Non-Destructive Part Characterization via Neural Network Transfer Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing non-destructive testing (NDT) methods using neural networks require extensive and time-consuming learning phases with large datasets to optimize estimation performance, particularly for structural health monitoring (SHM) in mechanical parts.
Innovation Solution
A method involving two stages of neural network training: an initial training on a model or real part to establish a robust feature extraction block, followed by a targeted adaptation of the classification block using a smaller dataset from the actual part, allowing for efficient characterization of defects and properties.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a neural network is trained extensively with large datasets to optimize estimation performance, then the prediction accuracy improves, but the training time and computational resources increase significantly
Solution Approach 1:
The patent applies preliminary action by pre-training a first neural network on a first database (which may be simulated or from a different part) before actual deployment. This pre-training establishes a robust feature extraction capability that can be transferred to the target application, reducing the need for extensive training on the actual part's data.
Solution Approach 2:
The patent uses copying by creating a first neural network model that replicates the essential feature extraction capabilities needed for the target application. This model is then adapted and fine-tuned on the actual part's data, rather than training from scratch, thus copying the successful approach while adapting it to specific needs.
2Reliability
If a comprehensive learning phase is conducted to optimize neural network estimation performance, then the diagnostic performance improves, but the time required for training increases
Solution Approach 1:
The patent performs preliminary training actions on a first neural network using a first database that may be simulated or collected from a different part. This preliminary learning phase establishes robust feature extraction capabilities before the network is deployed on the actual part, ensuring good diagnostic performance without requiring extensive training on the target part's data.
Solution Approach 2:
The patent extracts and transfers the feature extraction block from the first neural network to be used in the second neural network for the actual part. This extraction allows the diagnostic system to benefit from comprehensive learning on the first database while requiring only minimal adaptation on the actual part's data.
3Measurement precision
If a large amount of data is collected for neural network training, then the estimation performance is optimized, but the data collection time and resources increase
Solution Approach 1:
The patent copies the feature extraction capabilities from a first neural network trained on a first database to a second neural network for the actual part. This copying approach allows the system to achieve good estimation performance without collecting and training on large volumes of actual part data, as the essential patterns are already learned from the first database.
Solution Approach 2:
The patent extracts the feature extraction block from the first neural network and reuses it in the second neural network. This extraction eliminates the need to collect and process large amounts of training data for the actual part, as the feature extraction capabilities are already established from the first database.
Data Source
Figure 1A~1B
Figure 2A~2B
Figure 3A~3B
AI summary
A method for characterizing a part (10), comprising: a) performing non-destructive measurements using a sensor (11), the sensor being arranged on the part or facing the part; b) using the measurements as input data for a neural network (IMN2, CNIM2); c) characterizing the part on the basis of the value of each node of the output layer of the neural network; the method comprising, prior to steps b) and c): - forming a first database (DB1), on a first model part; - taking into account a first neural network (CNN1, CNN1, NN1), parameterized by a first learning operation, using the first database (DB1); - forming a second database (DB2), comprising experimental measurements of the physical quantity taken on the part to be characterized - a second learning operation, using the second database (DB2), so as to parameterize a second neural network (CNN2, NN2), using the parameterization of the first neural network (CNN1, CNN1, NN1). In step c), the neural network that is used is the second neural network (CNN2, NN2).