Explainable Time-Series Anomaly Detection for Equipment Data

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for anomaly detection in time series data, particularly in industrial equipment monitoring and cybersecurity, rely on artificial intelligence algorithms that operate as 'black boxes' and are not understandable by human operators, and current statistical tools fail to account for time series data effectively.

Innovation Solution

A method using descriptive vectors and machine learning-based anomaly detection, combined with explainable models like isolation forests and Shapley values, to identify anomalies and provide understandable explanations, enabling predictive maintenance and cybersecurity enhancements.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If artificial intelligence algorithms are used for anomaly detection in time series data, then detection capability is improved, but understandability by human operators deteriorates

Engineering Contradiction:

Improveanomaly detection capabilityVSAvoidunderstandability of detection results

Core Design Contradiction:

Measurement precisionVSLoss of information

Solution Approach 1:

The patent introduces an intermediary explanation layer that translates black-box AI anomaly detection results into human-understandable insights. This intermediary component analyzes the AI model's decision-making process and generates explanatory narratives that bridge the gap between complex algorithmic outputs and human comprehension, allowing operators to understand why anomalies were detected without sacrificing detection accuracy.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Ease of operation

If statistical tools are used for anomaly detection, then ease of operation is improved, but effectiveness in handling time series data deteriorates

Engineering Contradiction:

Improvesimplicity of detection methodVSAvoidanomaly detection accuracy

Core Design Contradiction:

Ease of operationVSMeasurement precision

Solution Approach 1:

The patent merges statistical tools with machine learning algorithms to create a hybrid anomaly detection system. The statistical components provide simplicity and interpretability, while the machine learning components handle the complexity of time series patterns. This combination allows the system to maintain ease of operation through statistical methods while achieving high detection accuracy through machine learning's ability to model temporal dependencies.

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentEP4303775B1Method and device for anomaly detection and associated explanation determination in data time series

Publication Date: 2025.09.17 THALES SA

Figure 1
Figure 2
Figure 3

AI summary

The invention relates to a method and device for detecting anomalies and determining associated explanations in time series of data, each time series having an associated duration and comprising a number N of operating values of equipment or a system.The process includes the following steps: - projection (40), for each time series, of the values of the time series onto a number M of basis functions and obtaining a descriptive vector of size M, - training (42) of a parameterized anomaly determination model, by machine learning, on calculated descriptive vectors, - application (44) of this model to the descriptive vectors and obtaining an anomaly score per time series, - determination (46) of a subset of time series containing an anomaly, and for each time series of said subset, determination (48) of at least a part of the values of said descriptive vector having an influence in obtaining the anomaly score of said time series.