Machine Tool Abnormality Detection via One-Class Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional abnormality diagnosis methods for machine tools face challenges in setting accurate thresholds for signals like spindle load, cutting force, and cutting vibration, which vary with cutting conditions, leading to low diagnostic accuracy.
Innovation Solution
An abnormality-detecting device and method that uses one-class machine learning to create a normal model from measurement data during normal machining, allowing for real-time diagnosis of abnormal conditions and updating the model based on non-abnormal data, with optional multi-class machine learning for classification and re-diagnosis using different methods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional threshold-based diagnosis is used, then the diagnostic system is simple to implement, but the diagnostic accuracy deteriorates due to difficulty in setting thresholds for varying cutting conditions
Solution Approach 1:
The patent replaces the conventional mechanical threshold-based diagnosis system with a machine learning-based diagnostic system. Instead of using fixed thresholds that require manual setting and adjustment, the system uses trained models (such as support vector machines, neural networks, or random forests) that automatically learn optimal decision boundaries from historical data, thereby improving diagnostic accuracy without requiring complex manual threshold management
Solution Approach 2:
The patent transforms the diagnostic approach by changing from fixed parameter thresholds to dynamic, data-driven parameter relationships. The system collects multiple measurement parameters (vibration, temperature, acoustic emission, etc.) and uses machine learning algorithms to analyze their complex interactions and temporal patterns, enabling accurate diagnosis that adapts to varying cutting conditions without requiring manual threshold adjustment for each condition
2Reliability
If fixed thresholds are used for diagnosis, then the system is easy to operate, but false positives and false negatives increase due to varying cutting conditions
Solution Approach 1:
The diagnostic system performs self-service by automatically learning and adapting to different cutting conditions through machine learning. The system continuously collects operational data, trains its models, and updates its diagnostic criteria without requiring manual intervention or expert knowledge. This self-learning capability improves reliability across varying conditions while maintaining ease of operation, as the system handles the complexity internally
Solution Approach 2:
The patent implements feedback mechanisms where the diagnostic system continuously monitors measurement data, compares it against learned patterns, and refines its diagnostic accuracy over time. The system uses feedback from historical diagnosis results and actual tool conditions to update its models, thereby reducing false positives and false negatives while maintaining simple operation through automated model refinement
3Measurement precision
If multiple measurement parameters are collected and analyzed using machine learning, then the diagnostic accuracy is improved, but the data processing complexity and computational requirements increase
Solution Approach 1:
The patent applies segmentation by dividing the complex diagnostic task into distinct processing stages: data acquisition from multiple sensors, pre-processing and feature extraction, model training phase, and real-time inference phase. This segmentation allows the system to handle complex multi-parameter analysis systematically, processing different types of data through specialized modules rather than attempting to analyze all parameters simultaneously in a monolithic system
Solution Approach 2:
The system performs preliminary actions by pre-processing measurement data and extracting relevant features before main diagnostic analysis. During the model training phase, the system pre-processes historical data to identify important patterns and relationships, storing pre-computed features that can be quickly applied during real-time diagnosis. This preliminary processing reduces the computational burden during actual operation while maintaining high diagnostic accuracy
Data Source
AI summary
An abnormality-detecting device for detecting abnormalities of a tool of a machine tool comprises: an acquiring unit for acquiring multiple measured values relating to the tool as measurement data (vibration information, cutting force information, sound information, main shaft load, motor current, power value); a normal model unit for learning the measurement data acquired during normal machining by one class machine learning and creating a normal model; an abnormality-diagnosing unit for acquiring measurement data during machining after creation of the normal model while diagnosing whether said measurement data is normal or abnormal on the basis of the normal model; and a re-diagnosing unit for re-diagnosing measurement data, which has been diagnosed to be abnormal by the abnormality-diagnosing unit, by a method different from the abnormality-diagnosing unit.