Anomaly Explanation Vector for Interpretable ML Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing anomaly detection methods using machine learning algorithms struggle to provide interpretable explanations for anomaly detection results, especially when dealing with complex time series data from sensors, leading to a lack of trust and understanding in the underlying models.

Innovation Solution

A computer-implemented anomaly determination explanation method that determines an explanation vector by calculating an anomaly measure from an input vector using a state evaluation model, normalizing the input vector to represent a normal state, and then estimating the contribution of each component to the anomaly determination.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If machine learning algorithms are used for anomaly detection, then detection accuracy is improved, but interpretability deteriorates

Engineering Contradiction:
Improveanomaly detection accuracyVSAvoidmodel interpretability
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the anomaly detection system into two distinct components: (1) a machine learning-based anomaly detection module that maintains high detection accuracy, and (2) a separate explanation generation module that produces human-interpretable explanations. This segmentation allows each component to optimize for its specific function without compromising the other.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary explanation generation mechanism that acts as a bridge between the black-box anomaly detection model and human users. This intermediary translates the model's internal decision logic into comprehensible explanations, thereby maintaining both detection accuracy and interpretability.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If complex time series data is analyzed, then anomaly detection capability is improved, but computational complexity increases

Engineering Contradiction:
Improveanomaly detection capabilityVSAvoidcomputational complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent extracts and processes only the essential features from complex time series data that are most relevant for anomaly detection. By selectively extracting key temporal patterns and dependencies rather than processing all raw data, the system maintains high detection capability while reducing computational burden.

Inventive Principle:
Principle #2Taking out (Extraction)

3Reliability

If explanations are generated for anomaly detection, then user trust is improved, but computational overhead increases

Engineering Contradiction:
Improveuser trustVSAvoidcomputational overhead
Core Design Contradiction:
ReliabilityVSUse of energy by stationary object

Solution Approach 1:

The patent implements a partial explanation generation approach where explanations are produced only for the specific features that contributed to anomaly detection, rather than generating comprehensive explanations for all input features. This selective explanation strategy reduces computational overhead while still providing sufficient interpretability to build user trust.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP4557179A1Framework for human-comprehensible explanation of detected anomalies
Publication Date: 2025.05.21 SIEMENS AG
  • EP4557179A1 patent drawingFigure 1~2
  • EP4557179A1 patent drawingFigure 1~2
  • EP4557179A1 patent drawing

AI summary

The invention relates to a method and a device for determining reasons for detection of an anomalous system state. To this end, a normalized system state is determined as being the closest state to the anomalous system state that is determined not to be an anomalous system state.