Adaptive Sliding Window for SHM Data Cleaning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing complexity and quality issues in structural health monitoring (SHM) data due to high velocity, variety, volume, low value density, and low veracity, particularly the presence of abnormal data, hinder accurate fault diagnosis in mechanical systems, as traditional signal processing methods are inadequate and machine learning algorithms struggle with data quality evaluation.
Innovation Solution
An adaptive method using adaptive sliding window (ASW) technology to divide SHM data into segments, followed by the calculation of weighted multiscale local outlier factor (WMLOF) values to detect anomalies, effectively refining data quality by identifying and eliminating abnormal data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional signal processing technology is used, then the processing is simple and easy to implement, but it becomes inapplicable for complex data sets with high velocity, variety, and volume
Solution Approach 1:
The patent segments the complex SHM data into multiple sub-datasets based on different time scales or feature dimensions. By dividing the large-scale complex data into manageable segments, the method enables traditional signal processing to be applied effectively to each segment while maintaining overall adaptability to the complex data set structure
Solution Approach 2:
The patent transforms the complex high-dimensional SHM data into multiple lower-dimensional views or projections. This dimensionality reduction approach allows traditional signal processing methods to operate on simplified representations while preserving essential features, thus improving adaptability without proportionally increasing processing complexity
2Measurement precision
If machine learning algorithms are used, then fault diagnosis accuracy is improved, but data quality issues such as low value density and low veracity cause garbage in, garbage out
Solution Approach 1:
The patent implements preliminary data cleaning and quality assessment steps before feeding data into machine learning algorithms. By pre-processing the SHM data to remove low-quality segments, handle missing values, and filter out erroneous measurements, the method ensures that only high-quality data enters the machine learning pipeline, preventing the garbage in, garbage out problem while maintaining high fault diagnosis accuracy
Solution Approach 2:
The patent incorporates feedback mechanisms where the output of machine learning algorithms is used to identify and flag potential data quality issues. This feedback loop allows the system to continuously improve data quality by identifying patterns of erroneous data and adjusting preprocessing parameters accordingly, thus enhancing both reliability and diagnostic accuracy
3Productivity
If fixed-size sliding window is used for data segmentation, then the processing is straightforward, but data leakage and redundancy occur
Solution Approach 1:
The patent employs dynamic sliding window sizes that adapt to the local characteristics of the SHM data. Instead of using a fixed window size, the method adjusts the window length based on signal variability, event density, and feature importance in different data regions. This dynamic approach prevents data leakage by ensuring windows are appropriately sized for local conditions while reducing redundancy through optimized overlap control
Solution Approach 2:
The patent applies different segmentation strategies to different regions of the SHM data based on local characteristics. In regions with high event density or rapid changes, smaller windows are used to capture local features accurately, while in stable regions, larger windows reduce redundancy. This local quality approach ensures optimal processing efficiency while minimizing information loss across the entire data set
Data Source
AI summary
An adaptive method of cleaning structural health monitoring (SHM) data based on local outlier factor is provided, including following steps: step 1: dividing SHM data to be analyzed into a series of data segments by using adaptive sliding window (ASW) technology; step 2: extracting time-domain statistical factors and frequency-domain statistical factors of each of the data segments to refine data information, thereby forming objects for study; step 3: calculating an outlier degree of each of the objects by using a weighted multiscale local outlier factor (WMLOF) based on feature factors; and step 4: detecting anomalies in SHM data by comparing a WMLOF value with a threshold value. The adaptive method can improve data quality effectively.