To improve the design efficiency of film cooling in turbine vane, a neural network constrained by Gaussian function was proposed to predict the cooling effectiveness distribution of single film row, and the distribution of multi film rows was predicted by combining the modified superposition principle. For a single film row on the flat-plate, three coefficients of the Gaussian function were predicted with the main flow turbulence degree, density ratio, blowing ratio, film inclined angle, length-width ratio and dimensionless flow direction distance taken as inputs, and then the dimensionless lateral distance was substituted into the predicted Gaussian function to calculate the cooling effectiveness. The prediction error of the Gauss function constrained neural network for the average cooling effectiveness of the test samples was only 5.70%, which was 67% lower than that of network without constraints. Based on the model of single film row and the correction coefficient of multi film rows superposition principle predicted according to the blowing ratio and dimensionless coordinates, the cooling effectiveness distribution of multi film rows was calculated. The prediction errors of the average cooling effectiveness were only 6.19% and 12.19% with the blowing ratio at 0.5 and 1, respectively. According to the results, the proposed method can predict the film cooling effectiveness distribution accurately.

The curvilinear coordinate system’s immersion boundary method (IBM) can be used to realize high-fidelity virtual-real mapping of the real aero-engine combustor in the design stage, and retain all the geometric structure information. Based on the IBM method, large eddy simulation (LES) combined with transported probability density functional turbulent combustion model (TPDF) was used to simulate the real structure of a double-swirler combustor, a single-head model of a throughflow combustor and 1/10 model of a slinger combustor, and test the effectiveness of key technologies of the digital twin system. Compared with the prediction and experiment results, the average errors of axial, radial and tangential velocities near the outlet of the swirler of the double-swirler combustor were 15.7%, 23.8% and 15.0%, respectively. The detailed turbulent combustion fields of the single-head model of a throughflow combustor and the slinger combustor were analyzed under the unsteady state, and the average relative errors of the outlet temperature distribution were 11.66% and 17.95%, respectively. Therefore, the combustor’s digital twin system obtained based on virtual real mapping has certain effectiveness, and the proposed method has potential engineering application prospects.

Aero-engine icing can affect flight safety. Studying the mechanism of aero-engine icing process can help improve the flight safety performance and accomplish airworthiness verification. The Eulerian method was used to calculate the motion of water droplet. The energy balance method was adopted to calculate the multiphase heat transfer and flow process of icing, and the one-dimensional heat conduction model was employed to correct the process of heat conduction with wall. The data exchange model based on phase interface heat conduction equilibrium equation was proposed. Based on this model, numerical verification of icing wind tunnel test data on the engine inlet part was accomplished. The maximum icing thickness error compared by simulation and test could be controlled at about 10%. The phase interface data exchange model can accurately simulate the water film flow and the temperature gradient near wall, and can also currently exchange the flow parameters between the user defined function (UDF) and the solver. The results showed that the model had good accuracy in calculating the maximum icing thickness.

Owing to the inaccuracies of the component characteristic maps, the traditional thermodynamic model for on-wing turbofans exhibited a significant modeling error. Moreover, when the model was iterated near the boundary points of the maps, it was prone to non-convergence, rendering it unreliable. To address these issues, a neural network modeling method for turbofans based on thermodynamic was put forward. This method improved modeling accuracy by fully considering the optimization of thermodynamic constraints during the training process of neural network models. By constructing a component-level network structure, implementing a components-cooperating loss function, and applying a fusion training process, the traditional iterative process of the thermodynamic model was transformed based on the component characteristic maps into a multi-objective optimization and training process of the component-level neural network. This approach improved the convergence and modeling accuracy of the model. The model was trained and tested using 26970 actual engine flight data. The results demonstrated that the maximum error of the proposed modeling method was approximately 7%, even under loose quasi-steady-state data, which was about 5% lower than that of the thermodynamic model based on the characteristic maps.

One of the effective measures to maintain efficient operation of aero-engine in practical engineering is to apply active tip clearance control technology, provided that an accurate tip clearance model is established for achieving tip clearance prediction. A simplified physical model and a mathematical model of tip clearance were established. The tip clearance calculation was transformed into a problem of thermal deformation and heat transfer. The operating parameters of the engine was extracted through machine learning model, and the boundary of heat transfer was solved using the effective features, thus realizing rapid prediction of real-time tip clearance based on the operating parameters of the engine. The cross validation accuracy of machine learning model was 98.9%, and the verification accuracy of the tip clearance model was 4.3%. The tip clearance calculation results under different working conditions and the change rule of the cold air flow rate were obtained, and the calculation time was less than 0.03 s.

