Anomaly Detection in Sensor Systems Using Subsequence Clustering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection methods in time series data from sensors are limited in detecting various types of operating anomalies, particularly failing to identify recurrent anomalies and requiring prior knowledge or supervision, leading to unreliable detection and high computational costs.
Innovation Solution
A method that constructs a normal operating model by classifying and grouping subsequences based on similarity, assigning weights to representative subsequences, and attributing normality scores to detect anomalies without prior knowledge or supervision, improving reliability and scalability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If discord-based anomaly detection is used, then isolated anomalies can be detected, but recurrent anomalies remain undetected
Solution Approach 1:
The patent segments the time series data into multiple subsequences of equal length and groups them by similarity. By analyzing the distribution of these grouped subsequences, the method can detect both isolated anomalies (outliers in the distribution) and recurrent anomalies (subsequences that form distinct clusters away from the normal mode). This segmentation approach resolves the contradiction by enabling detection of multiple anomaly types simultaneously.
Solution Approach 2:
The patent extracts more subsequences than traditionally analyzed, grouping them into multiple modes representing different operational states. By examining the distribution across these modes and identifying deviations from the predominant mode, the method achieves comprehensive anomaly detection. This excessive sampling ensures both isolated and recurrent anomalies are captured in the analysis.
2Measurement precision
If supervised anomaly detection methods are used, then detection accuracy improves, but prior knowledge and supervision are required
Solution Approach 1:
The patent implements an unsupervised method where the system automatically identifies the normal operational mode by finding the predominant subsequence group in the data. The anomaly detection then proceeds by comparing other subsequences against this self-identified normal mode, requiring no external supervision or prior knowledge. This self-service approach resolves the contradiction by achieving accurate detection through autonomous system behavior.
Solution Approach 2:
The patent performs preliminary analysis to automatically determine the normal operational mode by identifying the predominant group of subsequences before proceeding with anomaly detection. This preliminary action establishes the reference framework without requiring external supervision, enabling subsequent accurate detection of deviations from normal operation while maintaining system simplicity.
3Reliability
If comprehensive anomaly detection is implemented, then detection coverage improves, but computational time increases
Solution Approach 1:
The patent segments the large time series into multiple smaller subsequences of equal length, which can be processed independently and grouped by similarity. This segmentation enables parallel processing and reduces the computational burden compared to analyzing the entire time series sequentially, while maintaining comprehensive anomaly detection coverage through the distribution analysis of all grouped subsequences.
Solution Approach 2:
The patent merges similar subsequences into groups or modes based on their similarity, representing multiple identical or near-identical subsequences as a single representative group. This merging significantly reduces the number of individual comparisons needed while preserving the full detection capability, as the distribution characteristics of the original subsequences are maintained through their group representatives.
Data Source
AI summary
A method for determining a health status of a system of interest is proposed. The method comprises acquiring (S1) a time series, extracting (S2) subsequences, selecting (S3) a set of subsequences, classifying (S4) the subsequences of the set into several groups on the basis of at least one criterion of resemblance to at least one reference subsequence, and constructing (S5) a normal operating model of the system of interest. The construction includes, for each group, a modeling (S51) of a representative subsequence and a determination (S52) of an associated weight. The normal model is defined by the modeled subsequences and the associated weights. The method further includes an attribution (S6) of a normality score to each subsequence extracted by comparison with the normal model, an identification (S7) of at least one abnormal subsequence, and a determination (S8) of the health status of the system of interest.