Industrial Anomaly Detection Using Adaptive Cycle and Shape Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing anomaly detection methods in industrial systems struggle to handle diverse anomaly formats, inconsistent periodicity in sensor data, and require significant historical anomaly data, leading to inaccurate and sensitive performance due to fixed window sizes.
Innovation Solution
An innovative method using two modules: the first module identifies noisy cycles through noise scores, and the second module identifies anomalies based on differing shapes, employing adaptive ensemble techniques with Dynamic Time Warping distance to handle varying periodicities and requiring no historical anomaly data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If fixed predetermined window size is used for anomaly detection, then the detection process is simple, but the performance becomes sensitive to window size selection and cannot handle inconstant periodicity
Solution Approach 1:
The patent applies dynamics by transitioning from fixed window sizes to adaptive, variable window sizes that adjust based on the detected periodicity characteristics of the time series data. The system dynamically determines optimal window sizes for different segments of data, allowing the detection process to adapt to inconstant periodicity while maintaining reasonable complexity through automated adaptation.
Solution Approach 2:
The patent changes the parameter of window size from a fixed predetermined value to a variable parameter that is optimized based on the specific characteristics of the time series data. By calculating optimal window sizes according to detected periodicity patterns, the system improves detection accuracy without requiring manual tuning for each dataset.
2Ease of manufacture
If existing anomaly detection methods are applied, then the implementation is straightforward, but they cannot appropriately handle diverse formats of anomalies and inconstant periodicity
Solution Approach 1:
The patent segments the anomaly detection task into distinct modules: periodicity detection, optimal window size determination, and anomaly scoring. This segmentation allows each component to handle specific aspects of diverse anomaly formats independently, improving versatility while keeping the overall implementation manageable through modular design.
Solution Approach 2:
The patent creates a universal anomaly detection framework that can handle multiple anomaly formats and periodicity patterns through a single integrated system. The method universally applies to different time series datasets by automatically adapting to their specific characteristics, eliminating the need for separate implementations for different anomaly types.
3Measurement precision
If classification-based methods are used, then the model can distinguish normal from abnormal windows, but a relatively large number of anomalous data are required for training
Solution Approach 1:
The patent inverts the traditional classification approach by not requiring anomalous data for training. Instead, it trains on normal data to establish expected patterns and then identifies anomalies as deviations from these patterns. This inversion eliminates the need for scarce anomaly labels while maintaining classification accuracy through unsupervised or semi-supervised learning paradigms.
Solution Approach 2:
The system performs self-service by automatically determining optimal window sizes and detection parameters from the data itself without requiring external annotation of anomaly cases. The method uses the inherent structure of the time series data to guide the detection process, making the system self-adapting and reducing dependency on manual data preparation.
4Measurement precision
If small window size is used in matrix profiling, then more detailed anomalies can be detected, but the scoring becomes noisy with large number of prominent scores
Solution Approach 1:
The patent changes the window size parameter dynamically based on the detected periodicity of the time series data. By optimizing window size for each specific dataset rather than using small fixed sizes, the system achieves detailed anomaly detection while maintaining scoring stability through parameter adaptation to the data's inherent structure.
Data Source
AI summary
A method for detecting an anomaly in time series sensor data. The method may include identifying a noisiest cycle from the time series sensor data; for an evaluation of the noisiest cycle indicative of the anomaly being detected at a confidence level above a threshold, providing an output associated with the noisiest cycle as being the anomaly; and for the evaluation of the noisiest cycle indicative of the anomaly being detected at the confidence level not above the threshold: identifying a cycle from the time series sensor data having a most differing shape; and providing the output associated with the cycle having the most differing shape as being the anomaly.