Time-Series Data Overlay for Industrial Machine Learning Validation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies face challenges in confirming whether inappropriate data, acquired under different conditions or operations, are mixed with normal data for machine learning in industrial machines, leading to inappropriate learning models and reduced accuracy in anomaly detection.
Innovation Solution
A learning data confirmation support device that includes a data acquisition unit for acquiring measurement data, a display control unit for aligning and superimposing time-series data on a graph, and a data selection unit for excluding inappropriate data, facilitating the creation of accurate learning models for anomaly detection in industrial machines.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If measurement data is acquired continuously for machine learning, then the quantity of learning data is increased, but inappropriate data acquired under different conditions may be mixed in, reducing learning model accuracy
Solution Approach 1:
The system provides visual feedback by superimposing time-series data on graphs, allowing operators to identify inappropriate data patterns. The display control unit creates visual representations that show deviations from normal operation patterns, enabling feedback-based data quality assessment and selection.
Solution Approach 2:
The display control unit acts as an intermediary between the raw measurement data and the learning process. It transforms numerical time-series data into visual graph representations, allowing operators to intermediate-select appropriate data before it enters the machine learning model, thus preventing inappropriate data from degrading model accuracy.
2Manufacturing precision
If manual confirmation of each piece of time-series data is performed to exclude inappropriate data, then the accuracy of learning model is improved, but the time and labor required for data preparation is greatly increased
Solution Approach 1:
The system creates visual copies (graph representations) of the time-series data that preserve the essential patterns and characteristics. These visual copies allow operators to quickly assess data quality without examining raw numerical data point-by-point, significantly reducing the time required for data confirmation while maintaining accuracy.
Solution Approach 2:
The display control unit uses visual representation where different patterns and deviations can be distinguished through graphical variations. This visual differentiation allows operators to quickly identify inappropriate data patterns without manual numerical analysis, reducing preparation time while maintaining model accuracy.
3Quantity of substance
If inappropriate data is included in learning data, then the quantity of learning data is maintained, but the accuracy of anomaly detection is reduced
Solution Approach 1:
The visual display system provides feedback on data quality by showing superimposed time-series patterns. Operators can identify inappropriate data that would degrade anomaly detection accuracy and exclude it before training the learning model, thus maintaining both data quantity and detection accuracy.
Data Source
AI summary
To facilitate confirmation as to whether or not it is data measured by the same operation upon acquiring measurement data in an industrial machine. A learning data confirmation support device 3 that facilitates confirmation of contamination of inappropriate data when learning data including only normal data are acquired in advance, in order to detect an anomaly of an industrial machine using machine learning, includes a data acquisition unit 31 that acquires measurement data including time-series data representing at least one of a predetermined state quantity or control quantity relating to control when the industrial machine is made to perform a certain operation; and a display control unit 32 that aligns a plurality of pieces of time-series data acquired by the data acquisition unit in a direction of a time axis and, in this state, superimposes a same type of pieces of data of the time-series data to display in a graph.