Study on the denoising method of train on-board relay coil current based on 3 σ-EEMD

Railway relays are pivotal electrical components on trains, with the state of their contacts crucial for safe operation. Thus, it is essential to monitor the status of train relays to foresee potential failures. The coil current of the relay is a key parameter for diagnosing the state of the relay. Given the complex electromagnetic environment on-board, monitoring with current probes introduces significant noise, which affects data quality. Addressing how to process this noisy data to extract usable current information is a pressing challenge. To tackle the noise issue in coil currents, a novel denoising approach named $3 \sigma$-EEMD, which combines Outlier Detection (using the $3 \sigma$ method) and Adaptive Ensemble Empirical Mode Decomposition (EEMD) is proposed. Initially, the $3 \sigma$ method is employed to detect outliers in the relay coil current that deviate significantly from the original waveform, replacing these outliers with the average of the adjacent normal data. Subsequently, Ensemble Empirical Mode Decomposition (EEMD) is applied to the data cleansed of outliers. The variance of the autocorrelation function of the resulting Intrinsic Mode Functions (IMFs) is calculated. By setting a threshold for the variance of the autocorrelation function, IMFs exceeding this threshold, along with the residual component, are selected for reconstruction. The reconstructed data represents the denoised coil current. Compared to common wavelet threshold denoising methods, the $3 \sigma$-EEMD approach significantly enhances the denoising effect on train relay coil currents, with a $75 \%$ improvement in Signal-to-Noise Ratio (SNR) and a 95% reduction in Root Mean Square Error (RMSE), demonstrating a significant denoising effect and more intact waveform characteristics, thereby validating the effectiveness of this method.