Industrial Anomaly Diagnosis Using Sensor Contribution Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current anomaly detection techniques in industrial systems are inadequate for identifying the specific sensors responsible for anomalies, as they fail to effectively capture temporal effects and lack standardized methods for diagnosis after detection.
Innovation Solution
A system and method that convert multivariate time series data into finite-sized time windows, compute features, preprocess data, generate an anomaly detection model, calculate anomaly scores, identify reference instances, and sort diagnosis scores to pinpoint anomalous sensors, incorporating techniques like decision boundary analysis and substitution methods for accurate sensor identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If standard anomaly detection algorithms (1-class SVM, isolation forest, Gaussian mixture model) are used, then anomaly detection capability is improved, but anomaly diagnosis capability (identifying responsible sensors) deteriorates
Solution Approach 1:
The patent segments the anomaly detection task into two distinct phases: (1) anomaly detection using standard algorithms like 1-class SVM or isolation forest, and (2) anomaly diagnosis using a separate contribution analysis module. This segmentation allows each module to specialize - the detection module achieves high reliability while the diagnosis module recovers sensor contribution information that would be lost in a unified approach.
Solution Approach 2:
The patent introduces an intermediary contribution analysis module that bridges anomaly detection and sensor identification. This intermediary takes the anomaly detection results and systematically analyzes sensor contributions through methods like T2 statistics, Q statistics, or permutation importance, thereby recovering the lost sensor contribution information without compromising the detection capability.
2Measurement precision
If T2 statistics and Q statistics in PCA are used for anomaly diagnosis, then sensor contribution identification is improved, but temporal effect capture deteriorates
Solution Approach 1:
The patent applies dynamics by making the contribution analysis adaptive to temporal characteristics. Instead of using static PCA-based T2/Q statistics that treat all data points equally, the system dynamically adjusts the contribution analysis to account for temporal dependencies in time series data, thereby capturing temporal effects while maintaining precise sensor identification.
Solution Approach 2:
The patent changes the parameters of the contribution analysis by offering multiple methods (T2 statistics, Q statistics, permutation importance) with different temporal sensitivity characteristics. This allows the system to select or combine methods that best capture temporal effects for specific applications, transforming the static parameter set into a flexible, temporally-aware analysis framework.
3Ease of operation
If multivariate time series data is analyzed without time windowing, then data processing simplicity is improved, but temporal pattern recognition deteriorates
Solution Approach 1:
The patent segments continuous time series data into discrete time windows, creating manageable chunks that preserve temporal patterns while simplifying processing. Each window can be independently analyzed for sensor contributions, making the overall analysis more tractable while maintaining the temporal structure necessary for accurate pattern recognition.
Solution Approach 2:
The patent performs preliminary time windowing and feature extraction before the main contribution analysis. This preliminary action organizes the raw time series data into structured windows with extracted features, making subsequent analysis simpler and more effective at capturing temporal patterns without requiring complex real-time processing.
4Reliability
If comprehensive feature extraction and preprocessing are performed, then model accuracy is improved, but computational complexity deteriorates
Solution Approach 1:
The patent segments the computational workload into distinct preprocessing and analysis phases. Feature extraction and preprocessing are performed once on segmented time windows, and the resulting features are reused in the contribution analysis. This segmentation avoids redundant computations while maintaining model accuracy.
Solution Approach 2:
The patent performs comprehensive feature extraction and preprocessing as preliminary actions before the main anomaly detection and diagnosis tasks. By preparing features in advance and organizing data into structured windows, the system reduces the computational complexity of the actual analysis while maintaining high model accuracy through the use of rich, pre-processed features.
Data Source
AI summary
Industrial plants involve a large amount of equipment, which generate a large amount of data. By analyzing this data, the operator can diagnose anomaly in the plant. Analyzing this data is difficult and time taking task. A method and system for diagnosing anomaly in an industrial system in a time efficient and convenient manner has been provided. The system is configured to diagnose the anomaly by finding out one or more sensors responsible for the anomaly. The present disclosure treats the anomaly detection model as a score generating function. Whenever for a particular instance the score given by the anomaly detection model crosses a pre-determined threshold, anomaly is reported and the diagnosis algorithm is triggered. The system is configured to diagnose the anomaly predicted in case of time series as well as non-time series data.