AI Anomaly Detection With Reconstruction Error Explanation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing AI-based anomaly detection systems, particularly those using auto-encoders, lack transparency in explaining which input variables contribute to high reconstruction errors, hindering trust and effective error correction in industrial processes.

Innovation Solution

A method utilizing an auto-encoder for anomaly detection in technical devices, combined with an evaluation device and analysis device, to determine and classify reconstruction errors, and generate rules for improving the system based on historical data and variable importance, enabling accurate cause analysis and automated correction.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If an auto-encoder AI model is used for anomaly detection, then the detection effectiveness is improved, but the interpretability and explainability of the detection results deteriorate

Engineering Contradiction:
Improveanomaly detection effectivenessVSAvoidinterpretability of detection results
Core Design Contradiction:
ReliabilityVSLoss of information

Solution Approach 1:

The patent introduces an intermediary explanation module that acts as a bridge between the auto-encoder's internal representations and human interpreters. This module generates natural language explanations that translate the model's reconstruction error analysis into understandable insights about which input variables and their relationships caused the anomaly, thereby maintaining detection effectiveness while restoring interpretability.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent segments the anomaly explanation into distinct components: identification of problematic input variables, analysis of their states/values, examination of relationships between variables, and generation of actionable insights. This segmentation allows the complex interpretation task to be broken down into manageable parts that can be systematically processed and presented.

Inventive Principle:
Principle #1Segmentation

2Loss of information

If simpler AI models like decision trees are used in parallel with deep learning models, then the interpretability is improved, but the device complexity increases

Engineering Contradiction:
Improveinterpretability of detection resultsVSAvoidsystem architecture complexity
Core Design Contradiction:
Loss of informationVSDevice complexity

Solution Approach 1:

The patent extracts the interpretability function from separate parallel models and integrates it directly into the auto-encoder architecture through explanation modules. This extraction eliminates the need for maintaining separate simpler models, as the explanation capabilities are now embedded within the deep learning model itself, reducing overall system complexity while preserving interpretability.

Inventive Principle:
Principle #2Taking out (Extraction)

3Measurement precision

If more detailed analysis of input variables and their relationships is performed, then the diagnostic accuracy is improved, but the computational complexity increases

Engineering Contradiction:
Improvediagnostic accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent performs preliminary analysis during the training phase, where the model learns not only to minimize reconstruction error but also to identify and weigh the importance of different input variables and their relationships. This preliminary action embeds diagnostic knowledge into the model weights and structures, enabling accurate variable prioritization during runtime without requiring complex real-time computations.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent changes the parameters being optimized during training to include not only reconstruction error but also metrics related to variable importance and relationship strength. By modifying the objective function and training parameters, the model automatically learns to prioritize the most diagnostic variables and relationships, achieving high diagnostic accuracy with computationally efficient inference.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4505254B1Computer-implemented method and system for anomaly detection
Publication Date: 2025.12.31 SIEMENS AG OESTERR
  • EP4505254B1 patent drawingFigure 1~2
  • EP4505254B1 patent drawingFigure 3
  • EP4505254B1 patent drawing

AI summary

The invention relates to a computer-implemented method for anomaly detection during the control of a technical device (TD) by a control means (PLC), comprising the following steps: a) providing input data (IN), coding the input data (IN) with the aid of an auto-encoder to form coded data (C), and decoding the coded data (C) with the aid of the auto-encoder to form output data (OUT), and determining and classifying a reconstruction error (RE), b) determining a target variable (TV) from the reconstruction error (RE) if the reconstruction error (RE) exceeds a predefined reconstruction error threshold (RET), c) providing a model on the basis of artificial intelligence to an evaluation device (XPL), d) determining, by the evaluation device (XPL), a prediction for the target variable (TV) using the model and the classification of the reconstruction error (RE), and providing the prediction to an analysis device (REA), e) determining, by the analysis device (REA), a rule for improving the reconstruction error (RE) from the prediction and a predefined rule set, providing the rule to the control means (PLC) and controlling, by the control means (PLC), the technical device (TD) with the aid of the rule.