the impact of sudden failures on ship operation, and can also provide valuable basis for the design improvement of the turbocharger and the optimization of the rotor system, and improve the overall reliability and service life of the turbocharger.

[0004] Most traditional fault prediction methods are based on data-driven. This method relies on a large amount of fault data and requires multiple fault simulation tests on the machine to obtain a fault data set. However, for a marine turbocharger, which is a high-speed variable-condition operating machine and works in a harsh environment of high temperature, high humidity and strong vibration for a long time, there are many problems: (1) The structure of the marine turbocharger is complex and expensive. It usually operates under normal conditions, and fault data is scarce. Moreover, the cost of carrying out fault simulation tests is relatively high. The uncertainty and potential risks during the simulation experiment may cause equipment damage, further increasing the cost.

[0005] (2) Typical fault modes of marine turbochargers, such as rotor imbalance and bearing wear, are difficult to accurately simulate on actual machines, and the criterion standards for fault levels are not unified, making it difficult to obtain sufficient tagged fault samples. Especially for the simulation of severe faults, due to the harsh working environment of the turbocharger, permanent irreversible damage consequences such as engine burnout are likely to occur during the experiment, and even major accidents such as casualties of experimental personnel may be caused.

[0006] (3) The temperature of the marine turbocharger housing is as high as 200 °C, far exceeding the conventional tolerance range of vibration acceleration sensors, and the arrangement positions of the sensors are limited, making it difficult to achieve long-term reliable online monitoring of key parts of the turbocharger. Even if expensive high-temperature sensors are selected, continuous operation in a high-temperature environment for a long time will cause large temperature drifts in their performance and a significant decline in reliability.

[0007] (4) Due to the compact layout of the ship power system, the vibration generated by the engine during operation is transmitted to the measuring points of the sensors arranged on the marine turbocharger through mechanical connections, superimposed on the turbocharger vibration signal with other interference signals such as noise, affecting the accuracy of turbocharger fault diagnosis and condition prediction results. Summary of the Invention

[0008] The purpose of the present invention is to provide a fault prediction method for marine engine turbochargers based on digital twin to achieve efficient and accurate fault prediction of marine engine turbochargers.

[0009] To solve the above technical problems, the present invention provides a fault prediction method for marine engine turbochargers based on digital twin, including: S1. Obtain the operating parameters and vibration signals of the marine turbocharger; S2. Perform signal processing on the vibration signals; S3. Establish a three-dimensional solid model of the turbocharger; S4. Establish a fluid excitation reconstruction model and a vibration response model of the turbocharger according to the three-dimensional solid model of the turbocharger; S5. Conduct fault simulation based on the vibration response model to obtain a full-condition fault data set; the full-condition fault data set includes the operating parameters of the turbocharger under different fault types and fault levels, as well as the corresponding various vibration characteristic parameters; S6. Construct and use the full-condition fault data set to train a fault prediction model, and screen the categories of sensitive vibration characteristic parameters based on various vibration characteristic parameters; the input of the fault prediction model is the operating parameters and vibration characteristic parameters, and the output of the fault prediction model is the fault type and fault level; S7. Construct a digital twin model according to the data transfer relationship among the operating parameters, vibration signals, fluid excitation reconstruction model, vibration response model, and fault prediction model; S8. Input the operating parameters and sensitive vibration characteristic parameters into the digital twin model to obtain the fault type and fault degree output by the digital twin model.

[0010] According to the above solution, the operating parameters include the supercharger rotor speed, the inlet and outlet temperatures, pressures, mass flows, and ambient temperature of the compressor end and the turbine end; The vibration signals include the transverse vibration displacement signal of the rotor and the overall vibration acceleration signal of the machine; the transverse vibration displacement signal of the rotor is obtained by an eddy current displacement sensor, and the eddy current displacement sensor is arranged on both sides of the vertical center line of the upper half of the rotor bearing at 45°, and is radially installed in the same transverse plane perpendicular to the rotor axis, and the different eddy current displacement sensors are 90° apart from each other; the overall vibration acceleration signal of the machine is obtained by an acceleration sensor, and the acceleration sensor is arranged on the machine base.

[0011] According to the above solution, the step S2 includes: S201. Perform empirical mode decomposition on the vibration signal, remove the engine vibration signal components with lower frequencies in the vibration signal, and retain the supercharger vibration signal components with higher frequencies; S202. Remove the abnormal data in the vibration signal based on the 3σ criterion; S203. Use a low-pass filter to filter the frequencies in the vibration signal that exceed the measurement range of the sensor.

[0012] According to the above solution, the step S3 includes: S301. Obtain the structural characteristic parameters and material property parameters of different components of the supercharger; S302. Establish a three-dimensional solid model of the supercharger based on the material property parameters; Among them, the components of the supercharger include a housing and a rotor; the housing includes a turbine housing, a compressor housing, a bearing housing, a diffuser, a nozzle ring, and a machine base; the rotor is located inside the supercharger housing and includes a rotor shaft, a turbine, and an impeller; The structural characteristic parameters include geometric dimensions, structural characteristics, contact types, friction characteristics, and clearance dimensions; the material property parameters include density, Poisson's ratio, elastic modulus, stiffness, and damping.

[0013] According to the above solution, the process of establishing the fluid excitation reconstruction model of the supercharger includes: S4011. Perform tetrahedral finite element mesh division on the three-dimensional solid model to obtain the solid domain and fluid domain of the supercharger; S4012. Establish a fluid excitation model of the fluid domain in the computational fluid dynamics software and set the boundary conditions of the fluid excitation model; S4013. Use the optimal Latin hypercube experimental design to select the operating parameters of the supercharger under different working conditions; S4014. Input the selected operating parameters into the fluid excitation model to obtain the output time-domain response value of the fluid excitation model; S4015. Convert the output time-domain response value into a frequency-domain response value, and extract specific frequency-domain features of the frequency-domain response value; S4016. Construct a data set with the selected operating parameters and the specific frequency-domain features of the frequency-domain response value, and divide the data set into a training set and a test set; S4017. Based on the training set and the test set, use a BP neural network to construct a surrogate model, and verify the accuracy of the surrogate model to ensure that the accuracy of the surrogate model meets the requirements; S4018. Reconstruct the output of the surrogate model using the Fourier series method, and use the reconstruction result as the output of the fluid excitation reconstruction model.

[0014] According to the above scheme, the process of establishing the vibration response model of the supercharger includes: S4021. Establish a vibration response model based on the set boundary conditions; S4022. Input the output of the fluid excitation reconstruction model under typical working conditions into the vibration response model to obtain the vibration response output by the vibration response model; S4023. Modify the vibration response model by comparing the vibration signal and the vibration response under typical working conditions; the vibration signal under typical working conditions is obtained through experimental acquisition.

