Research On Prediction Of Aeroengine Status Data And Optimization Method Of Aeroengine Washing Schedule

After the aeroengine runs for a long time,the components will accumulate pollutants such as air dust and heavy oil dirt,which will reduce the engine performance and increase the carbon emissions and maintenance costs.Wing cleaning refers to the method of removing pollutants without disassembling the engine,which can solve the problem caused by engine pollution to a certain extent.This method is simple and efficient,with short parking time,and is the first choice for airlines.At present,most airlines adopt the regular cleaning strategy,that is,cleaning at a fixed time interval during the wing period.The regular cleaning strategy is simple to formulate,but the disadvantage is that insufficient cleaning often occurs.Insufficient cleaning will make the aeroengine run in a state of low efficiency and high fuel consumption,and shorten the service life of the engine.Due to the high cost of wing cleaning,frequent cleaning in order to improve performance will bring huge maintenance costs.The key to balancing engine performance and maintenance cost is to determine the optimal cleaning time.Therefore,the paper aims to explore the best cleaning time for aeroengines,and studies the key technologies such as aeroengine status data noise reduction,aeroengine performance prediction,aeroengine cleaning effect evaluation,and aeroengine cleaning time optimization.In order to solve the problem that the current noise reduction methods can not give consideration to both good noise reduction effect and the step of unsmoothed data when denoising aeroengine status data,a Taylor noise reduction self encoder model is proposed.The model extracts features from the state data by using the gated cycle unit,uses a new type of neuron combined with Taylor formula to screen the useful information in the features,and finally reconstructs the useful information into pure aeroengine state data through the gated cycle unit.Compared with the current popular methods,the proposed method can achieve better noise reduction effect on the basis of retaining the original step amplitude of aeroengine state data.In order to solve the problem that Levenberg Marquardt(LM)neural network is easy to fall into local minima in the process of network training,a reconstruction dual regulation LM neural network is proposed.First,the aeroengine fuel flow sequence is decomposed based on phase space reconstruction technology,which reduces the nonlinearity of the original data and improves the prediction accuracy of the neural network model.Secondly,a dual regulation optimization mechanism is proposed,which adjusts the trust region through the convergence step and adjusts the gradient descent direction through self adaptation.The double regulation mechanism can avoid the neural network falling into local optimum easily.Finally,with the reconstructed phase space as the input and the dual regulation optimization mechanism combined with LM algorithm as the descent gradient calculation method,the reconstructed dual regulation LM neural network model is built.The experimental results show that the proposed model can significantly improve the convergence speed and achieve accurate state parameter prediction.Thirdly,carry out parameter change prediction after aero-engine washing,which can provide a theoretical basis for evaluating the effect of energy conservation and emission reduction after washing and formulating a reasonable maintenance plan.There is continuity between the prediction trend of traditional methods and the previous data signals,so it is difficult to be applied to predict the step after washing.A step parameter prediction model based on process neural network is developed in this paper.Taking exhaust gas temperature margin(EGTM)as the research object,this paper analyzes the step related parameters,so that the model focuses on predicting the data step instead of extending the data trend.Because there are few washing records and few data mutations for model learning,transfer learning is introduced to complete model training.Experimental results show the effectiveness of the proposed method.Finally,based on the above work,an optimization method of aeroengine cleaning time based on reinforcement learning is proposed.A reinforcement learning framework for optimizing cleaning strategies is proposed to maximize the economic benefits of cleaning and engine performance.Use the data obtained from airlines to evaluate the carbon emissions after cleaning and the company’s benefits for the reinforcement learning framework.In addition,the proposed framework includes a hybrid migration process neural network for updating engine status,which is composed of multiple network models,and can acquire knowledge from other engine status data to learn the cumulative effect of multiple cleaning.The experimental results verify the effectiveness of the proposed reinforcement learning framework and the hybrid transfer process neural network in the framework.This paper has certain practical application value for enriching the optimization of aeroengine washing decision and improving the refinement level of aeroengine management.This study can also provide theoretical support for operation and maintenance optimization and management optimization of other complex equipment.