WAVELET SIGNAL DENOISING: IMPROVED MULTI-SCALE PRINCIPAL ANALYSIS

This study presents a fault diagnosis method and system designed to overcome the challenges associated with noise and randomness in process data using an improved wavelet threshold denoising technique. Modern industrial systems have become increasingly complex, necessitating accurate and timely fault detection to ensure production stability and safety. While Principal Component Analysis (PCA) has been a popular multivariate statistical technique for handling high-dimensional, noisy, and highly correlated data, its single-scale limitation restricts its effectiveness in analyzing multi-scale process data. The proposed method integrates wavelet transform and multiscale principal component analysis (MSPCA) to address these limitations. The process involves applying a wavelet transform to decompose the original data into different scales, followed by adaptive noise reduction using enhanced wavelet thresholding. The method combines the advantages of hard and soft thresholding techniques to minimize oscillations and distortions in denoised signals. The denoised data is then reconstructed by selecting low-frequency components containing fault information and removing high-frequency noise. Subsequently, MSPCA is applied at each scale, enabling fault diagnosis across multiple dimensions of process data. This approach effectively mitigates the impact of noise introduced by human error, sensor degradation, and environmental interference, which commonly affect industrial processes. By improving the continuity and accuracy of wavelet denoising, the system enhances fault detection precision and reliability. The method provides a robust solution for fault diagnosis in modern industrial systems, with applications across a wide range of industries requiring multi-scale data analysis. Keywords: Noise Reduction, Signal processing, Fault Diagnosis, Signal Reconstruction, Multiscale Principal Component Analysis (MSPCA), Wavelet Threshold Denoising.