[0015] According to the above scheme, the fault working conditions corresponding to the full working condition fault data set include rotor imbalance fault and bearing wear fault; The simulation method for rotor imbalance fault includes: loading the unbalanced forces in two vertical directions onto the equivalent axis center weightless nodes of the rotor system; The simulation method for bearing wear fault includes: setting the clearance between the bearing and the rotor shaft; and setting the stiffness coefficients and damping coefficients of the bearing oil film in different directions.

[0016] According to the above scheme, the step S5 includes: S501. Perform modal reduction on the vibration response model; S502. Use the vibration response model after modal reduction to perform fault simulation to obtain a full working condition fault data set.

[0017] According to the above scheme, the categories of the vibration characteristic parameters include: peak-to-peak value, mean value, root mean square, standard deviation, skewness, kurtosis, waveform factor, peak factor, impulse factor, margin index, kurtosis index, skewness index, center frequency, mean square frequency, frequency standard deviation.

[0018] The present invention also provides a digital twin model for predicting the faults of a marine engine turbocharger, including: Information sensing and transmission module, used to obtain the operating parameters and vibration signals of the marine turbocharger; The fault prediction model establishment module is used to process the vibration signal and establish a three-dimensional solid model of the supercharger, a fluid excitation reconstruction model, and a vibration response model. Then, fault simulation is performed based on the vibration response model to obtain a full-condition fault data set. Then, a fault prediction model is constructed and trained using the full-condition fault data set. The fault prediction model outputs the fault type and fault degree of the marine supercharger based on the input operating parameters and vibration characteristic parameters.

[0019] Beneficial Effects The present invention solves the problems of high cost, high risk, scarce fault data and mixed vibration signals of marine turbocharger fault simulation by constructing a fault prediction method system based on digital twins. Specifically, by establishing a fluid excitation reconstruction model, a vibration response model and a fault prediction twin model, and combining multi-operating condition fault simulation to generate a full-operating condition data set, the cost and risk of physical experiments are significantly reduced; at the same time, signal processing technology is used to separate mixed signals in vibration signals and extract sensitive features, and the digital twin model is combined to realize virtual monitoring of key components, which improves the accuracy of fault prediction and engineering applicability, and provides data support for optimizing maintenance strategies.

[0020] Furthermore, the optimal Latin hypercube design (OLHD) is used to optimize sample generation, and a proxy model is constructed through the BP neural network to achieve rapid reconstruction of fluid excitation characteristics, which significantly reduces the time cost of traditional simulation calculations and solves the problem of low efficiency in fault data generation. At the same time, the Fourier series method is combined to ensure the accuracy of excitation reconstruction, providing an efficient data basis for multi-condition fault simulation.

[0021] The present invention reduces the order of the vibration response model through modal reduction, retaining only key low-order modes for calculation. It can improve the computing efficiency to within the real-time requirement while ensuring the simulation accuracy, solves the problem of time-consuming generation of full-condition fault data sets, and provides large-scale, multi-label fault data for the training and verification of fault prediction models. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is a flow chart of a method for predicting faults of a marine engine turbocharger based on digital twins according to Embodiment 1 of the present invention; Figure 2 This is the technical route of the fluid excitation reconstruction method of the marine engine turbocharger of the second embodiment of the present invention; Figure 3 A technical route for constructing a vibration response model of a marine engine turbocharger according to the second embodiment of the present invention; Figure 4This is the technical route of the fault prediction twin model for the marine engine turbocharger in Embodiment 2 of the present invention; Figure 5 This is the overall technical route of the fault prediction for the marine engine turbocharger based on digital twin in Embodiment 2 of the present invention. Specific implementation manners

[0023] To make the objectives, technical solutions and advantages of the embodiments of the present disclosure clearer, the technical solutions of the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present disclosure. Apparently, the described embodiments are some but not all of the embodiments of the present disclosure. All other embodiments obtained by those of ordinary skill in the art based on the described embodiments of the present disclosure without creative efforts shall fall within the scope of protection of the present disclosure.

[0024] Embodiment 1: Refer to Figure 1 , this embodiment discloses a method for fault prediction of a marine engine turbocharger based on digital twin, including: S1. Obtain the operating parameters and vibration signals of the marine supercharger; S2. Perform signal processing on the vibration signals; S3. Establish a three-dimensional solid model of the supercharger; S4. Establish a fluid excitation reconstruction model and a vibration response model of the supercharger according to the three-dimensional solid model of the supercharger; S5. Perform fault simulation based on the vibration response model to obtain a full-condition fault data set; the full-condition fault data set includes the operating parameters of the supercharger under different fault types and degrees of faults, as well as the corresponding various vibration characteristic parameters; S6. Construct and train a fault prediction model using the full-condition fault data set, and screen the categories of sensitive vibration characteristic parameters based on various vibration characteristic parameters; the input of the fault prediction model is the operating parameters and vibration characteristic parameters, and the output of the fault prediction model is the fault type and degree of fault; S7. Construct a digital twin model according to the data transfer relationship among the operating parameters, vibration signals, fluid excitation reconstruction model, vibration response model and fault prediction model; S8. Input the operating parameters and sensitive vibration characteristic parameters into the digital twin model to obtain the fault type and degree of fault output by the digital twin model.

[0025] Further, the operating parameters include the supercharger rotor speed, the inlet and outlet temperatures, pressures, mass flows of the compressor end and the turbine end, and the ambient temperature; The vibration signal includes the transverse displacement signal of the rotor and the vibration acceleration signal of the whole machine; the transverse displacement signal of the rotor is obtained by an eddy current displacement sensor, which is arranged at 45° on both sides of the vertical center line of the upper half of the rotor bearing, and is radially installed in the same transverse plane perpendicular to the rotor axis, with different eddy current displacement sensors being 90° apart; the vibration acceleration signal of the whole machine is obtained by an acceleration sensor, which is arranged on the machine base (because the temperature of the supercharger housing is as high as 200°C, far exceeding the tolerance range of the vibration acceleration sensor, the acceleration sensor is arranged on the machine base with a lower temperature).

[0026] Furthermore, the step S2 comprises: S201, performing empirical mode decomposition on the vibration signal, eliminating engine vibration signal components with lower frequencies in the vibration signal, and retaining supercharger vibration signal components with higher frequencies; Specifically, the supercharger bench signal is defined as X(t), the engine vibration signal is defined as Y(t), and the white noise is defined as Z(t). The supercharger vibration signal W(t)=X(t)+Y(t)+Z(t) is obtained by signal superposition. The number of moving parts in the engine is large, the vibration propagation is complex, and the vibration signal expression is complex, containing the status characteristic information of many devices; according to the working principle of the engine, the excitation sources of the vibration signal can be divided into valve seating impact, exhaust valve throttling impact, combustion excitation source and piston knocking excitation; when different excitation sources act on the engine, the vibration induced has its own specific frequency and amplitude characteristics, and these frequencies are usually positively correlated with the engine speed, especially with the integer multiple frequency of the speed; generally speaking, the engine speed is significantly lower than the supercharger speed, so the empirical mode decomposition (EMD) of W(t) can remove the components of the engine vibration signal with lower frequencies, while retaining the components of the supercharger vibration signal with higher frequencies; S202, eliminating abnormal data in the vibration signal based on the 3σ criterion (abnormal data includes non-numeric values, outliers, noise, etc.); S203: Filter out the frequencies in the vibration signal that are beyond the measurement range of the sensor using a low-pass filter.