To improve the generalizability of the ResNet deep learning model used for optimal design of the blade profile, a parametric design of a compressor blade profile suitable for subsonic and transonic speeds was carried out. The blade profile was constructed based on the geometric model in Matlab, and the design variables included the maximum thickness of the blade profile, the position of the maximum thickness, the mesh distance and the angle between the trailing edge of the blade profile and the axial direction, and the geometric model was meshed in batch by Pointwise software with a mesh size of 300 000, and simulation was carried out by OpenFOAM fluid simulation. The final simulation was performed by parametric modeling of 4 design variables to obtain the fluid simulation dataset of blade profile, which contained 22331 simulation cases of blade profile and can provide training and test datasets for ResNet deep learning models, helping to improve the generalization of the model.

In view of the traditional aero-engine combustor design process with long calculation cycle, high processing test and cost which restricts the engine design cycle, based on the aero-engine combustor model, POD-PCE-Kriging (proper orthogonal decomposition-polynomial chaotic expansion-Kriging) model and particle swarm optimization (PSO) algorithms were combined to construct the combustion performance surrogate model and carry out multi-objective optimization design. Through the test, the predicted results of POD-PCE-Kriging model were compared with the calculated results of one-dimensional program, and the root mean square errors of the predicted values of combustion efficiency and total pressure loss were 0.0063% and 0.1227%, respectively. Optimization search was carried out for the design variables, and the obtained Pareto optimal solution set was analyzed to provide physical insight into the design of advanced aero-engine combustor to meet the performance specifications, which can quickly and accurately obtain the design parameters to meet the optimal performance and accelerate the development cycle of aero-engine.

It is difficult to modify and broaden the scope of application for empirical models because of its inadequate ability to fit strong nonlinear function relationship. In order to solve these problems, a physics-informed deep learning cascade loss model of embedding the pressure distribution of cascade into neural networks was proposed. The loss prediction error decreased by 22.3% compared with empirical model for end-to-end neural networks and 37.9% compared with physics-informed model.

A general framework for solving thermal conductivity problems by physics-informed neural networks (PINNs) was developed and the treatment methods of three-dimensional unsteady problems, initial conditions, three types of boundary conditions, and surface boundaries were described. A one-dimensional thermal conductivity problem was solved by PINNs. The maximum relative error and average relative error between the solution and the theoretical solution were 0.0017% and 0.0011%, respectively. Using a simplified blade thermal conductivity problem as a case study, PINNs was compared with conventional finite element methods and the effects of different network architectures and hyperparameters of PINNs on the results were explored. The results showed that for the thermal con-ductivity problem of a simplified blade, the finite element method solved the problem in 11.7 s and PINNs solved in 8.96 s, with a maximum error of 1.03% and an average error of 0.139%. The internal cold source intensity of the solid blade was adjusted slightly, and the new convergence time was 1.41 s based on the PINNs training convergence, proving that PINNs method has the ability of fast calculation in case of minor change of the design condition.

A data-driven based hub loss prediction model was developed for the fan rotor under complex inflow conditions. The key aerodynamic parameters were extracted as input parameters and the entropy loss as output parameter. The sample database was constructed based on computation-efficient single-blade-passage steady computational method. Different boundary conditions were set and combined to make the database samples cover a wide range of complex inflows as far as possible. The RBF neural network was used to construct the mapping between input and output parameters to realize rapid prediction of hub loss. Results showed that the loss model can accurately capture the radial distributions of hub loss and significantly improve the prediction accuracy. Meanwhile, the averaged loss prediction error in the rotor hub region was mostly lower than 10% under different inlet mass flow, inlet swirl angle and inflow distortion conditions.

