Press Machine Sensor Modeling for Early Abnormality Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing press machine monitoring systems are limited in their ability to comprehensively predict various abnormalities, often failing to detect signs of failure effectively.
Innovation Solution
A press machine equipped with multiple sensors and an information processing device that performs machine learning to generate learning models from collected data, calculates predicted values, and determines the degree of abnormality based on differences between actual and predicted values, enabling early detection of potential failures.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple sensors and machine learning models are used to comprehensively monitor press machine abnormalities, then the reliability of abnormality detection is improved, but the device complexity increases
Solution Approach 1:
The patent applies universality by creating a comprehensive monitoring system that uses multiple sensors (vibration, temperature, pressure, acoustic emission) to detect various types of abnormalities simultaneously. The machine learning model serves as a universal analytical engine that processes data from all sensor types through a common framework, enabling the system to detect mold abnormalities, machine component failures, and operational anomalies with a single integrated approach rather than separate specialized systems for each abnormality type.
2Measurement precision
If machine learning models process multiple sensor data to predict abnormalities, then the measurement precision of abnormality detection is improved, but the loss of time for data processing increases
Solution Approach 1:
The patent applies preliminary action by pre-training machine learning models during an initial phase using historical sensor data from normal and abnormal operations. This pre-processing of data and training of models enables the system to make rapid real-time predictions during actual operation, as the computational heavy lifting has already been performed during the offline training phase. The model learns patterns and thresholds in advance, allowing for quick anomaly detection without extensive real-time computation.
Solution Approach 2:
The patent applies skipping by implementing a two-stage detection approach: first, simple threshold-based filtering is applied to rapidly identify obviously normal or abnormal conditions, and only cases that require detailed analysis are passed to the full machine learning model. This allows the system to quickly process most data points with minimal computation and reserve intensive processing only for borderline cases, thereby reducing overall processing time while maintaining detection precision.
Data Source
Figure 1~2
Figure 3
Figure 4~5
AI summary
A press machine includes: a learning-model generating unit (101) that uses one data from among a plurality of data collected from a plurality of sensors, as an objective variable, and uses data other than the one data as an explanatory variable to perform machine learning to generate a learning model for the one data, the generation being performed for all of the plurality of data; a predicted-value calculating unit (102) that inputs an actually measured value of data other than one data from among the plurality of data collected from the plurality of sensors, into the learning model for the one data to calculate a predicted value of the one data, the calculation being performed for all of the plurality of data; a degree-of-abnormality calculating unit (103) that calculates a degree of abnormality based on a difference between an actually measured value and a predicted value of the plurality of data; and a degree-of-abnormality outputting unit (104) that outputs the calculated degree of abnormality.