[0027] Compared with X(t) after signal processing, the time domain and frequency domain diagrams of W(t) and X(t) are basically consistent, which proves the effectiveness of the above method.

[0028] Furthermore, the step S3 comprises: S301. Obtain structural characteristic parameters and material characteristic parameters of different components of the supercharger (through design drawings, manuals, nameplates, actual measurements, etc.); S302, establishing a three-dimensional solid model of the supercharger (in a non-fault mode) based on the material characteristic parameters; Among them, the components of the supercharger include a housing and a rotor; the housing includes a volute end housing, a compressor end housing, a bearing housing, a diffuser, a nozzle ring, and a frame; the rotor is located inside the supercharger housing and includes a rotor shaft, a turbine, and an impeller. The structural characteristic parameters include geometric dimensions, structural characteristics, contact types, friction characteristics, and clearance dimensions; the material characteristic parameters include density, Poisson's ratio, elastic modulus, stiffness, and damping.

[0029] Furthermore, the process of establishing the fluid excitation reconstruction model of the supercharger includes: S4011. Perform tetrahedral finite element mesh division on the three-dimensional solid model to obtain the solid domain and fluid domain of the supercharger. S4012. Establish a fluid excitation model for the fluid domain in computational fluid dynamics software (such as Fluent) and set the boundary conditions of the fluid excitation model. Specifically, the boundary conditions of the fluid excitation model include: using the Moving mesh sliding mesh method to solve the flow field data transmission problem between the stationary and moving flow domains, that is, setting the wall region where the fluid in each flow passage contacts the blade wall as MovingWall, and simplifying the remaining solid walls as impermeable, non-slip, and adiabatic boundary conditions; setting the inlet of the flow passage as a mass flow inlet, the outlet as a pressure outlet, and giving the actual parameter values according to the monitoring results of the supercharger operating parameters, and performing transient solution to output the fluid excitation. The convergence criterion for calculation is that the root mean square residual is less than 10 -4 ; S4013. Use the optimal Latin hypercube experimental design to select the operating parameters of the supercharger under different working conditions. In this embodiment, the optimal Latin hypercube experimental design (OLHD) adopted makes full use of the characteristics of various operating parameters to construct the experimental scheme, and reduces the random error through reasonable washing steps to obtain more accurate results. S4014. Input the selected operating parameters into the fluid excitation model to obtain the output time-domain response value (i.e., the time-domain excitation force) of the fluid excitation model. S4015. Convert the output time-domain response value into a frequency-domain response value (i.e., the frequency-domain excitation force), and extract the specific frequency-domain characteristics (including frequency and amplitude) of the frequency-domain response value. S4016. Construct a data set with the selected operating parameters and the specific frequency-domain characteristics of the frequency-domain response value, and divide the data set into a training set and a test set. S4017. According to the training set and the test set, use a BP neural network to construct a surrogate model and verify the accuracy of the surrogate model to ensure that the accuracy of the surrogate model meets the requirements. S4018. Reconstruct the output of the surrogate model using the Fourier series method, and use the reconstruction result as the output of the fluid excitation reconstruction model. In this embodiment, the rotor system rotates around the x-axis, and the fluid excitation mainly includes the turbine end casing forces F y1 , F z1 , the turbine forces F y2 , F z2 , the compressor end casing forces F y3 , F z3 , the impeller forces F y4 , F z4 ; Based on the above fluid excitation model, the time-domain diagram of the fluid excitation with the rotation of the rotor is output, and each time-domain excitation is converted into a frequency-domain excitation through Fourier transform : :

[0030] It is found through analysis that the excitation takes one rotation of a blade of the impeller / turbine as a cycle, and although the frequency components are complex, the fundamental frequency , the first harmonic frequency , the second harmonic frequency and the third harmonic frequency occupy the main components, and the amplitude is the largest at the first harmonic frequency; therefore, the phase and amplitude of the main frequency components in the frequency spectrum can be solved through inverse Fourier transform, and the Fourier series expression is used to reconstruct the fluid excitation :

[0031]

[0032] .

[0033] Furthermore, the process of establishing the vibration response model of the supercharger includes: S4021. Establish the vibration response model based on the set boundary conditions; In this embodiment, the vibration response model is established using the Power Unit module of AVL Excite software; the boundary conditions of the vibration response model include the connection, constraint, and friction between the components of the turbocharger; S4022. Input the output of the fluid excitation reconstruction model under typical working conditions into the vibration response model to obtain the vibration response output by the vibration response model; S4023. Modify the vibration response model by comparing the vibration signal and the vibration response under typical working conditions; the vibration signal under typical working conditions is obtained through experimental acquisition; In this embodiment, it should be ensured that the error between the vibration response output by the vibration response model and the vibration signal obtained through experiments does not exceed 5%.

[0034] Furthermore, the fault conditions corresponding to the full-condition fault dataset include rotor imbalance fault and bearing wear fault; The simulation method for rotor imbalance fault includes: loading the unbalanced forces in two perpendicular directions onto the equivalent axis weightless nodes of the rotor system; The rotor of the turbocharger rotates at a high speed. The asymmetric mass distribution under the high-speed rotation of the rotor generates centrifugal force, resulting in severe vibration of the supercharger and rotor imbalance. Based on this, the response effect of the imbalance is equivalent to applying two mutually perpendicular harmonic forces in the Y and Z directions of the rotor axis. In this embodiment, the 、 method of loading the unbalanced forces in two perpendicular directions onto the equivalent axis weightless nodes of the rotor system is adopted to simulate the rotor imbalance fault; define the rotor unbalanced force as F m ,F m The projections in the two directions of the rotor system except the rotation axis x are respectively:

[0035] In the above formula, F represents the unbalanced force, unit N; m represents the unbalanced mass, unit g; e represents the eccentricity between the unbalanced mass and the rotation center, unit mm; represents the rotational speed of the rotor system, unit rad / s; t represents a certain moment of the rotor system rotation, unit s; The simulation method for bearing wear fault includes: setting the clearance between the bearing and the rotor shaft; and setting the stiffness coefficients and damping coefficients of the bearing oil film in different directions; For the bearing wear fault, in addition to directly changing the clearance between the bearing and the rotor shaft, it mainly has a greater impact on the lubrication condition of the bearing; the dynamic characteristic coefficients of the bearing oil film are closely related to parameters such as the inlet oil temperature, viscosity, structural size, and rotational speed; when the journal is slightly disturbed at the static equilibrium position, the oil film force can be approximately regarded as a function of the small displacement and velocity of the journal, and retaining the first order gives:

