Based on Acoustic Emission and Neural Network Pattern Recognition to Reduce the Noise of In‐service Reinforced Concrete Columns

Abstract The large‐scale and complex structure of reinforced concrete columns results in a pronounced distortion of acoustic emission waves, which in turn distorts the monitoring data. The vibrations from the adjacent slabs of in‐service reinforced concrete columns create a complex noise environment, complicating the differentiation of background noise from damage signals using traditional hardware filtering methods. Previous Wavelet Transform and Empirical Mode Decomposition methods have been particularly challenging for the processing of high‐dimensional data. This paper introduces a machine learning‐based noise reduction method capable of significantly enhancing the signal‐to‐noise ratio. The paper takes into account the frequency domain characteristics of the waveform signal and employs principal component analysis for dimensionality reduction, extracting the principal components PC1 to PC3 from the eigenvalues. Ultimately, the feature vectors derived from principal components PC1 to PC3 were fed into the neural network for pattern recognition. We achieved an 88.2% accuracy rate in pattern recognition of acoustic emission signals using neural networks.