Sensor Anomaly Detection With Hierarchical Material Flow Forecasting

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing sensor anomaly detection algorithms in complex systems suffer from high false positive rates and require substantial labeled data, making them inefficient and unreliable.

Innovation Solution

A machine learning model incorporating relational constraints of the system's structure, using a Gradient-based Reconciling Propagation algorithm to learn a projection matrix that projects base forecasts into a hierarchically-coherent solution space, reducing false positives and improving training efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If simple parametric models like Gaussian distributions are used for anomaly detection, then the method is easy to implement, but the false positive rate is high

Engineering Contradiction:
Improveease of implementationVSAvoidfalse positive rate
Core Design Contradiction:
Ease of manufactureVSReliability

Solution Approach 1:

The patent transforms the anomaly detection problem from direct parameter comparison to a reconstruction error-based approach. By changing the fundamental parameter being analyzed (from raw sensor values to reconstruction residuals), the system achieves both simplicity and reliability. The autoencoder architecture automatically learns meaningful parameter transformations without requiring manual feature engineering.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent replaces traditional statistical mechanical models (Gaussian distributions, mean-variance calculations) with a neural network-based reconstruction system. This substitution allows the system to capture complex non-linear relationships in sensor data while maintaining computational efficiency through the autoencoder's inherent dimensionality reduction capabilities.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If machine learning algorithms are trained with labeled data to detect anomalies, then detection accuracy is improved, but data requirements and training complexity increase

Engineering Contradiction:
Improvedetection accuracyVSAvoiddata requirements
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The autoencoder performs self-service by using unlabeled normal operation data to train itself. The model learns to reconstruct normal patterns automatically, and any deviation from these learned patterns is flagged as an anomaly. This self-supervised approach eliminates the need for manual labeling while maintaining high detection accuracy.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The system performs preliminary action by pre-training the autoencoder on extensive normal operation data before deployment. This preliminary training phase allows the model to establish a robust baseline of normal behavior, enabling it to detect anomalies with high accuracy using minimal additional labeled data during actual operation.

Inventive Principle:
Principle #10Preliminary action

3Adaptability or versatility

If historical data is used to establish normal behavior models, then the system adapts to actual operations, but the model may not capture complex system relationships

Engineering Contradiction:
Improvesystem adaptabilityVSAvoidmodel complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The autoencoder employs a nested structure where hidden layers are contained within the overall network architecture. This nesting allows the model to capture hierarchical relationships in sensor data, with each layer learning progressively more abstract features. The nested structure enables the system to adapt to complex system relationships while maintaining a manageable model through systematic organization.

Inventive Principle:
Principle #7Nested doll (Nesting)

Solution Approach 2:

The patent applies dimensionality change by transforming high-dimensional sensor data into a lower-dimensional latent space during encoding, then reconstructing it during decoding. This dimensional transformation allows the model to capture essential system relationships in a compressed form, adapting to complex patterns without requiring proportionally complex model structures.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentEP4286970B1Method and system for detecting sensor anomalies
Publication Date: 2025.09.24 SIEMENS AG
  • EP4286970B1 patent drawingFigure 1~2
  • EP4286970B1 patent drawingFigure 3
  • EP4286970B1 patent drawingFigure 4

AI summary

For detecting sensor anomalies, a machine learning model models a material flow in an industrial system, in particular in a production line, as a hierarchical time series, wherein the hierarchical time series represents a structure of the material flow using a directed acyclic graph with a set of nodes and a set of edges, wherein each node is associated to a time series, and wherein the edges represent parent-child relations where each value of a time series at a parent node equals the sum of the respective values of its child nodes. The machine learning model forecasts (OP1) predicted time series values for all nodes. Current sensor measurements received (OP2) from sensors placed in the industrial system are compared (OP4) to the predictions of the machine learning model. An anomaly is detected (OP5) if the difference exceeds a threshold. The described embodiments improve the performance of sensor anomaly detection by incorporating additional domain knowledge about the structure of the system in the form of relational constraints, reducing the prediction error for anomaly detection in problems involving material flow (lower false positive rate), reducing costs that are associated with false or missed anomalies, and increasing the trustworthiness and ease of use in adopting AI-based algorithms. Increased training efficiency is achieved by leverag-ing domain knowledge, requiring less data to achieve a highly performant model.