Production Process Data Similarity Search Using Metadata Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for analyzing production process deviations in industrial plants are inefficient due to the large amount of data that needs to be searched, making it tedious and potentially excluding relevant segments, especially when trying to identify similar process data sequences for quality issues or unwanted behavior.
Innovation Solution
A method that involves providing a process data sequence with a specific data pattern, generating metadata, determining relevant process data sequences based on this metadata, calculating a similarity value, and using this value to identify and sort similar sequences in a database, thereby reducing the search space and improving the quality of search results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a plurality of industrial sensors is used to characterize the production process, then measurement precision and data quality are improved, but the amount of data to be searched increases, making similarity search more difficult and time-consuming
Solution Approach 1:
The patent segments the large multivariate process data into multiple univariate time series, each representing a specific process variable from individual sensors. This segmentation allows the similarity search to be performed separately on each univariate series, dramatically reducing the computational complexity compared to searching the entire multivariate dataset at once, while preserving the measurement precision of each individual sensor.
Solution Approach 2:
The patent extracts and identifies similar patterns across different univariate time series by comparing their temporal characteristics. It separates the similarity search problem from the multivariate context and applies it to individual univariate series, then aggregates the results to identify overall similar process segments, thus reducing the search space while maintaining comprehensive analysis.
2Loss of time
If manual selection of time ranges is used to reduce data amount, then search time is reduced, but relevant segments may be excluded and measurement quality deteriorates
Solution Approach 1:
The patent implements a self-service mechanism where the similarity search algorithm automatically identifies and selects relevant time ranges by comparing temporal patterns across univariate time series. The system autonomously determines which segments are similar without requiring manual intervention, thus reducing search time while ensuring that relevant process deviations are not excluded due to arbitrary manual time range selection.
3Reliability
If the entire multivariate time series is searched for similar patterns, then completeness of analysis is improved, but the complexity of the search process and computational requirements increase significantly
Solution Approach 1:
The patent divides the complex multivariate time series search into multiple simpler univariate time series searches. By segmenting the multivariate data into individual sensor readings and performing similarity search on each univariate series separately, the system maintains analysis completeness across all process variables while significantly reducing the computational complexity and resource requirements compared to searching the entire multivariate dataset as a single entity.
Data Source
AI summary
A method for analysing process data related to a segment of a production process includes providing a process data sequence of the segment of the production process exhibiting a data pattern of at least one process variable to be analyzed; providing a set of metadata; determining process data sequences, which are stored in a first database; determining a start timestamp and end timestamp of each of the determined process data sequence, based on the first database; and calculating a similarity value for each of the determined process data sequences compared to the provided process data sequence, based on the data pattern of the at least one process variable, wherein the determined process data sequences for the calculation are provided, based on the related start timestamps and end timestamps, by accessing a second database comprising the process data sequences, for analysing the process data.