Multivariate Performance Analysis for Real-Time Anomaly Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques for analyzing system data are inefficient, computationally expensive, and fail to recognize complex or nuanced conditions, often requiring large amounts of labeled training data and resource-intensive computations, especially when dealing with high-dimensional data from sensors.
Innovation Solution
Utilize supervised machine learning to determine significant parameters for a target variable, generate a multivariate baseline excluding irrelevant data, and employ a modified Mahalanobis distance computation to efficiently identify anomalous system states.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If conventional threshold-based techniques are used to analyze sensor data, then implementation is simple, but the techniques fail to recognize complex or nuanced conditions and are inefficient with large amounts of data
Solution Approach 1:
The patent transforms the analysis approach from simple threshold comparisons to multivariate statistical analysis by changing the parameters from single-value thresholds to dynamic control charts that incorporate multiple sensor parameters simultaneously. This allows recognition of complex conditions while maintaining computational efficiency through established statistical methods.
Solution Approach 2:
The patent adds a new dimension to data analysis by moving from univariate threshold checking to multivariate control charting. This dimensional transformation enables the system to analyze relationships between multiple parameters simultaneously, recognizing nuanced conditions that single-parameter analysis would miss.
2Measurement precision
If machine learning models with labeled training data are used to classify anomalous data, then classification accuracy improves, but the process becomes expensive, time-consuming, and resource-intensive
Solution Approach 1:
The patent enables the system to automatically establish its own baseline and detection rules from historical process data without requiring external labeled training data. The control charts self-adapt to normal process variation, eliminating the need for resource-intensive supervised learning while maintaining high anomaly detection accuracy.
Solution Approach 2:
Instead of training custom machine learning models, the patent uses established statistical control chart methodologies that have been proven effective for anomaly detection. This approach copies the success of traditional statistical process control while avoiding the computational costs of modern machine learning training.
3Reliability
If k-NN and z score techniques are used for anomaly detection, then predictive capability is provided, but the techniques perform poorly with high-dimensional data and require extensive training data
Solution Approach 1:
The patent extracts and focuses only on the most relevant parameters for anomaly detection using control chart methodology, rather than attempting to process all high-dimensional data with k-NN. This selective extraction reduces computational complexity while maintaining detection reliability by concentrating on critical process parameters.
4Measurement precision
If Mahalanobis distance computation is performed on all multivariate data, then comprehensive anomaly detection is achieved, but the computation becomes expensive and time-consuming
Solution Approach 1:
The patent segments the multivariate analysis into individual control charts for each critical parameter, rather than computing Mahalanobis distance across all parameters simultaneously. This segmentation reduces computational burden while maintaining detection accuracy by focusing on the most influential parameters separately.
Solution Approach 2:
The patent applies control charting to only the most critical parameters identified through domain knowledge and initial analysis, rather than applying comprehensive Mahalanobis distance to all available parameters. This partial action approach achieves sufficient detection accuracy with significantly reduced computation time.
Data Source
AI summary
The embodiments described herein generally relate to automated performance analysis of a system. Embodiments include receiving parameter values for a plurality of parameters captured during a time period. Embodiments include providing inputs based on the data set to a supervised machine learning model configured to determine significant parameters with respect to a target variable. Embodiments include receiving, from the supervised machine learning model in response to the inputs, an indication of two or more significant parameters from the plurality of parameters with respect to the target variable. Embodiments include generating a multivariate cluster for the target variable based on the two or more significant parameters and determining an anomalous state of the system with respect to the target variable based on the multivariate cluster for the target variable and data captured after the time period.