Optical Fibre Sensing Model Training with Unlabelled Data Transfer
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge with applying machine learning techniques to optical fibre sensing data is the limited availability of ground truth data, leading to low accuracy in machine learning models due to differences in site conditions and train tracks, and the large volume of unlabelled data being unused.
Innovation Solution
A two-step training method using a first machine learning model trained with unlabelled optical fibre sensing data, followed by a transfer learning process with a smaller dataset to adapt the model for specific tasks, leveraging the large volumes of unlabelled data for improved accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are trained using conventional techniques with limited ground truth data, then training time and computational resources are reduced, but model accuracy deteriorates due to limited data availability and site condition variations
Solution Approach 1:
The patent applies preliminary action by training a first machine learning model in advance using large volumes of unlabelled optical fibre sensing data. This pre-trained model serves as a foundation that can be later adapted to specific tasks with smaller labelled datasets, thereby improving overall model accuracy while managing data limitations
Solution Approach 2:
The patent uses an intermediary approach by introducing a two-stage training process where unlabelled data acts as an intermediary resource. The first model trained on unlabelled data serves as an intermediate representation that bridges the gap between abundant unlabelled data and the need for accurate classification with limited labelled ground truth data
2Quantity of substance
If large volumes of unlabelled optical fibre sensing data are collected, then data availability for training improves, but data processing and storage requirements increase
Solution Approach 1:
The patent applies segmentation by dividing the training process into two distinct stages: first training a model on unlabelled data, then adapting it with labelled data. This segmentation allows the system to handle large volumes of unlabelled data efficiently without requiring complex end-to-end processing of all data simultaneously
Solution Approach 2:
The patent extracts value from unlabelled data by using it specifically for initial model training, separating this function from the subsequent fine-tuning stage with labelled data. This extraction approach allows the system to leverage large datasets without proportionally increasing processing complexity for the entire training pipeline
3Measurement precision
If machine learning models are trained for specific site conditions and train tracks, then model specialization improves, but model adaptability to different locations deteriorates
Solution Approach 1:
The patent applies universality by training the first machine learning model on diverse unlabelled data from multiple sources and conditions. This creates a universal base model that can be adapted to various specific applications and site conditions, thereby maintaining both generalizability and the ability to achieve specialization when needed
Data Source
AI summary
A machine learning method where, in a first step, a first (general) machine learning model is trained using a first training dataset including unlabelled optical fibre sensing data. Then, in a second step, a transfer learning process is applied to adapt or fine-tune the first machine learning model to a more specific application (e.g. to perform a specific type of detection or classification). Due to the large volumes of optical fibre sensing data available, the first machine learning model may provide a general machine learning model which has a high level of generality and is highly adaptable.