[0036] In the formula, F x 、F y represent the components of the oil film force in the X and Y directions, F x0 、F y0 represent the components of the oil film force at the static equilibrium position in the X and Y directions, unit N; 、 、 、 respectively represent the oil film stiffness coefficients of the coupling action in the x direction, y direction, and two mutually perpendicular directions of x and y, unit N / mm; 、 、 , represents the oil film damping coefficient in the corresponding direction, with the unit of N·s / mm; Based on the existing calculation method of bearing oil film characteristic parameters, look up the table according to the Murphy number S, and calculate the dynamic characteristic coefficients of the bearing oil film by interpolation method; then according to the calculated oil film characteristic parameters , , , , , , , , establish a spring connection between the bearing and the rotor shaft, and assign the stiffness coefficient and damping coefficient in the corresponding direction to simulate the oil film characteristics under different bearing wear faults:

[0037] In the formula, represents the lubricating oil viscosity, with the unit of N·s / mm2; W represents the stable static load on the journal of the rotor shaft, with the unit of N; D = 2R represents the journal diameter, with the unit of mm; N = Ω / 2π represents the journal speed, with the unit of rad / s; L represents the length of the rotor shaft, with the unit of mm; C represents the radial clearance of the bearing, with the unit of mm.

[0038] Through experimental measurement, the unbalanced mass of the rotor under normal operating conditions of the supercharger is 3 g·mm, and the bearing clearance is 0.03 mm. According to the above methods for simulating rotor unbalance faults and bearing wear faults, use the supercharger vibration response model to simulate and analyze the vibration of the supercharger under different speeds and different fault degrees. It is found that with the increase of speed and the deepening of rotor unbalance faults and bearing wear faults, the area of the rotor center orbit, the effective value, peak-to-peak value, and spectral peak value of the vibration acceleration of the engine feet will all increase accordingly, and the increasing amplitude is continuously rising with the increase of speed. In addition, although the frequency components of the two types of typical faults are complex, the fundamental frequency occupies the main component. Among them, the vibration characteristics of the rotor dynamic unbalance amount change gently between 3 - 12 g·mm, and increase significantly when it exceeds 12 g·mm; the vibration characteristics of the bearing clearance change gently between 0.03 - 0.08 mm, and increase significantly when it exceeds 0.08 mm. This discovery provides a quantitative reference basis for the division of fault degrees.

[0039] Furthermore, the step S5 includes: S501. Perform modal reduction on the vibration response model; Although the simulation results of the supercharger vibration response model are relatively comprehensive and accurate, the simulation calculation time is too long. Considering the real-time requirement of the digital twin model, a modal reduction method is adopted to achieve the purpose of model order reduction and rapid response output. Usually, only a few low-order modes of each component of the mechanical system account for the main components in the vibration response. Therefore, low-order modes can be selected for solution on the premise of meeting the accuracy requirements, which can reduce the calculation cost and significantly improve the solution efficiency at the same time. This process is called modal reduction.

[0040] The characteristic equation of a system with n degrees of freedom is:

[0041] In the formula, represents the stiffness matrix; represents the modal matrix; represents the mass matrix; represents the eigenvalue matrix, and it is a diagonal matrix.

[0042] Take the first m-order modes to perform modal reduction on the model. Substitute the first m-order modal parameters into the system characteristic equation to obtain:

[0043] Use the finite element analysis software Ansys as a solver to perform modal reduction on the supercharger substructure body and rotor system respectively. Compare the modal analysis frequencies of the structure before and after modal reduction to verify the accuracy. If the error of each order modal frequency does not exceed 5%, the vibration response model after modal reduction can be directly used to simulate the full-condition fault data to obtain various types of fault data.

[0044] S502. Use the vibration response model after modal reduction to perform fault simulation to obtain a full-condition fault data set.

[0045] Furthermore, the categories of the vibration characteristic parameters include: peak-to-peak value, mean value, root mean square, standard deviation, skewness, kurtosis, waveform factor, peak factor, impulse factor, margin index, kurtosis index, skewness index, center frequency, mean square frequency, frequency standard deviation.

[0046] Furthermore, the construction and training methods of the fault prediction model include: S601. Based on the deep neural network, construct a fault prediction model, take the operating parameters and vibration characteristic parameters as the inputs of the fault prediction model, and take the fault type and fault degree as the outputs of the fault prediction model; S602. Randomly divide the full-condition fault data set into a training set and a test set, and train the fault prediction model until the fault prediction model meets the expected requirements; Specifically, during training, the Adam optimization algorithm is used to adaptively adjust the learning rate to optimize the weights and biases of the deep neural network, and the backpropagation algorithm is used to minimize the loss function. A regularization term is added to the loss function, and the L1 regularization method is used for sensitive feature selection while preventing overfitting to ensure the generalization ability of the model. During training, when indicators such as the loss values, mean squared error (MSE), mean absolute error (MAE), and coefficient of determination (R²) of the training set and test set reach the expected range, the training ends. Analyze the L1 regularization results and screen out non-zero weight features, specifically including peak-to-peak value, root mean square, standard deviation, and mean square frequency. These non-zero weight features have a greater influence on the prediction results, that is, the sensitive features required.

[0047] This embodiment also provides a digital twin model for predicting the faults of marine engine turbochargers, including: An information perception and transmission module for acquiring the operating parameters and vibration signals of the marine supercharger; A fault prediction model establishment module for performing signal processing on the vibration signals, establishing a three-dimensional entity model, a fluid excitation reconstruction model, and a vibration response model of the supercharger, then performing fault simulation based on the vibration response model to obtain a full-condition fault data set, and then constructing and training a fault prediction model using the full-condition fault data set; where the fault prediction model outputs the fault type and fault degree of the marine supercharger according to the input operating parameters and vibration characteristic parameters.

[0048] Embodiment 2: This embodiment discloses a method for predicting the faults of marine turbochargers based on digital twin, mainly including: a fluid excitation reconstruction method for marine engine turbochargers, a method for constructing a whole-machine vibration response model, a method for simulating typical faults, a mapping method for fault characteristics, operating parameters, and fault modes and their degrees, a method for constructing a fault prediction twin model, etc.

[0049] As Figure 2 shown, the fluid excitation reconstruction method for marine engine turbochargers is based on the fast reconstruction technology of Fourier series and combines with a surrogate model to realize the prediction and reconstruction of fluid excitation characteristics. This method relies on the fluid excitation model of the supercharger. First, determine the range of each operating parameter of the supercharger, use the optimal Latin hypercube design of experiments (OLHD) to optimize the experimental scheme, simulate and solve the excitation characteristics under different operating parameters, make full use of the characteristics of each operating parameter to generate an efficient data set, reduce the calculation cost and random error; then, design a surrogate model based on the BP neural network, use the data set to train the model and verify its accuracy to ensure that the model has good prediction ability; finally, through the trained surrogate model, predict the excitation characteristics according to the input operating parameters, and combine with the Fourier series method to realize the fast reconstruction of fluid excitation.

