Machine Anomaly Root Cause Analysis Using Reconstruction Errors

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current machine monitoring methods detect anomalies but fail to consistently identify their causes, leading to time-consuming manual root cause analysis and potential machine downtimes, especially in large systems where complex sensor data relationships are involved.

Innovation Solution

A method using machine learning-based models that reconstruct operating parameters to determine reconstruction errors, allowing for consistent cause analysis by selecting operating parameters with significant errors, thereby pinpointing the likely faulty machine part.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If machine learning methods are used to detect anomalies, then anomaly detection capability is improved, but consistent root cause analysis capability deteriorates

Engineering Contradiction:
Improveanomaly detection capabilityVSAvoidroot cause analysis consistency
Core Design Contradiction:
Measurement precisionVSLoss of information

Solution Approach 1:

The patent segments the root cause analysis by computing individual reconstruction errors for each operating parameter separately. Instead of treating the anomaly detection as a single black-box output, the method breaks down the overall reconstruction error into parameter-specific components, allowing consistent identification of which specific parameter contributes most to the anomaly.

Inventive Principle:
Principle #1Segmentation

2Loss of information

If manual root cause analysis is performed using expert knowledge, then cause identification is possible, but analysis time increases

Engineering Contradiction:
Improvecause identification capabilityVSAvoidanalysis time
Core Design Contradiction:
Loss of informationVSLoss of time

Solution Approach 1:

The system performs self-service root cause analysis by automatically computing reconstruction errors for each operating parameter and identifying the parameter with the largest error. This eliminates the need for manual expert analysis, as the system autonomously determines the likely cause of the anomaly based on the model's reconstruction performance.

Inventive Principle:
Principle #25Self-service

3Measurement precision

If separate anomaly detection and root cause analysis methods are used, then detection accuracy is improved, but result consistency deteriorates

Engineering Contradiction:
Improveanomaly detection accuracyVSAvoidresult consistency
Core Design Contradiction:
Measurement precisionVSStability of the object's composition

Solution Approach 1:

The patent merges anomaly detection and root cause analysis into a unified framework. The same autoencoder model used for anomaly detection is leveraged to compute parameter-specific reconstruction errors for root cause analysis. This ensures that both functions use the same learned representations and consistency in their results.

Inventive Principle:
Principle #5Merging (Combining)

4Loss of information

If numerous sensors are used to capture operating parameters, then monitoring comprehensiveness is improved, but data processing complexity increases

Engineering Contradiction:
Improvemonitoring comprehensivenessVSAvoiddata processing complexity
Core Design Contradiction:
Loss of informationVSDevice complexity

Solution Approach 1:

The patent extracts only the most relevant information from the complex sensor data by identifying the single operating parameter with the largest reconstruction error. Instead of processing and analyzing all sensor data equally, the method extracts the key diagnostic information needed for root cause identification, simplifying the analysis while maintaining comprehensiveness.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP3876060B1Method and computing unit for analyzing the reason for an anomalous state of a machine
Publication Date: 2023.07.05 ROBERT BOSCH GMBH
  • EP3876060B1 patent drawingFigure 1
  • EP3876060B1 patent drawingFigure 2
  • EP3876060B1 patent drawingFigure 3

AI summary

The invention relates to a method for analyzing the cause of an anomalous state of a machine in a model-based monitoring method that detects anomalous states of the machine by determining an anomaly measure, wherein the model models machine states by reconstructing at least two operating parameters of the machine, which serve as input values ​​of the model, as output values, wherein in the model the output values ​​have deviations from the respective corresponding input values ​​and the model has been trained by machine learning to minimize the deviations.If an anomalous condition is detected by the monitoring procedure, the procedure includes: determining a reconstruction error for each of the operating parameters based on the input value and the output value of the respective operating parameter reconstructed by the model, selecting at least one operating parameter based on the reconstruction errors, and outputting the at least one selected operating parameter.