Time-Series Signature Labeling from System State Changes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The manual process of labeling time series data for machine learning methods is labor-intensive and restricts the use of labeled data, as it requires significant human effort and expertise, limiting the generation of sufficient training data for pattern recognition in technical processes.
Innovation Solution
A method for automatically generating labeled signatures in time series data by processing recorded data from technical installations, adjusting quantization based on status changes, and assigning identifiers to signature sections, enabling automated labeling and reducing manual effort.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling is performed by personnel with plant and process knowledge, then labeling accuracy is improved, but time consumption and effort increase significantly
Solution Approach 1:
The system performs automatic self-labeling by utilizing its own recorded state values and data time series. The labeling process does not require external human intervention but is executed autonomously by the system through automated analysis of its operational data, thereby eliminating time-consuming manual labeling while maintaining accuracy through systematic processing
Solution Approach 2:
The system changes the parameter of labeling from manual human operation to automated computational processing. By transforming the labeling task into an automated data processing operation that analyzes state values and time series data, the system achieves both speed and accuracy without requiring human time investment
2Reliability
If manual labeling is performed, then contextual information can be accurately input, but the process becomes very time-consuming and limits the use of machine learning methods
Solution Approach 1:
The system generates its own training data autonomously by processing its operational data time series and state values. This self-service capability enables unlimited generation of labeled training data without human intervention, dramatically increasing productivity while maintaining reliability through systematic automated processing
Solution Approach 2:
The system performs preliminary data preparation by recording and storing state values and data time series during normal operation. This preliminary action creates a ready pool of raw data that can be automatically processed into labeled training data on demand, enabling high throughput without compromising data quality
3Quantity of substance
If synthetic test data is generated from previously labeled data, then new data sets can be derived, but fully automatic assignment is still not achieved
Solution Approach 1:
The system completely automates the labeling process by having it perform its own labeling without relying on previously manually labeled data. The system independently analyzes its operational data and assigns labels automatically, achieving full automation while generating abundant training data for machine learning applications
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The invention relates to a method for automatically generating labelled signatures (30), wherein: firstly a quantisation of a state time series (16) recorded with regard to a system (12) is adapted so that all states included in the state time series (16) have at least one state duration longer than a minimum time; subsequently, system states (32) are determined on the basis of state changes in at least one state time series (16); and the sections of a synchronous data time series (14) recorded with regard to the system (12), said sections being between the locations of two successive system states (32), are determined as signatures (30), and these are provided with an identifier (34) which comprises at least one system state (32) which determines the signature (30).