Anomaly Detection Overview

Falkonry TSI Platform supports AI-based anomaly detection method for operational time series and metrics data to reveal unexpected trends, waveforms, shapes, and noise levels in both time series as well as metrics data. Its primary purpose is to automatically detect deviations from normal operating behavior without requiring any manual setup or supervision. The findings enable engineers to expedite root cause analysis of anomalous and abnormal behaviors.

AI-Driven and Self-Supervised**

Falkonry's anomaly detection leverages Falkonry's globally patented PatternIQ™ technology, which uses a self-supervised approach to discover and explain patterns and anomalies from high-resolution time series data. This is crucial because real-world systems often have scarce actual behavior data and require the detection of novel patterns that lack known signatures.

Deep Neural Network Architecture**

PatternIQ™ incorporates a deep neural network (DNN) learning architecture designed to learn multi-timescale embeddings. These embeddings capture common signal shapes, waveforms, and value distributions.

Anomaly Scoring**

Anomaly detection is performed using a normalized reconstruction error, which provides a unified comparison scale across various sensor data, regardless of their heterogeneous physical quantities. The process produces a real-valued anomaly score, where scores above 3 typically indicate rare or significant deviations from expected behavior.

Purpose and Benefits

No-Code and Automated

Falkonry's Anomaly Detection is designed to operate without requiring any code or machine learning (ML) expertise from the user. It gets to work as soon as enough data accumulates, finding anomalies effortlessly wherever real-time data is available.

Visualization and Prioritization

Detected anomalies are presented to engineers or operators as a heatmap on the Falkonry interface. This heatmap highlights unusual behavior and helps users prioritize which anomalies to investigate based on their severity and persistence. Different color gradients indicate varying degrees of unusualness, from normal (purple) to significant deviations (yellow).

Contextual Understanding

The findings from anomaly detection enable engineers to expedite root cause analysis of anomalous and abnormal behaviors by providing explanations and comparable scores across signals.

Integration with Rules

The anomaly scores generated by AI can be used as inputs for Falkonry's Rules engine to create smart alerts. For example, a rule can be set to alert when any anomaly score is above a certain value for a specified duration, or when multiple signals within an asset show high anomaly scores.

In essence, Falkonry's Anomaly Detection automates the complex task of finding unexpected behaviors in your data, providing clear, actionable anomaly scores and visualizations to guide timely interventions.