A Novel Classification Method for Flutter Signals Based on the CNN and STFT

Abstract

Necessary model calculation simplifications, uncertainty in actual wind tunnel test, and data acquisition system error altogether lead to error between a set of actual experimental results and a set of theoretical design results; wind tunnel test flutter data can be utilized to feedback this error. In this study, a signal processing method was established to use the structural response signals from an aeroelastic model to classify flutter signals via deep learning algorithm. This novel flutter signal processing and classification method works by combining a convolutional neural network (CNN) with time-frequency analysis. Flutter characteristics are revealed in both time and frequency domains, which are harmonic or divergent in the time series; the flutter model energy is singular and significantly increases in the frequency view, so the features of the time-frequency diagram can be extracted from the dataset-trained CNN model. As the foundation of the subsequent deep learning algorithm, the datasets are placed into a collection of time-frequency diagrams calculated by short-time Fourier transform (STFT) and labeled with two artificial states, flutter or no flutter, depending on the source of the signal measured from a wind tunnel test on the aeroelastic model. After preprocessing, a cross-validation schedule is implemented to update (and optimize) CNN parameters though the trained dataset. The trained models were compared against test datasets to validate their reliability and robustness. Our results indicate that the accuracy rate of test datasets reaches 90%. The trained models can effectively and automatically distinguish whether or not there is flutter in the measured signals.

Aeroengine Control System Sensor Fault Diagnosis Based on CWT and CNN

“…is significantly reduces overfitting and simulates a neural network with a large number of different network structures and in turn make the nodes in the network more robust. e trainable parameters of the CNN model were first initialized, and then optimized by the back-propagation (BP) algorithm and the adaptive moment estimation (Adam) algorithm [25] in the PyTorch deep learning framework. e optimization process serves to calculate the error between the real value and the predicted value.…”

End-to-End Deep Learning by MCU Implementation: An Intelligent Gripper for Shape Identification

“…STFT and CNN are used in fault diagnosis of bearings and also implemented for gearbox faults diagnosis [ 24 , 25 ]. Here are some methods based on STFT and CNN for classification, such as the flutter in the aircraft [ 26 ], and ECG arrhythmia classification [ 27 ]. As mentioned previously, STFT preprocessing provides frequency information of signals but also tracks their variation over time.…”

Intelligent Fault Diagnosis of Aeroengine Sensors Using Improved Pattern Gradient Spectrum Entropy

“…The CNN can solve gradient decay problems in addition to its stronger feature learning and characterization abilities than the shallow network [15,16]. The CNN preprocessing procedure is relatively brief, so the learning time is brief-there is less data necessary to learn free parameters, which relieves the memory burden for network operation and allows for the construction of a very powerful neural network [17]. The CNN also has strong antinoise capability and preserves all necessary information during feature selection.…”