Machine-Learning Fault Detection With Dynamic Time-Sequential Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for predicting machine errors in smart factories rely on fixed, time-irrelevant thresholds set by operators, making it difficult to accurately detect machine malfunctions or product defects, especially in dynamic operational data patterns.
Innovation Solution
The method involves collecting time-sequential operation data, dividing it into predetermined intervals, and using machine-learning techniques to generate dynamic threshold data, which are then used to detect deviations and determine error events, providing real-time feedback to operator devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If fixed thresholds are used for error detection, then the system is simple to operate, but the detection accuracy deteriorates because thresholds cannot adapt to dynamic operational patterns
Solution Approach 1:
The patent applies dynamics by transforming fixed thresholds into dynamic, time-sequential thresholds that automatically adapt to changing operational patterns. The system collects historical operation data, divides it into time intervals, and generates thresholds that evolve with the machine's operational state, thereby improving detection accuracy without requiring manual intervention.
Solution Approach 2:
The system implements self-service by automatically generating and updating thresholds using machine learning algorithms without operator intervention. The threshold generation unit autonomously processes operation data and produces optimized thresholds, eliminating the need for manual threshold setting and adjustment while maintaining high detection accuracy.
2Measurement precision
If time-sequential threshold data are generated using machine-learning techniques, then error detection accuracy is improved, but the computational complexity increases
Solution Approach 1:
The patent applies segmentation by dividing the continuous operation data into discrete time intervals and processing each segment separately. This approach breaks down the complex computational task into manageable chunks, allowing the system to generate time-sequential thresholds without overwhelming computational burden while maintaining detection accuracy.
Solution Approach 2:
The system performs preliminary action by pre-processing and organizing operation data into standardized time intervals before threshold generation. This preparatory step structures the data in advance, making subsequent machine learning computations more efficient and reducing the overall computational complexity of the error detection process.
3Reliability
If fixed thresholds are set by operators, then the system is easy to implement, but it cannot accurately detect errors in dynamic operational patterns
Solution Approach 1:
The system implements self-service by automatically generating thresholds through machine learning algorithms without requiring operator expertise in threshold setting. The threshold generation unit autonomously analyzes operation data and produces optimized thresholds, maintaining ease of operation while dramatically improving detection reliability through adaptive, data-driven threshold selection.
4Measurement precision
If separate thresholds are set for each production time section, then detection accuracy improves, but the number of thresholds to manage increases
Solution Approach 1:
The patent applies dynamics by creating a single set of time-sequential thresholds that automatically adapt to different production time sections based on the temporal patterns learned from data. Instead of manually managing separate static thresholds for each section, the system generates dynamic thresholds that evolve with time, achieving high detection precision while simplifying threshold management.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
According to the present disclosure, time-sequential threshold data can be automatically detected by a server and thus can be compared with operation data in all of time domains. Therefore, it is not necessary for an operator to input threshold data by hand. Further, according to the present disclosure, it is possible to precisely detect an error of a machine or a defect of a product which has not been conventionally recognized at the time of setting a threshold (absolute value).