Machined Surface Image Learning for Accurate Tool Wear Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for determining tool replacement time based on numerical data, such as contrast in machined surface images, are unreliable due to variations with workpieces and imaging conditions, making it difficult to accurately estimate tool state.
Innovation Solution
A tool state learning device that performs supervised learning using annotated images of machined surfaces to generate a trained model for estimating tool state, which includes a teacher data acquisition unit, a tool state learning unit, and a tool state estimation unit to determine the tool's state from captured images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If numerical data such as contrast is used as a reference to determine tool replacement time, then tool state can be estimated, but the estimation accuracy deteriorates because the numerical data varies depending on workpieces and imaging conditions
Solution Approach 1:
The patent creates a virtual model (trained model) that copies the relationship between machined surface images and tool states from training data. Instead of directly measuring tool state from variable numerical data, the system learns a mapping from images to tool states through supervised learning, making the estimation robust to variations in workpieces and imaging conditions.
Solution Approach 2:
The patent transforms the approach from using direct numerical measurements (contrast values) to using image data processed through a trained model. The model learns optimal parameter transformations during training, converting variable image inputs into stable tool state predictions that are independent of specific workpiece or imaging condition variations.
2Extent of automation
If threshold values are set based on numerical data to determine tool replacement timing, then automated decision making is achieved, but reliability deteriorates because appropriate threshold values cannot be determined due to data variability
Solution Approach 1:
The system replaces manual threshold setting with a trained model that automatically learns the relationship between images and tool states. The model copies expert knowledge from training data and applies it automatically to new images, eliminating the need for manual threshold determination while maintaining high reliability.
Solution Approach 2:
The supervised learning process uses labeled training data (images paired with actual tool states) to provide feedback during model training. This feedback mechanism allows the model to automatically adjust its parameters to improve prediction accuracy, achieving reliable automated decision-making without manual intervention.
3Measurement precision
If supervised learning is performed using annotated images to generate a trained model, then tool state estimation accuracy is improved, but device complexity increases due to the need for training data acquisition and model generation
Solution Approach 1:
The system performs preliminary actions by collecting and annotating training data in advance, then training the model before actual tool state estimation is needed. This preliminary preparation creates a ready-to-use trained model that can quickly and accurately estimate tool states during operation without requiring complex real-time processing.
Solution Approach 2:
The trained model becomes a self-service component that automatically performs tool state estimation without requiring external intervention. Once trained, the model independently processes images and outputs tool state predictions, reducing the operational complexity of the system.
Data Source
AI summary
A tool state learning device has a storage unit for storing an arbitrary image captured by an imaging device imaging a machined surface of an arbitrary workpiece cut using an arbitrary tool, and a teacher data acquisition unit for acquiring, as input data, an arbitrary image stored in the storage unit, and acquiring, as a label, the state of the tool annotated according to prescribed levels indicating the degree of wear of the tool on the basis of the arbitrary image. The tool state learning device also has a tool state learning unit for using the acquired label and input data to perform supervised learning, and generating a learned model in which a machined surface image of the machined surface of a workpiece imaged by the imaging device is inputted and the state of the tool that cut the machined surface of the workpiece is outputted.