Digital Twin Anomaly Detection for Industrial Time-Series Monitoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current anomaly detection systems in complex technical systems face challenges in achieving a balance between detection sensitivity and false alarms, particularly in large-scale industrial monitoring systems, where rule-based methods are resource-intensive and prone to false positives, and machine learning approaches require significant labeled data.
Innovation Solution
A semi-supervised deep anomaly detection method using digital twin simulation data, which generates a large amount of labeled data with minimal human interaction, employing a Siamese auto-encoder neural network to differentiate between regular and anomalous operating states by processing triples of data samples from both simulated and real-world sensor data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If rule-based monitoring systems are used to detect anomalies in complex technical systems, then detection coverage can be improved by capturing relevant anomalous signals, but the system generates many false alarms and requires significant expert knowledge to formulate rules
Solution Approach 1:
The patent creates a digital twin (virtual copy) of the complex technical system that replicates its behavior and operational states. This digital twin is trained on historical sensor data to learn normal operational patterns. During monitoring, the digital twin generates expected sensor readings that are compared against actual readings, automatically detecting anomalies without requiring manual rule formulation. This copying approach eliminates the need for complex expert-defined rules while maintaining high detection accuracy.
Solution Approach 2:
The patent replaces the mechanical rule-based monitoring system with a data-driven machine learning approach. Instead of using manually crafted rules and thresholds, the system employs neural networks and statistical models that automatically learn from historical data. This substitution transforms the monitoring mechanism from a static rule-based system to a dynamic adaptive system that improves over time without increasing rule complexity.
2Reliability
If machine learning approaches are used for anomaly detection, then detection performance can be improved, but significant labeled data is required which increases resource demand
Solution Approach 1:
The patent performs preliminary action by pre-training the digital twin on extensive historical sensor data during an offline phase. This pre-training establishes a robust baseline understanding of normal operational patterns before actual anomaly detection begins. By preparing the model in advance with available historical data, the system reduces the need for continuous labeled anomaly data during operational monitoring, as the digital twin already possesses learned knowledge of system behavior.
Solution Approach 2:
The patent implements self-service by enabling the digital twin to continuously learn and update its understanding of normal operations from incoming sensor data without requiring external labeled anomaly data. The system serves itself by using its own operational data to refine its detection capabilities, automatically adapting to new patterns and conditions while maintaining high detection performance without additional resource-intensive labeled data collection.
Data Source
AI summary
A computer-implemented method for detecting an anomalous operating status of a technical system. A training phase obtains a first set of time-series values generated by a digital twin simulation of the technical system for a regular operating status and a second set of time-series values measured by sensors in an anomalous operating status, and adjusts parameters of a machine learning model for detecting the regular operating status and for discriminating data samples of the regular operating status from data samples of the anomalous operating status to generate a trained machine learning model. A monitoring phase obtains a set of multivariate time-series values measured by the sensors, calculates an anomaly score value for determining whether the technical system is in an anomalous operating status based on the obtained set of multi-variate time-series values and the trained machine learning model, and outputs a signal including information on the determined anomalous operating status.