[0050] AsFigure 3 As shown in Figure 3 , for the vibration response simulation requirements of marine engine turbochargers, after constructing a high-precision finite element model of the turbocharger, the boundary conditions are set using AVL Excite software, and the results of the fluid excitation reconstruction model are used as inputs to establish a vibration response model of the entire turbocharger. By conducting vibration response calculation and analysis, the vibration responses of the key components of the turbocharger can be obtained.

[0051] The simulation of typical faults of marine engine turbochargers is based on the fault mechanism. By applying two mutually perpendicular harmonic forces perpendicular to the rotor rotation direction at the equivalent axis center weightless node of the rotor system, the simulation of rotor imbalance faults is achieved; by changing the clearance between the bearing and the rotor shaft and calculating the stiffness coefficient and damping coefficient of the bearing oil film under different working conditions, the simulation of bearing wear faults is achieved.

[0052] The mapping method of fault characteristics, operating parameters, fault modes and their degrees of marine engine turbochargers is based on a deep neural network model. The Adam optimization algorithm is used to adjust the weights, and L1 regularization is combined for sensitive feature selection and to prevent overfitting. The model is optimized through the training set and the test set to ensure that the relevant performance indicators meet the expected requirements.

[0053] As Figure 4 shown in Figure 4 , the construction of the typical fault prediction twin model of marine engine turbochargers is based on the mapping relationship between the fault characteristics, operating parameters, fault modes and their degrees of the turbocharger. The operating parameters and vibration signals of the turbocharger are obtained through the information perception and transmission module, realizing the effective connection between the physical space and the digital space. The model inputs are the sensitive features of the operating parameters and vibration signals, and the outputs are the prediction results of the fault type and degree.

[0054] As Figure 5As shown in the figure, the specific technical scheme is as follows: obtain the operating parameters and vibration data of the marine supercharger; determine the three-dimensional solid model according to the structural characteristics and material parameters of the supercharger; mesh the three-dimensional solid model, extract the fluid domain and the solid domain, and create the supercharger fluid excitation model and the whole machine vibration response model; combine the operating parameters under different working conditions and the excitation characteristics of the supercharger fluid excitation model output under the corresponding working conditions, and use the optimal Latin hypercube experimental design method and BP neural network algorithm to design the fluid excitation reconstruction model; reconstruct the fluid excitation of the typical working condition, input it into the supercharger vibration response model for simulation calculation, and compare the measured supercharger vibration data. According to the vibration simulation data, the model parameters are continuously corrected to ensure that the vibration response model can accurately reflect the actual vibration characteristics of the supercharger; vibration simulation calculations of typical fault modes and fault degrees of the supercharger under multiple working conditions are carried out, and the mapping relationship between the supercharger operating parameters, fault vibration characteristics, and fault modes and degrees is established based on the deep neural network model, and the sensitive vibration characteristics are selected in combination with L1 regularization; by integrating the above models and analysis results, the information perception and transmission module is designed, and a digital twin model for supercharger fault prediction is established. The collected supercharger operating parameters and vibration signals are input to predict the supercharger fault type and fault degree in real time.

[0055] The method described in this embodiment includes the following steps: 1) Obtaining marine turbocharger operating parameters and turbocharger vibration signals The developed marine turbocharger monitoring and diagnosis system is used to measure the turbocharger operating parameters in different states on the turbocharger test bench, and tests are carried out at multiple operating points. The operating parameters include the turbocharger rotor speed, compressor and turbine inlet and outlet temperatures, pressure, mass flow rate, and ambient temperature, which are used as boundary conditions for the turbocharger fluid excitation model and vibration response model, and as inputs for the fluid excitation reconstruction model and the turbocharger fault prediction twin model.

[0056] The supercharger vibration signal mainly includes the transverse displacement signal of the rotor and the vibration acceleration signal of the whole machine. The eddy current displacement sensor is arranged at 45° on both sides of the vertical center line of the upper half of the rotor bearing, and is installed radially in the same transverse plane perpendicular to the rotor axis. The sensors are 90° to each other to collect the transverse displacement data of the supercharger rotor. The temperature of the supercharger housing is as high as 200°C, which is far beyond the tolerance range of the vibration acceleration sensor. Therefore, the acceleration sensor is arranged on the base with a lower temperature to collect the supercharger vibration data. The engine foot vibration signal and the rotor transverse vibration signal are mainly used to verify the accuracy of the supercharger vibration response model and to provide sensitive features as input parameters for the digital twin model.

[0057] Considering that the actual supercharger vibration signal contains various interference signals such as engine vibration signals and high-frequency noise compared to the signals measured on the test bench. To improve the accuracy and development prospects of the present invention in engineering practical applications, a set of interference signal processing methods is proposed by simulating the supercharger vibration signal W(t) by loading interference signals such as engine vibration and noise. Define the supercharger bench signal as X(t), the engine vibration signal as Y(t), and the white noise as Z(t), and the signal superposition gives W(t)=X(t)+Y(t)+Z(t). The engine has a large number of moving parts, complex vibration propagation, and complex vibration signal manifestations, including the state characteristic information of many devices. According to the engine working principle, the excitation sources of the vibration signal can be divided into valve seat impact, exhaust valve throttling impact, combustion excitation source, and piston knock excitation. When different excitation sources act on the engine, the induced vibrations have their own specific frequency and amplitude characteristics, and these frequencies are usually positively correlated with the engine speed, especially with integer multiples of the speed. Since the engine speed is usually significantly lower than the supercharger speed, empirical mode decomposition (EMD) is performed on W(t) to remove the lower-frequency engine vibration signal components while retaining the higher-frequency supercharger vibration signal components. For other abnormal components such as non-numerical values, outliers, and noise in W(t), abnormal data is removed based on the 3σ criterion, and a low-pass filter is used to filter out the frequencies outside the sensor measurement range in the signal. Comparing the processed W(t) and X(t) signals, the time-domain diagram and frequency-domain diagram are basically the same, proving the effectiveness of the method.

[0058] 2) Establish a three-dimensional solid model of the supercharger The supercharger is mainly divided into two parts: the housing and the rotor. The housing includes components such as the volute end housing, compressor end housing, bearing housing, diffuser, nozzle ring, and frame. The rotor is located inside the supercharger housing and includes the rotor shaft, turbine, and impeller. By referring to design drawings, manuals, nameplates, and actual measurements, the geometric dimensions, structural characteristics, contact types, friction characteristics, clearance dimensions, and material property parameters such as density, Poisson's ratio, elastic modulus, stiffness, and damping of different components of the supercharger are obtained, and a three-dimensional solid model of the supercharger in the fault-free mode is established.