To explore the matching between the simulation data and test data of the engine and components, a data-driven serial collaborative simulation technology for aero-engine components was developed based on MATLAB/SIMULINK, Python and general commercial CFD software, which mainly consists of an integrated simulation platform for aero-engine components and a serial collaborative simulation platform. The integrated simulation platform adopted a modular approach to model the engine components, combined the cooperating equations and boundary constraints, and implemented the simulation of the whole engine characteristics; the serial collaborative simulation platform of engine components adopted a self-developed program combined with the underlying solver, proposed the data transfer and processing method of the intersection interface in the overlap region, finally implemented the serial simulation and boundary iterative solution between components, and completed the coupled solution of 0D and 3D simulation. Taking a small turbojet engine as the research object, the case verification was carried out based on available test data, the maximum error of calculation was less than 5%, indicating the accuracy and engineering application value of the integrated simulation platform and serial collaborative simulation platform.

To realize the prediction of 3D physical field, a method for rapid evaluation of 3D thermal stress on the outer wall of double-wall cooling structure using convolutional neural network (CNN) was presented. For the structural characteristics of the flat plate shape of the double-wall cooling structure outer wall, the temperature field was sliced into multiple sections along the wall thickness direction. The temperature was used as the basic element of the input tensor of the CNN, and the sections at different thickness positions corresponded to channel dimensions of the input tensor. The 3D temperature field was input into the CNN for output of the 3D von-Mises stress field under thermal load. The average absolute error of the trained converged network on the testing set was 1.23 MPa, with an average relative error of 15.10%, and the average absolute error for peak stresses was 16.10 MPa, with an average relative error of 11.81%. Results showed that for the thermal stress prediction problem of double-wall cooling structures, CNN can complete the temperature-to-stress mapping well. Exploring potential mechanisms of the thermoelasticity problems by using deep learning methods is expected to be realized.

The heat transfer process of pin-fin channel was studied by simulation, and a prediction model of the internal convective heat transfer coefficient was constructed. Firstly, several flow resistance elements were constructed to predict the cooling air flow rate, and then the Reynolds number in the channel was calculated according to the flow rate. Secondly, the Reynolds number and geometric parameters were combined and fed into a genetic algorithm and back propagation neural network to predict the average convective heat transfer coefficient of elements in the pin-fin channel respectively. Finally, a conversion method of equivalent heat transfer coefficient of pin-fin heat conduction was established based on ribbed heat transfer model, so as to apply the model to the cooling effect prediction of actual double-wall turbine blades. Numerical simulation results showed that the model can predict the flow rate of cool air and the convective heat transfer coefficient in the channel, and the relative error was controlled within 5%.

In order to improve the prediction accuracy of compressor performance, a surrogate model of deviation angle and total pressure loss based on machine learning methods was built. A two-stage compressor was used as the research object, and the elementary cascade database was established using experimental data of the flow field and the geometric parameters under multiple rotation speed conditions. The sensitivity analysis method was used to screen out the input parameters that have the greatest impact on the deviation angle and total pressure loss. Two machine learning algorithms, i.e.: Gaussian process regression and artificial neural network, were used to establish the deviation angle and total pressure loss model, and Bayesian optimization algorithm was introduced to search for the best hyperparameters of the model. For the optimization iteration and generalization problems faced by artificial neural networks, the Adam algorithm was introduced to adjust the model learning rate and modify the parameter gradient to accelerate the convergence. At the same time, the regularization method was used to enhance the generalization of the model. The cross validation scheme was used in model training process to reduce the risk of overfitting, and the optimal surrogate models were integrated into the throughflow program. The comparison of the calculation results showed, the total pressure ratio prediction error of the surrogate model on low speed conditions was significantly lower than that of the empirical model, among which the artificial neural network modeling had the most significant improvement, and the prediction error was reduced by 0.1 compared with the empirical model. Through comparison of surrogate models, the surrogate model based on artificial neural network had higher prediction accuracy and stronger robustness than the surrogate model based on Gaussian process.

A machine learning detection method for aero-engine system false data injection attacks based on Gramian angular field (GAF) and densely connected convolutional networks (DenseNet) was proposed. Firstly, two attack models of continuous and interval spurious data injection were constructed based on the simulation dataset of NASA’s commercial modular aero-propulsion system simulation (C-MAPSS). Secondly, the GAF method was proposed to transform the timing signal obtained by the aero-engine sensors into the image signal, and a DenseNet-121 network was designed to detect whether the aero engine was subject to false data injection attack and the type of attack was identified. Finally, the average classification accuracy of GAF-DenseNet method on T24, T50, and P30 sensors was 98.46%, which was 1.91%, 3.82%, and 0.38% better compared with long and short-term memory, gated recurrent units, and convolutional neural networks, respectively.

