Supervised Learning Labeling for Intermediate Tool Wear States
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine tool monitoring systems struggle to accurately determine the state of tools and bearings during intermediate states of wear, leading to inefficiencies and increased costs due to inappropriate inference results and the inability to adapt to varying processing conditions, resulting in decreased productivity and potential machine damage.
Innovation Solution
A method for labeling a data set of a supervised learning machine that uses normalized deviation degrees to map the state transition between reference states, allowing for accurate labeling and improved inference in intermediate states, thereby enhancing the learning model's ability to adapt to varying conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a learning model is trained only on processing information where tool life sufficiently remains, then the model can accurately identify normal states, but it cannot reliably determine states where tool life does not remain
Solution Approach 1:
The patent applies preliminary action by pre-defining reference states (reference state A for normal conditions and reference state B for abnormal conditions) and their corresponding output values before actual tool state estimation. This allows the supervised learning machine to have predetermined benchmarks for comparison, enabling it to accurately estimate tool states even in intermediate conditions where tool life does not sufficiently remain, by comparing current states against these pre-established references.
2Ease of manufacture
If fixed criteria are used to determine tool states, then implementation is simple, but the system cannot adapt to new processing conditions or varying tool wear patterns
Solution Approach 1:
The patent applies dynamics by using a supervised learning machine that can dynamically adapt to new processing conditions while maintaining a structured framework. The system dynamically learns from training data consisting of processing information paired with reference state labels, allowing it to adjust to varying tool wear patterns and new processing conditions without requiring complete system redesign, thus balancing simplicity with adaptability.
3Loss of information
If unsupervised learning is used to create clusters based on normal processing data, then the system can identify deviations from normal states, but it cannot accurately label intermediate states between normal and abnormal
Solution Approach 1:
The patent applies the intermediary principle by introducing reference states (A and B) as mediator concepts that bridge the gap between normal and abnormal tool states. These reference states serve as intermediate benchmarks that allow the system to accurately label and understand intermediate states during tool wear progression, rather than simply detecting binary normal/abnormal transitions.
Data Source
AI summary
A method for labeling a data set of a supervised learning machine includes: calculating a deviation degree from a reference state A for each corresponding data in cases where a plurality of state estimation targets are in the reference state A and a reference state B; calculating a normalized deviation degree by taking a map of the deviation degree from A for each of the corresponding data with a map set for each of the state estimation targets. The map is allowed to map the deviation degree from A in A on a specified value to be output by the supervised learning machine in A; allowed to map the deviation degree fromA in B on a specified value to be output by the supervised learning machine in B.