Battery Diagnostic Data Denoising for Field-Representative Model Training
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge lies in the significant difference between experimental data from batteries and field data from vehicles, leading to concerns about the reliability of diagnostic results when applying machine learning-based state diagnosis models to actual vehicle batteries.
Innovation Solution
A data processing apparatus and method that generates synthetic data reflecting field data characteristics, removes noise using a denoising model, and evaluates the model's performance by comparing state diagnosis models trained on denoised and synthetic data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If state diagnosis models are trained using experimental data from batteries, then the models can be developed and applied, but the diagnostic reliability deteriorates when applied to field data from vehicles due to significant differences in characteristics
Solution Approach 1:
The patent introduces synthetic data as an intermediary between experimental data and field data. This synthetic data is generated to reflect field data characteristics while being derived from controlled experimental data, serving as a bridge that enables model training with field-representative data without requiring access to actual field conditions during the training phase.
Solution Approach 2:
The patent applies parameter changes by modifying experimental data through noise addition and transformation processes to generate synthetic data with altered characteristics that match field data distributions. This involves changing statistical parameters, noise levels, and data distributions to bridge the gap between controlled experimental conditions and real-world field conditions.
2Reliability
If synthetic data is generated to reflect field data characteristics, then diagnostic reliability improves, but noise is introduced into the data
Solution Approach 1:
The patent converts the harmful effect of noise into a beneficial feature by intentionally adding controlled noise to experimental data during synthetic data generation. This noise, which would normally be considered detrimental, is used to make the synthetic data more representative of real field conditions, thereby improving model generalization and diagnostic reliability when deployed in actual vehicle environments.
3Measurement precision
If noise is removed from synthetic data using a denoising model, then data quality improves, but additional processing steps and model complexity are required
Solution Approach 1:
The patent applies preliminary action by training a denoising model in advance on synthetic data with known noise characteristics. This pre-trained denoising model can then be applied to remove noise from new synthetic data without requiring complex real-time processing during deployment. The denoising operation is performed as a straightforward application of the pre-trained model rather than through complex iterative optimization.
4Adaptability or versatility
If multiple learning models (extraction model, denoising model, state diagnosis model) are trained, then comprehensive data processing capability is achieved, but training time and computational resources increase
Solution Approach 1:
The patent segments the complex data processing task into distinct functional modules: an extraction model for identifying field data characteristics, a denoising model for removing noise from synthetic data, and a state diagnosis model for actual diagnostic tasks. Each model is trained independently on specific datasets with particular characteristics, allowing for specialized optimization while reducing the overall training burden compared to training a single comprehensive model on all data types simultaneously.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A data processing apparatus according to an embodiment disclosed in this document may include an acquisition unit configured to acquire experimental data of a battery and field data of a vehicle, a generation unit configured to generate synthetic data based on the experimental data and the field data, and a first learning unit configured to train a denoising model to remove noise from the synthetic data based on the experimental data and the synthetic data.