Machine-Learning Abnormality Diagnosis for Manufacturing Signals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining abnormalities in manufacturing facilities are unreliable due to the difficulty in setting threshold values and the lack of robustness in distinguishing between normal and abnormal states, often leading to inconsistent results across different analysis methods and requiring extensive data collection for abnormal cases.
Innovation Solution
An abnormality determination support apparatus that uses a combination of data preparation, primary determination, and secondary determination units, employing machine learning with multiple analysis methods to convert data into numerical indices, and utilizing a machine learning device for accurate abnormality detection independent of threshold values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If traditional threshold-based determination methods are used, then the determination process is simple, but the reliability of abnormality determination deteriorates due to inconsistent results across different analysis methods and difficulty in setting threshold values
Solution Approach 1:
The patent introduces a determination support device as an intermediary system that collects determination results from multiple analysis methods and uses machine learning to synthesize a final determination. This mediator integrates outputs from different analysis approaches (vibration analysis, temperature analysis, etc.) and applies learned patterns to achieve reliable abnormality determination without requiring manual threshold setting for each method.
Solution Approach 2:
The patent transforms the determination problem from threshold-based parameter comparison to machine learning-based pattern recognition. Instead of setting fixed thresholds for various parameters (vibration amplitude, temperature, etc.), the system uses machine learning models that automatically learn optimal determination parameters from training data, adapting to different operating conditions and eliminating the need for manual threshold adjustment.
2Measurement precision
If multiple analysis methods are used to improve determination accuracy, then the measurement precision improves, but the device complexity increases due to multiple analysis methods and threshold settings
Solution Approach 1:
The patent merges multiple analysis methods into a unified determination framework. Instead of maintaining separate threshold-based determination systems for each analysis method (vibration, temperature, pressure, etc.), the system collects results from all methods and combines them through a single machine learning model that processes multiple input features simultaneously, achieving accurate determination while simplifying the overall system architecture.
Solution Approach 2:
The determination support device serves as a universal system that handles multiple analysis methods through a single machine learning model. The model is designed to accept various types of input data (time-series data, spectral data, statistical parameters) from different analysis methods and produces a unified determination output, making the system multi-functional without requiring separate processing pipelines for each method.
3Difficulty of detecting and measuring
If threshold values are set to improve abnormality detection, then the detection capability improves, but the loss of information occurs due to arbitrary threshold settings that may misclassify normal or abnormal states
Solution Approach 1:
The system implements feedback through machine learning where determination results are continuously refined based on training data. The machine learning model learns from labeled examples of normal and abnormal states, automatically adjusting its decision boundaries without arbitrary threshold setting. This feedback mechanism ensures that the determination process preserves information about the actual state by learning from real operational data rather than relying on predetermined thresholds.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The abnormality determination support apparatus comprises an analysis object data preparation unit, a primary determination unit, and a secondary determination unit. The analysis object data preparation unit acquires a time-series signal representing at least one of a state of the manufacturing facility and a product quality from a data collection apparatus of the manufacturing facility, and extracts analysis object data from the time-series signal. The primary determination unit derives a plurality of primary determination results from common analysis object data by applying a plurality of different types of analysis methods to the analysis object data extracted by the analysis object data preparation unit. The secondary determination unit includes a machine learning device which is learned using a pair of a primary determination result obtained by the primary determination unit and a corresponding set of an abnormality determination result and an abnormality cause, which is a correct answer, as a teacher signal, and inputs a plurality of primary determination results obtained from common analysis object data in the primary determination unit to the machine learning device, and outputs a secondary determination result and an estimated abnormality cause output from the machine learning device as the determination information.