The optimization design method of controllable morphing blade profile considering both morphing cost and aerodynamic benefits was discussed. A machine learning algorithm was used to build a prediction model to predict key aerodynamic parameters of morphing blade profiles. The morphing cost and aerodynamic benefits were quantified, and a Bayesian optimization framework was built for optimization. Results showed that the prediction and optimization framework based on machine learning can accurately predict the aerodynamic performance of the fan after morphing, and evaluate the profit boundary of the blade profile morphing considering the morphing cost. The main conclusion indicated that using machine learning algorithm and Bayesian optimization framework can obtain a morphing scheme taking into account both morphing cost and aerodynamic benefits. This scheme can reduce the maximum stress of blade by 14.17% and the energy consumption of piezoelectric actuator by 67.45% while ensuring the improvement of aerodynamic performance compared with the scheme only considering aerodynamic benefits.

In view of the problems of poor reliability, low efficiency and high cost in planning space installation of aviation cable, an optimized method of route planning of aviation cable layout based on improved ant colony algorithm was proposed. The wiring installation space was rasterized, and the real environment modeling was carried out for the wiring installation space by analyzing the requirements and constraints of aviation wiring. The modeling space obtained was used to optimize the two-dimensional routing path of aviation cables. The heuristic function was improved by using the transition rule guided to the destination direction and increasing the penalty factor of turning corner, which reduced the blindness of path search and improved the smoothness of the planned path. An adaptive adjustment of pheromone volatile factor was used to improve the search efficiency and convergence speed of the algorithm. Genetic variation was introduced to avoid the algorithm falling into local optimum. In the simulation experiment, compared with other algorithms, the proposed method can significantly reduce overall cable path layout and the number of inflection points, indicating that the cable length was reduced and the cable electrical performance was better, enabling to provide the stability of the aircraft engine system. The feasibility and effectiveness of the proposed algorithm were verified.

Abnormal vibration of a core engine rotor system was observed in the experimental results under different working conditions. A simplified rotor model with looseness in its front squirrel cage supporting was established based on the experimental results. Based on the rotor dynamic theory and the experimental signal characteristics, the bolt looseness of the front squirrel cage supporting was modeled by a generalized Dahl friction model following Masing rules. The orbit of shaft center, displacement of front supporting and supporting anti-force were carefully analyzed for the following three types: linear rotor system without fault; nonlinear rotor system with bolt looseness but without gyroscopic effect; nonlinear rotor system with bold looseness and gyroscopic effect. The simulation results showed that, in a rotor system with dry friction, the response of the system had rich harmonic components; the gyroscopic effect showed important influence on the harmonic components in the dynamic responses. The research results provide a reference for the monitoring and diagnosis of this type of faulty rotor system.

Based on the high and low cycle fatigue test results of turbine attachment, the fretting fatigue damage control parameters were determined and the life prediction model was established. Parametric model of turbine attachment was established. The correlation between various damage parameters such as equivalent stress, contact pressure, friction stress, related slip distance and the fretting fatigue life was further analyzed by finite element analysis, finding that the FFD (fretting fatigue damage) parameters with integrated consideration of fatigue and wear had the highest correlation with the life, and the correlation coefficient could reach 97%. Based on the power function, the relationship between the FFD parameters and the fretting fatigue life was fitted, and the predicted fretting fatigue life of the turbine attachment was within 1.5 times scatter band. The sensitivity analysis of FFD parameters to geometric parameters was carried out, and the effects of different geometric parameters on fretting fatigue life were obtained from the perspective of numerical simulation: the fretting fatigue life was most sensitive to the pressure angle, which showed a negative correlation.

