Process State Visualization Using Cluster Trends and Evaluation Values
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In manufacturing processes, changes in process states such as quality of materials, equipment conditions, and worker skills lead to frequent and discontinuous changes, causing false alarms in abnormality diagnosis and missing potential quality issues due to unmeasured factors, which existing multi-dimensional data analysis methods struggle to detect.
Innovation Solution
A process state analysis device with an evaluation value calculation unit and a graph creation unit that calculates evaluation values for clusters based on multi-dimensional process data and creates graphs representing the number of nodes in each cluster over time, allowing for intuitive visualization of changes and detection of unmeasured factors by adjusting the evaluation value calculation range and color reference values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If abnormality diagnosis is performed using evaluation value based on measured value, then abnormality detection capability is improved, but false alarms increase and unmeasured factor changes are missed
Solution Approach 1:
The patent transitions from one-dimensional measured value analysis to two-dimensional analysis by introducing cluster identification information as an additional dimension. This allows the system to detect both measured value abnormalities and unmeasured factor changes by analyzing the relationship between evaluation values and cluster classifications, thereby reducing false alarms while maintaining detection capability.
2Reliability
If multi-dimensional data analysis is used to detect process changes, then abnormality detection is improved, but changes in unmeasured factors cannot be directly detected
Solution Approach 1:
The patent introduces cluster identification information as an intermediary that bridges measured values and unmeasured factors. By classifying data into clusters based on multiple parameters and tracking cluster identification over time, the system can infer unmeasured factor changes even though these factors are not directly measured, thus preventing information loss.
3Adaptability or versatility
If clustering is performed on process data to identify patterns, then process state classification is improved, but false alarms occur due to frequent discontinuous changes
Solution Approach 1:
The patent implements feedback by comparing current cluster identification with historical cluster identification trends. When discontinuous changes occur in the process, the system uses the time-series information of cluster identification to distinguish between normal process variations and actual abnormalities, providing feedback that reduces false alarms while maintaining adaptive classification capability.
Data Source
AI summary
An embodiment of the process state analysis device of the present invention is provided with an evaluation value calculation unit and a graph creation unit. The evaluation value calculation unit calculates, within an evaluation value calculation range indicating a target range for calculating evaluation values, an evaluation value for each cluster that is classified on the basis of multi-dimensional process data output from each measurement device. The graph creation unit determines a hue for a graph element for each cluster on the basis of the evaluation value for the cluster as calculated within the evaluation value calculation range, and on the basis of a color reference evaluation value corresponding to a reference hue for the graph element, and creates and outputs a graph representing, for each aggregation unit time interval in a specified display period, the number of nodes belonging to each cluster.