[0059] 3) Establish a fluid excitation model, a fluid excitation reconstruction model, and a vibration response model of the supercharger Marine turbochargers use the energy of the exhaust gas discharged from the engine to drive the turbine to rotate, driving the compressor impeller coaxial with the turbine to rotate at high speed. At the same time, the compressor presses air into the engine cylinder, increasing the air volume of the engine and making the fuel utilization more sufficient. Tetrahedral finite element meshing is performed on the three-dimensional solid model of the turbocharger to obtain the solid domain, and the fluid domain is extracted according to the working principle of the turbocharger. To ensure the accuracy of the subsequent model establishment, it is necessary to conduct quality assessment and qualification rate on parameters such as warpage, aspect ratio, twist, chord difference, and Jacobian of the finite element meshes of the solid domain and fluid domain of the turbocharger to ensure that they meet the engineering requirements.

[0060] The fluid domain is divided into a compressor flow passage and a turbine flow passage according to different gas media. The compressor flow passage consists of an air inlet, a compressor impeller, a diffuser section, and an outlet, and the flow passage medium is fresh air. The turbine flow passage consists of an air inlet, an axial flow passage, a turbine, and an outlet, etc., and the flow passage medium is the exhaust gas discharged from the engine. In the computational fluid dynamics software Fluent, a fluid excitation model of the compressor flow passage and the turbine flow passage is established, and the Moving mesh sliding mesh method is used to solve the problem of flow field data transmission between the static and dynamic flow domains, that is, the wall region where the fluid in each flow passage contacts the blade wall is set as the Moving Wall, and the remaining solid walls are simplified to impermeable, non-slip, and adiabatic boundary conditions. The inlet of the flow passage is set as a mass flow inlet, and the outlet is set as a pressure outlet, and the actual parameter values are given according to the monitoring results of the turbocharger operation parameters, and transient solution is performed to output the fluid excitation, and the convergence criterion for calculation is that the root mean square residual is less than 10 -4 . The rotor system rotates around the x-axis, and the fluid excitation mainly includes the turbine end shell forces F y1 、F z1 , the turbine forces F y2 、F z2 , the compressor end shell forces F y3 、F z3 , the impeller forces F y4 、F z4 .

[0061] The above fluid excitation model outputs the time-domain diagram of the fluid excitation with the rotation of the rotor. Through Fourier transform, each time-domain excitation is converted into a frequency-domain excitation , and the excitation characteristics are analyzed:

[0062] It is found that each excitation takes one rotation of a blade of the impeller / turbine as a cycle, and although the frequency components are complex, the fundamental frequency , the first harmonic , the second harmonic and the third harmonic It occupies the main component, and the amplitude is the largest at the first blade passing frequency. Therefore, the phase of the main frequency components in the spectrum can be solved by inverse Fourier transform. and amplitude , and use the Fourier series expression to reconstruct the fluid excitation :

[0063]

[0064]

[0065] The core requirement of digital twin is to be able to provide near-real-time feedback, but the calculation time of using the fluid excitation model is too long. Based on the method of quickly reconstructing fluid excitation by Fourier series, a fluid excitation reconstruction model is designed using surrogate model technology. Considering the calculation cost problem, it is difficult to obtain a large number of characteristic samples using the fluid excitation model. The optimal Latin hypercube design (OLHD) method is used to solve this problem. OLHD is improved on the basis of the traditional Latin hypercube design LHD, makes full use of the characteristics of various operating parameters to construct the test scheme, and reduces the random error through reasonable washing steps to obtain more accurate results. Use the OLHD design scheme to set the boundary conditions of the fluid excitation model, that is, the operating parameters of the supercharger under different working conditions, and record the characteristics of the main frequency components of the output excitation to construct a data set. Divide the data set into a training set and a test set, construct a surrogate model based on the BP neural network algorithm, and verify its accuracy to ensure that the surrogate model has good prediction performance. Finally, use the trained surrogate model to output the excitation characteristics according to the input operating parameters, and combine the Fourier series method to realize the rapid reconstruction of the excitation characteristics.

[0066] Using the Power Unit module of AVL Excite software, combined with the engineering practice, define the connection, constraint, friction and other boundary conditions between the components of the turbocharger, and establish a vibration response model of the supercharger. Input the fluid excitation output under the typical working conditions of the fluid excitation reconstruction model into the vibration response model, and carry out the vibration response calculation and analysis of the supercharger, and the vibration responses of all key components of the supercharger can be obtained. Compare the experimental data and simulation data of the rotor center orbit and the vibration of the engine mounts of the supercharger, correct the parameter settings of the boundary conditions and verify the correctness of the model to ensure that the error does not exceed 5%.

[0067] 4) Obtain the full-condition fault data set Use the verified vibration response model to simulate the rotor imbalance fault and bearing wear fault, carry out the typical fault simulation of the variable-speed supercharger under multiple working conditions, obtain the fault data set, and provide data support for the fault prediction model.

[0068] The rotational speed of the turbocharger rotor is relatively high. The asymmetric mass distribution during the high-speed rotation of the rotor generates centrifugal force, resulting in severe vibration of the supercharger and rotor imbalance. The response effect of the imbalance is equivalent to applying two mutually perpendicular harmonic forces in the Y and Z directions of the rotor axis. By using the method of , loading the unbalanced forces in two perpendicular directions onto the weightless node at the equivalent axis center of the rotor system, the simulation of rotor imbalance faults is achieved. Define the rotor unbalanced force as F m , F m The projections in the two directions of the rotor system except for the rotating shaft x direction are respectively:

[0069] In the formula, F represents the unbalanced force, unit N; m represents the unbalanced mass, unit g; e represents the eccentricity between the unbalanced mass and the rotation center, unit mm; represents the rotational speed of the rotor system, unit rad / s; t represents a certain moment of the rotor system rotation, unit s.

[0070] For bearing wear faults, in addition to directly changing the clearance between the bearing and the rotor shaft, it mainly has a greater impact on the lubrication condition of the bearing. In order to better simulate the actual bearing fault situation, the oil film dynamic characteristics of the bearing are analyzed in depth. The dynamic characteristic coefficients of the bearing oil film are closely related to parameters such as the inlet oil temperature, viscosity, structural dimensions, and rotational speed. When the journal is slightly disturbed at the static equilibrium position, the oil film force can be approximately regarded as a function of the small displacement and velocity of the journal. Retaining the first order, we can get: (3) In the formula, Fx and Fy represent the components of the oil film force in the X and Y directions, F x0 , F y0 represent the components of the oil film force at the static equilibrium position in the X and Y directions, unit N; , , , respectively represent the oil film stiffness coefficients of the coupling effects in the x direction, y direction, and two mutually perpendicular directions of x and y, unit N / mm; , , , represent the oil film damping coefficients in the corresponding directions, unit N·s / mm.

[0071] Referring to the calculation method of bearing oil film characteristic parameters in Zhong Yie's "Rotor Dynamics", look up the table according to the Murphy number S, and calculate each dynamic characteristic coefficient of the bearing oil film by the interpolation method. According to the calculated oil film characteristic parameters , , , , , , and , a spring connection is established between the bearing and the rotor shaft, and the corresponding stiffness coefficient and damping coefficient are assigned to simulate the oil film characteristics under different bearing wear faults.

