Frequency Domain Anomaly Detection for IT Infrastructure
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional anomaly detection systems in IT infrastructure are complex, resource-intensive, and often detect incidents too late, as they rely on labeled time domain data and machine learning algorithms that require significant processing power and memory, making them unsuitable for real-time monitoring and missing subtle incidents that develop over minutes or hours.
Innovation Solution
The system transforms measurement signals from IT infrastructure elements into the frequency domain using Discrete Fourier Transformation (DFT) and sliding windows to detect point anomalies based on energy content, employing adaptive thresholds and statistical models to identify incidents by analyzing high-frequency components and their density over time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning algorithms and pattern recognition techniques are used for anomaly detection, then detection accuracy is improved, but processing power and memory requirements increase significantly
Solution Approach 1:
The patent extracts only the essential features needed for anomaly detection by transforming signals to the frequency domain and analyzing specific frequency components. This extraction approach eliminates the need for complex machine learning models while maintaining detection accuracy, as the frequency transformation reveals characteristic patterns that are sufficient for identifying anomalies without requiring extensive computational resources.
Solution Approach 2:
The patent replaces expensive, resource-intensive machine learning algorithms with simpler, computationally efficient frequency domain analysis methods. This substitution uses lightweight mathematical transformations (Fourier transform) that require minimal processing power and memory, making the system suitable for real-time monitoring applications where resources are constrained.
2Reliability
If machine learning techniques are used for real-time anomaly detection, then detection capability is improved, but system complexity increases
Solution Approach 1:
The patent replaces complex machine learning systems with a simpler frequency domain analysis approach. By substituting the mechanical complexity of ML algorithms with mathematical signal processing techniques, the system achieves real-time detection capability with reduced complexity. The frequency transformation and energy calculation provide a straightforward, deterministic method that is easier to implement and maintain than ML-based solutions.
3Reliability
If conventional anomaly detection systems wait for incidents to occur before detection, then false positives are reduced, but detection time increases
Solution Approach 1:
The patent performs preliminary analysis by continuously transforming signals to the frequency domain and monitoring frequency components in real-time. This preliminary action enables the system to detect anomalies as they begin to develop, rather than waiting for full incident formation. By maintaining continuous frequency domain monitoring, the system can identify characteristic frequency patterns that indicate emerging issues, reducing detection time while maintaining reliability.
Data Source
Figure 1
AI summary
Method for computing of anomalies based on frequency driven transformation of a at least one measured signal from a information technology infrastructure element, wherein the measured signal is continuously processed as a digitized signal having a discrete number of measurement points, each measurement point having a same predetermined distance on a time axis and the digitized signal is transformed into a Fourier signal in a frequency space, wherein the computing of anomalies comprises repeating the following steps: calculating an energy content of a predetermined number of high frequency components of the Fourier signal, determining whether or not a point anomaly has occurred based on a comparison wherein of the calculated energy content with a learned threshold, and storing a first predetermined number of calculated energy content.