Outlier Detection via Cluster Modeling for Sensor Data

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing automated system monitoring techniques require detailed engineering simulations, expert knowledge, and significant computer resources, making them unsuitable for real-time anomaly detection in systems where off-nominal data is unavailable and computer resources are limited.

Innovation Solution

A system and method for building monitoring knowledge bases using nominal system data, forming clusters of expected sensor values, and analyzing query points to detect anomalies without needing off-nominal data, utilizing a cluster database that weights distances based on the number of training points and standardizes results for consistent interpretation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If detailed engineering simulations and model based reasoning are used for system monitoring, then measurement precision and anomaly detection capability are improved, but device complexity and computational resource requirements increase significantly

Engineering Contradiction:

Improveanomaly detection capabilityVSAvoidcomputational resource requirements

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent creates simplified copies of system behavior through cluster models that capture nominal operating patterns without requiring full engineering simulations. These cluster models are trained on historical data to represent expected system behavior, enabling anomaly detection through comparison rather than through complex simulation execution.

Inventive Principle:

Principle #26Copying

Solution Approach 2:

The patent replaces expensive, computationally intensive simulation models with cheaper, data-driven cluster models that can be rapidly trained and deployed. These cluster models require minimal computational resources during operation, making them suitable for real-time monitoring in resource-constrained environments.

Inventive Principle:

Principle #27Cheap short-living objects (Disposable)

2Measurement precision

If model based reasoning with detailed system simulation is implemented, then anomaly detection accuracy is improved, but speed of execution deteriorates making real-time monitoring infeasible

Engineering Contradiction:

Improveanomaly detection accuracyVSAvoidexecution speed

Core Design Contradiction:

Measurement precisionVSSpeed

Solution Approach 1:

The patent performs preliminary analysis by training cluster models on historical nominal data during an offline phase. This pre-processing creates ready-to-use models that capture system behavior patterns, eliminating the need for complex simulations during real-time operation. The clustering algorithm groups similar data points to establish expected behavior baselines in advance.

Inventive Principle:

Principle #10Preliminary action

3Reliability

If classification models requiring both nominal and off-nominal data are used, then anomaly detection reliability is improved, but ease of operation deteriorates due to data availability constraints

Engineering Contradiction:

Improveanomaly detection reliabilityVSAvoiddata availability requirements

Core Design Contradiction:

ReliabilityVSEase of operation

Solution Approach 1:

The patent extracts and removes the requirement for off-nominal training data from the anomaly detection process. By using only nominal data to train cluster models and then measuring deviation from these clusters, the system achieves reliable anomaly detection without needing examples of actual anomalies during training, significantly simplifying data collection and model training.

Inventive Principle:

Principle #2Taking out (Extraction)

4Speed

If single best matching data point is used for outlier detection as in prior art, then computational speed is improved, but measurement precision deteriorates due to vulnerability to outlier contamination

Engineering Contradiction:

Improvecomputational speedVSAvoidoutlier detection accuracy

Core Design Contradiction:

SpeedVSMeasurement precision

Solution Approach 1:

The patent merges multiple data points into cluster models rather than relying on single best matching points. Each cluster represents a group of similar nominal observations, and anomaly detection is performed by measuring distance to the cluster centroid or boundary. This aggregation approach makes the system robust to individual outlier contamination while maintaining computational efficiency through vectorized operations.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS9336484B1System and method for outlier detection via estimating clusters

Publication Date: 2016.05.10 UNITED STATES OF AMERICA AS REPRESENTED BY THE ADMINISTRATOR NAT AERONAUTICS & SPACE ADMINISTRATION

AI summary

An efficient method and system for real-time or offline analysis of multivariate sensor data for use in anomaly detection, fault detection, and system health monitoring is provided. Models automatically derived from training data, typically nominal system data acquired from sensors in normally operating conditions or from detailed simulations, are used to identify unusual, out of family data samples (outliers) that indicate possible system failure or degradation. Outliers are determined through analyzing a degree of deviation of current system behavior from the models formed from the nominal system data. The deviation of current system behavior is presented as an easy to interpret numerical score along with a measure of the relative contribution of each system parameter to any off-nominal deviation. The techniques described herein may also be used to “clean” the training data.