Anomaly Detection in Sensor Recordings Using Normalizing Flows
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Technical systems face challenges in detecting anomalies in sensor recordings, as erroneous signals from defective sensors can disrupt system operation, and existing methods lack accuracy in identifying the source of anomalous behavior.
Innovation Solution
A computer-implemented method using machine learning techniques, specifically normalizing flows, to ascertain first and second anomalous values from sensor recordings, allowing for the detection and identification of anomalous sensors, which can be temporarily or permanently disregarded to maintain system correctness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If redundant sensors are used to detect anomalies, then reliability of anomaly detection is improved, but device complexity increases
Solution Approach 1:
The anomaly detection function is segmented into two distinct evaluation stages: (1) individual sensor recording assessment using first anomaly detection models, and (2) cross-sensor consistency verification using second anomaly detection models. This segmentation allows the system to process sensor data in a structured manner, improving detection reliability while managing complexity through modular processing.
Solution Approach 2:
The patent introduces an intermediary evaluation mechanism where second anomaly detection models act as mediators to assess whether sensor recordings are consistent with other sensor measurements. This intermediary layer resolves conflicts between individual sensor readings and collective sensor behavior, enhancing reliability without requiring direct complex interactions between all sensor pairs.
2Measurement precision
If multiple anomaly detection models are applied to each sensor recording, then measurement precision of anomaly identification is improved, but computational resources required increase
Solution Approach 1:
The first anomaly detection models perform preliminary assessment of each sensor recording individually before the more computationally intensive second anomaly detection models are applied. This preliminary filtering ensures that only recordings requiring further verification undergo cross-sensor consistency checks, reducing overall computational energy consumption while maintaining high identification precision.
Solution Approach 2:
Different levels of computational analysis are applied to different sensor recordings based on their individual anomaly scores. Recordings with high anomaly scores from the first model undergo the full two-stage evaluation, while normal recordings require minimal processing. This local quality approach optimizes energy consumption by concentrating computational resources where they are most needed for precise anomaly identification.
Data Source
AI summary
A computer-implemented method for detecting anomalies in a plurality of sensor recordings of a technical system. The method includes: ascertaining a first anomalous value which, with regard to all sensor recordings of the plurality of sensor recordings, characterizes whether or not an anomaly is present; ascertaining a plurality of second anomalous values, each second anomalous value corresponding to a sensor recording of the plurality of sensor recordings, and with regard to the sensor recording characterizing whether or not an anomaly is present in other sensor recordings of the plurality of sensor recordings; detecting an anomaly in a sensor recording if the first anomalous value characterizes the presence of an anomaly, and the second anomalous value corresponding to the sensor recording characterizes no anomaly, and the second anomalous value differs beyond a predefined extent from other second anomalous values of the plurality of second anomalous values.