According to the defect of high flow resistance in current horizontal exhaust trailing edge slots, the scheme of radially tilted ones was proposed. Based on the improvements, the linear and curved tilted designs were further put forward. 3D numerical simulation methods were applied to these typical structures, then the internal flow fields were obtained and comprehensively compared. The results uncovered the mechanisms on flow resistance reduction of the new structures, by decreasing the coolant turning angle and suppressing the vortex in venting slots. In detail, the total pressure loss of typical linear and curved tilted designs could be reduced by about 10%−12% and 13%−15% compared with the horizontal one. The mixing processes of coolant injection and gas flow were simulated, which confirmed the previous conclusions, although the total losses of the novel cascade flow slightly increased. These facts and characteristic curves provided a guidance for the turbine blade’s comprehensive performance optimization.

Ignoring the radiation error, the velocity error and heat conduction error of a shielded total temperature probe under different working conditions were predicted by using the conjugate heat transfer numerical simulation method. Results showed that the thermal conductivity error was kept at a small value in the range of Mach number from 0.2 to 0.6; more than 90% of the total error was contributed by velocity error; the maximum measurement error was 1.13 K, which was 331.62%, 119.4%, 61.6% and 59.5% lower than the common structure A, B, C and D, respectively; velocity error and conduction error affected each other, and there was an optimal solution to minimize the total error. It was applicable to the total temperature measurement of isothermal incoming flow and high-speed non-isothermal incoming flow with Mach number greater than 0.5. Finally, the influence of thermocouple junction position on measurement accuracy was discussed. When the junction was kept away from the support from the design point, the measurement error followed the change law of first increasing and then decreasing.

Validated numerical approach and shock polar diagrams method were employed to investigate the flow structure and aerodynamic heating environment around high-mach flat- sweepback leading edge. The results demonstrated that with the increase of sweepback, the flat angle shock and leading edge shock crossed, and then Ⅳ, Ⅴ, Ⅵ shock-shock interactions were formed in turn. In general, the local pressure and heat flux increment caused by shock-shock interaction decreased with the increase of leading edge, but heat flux and pressure of interference zone under the transition Ⅳ shock-shock interaction might be lower than that under the typical Ⅴ shock-shock interaction. Moreover, it was found that the induced heat flux increased with the increase of attack angle, and decreased with the increase of Reynolds number. Also, a convenient discriminant method for high-Mach leading edge shock-shock interaction was provided based on critical deflection angle analysis. The discriminant results were validated against numerical results. The shock-shock interaction relation, interaction location and interference type graph for high-Mach leading edge was presented. The findings could benefit not only high-Mach aircraft integrated design, but also aerodynamic configuration optimization.

To investigate the combustion performance of swirling-flow single trapped vortex combustor, three swirler schemes (baseline type, small flow area and additional flare) were designed and thermal tests were carried out at an inlet temperature of 493 K and at atmospheric pressure, using RP-3 aviation kerosene as fuel and a cavity equivalent ratio between 0.8 and 1.8. Results showed that the flame in the main combustion zone of the baseline and small flow area schemes was a deconfined flame, while that of the plus flare scheme was a V-shaped standing flame, and the flame in the main combustion zone of the small flow area and plus flare schemes was more concentrated. In terms of combustion efficiency, the base type scheme was the lowest while the small flow area scheme was the highest, and the highest combustion efficiency obtained in the test was 96.3%. The combustion efficiency of the baseline model increased with the increase of the cavity equivalent ratio, while that of the small flow area solution remained basically the same at first and then increased slightly, and that of the plus flare solution increased rapidly and then slowly. The total pressure loss of the small circulation area scheme was higher than that of the baseline type, which was 7.2% and 4.5% respectively, at an inlet Mach number of 0.31.

In view of the problems that traditional fault diagnosis methods are prone to over-fitting under the condition of insufficient fault samples, and the weak fault features are difficult to be extracted under strong noise conditions, an aero-engine fault diagnosis method based on Siamese reduced-neuron attention networks was proposed. According to the principle of Siamese neural network, pairwise coupling of the samples in the training dataset was conducted, so that the input was changed from samples to sample pairs, and the diversity of input was improved. A reduced-neuron attention mechanism was integrated into the feature extraction module. Among them, the attention mechanism can quickly find useful features through global scanning, and suppress redundant features, which was in good agreement with the situation where the weak gas path fault features of aero-engines were submerged by noise; the reduced-neuron operation can reduce the amount of parameters and alleviate overfitting. The results show that this method achieves an average accuracy of 88.39% on the real monitoring data of CMF56-5B/7B series engines of an airline.