[0072]

[0073] In the formula, represents the lubricating oil viscosity, with the unit of N·s / mm2; W represents the stable static load on the journal of the rotor shaft, with the unit of N; D = 2R represents the journal diameter, with the unit of mm; N = Ω / 2π represents the journal speed, with the unit of rad / s; L represents the length of the rotor shaft, with the unit of mm; C represents the radial clearance of the bearing, with the unit of mm.

[0074] Through experimental measurement, the unbalanced mass of the rotor under normal operating conditions of the supercharger is 3 g·mm, and the bearing clearance is 0.03 mm. According to the above method for simulating rotor unbalance faults and bearing wear faults, the vibration response model of the supercharger is used to simulate and analyze the vibration of the supercharger under different speeds and different fault degrees. It is found that with the increase of speed and the deepening of rotor unbalance faults and bearing wear faults, the area of the rotor center orbit, the effective value, peak-to-peak value, and spectral peak value of the vibration acceleration of the engine mounts will all increase, and the increasing amplitude is continuously rising with the increase of speed. In addition, although the frequency components of the two types of typical faults are complex, the fundamental frequency occupies the main component. Among them, the vibration characteristics of the dynamic unbalance of the rotor change gently when the dynamic unbalance is between 3 - 12 g·mm, and rise significantly when it exceeds 12 g·mm; the vibration characteristics change gently when the bearing clearance is between 0.03 - 0.08 mm, and rise significantly when it exceeds 0.08 mm. This discovery provides a quantitative reference basis for the division of fault degrees.

[0075] Although the simulation results of the supercharger vibration response model are relatively comprehensive and the accuracy is high, the simulation calculation time is too long. Considering the real-time requirements of the digital twin model, the method of modal reduction is adopted to achieve the purpose of model order reduction and rapid response output. Usually, only a few low-order modes of each component of the mechanical system play a major role in the vibration response contribution. Therefore, low-order modes can be selected for solution on the premise of meeting the accuracy requirements, which can reduce the calculation cost and significantly improve the solution efficiency at the same time. This process is called modal reduction.

[0076] The characteristic equation of a system with n degrees of freedom is:

[0077] In the formula, represents the stiffness matrix; represents the modal matrix; represents the mass matrix; represents the eigenvalue matrix, and it is a diagonal matrix.

[0078] Take the first m modal orders to perform modal reduction on the model, and substitute the first m modal parameters into the system characteristic equation to obtain:

[0079] Use the finite element analysis software Ansys as the solver to perform modal reduction on the supercharger substructure body and the rotor system respectively. Compare the modal analysis frequencies of the structure before and after modal reduction to verify the accuracy. If the error of each modal frequency does not exceed 5%, the vibration response model after modal reduction can be directly used to simulate the full-condition fault data to obtain various fault data. Extract the time and frequency characteristics of the fault data to construct a fault data set, which mainly includes 15 common characteristic parameters such as peak-to-peak value, mean value, root mean square, standard deviation, skewness, kurtosis, waveform factor, peak factor, impulse factor, margin index, kurtosis index, skewness index, center frequency, mean square frequency, and frequency standard deviation.

[0080] 5) Establish a supercharger fault prediction digital twin model To construct the mapping relationship between fault characteristics, operating parameters, and fault severity, based on the deep neural network theory, the operating parameters and fault characteristics are used as inputs, and the fault type and fault severity are used as outputs. The Adam optimization algorithm is used to adaptively adjust the learning rate to optimize the weights and biases of the network, and the backpropagation algorithm is used to minimize the loss function. A regularization term is added to the loss function, and the L1 regularization method is used for sensitive feature selection while preventing overfitting to ensure the generalization ability of the model. Randomly divide the training set and the test set to train the neural network model until the loss value, mean square error (MSE), mean absolute error (MAE), coefficient of determination (R²) and other indicators of the training set and the test set reach the expected range. Analyze the L1 regularization results and screen out the non-zero weight features, specifically including peak-to-peak value, root mean square, standard deviation, and mean square frequency. These non-zero weight features have a greater influence on the prediction results, that is, the sensitive features required.

[0081] Based on the above model design and algorithm research, construct a supercharger fault prediction digital twin model. The model includes: an information perception and transmission module to obtain the operating parameters (speed, pressure, temperature, etc.) and vibration signals of the supercharger; a supercharger fault prediction model establishment module to determine the three-dimensional model of the supercharger, the fluid excitation model, the fluid excitation reconstruction model, the vibration response model after modal reduction, and the fault prediction model according to the operating parameters and vibration signals respectively. Establish a digital twin model according to the parameter transfer relationship between each model, and use the operating parameters and the sensitive features of the processed supercharger vibration signals extracted as the input of the digital twin model to output the corresponding fault type and severity prediction results.

[0082] In view of the problems such as high cost, high risk, great difficulty in diagnosis at variable speeds, difficulty in obtaining fault data samples, fuzzy division of fault degrees, limited positions for sensor arrangement where key components cannot be monitored, and mixed vibration signals of actual superchargers, the present invention combines various methods and technologies such as sensor measurement, finite element calculation, signal processing, neural network algorithms, and digital twin, and proposes a supercharger fault prediction method based on digital twin. The main effects are as follows: 1. The designed fluid excitation reconstruction model inputs the operating parameters of the supercharger, and the model automatically calculates the amplitude and phase of the main frequencies, and quickly reconstructs the fluid excitation through the Fourier series fitting method, reducing the calculation cost and improving the calculation efficiency while ensuring accuracy, accelerating the construction of the fault data set, and providing strong support for supercharger fault prediction.

[0083] 2. The modal reduction method is used to process the supercharger vibration response model, accelerating the vibration response calculation process, facilitating the simulation of the vibration responses of superchargers with different fault degrees under multiple working conditions, providing a large amount of labeled fault data, providing a rich data basis for the analysis of fault feature laws and the training of fault prediction models, helping to better divide different degrees of faults, and improving the accuracy and robustness of prediction.

[0084] 3. The supercharger vibration response model verified for accuracy can accurately reflect the vibration conditions of each part of the supercharger, realize real-time tracking of the vibration of each part of the supercharger, fully overcome the limitations of sensor use in actual monitoring, avoid data omission caused by sensor limitations, contribute to the comprehensive monitoring and management of the supercharger working state, and also provide data support for the optimal design and improvement of the supercharger.

[0085] 4. Through the interference signal processing technology, the interference of the supercharger signal mixing problem on the fault prediction result is avoided, which helps the health monitoring and fault warning of the supercharger, and ensures the accuracy and reliability of the twin model in a complex operating environment. 5. The establishment of the supercharger fault prediction model realizes the real-time prediction of two typical faults, namely rotor imbalance and bearing wear of the supercharger, provides a warning for the maintenance of the supercharger, effectively avoids the occurrence of common supercharger faults such as water leakage, oil leakage, surge, overheating, abnormal noise, and abnormal boost pressure, reduces the equipment downtime, and reduces the maintenance cost and potential safety risks, providing support for the operation and management strategy of the supercharger.

