Sensor Signal Denoising with Iterative SSA Window Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current noise removal methods for industrial process monitoring, such as laser welding, face challenges in accurately separating trend components from noise due to the lack of a general criterion for selecting window lengths and grouping strategies in Singular Spectrum Analysis (SSA), leading to poor separation and potential mixing of trend and noise components.
Innovation Solution
A modified SSA method is introduced, which iteratively adjusts the window length and applies single value decomposition to a trajectory matrix, calculating root mean square values and halting iterations when a minimum threshold is met, ensuring effective noise removal and reconstruction of a denoised signal without initial conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional Singular Spectrum_analysis is used for noise removal, then noise can be removed from the signal, but the separation between trend components and noise is poor due to lack of general criterion for selecting window lengths and grouping strategies
Solution Approach 1:
The method enables the signal processing system to automatically determine optimal window lengths and grouping strategies through iterative computation and automatic selection criteria, eliminating the need for manual parameter tuning and expert knowledge while achieving optimal noise removal performance
Solution Approach 2:
The patent systematically varies window length parameters and grouping strategies through iterative computation, automatically selecting optimal parameter combinations based on signal characteristics rather than relying on fixed or manually selected parameters
2Measurement precision
If iterative SSA with automatic window length selection is applied, then optimal separation between trend and noise is achieved, but computational complexity increases
Solution Approach 1:
The method employs feedback mechanisms where the results of each iterative SSA computation are evaluated against automatic selection criteria, and this feedback guides the selection of subsequent window lengths and grouping strategies, enabling the system to converge toward optimal parameters efficiently
Solution Approach 2:
The patent applies partial iterations of the SSA process with progressively refined parameters, performing computations only to the extent necessary to achieve optimal separation without exhaustive search of all possible parameter combinations, thus balancing accuracy with computational efficiency
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
A method for performing (200) a noise removal operation (200) on a signal (S) acquired by a sensor obtaining a denoised signal (x(t)), said noise removal operation including a Singular Spectrum Analysis (SSA), said Singular Spectrum Analysis (SSA) including performing iteratively an operation of decomposition (210) of said acquired signal (S) considered as one dimensional series, an operation (220) of construction of a trajectory matrix (X), transforming said trajectory matrix (X) in a form (XXT) to which single value decomposition is applicable, an operation (230) of single value decomposition on said transformed matrix (XXT) computing eigenvalues and eigenvectors of said matrix, an operation of reconstruction (250) of a one dimensional series corresponding to said denoised signal based on selected among said eigenvalues, wherein after the the single value decomposition operation is provided applying (242) sequentially a single value decomposition starting from a given window (L) value, in particular a value of three, for each iteration, calculating (244) the root mean square value between the current and previous eigenvalue, calculating a minimum and its position of said root mean square value halting (246) the iterations if said minimum is lower than a determined threshold value (ε), in particular lower than 1, otherwise increasing the window value and returning to the operation of decomposition (210) of said acquired signal (S).