Measurement Data Event Detection Using Segmented ML Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for event detection in measurement data, particularly in optical time domain reflectometry (OTDR) measurements, require extensive training data and expert evaluation, and data augmentation techniques can lead to errors and misclassification.
Innovation Solution
A method utilizing separate feature extraction and detection models, trained with limited data, and a continuous feedback loop for optimization, reduces the need for extensive training data and minimizes overfitting, enabling scalable and flexible event detection across various use cases.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex machine learning algorithms are used for event detection, then detection accuracy is improved, but training data requirements increase
Solution Approach 1:
The patent segments the machine learning task into separate sub-models (feature extraction model and detection model) that are trained and implemented independently. This segmentation allows each sub-model to be trained on smaller, more focused datasets rather than requiring a single large dataset for the entire detection pipeline, thereby maintaining detection accuracy while reducing overall training data requirements.
2Quantity of substance
If data augmentation techniques are used to generate additional training data, then training data volume increases, but error propagation and misclassification risk increase
Solution Approach 1:
The patent extracts and removes the problematic data augmentation step from the training pipeline. Instead of generating synthetic training data through augmentation techniques that may introduce errors, the approach uses separate sub-models that can achieve effective training with limited real data, thereby eliminating the source of error propagation while still providing sufficient training coverage.
3Adaptability or versatility
If a single comprehensive model is used for event detection, then model capability is improved, but computational resources and training complexity increase
Solution Approach 1:
The patent divides the comprehensive event detection model into separate sub-models (feature extraction model and detection model) that are trained and deployed independently. This segmentation reduces the complexity of each individual model, making them easier to train and compute, while the ensemble of sub-models maintains the overall capability to handle diverse event detection tasks.
4Measurement precision
If extensive training data is collected and stored, then model training quality is improved, but storage requirements and processing time increase
Solution Approach 1:
The patent segments the training process into separate sub-models that can be trained independently on smaller datasets. This approach maintains model training quality by ensuring each sub-model receives appropriate training data for its specific function, while significantly reducing the total time required to train the complete system compared to training a single comprehensive model on extensive data.
Data Source
Figure 1
Figure 2
AI summary
The present invention relates to a method for detecting an event based on data acquired by a measurement device, wherein the method (1) comprises the steps of: - Acquiring data by a measurement device (2); - Extracting features from the acquired data utilizing a feature extraction model (3); - Detecting an event based on the extracted features utilizing a detection model (4); and - Providing information about the detected event (5).