[0086] It should be noted that according to the needs of implementation, each step / component described in this application can be split into more steps / components, or two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0087] Those skilled in the art can easily understand that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for fault prediction of a marine engine turbocharger based on digital twin, characterized in that, Including: S1. Obtain the operating parameters and vibration signals of the marine supercharger; S2. Perform signal processing on the vibration signals; S3. Establish a three-dimensional solid model of the supercharger; S4. Establish a fluid excitation reconstruction model and a vibration response model of the supercharger based on the three-dimensional solid model of the supercharger; S5. Conduct fault simulation based on the vibration response model to obtain a full-condition fault data set; the full-condition fault data set includes the operating parameters of the supercharger under different fault types and degrees of faults, as well as the corresponding various vibration characteristic parameters; S6. Construct and use the full-condition fault data set to train a fault prediction model, and screen the categories of sensitive vibration characteristic parameters based on various vibration characteristic parameters; the input of the fault prediction model is the operating parameters and vibration characteristic parameters, and the output of the fault prediction model is the fault type and degree of fault; S7. Construct a digital twin model according to the data transfer relationship among the operating parameters, vibration signals, fluid excitation reconstruction model, vibration response model, and fault prediction model; S8. Input the operating parameters and sensitive vibration characteristic parameters into the digital twin model to obtain the fault type and degree of fault output by the digital twin model.

2. The method for predicting faults of a marine engine turbocharger based on digital twin according to claim 1, characterized in that, The operating parameters include the supercharger rotor speed, inlet and outlet temperatures, pressures, mass flows, and ambient temperature at the compressor end and turbine end; The vibration signals include the transverse vibration displacement signal of the rotor and the overall vibration acceleration signal of the machine; the transverse vibration displacement signal of the rotor is obtained by an eddy current displacement sensor, and the eddy current displacement sensor is arranged on both sides of the vertical center line of the upper half of the rotor bearing at 45°, and is radially installed in the same transverse plane perpendicular to the rotor axis, and the different eddy current displacement sensors are 90° apart from each other; the overall vibration acceleration signal of the machine is obtained by an acceleration sensor, and the acceleration sensor is arranged on the machine base.

3. The method for predicting faults of a marine engine turbocharger based on digital twin according to claim 1, characterized in that The step S2 includes: S201. Perform empirical mode decomposition on the vibration signals, remove the components of the engine vibration signals with lower frequencies in the vibration signals, and retain the components of the supercharger vibration signals with higher frequencies; S202. Remove the abnormal data in the vibration signals based on the 3σ criterion; S203. Use a low-pass filter to filter the frequencies in the vibration signals that exceed the measurement range of the sensor.

4. The method for predicting faults of a marine engine turbocharger based on digital twin according to claim 1, wherein, The step S3 includes: S301. Obtain the structural characteristic parameters and material property parameters of different components of the supercharger; S302. Establish a three-dimensional solid model of the supercharger based on the material property parameters; Among them, the components of the supercharger include a housing and a rotor; the housing includes a turbine end housing, a compressor end housing, a bearing housing, a diffuser, a nozzle ring, and a machine base; the rotor is located inside the supercharger housing and includes a rotor shaft, a turbine, and an impeller; The structural characteristic parameters include geometric dimensions, structural characteristics, contact types, friction characteristics, and clearance dimensions; the material property parameters include density, Poisson's ratio, elastic modulus, stiffness, and damping.

5. The method for predicting faults of a marine engine turbocharger based on digital twin according to claim 1, wherein The process of establishing the fluid excitation reconstruction model of the supercharger includes: S4011. Perform tetrahedral finite element mesh division on the three-dimensional solid model to obtain the solid domain and fluid domain of the supercharger; S4012. Establish a fluid excitation model of the fluid domain in the computational fluid dynamics software and set the boundary conditions of the fluid excitation model; S4013. Using the optimal Latin hypercube experimental design, select the operating parameters of the supercharger under different working conditions; S4014. Input the selected operating parameters into the fluid excitation model to obtain the output time-domain response value of the fluid excitation model; S4015. Convert the output time-domain response value into a frequency-domain response value and extract the specific frequency-domain characteristics of the frequency-domain response value; S4016. Construct a data set with the selected operating parameters and the specific frequency-domain characteristics of the frequency-domain response value, and divide the data set into a training set and a test set; S4017. According to the training set and the test set, use a BP neural network to construct a surrogate model and verify the accuracy of the surrogate model to ensure that the accuracy of the surrogate model meets the requirements; S4018. Reconstruct the output of the surrogate model using the Fourier series method, and use the reconstruction result as the output of the fluid excitation reconstruction model.

6. The method for predicting the faults of a marine engine turbocharger based on digital twin according to claim 1, wherein, The process of establishing the vibration response model of the supercharger includes: S4021. Establish a vibration response model based on the set boundary conditions; S4022. Input the output of the fluid excitation reconstruction model under typical working conditions into the vibration response model to obtain the vibration response output by the vibration response model; S4023. Modify the vibration response model by comparing the vibration signal and the vibration response under typical working conditions; the vibration signal under typical working conditions is obtained through experimental acquisition.

7. The method for fault prediction of a marine engine turbocharger based on digital twin according to claim 1, wherein, The fault working conditions corresponding to the full working condition fault data set include rotor imbalance fault and bearing wear fault; The simulation method for the rotor imbalance fault includes: loading the unbalanced forces in two vertical directions onto the equivalent axisless weightless nodes of the rotor system; The simulation method for the bearing wear fault includes: setting the clearance between the bearing and the rotor shaft; and setting the stiffness coefficients and damping coefficients of the bearing oil film in different directions.

8. The method for predicting the faults of a marine engine turbocharger based on digital twin according to claim 1, wherein The step S5 includes: S501. Perform modal reduction on the vibration response model; S502. Use the vibration response model after modal reduction to perform fault simulation to obtain a full working condition fault data set.

9. The method for predicting faults of a marine engine turbocharger based on digital twin according to claim 1, wherein The categories of the vibration characteristic parameters include: peak-to-peak value, mean value, root mean square, standard deviation, skewness, kurtosis, waveform factor, peak factor, impulse factor, margin index, kurtosis index, skewness index, center frequency, mean square frequency, frequency standard deviation.

10. A digital twin model for predicting faults of a marine engine turbocharger, characterized in that, including: An information perception and transmission module for obtaining the operating parameters and vibration signals of the marine supercharger; A fault prediction model establishment module for performing signal processing on the vibration signal, establishing a three-dimensional solid model, a fluid excitation reconstruction model, and a vibration response model of the supercharger, then performing fault simulation based on the vibration response model to obtain a full working condition fault data set, and then constructing and training a fault prediction model using the full working condition fault data set; wherein the fault prediction model outputs the fault type and fault degree of the marine supercharger according to the input operating parameters and vibration characteristic parameters.

Images