Machining Transfer Learning with Importance-Based Data Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Generating a machine learning model for setting machining conditions in machining machines is time-consuming due to the large amount of data collected from the machining site, even when using transfer learning techniques.
Innovation Solution
A transfer learning device that acquires machining data, calculates importance for each data piece, extracts relevant transfer learning data, and performs transfer learning using a pre-existing machine learning model to generate a new model efficiently.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If transfer learning is used to generate a machine learning model for current machining situation, then model generation time is reduced, but the amount of data required for learning becomes insufficient
Solution Approach 1:
The patent segments the large volume of collected machining data into multiple data sets with different characteristics (e.g., normal operation data, abnormal data, data from different machining conditions). This segmentation allows the system to selectively use appropriate data segments for transfer learning, reducing the total data processing time while ensuring sufficient representative data is available for effective model generation.
Solution Approach 2:
The patent applies local quality by selecting and weighting different data segments based on their relevance and quality for the specific transfer learning task. Instead of uniformly processing all collected data, the system identifies and prioritizes data segments that are most valuable for adapting the pre-trained model to the current machining situation, thereby reducing overall data processing time while maintaining learning effectiveness.
2Measurement precision
If all collected machining data is used for transfer learning, then model accuracy is improved, but processing time increases significantly
Solution Approach 1:
The patent extracts essential features and key data segments from the large volume of collected machining data. By identifying and extracting only the most relevant information needed for transfer learning (such as critical machining parameters, representative operation patterns, and key anomaly indicators), the system achieves sufficient model accuracy while dramatically reducing the time required to process and learn from the complete data set.
Solution Approach 2:
The patent applies partial action by using a carefully selected subset of the collected data that provides sufficient learning signal for accurate model adaptation. Rather than processing all available data, the system identifies the minimum necessary data volume and composition required to achieve the desired model accuracy for the specific transfer learning scenario, thereby optimizing the balance between accuracy and processing time.
Data Source
AI summary
Included are: a machining data acquiring unit to acquire machining data including data related to time-series machining conditions to perform machining for a plurality of steps and data related to a machining state for the plurality of steps on a basis of the machining conditions; an analysis unit to calculate an importance with respect to machining data; an extraction unit to extract transfer learning data for performing transfer learning of a first machine learning model on a basis of the importance; and a learning unit to perform the transfer learning of the first machine learning model by using the transfer learning data, and generate a second machine learning model that receives, as an input, data related to the machining conditions in time series for the plurality of steps and outputs data related to a machining state after execution for the plurality of steps.