Machine Tool Diagnosis Using Context-Combined Learning Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine tool diagnosis systems require multiple occurrences of abnormalities to improve model accuracy, leading to prolonged training times and inefficiencies in determining operational abnormalities.
Innovation Solution
A diagnosis system comprising a first acquiring unit for context information, a second acquiring unit for detection information, a first transmitting unit for context data, a second transmitting unit for detection data, and a determining unit that uses a learning device to assess normal operations by combining similar context and detection information models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a learning model is generated using abnormal data to improve diagnosis accuracy, then the accuracy in determining abnormalities improves, but it requires multiple abnormal occurrences which extends the training time
Solution Approach 1:
The system performs preliminary actions by collecting and storing detection information during normal operations before abnormalities occur. The learning model is pre-trained using normal operation data, and when an abnormality is detected, the system can quickly adapt by incorporating the abnormal data without requiring multiple abnormal occurrences, thus resolving the contradiction between accuracy and training time
Solution Approach 2:
The learning model is segmented into two distinct components: a normal operation model trained on normal detection information, and an abnormality detection mechanism that activates when deviations are detected. This segmentation allows the system to maintain high accuracy by specializing each model for its respective purpose while avoiding the need for extensive abnormal data collection
2Measurement precision
If the learning model is re-input for each individual facility to improve accuracy, then the maturity and accuracy of the learning model improve, but it takes a long time to improve the learning model
Solution Approach 1:
The learning model is designed with universality to function across multiple facilities and device types. By abstracting the core detection patterns and using standardized processing methods, the same learning model framework can be applied to different facilities without requiring complete re-training, thus improving both accuracy and efficiency simultaneously
Solution Approach 2:
Instead of creating entirely new learning models for each facility, the system creates copies or adaptations of a base learning model. The base model is trained on aggregated data from multiple facilities, and facility-specific adjustments are made through parameter tuning rather than complete re-training, significantly reducing the time required to improve model accuracy for individual facilities
Data Source
AI summary
A diagnosis device includes a first acquiring unit acquires, from a target device, context information corresponding to a current operation; a second acquiring unit acquires detection information output from a detecting unit that detects a physical quantity that changes according to operations performed by the target device; a first transmitting unit transmits the acquired context information to a learning device; a second transmitting unit transmits the acquired detection information to the learning device; a third acquiring unit acquires a model corresponding to the transmitted context information, from the learning device that determines whether any pieces of context information are identical or similar to each other and combines models generated from pieces of the detection information corresponding to the pieces of identical or similar context information; and a first determining unit determines whether an operation performed by the target device is normal by using the detection information and the model.