Advanced Noise Reduction for In-Cylinder Combustion Pressure Data Using ICEEMDAN and Optimal Wavelet Selection

This study introduces a robust approach for denoising pressure signals by integrating Improved Complete Ensemble Empirical Mode Decomposition (ICEEMDAN), Continuous Mean Square Error (CMSE) analysis, optimal wavelet selection, and wavelet thresholding techniques. ICEEMDAN decomposes noisy signals into Intrinsic Mode Functions (IMFs), which are classified via CMSE to distinguish noise-dominated IMFs from signal-dominated IMFs. A key innovation is the sparsity-based optimal wavelet selection strategy, which dynamically identifies the most effective wavelet basis and decomposition levels for each noise-dominated IMF. These IMFs undergo wavelet thresholding for noise suppression, while signal-dominated IMFs are preserved to ensure signal integrity during reconstruction. The proposed approach was validated on simulated signals and experimental data from lean syngas/air combustion at equivalence ratios. Results consistently demonstrate superior denoising performance across various metrics compared to traditional infinite impulse response (IIR) filters, empirical mode decomposition (EMD) and conventional ICEEMDAN-wavelet methods. By dynamically adapting wavelet selection and decomposition levels to varying signal and noise characteristics, this method achieves optimal noise suppression with high signal fidelity. Heursure and Minimaxi methods use hard thresholding and reveal consistent outperformance compared to Sqtwolog and Rigrsure in both simulated and real-world scenarios. This approach offers significant advancements in engine performance optimization, fuel efficiency improvement, combustion diagnostics, and industrial pressure monitoring. By bridging theoretical innovation with practical applications, the proposed method provides a scalable and precise solution for signal reconstruction in complex environments.