Material Test Noise Removal via Wavelet Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing noise removal methods for material test data, such as low-pass filtering, fail to accurately distinguish between test force changes and noise, leading to inaccurate determination of breakage points and elastic modulus calculation in high-speed tensile tests, and often result in data loss or unnatural filtering effects.
Innovation Solution
A noise removal method that divides raw data around change points, applies discrete wavelet transform and inverse transform for low-pass filtering, and reconnects filtered data while interpolating missing sections to maintain time series integrity and natural waveform characteristics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If low-pass filtering is applied to remove high-frequency noise from test force data, then noise is reduced, but change point characteristics (breakage points, elastic modulus) become inaccurate or lost
Solution Approach 1:
The test force data is segmented into multiple sections based on detected change points (breakage points, elastic modulus regions). Different filtering strategies are applied to different segments: strong low-pass filtering in regions away from change points, and minimal or no filtering near change points. This segmentation allows noise removal while preserving critical measurement characteristics.
Solution Approach 2:
The filtering strength is made local rather than uniform across the entire dataset. The patent applies adaptive filtering where the filtering coefficient varies depending on the local characteristics of the data - stronger filtering in stable regions and weaker or no filtering near change points. This local quality approach ensures that noise is removed where safe while preserving measurement accuracy where critical.
2Object-affected harmful factors
If aggressive low-pass filtering is used to remove noise, then noise reduction is improved, but data loss occurs and natural waveform characteristics are distorted
Solution Approach 1:
Change points (breakage points, elastic modulus regions) are detected and marked before the filtering process is applied. This preliminary identification allows the filtering algorithm to avoid aggressive filtering in critical regions, preventing data loss and waveform distortion before they can occur. The preliminary detection of what needs to be protected guides the subsequent filtering action.
3Ease of operation
If uniform filtering is applied across all data, then processing simplicity is maintained, but unnatural connections and artifacts appear in filtered data
Solution Approach 1:
The data is divided into segments with different filtering characteristics, and the filtering is applied piecewise rather than uniformly. This segmentation approach, while more complex than uniform filtering, prevents unnatural connections by ensuring that filtering transitions occur only in appropriate regions, maintaining the natural waveform composition throughout the dataset.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This method effectively removes high-frequency noise while preserving change point characteristics, allowing for accurate determination of breakage points and elastic modulus calculation, and reduces unnatural connections in filtered data.
Implementation Method 1
a wavelet transformation part which performs wavelet transformation on an input waveform
Implementation Method 2
a correction calculation part which calculates a wavelet transformation result of a next-period driving waveform from a result of wavelet transformation on the driving waveform, a result of wavelet transformation on a response waveform, and a result of wavelet transformation on a target waveform
Implementation Method 3
a reverse wavelet transformation part which performs reverse wavelet transformation on the wavelet transformation result of the next-period driving waveform found by the calculation so as to obtain a next-period driving waveform of time-series data
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A breakage point is detected as a change point from raw data and the data is divided before and after the breakage point to obtain divided data D1 and D2. When the low-pass filtering is performed on each of the divided data D1 and D2 and the filtering for all divided data ends, time-series data whose natural frequency is removed is reconstructed before and after the breakage point. When the reconstruction data are connected at the breakage point, it is possible to restore the time-series data of the test force to the time-series data whose natural vibration of the test machine body is removed while taking advantage of a change in test force at the breakage point.