Research On The Evaluation Method Of Manipulation Skills Of Teardrop Pattern Subjects Based On Competency

Aircraft engines are important components of military and civilian aircraft,and their health status directly affects the safety of the aircraft.In the past,maintenance personnel used periodic maintenance methods,which could reduce the probability of engine failure to a certain extent,but this method paid a high price in terms of maintenance costs and manpower resources and was not the optimal solution.In recent years,with the rapid development of the big data era and sensor technology,data-driven aircraft engine performance degradation prediction has become the mainstream solution.Technical personnel can obtain more real engine operating data,thereby determining the health status of the engine faster and more accurately.However,the sensor data collected in actual industrial production has the characteristics of high dimensionality and complex working scenarios,which brings problems such as insufficient feature utilization and inaccurate prediction.Therefore,how to establish accurate performance degradation prediction models based on different working scenarios is a major challenge at present.The main work of this thesis is as follows:(1)This thesis proposes an XGBoost-based outlier data removal method using the L1-norm regularization.Due to the reconstruction of time windows,short-term changes in features may be obscured,resulting in noisy data.XGBoost is used to predict the training set,and the residual Z-Score between the predicted and actual values is standardized to fit a normal distribution.Values greater than 3σ are removed to reduce noise data.(2)This thesis proposes a performance degradation model called XGBoost-DCNN.To reduce data loss,convolutional layers are used instead of pooling layers to extract key degradation features from sensor data while retaining global features.Since adjacent features have weak correlations,1D convolution filters are used to avoid overemphasis on certain features.(3)This thesis proposes a performance degradation model called XGBoost-Res NetDCNN.To further extract deep degradation features from sensor data,the depth of the DCNN network is increased,and more convolutional layers are added to better extract features.However,it was found that the training process may result in gradient disappearance or explosion,leading to reduced model accuracy.Therefore,Res Net is introduced to solve this problem.Validation experiments were conducted in different fault and operating conditions,and the designed network was compared with classical performance degradation prediction models,demonstrating its effectiveness.