Episodic Feature Extraction for Industrial AI Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Industrial process data, particularly in the form of episodic time series, poses challenges due to its non-consumable format, lack of canonical representation, high volume, continuous generation, and need for real-time updates, which complicates AI model development and maintenance, especially for automated tools that primarily focus on tabular data.
Innovation Solution
A system and method for standardizing industrial process data as episodic time series, involving an episode data generator that transforms raw data into a standardized format, a feature learning operations module that assigns feature tags based on key performance indicators, and an integrated machine learning operations module for optimized feature extraction and model readiness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If extensive feature extraction is performed on hundreds of time series features for automated model discovery, then model accuracy can be improved, but the process becomes time-consuming and computation-intensive
Solution Approach 1:
The patent extracts only the most relevant features from hundreds of time series features using domain knowledge and statistical analysis. Instead of processing all features, the system identifies and extracts a精简 subset of key features that are most predictive for industrial processes, significantly reducing computation time while maintaining model accuracy.
Solution Approach 2:
The patent applies different feature extraction strategies to different parts of the time series data based on local characteristics. Different subprocesses and signal types receive customized feature extraction treatments rather than uniform processing, optimizing the balance between feature quality and computational efficiency for each specific data segment.
2Reliability
If all time series features are extracted and stored for future model training, then model training can be more comprehensive, but storage requirements increase significantly
Solution Approach 1:
The patent extracts and stores only the essential aggregated features and statistical summaries rather than all raw time series data. By computing and retaining key metrics such as mean, variance, and domain-specific features, the system maintains model training capability while dramatically reducing storage requirements.
Solution Approach 2:
Instead of storing all raw data and extracting features later, the patent inverts the approach by pre-computing and storing only the extracted features and aggregated statistics. This inversion eliminates the need to store voluminous raw time series data while preserving all necessary information for model training.
3Adaptability or versatility
If feature extraction is performed continuously on incoming industrial process data, then models can stay up-to-date with new data distributions, but computational overhead increases
Solution Approach 1:
The patent implements periodic feature extraction and model updating rather than continuous processing. The system extracts features at regular intervals or when triggered by significant data distribution changes, reducing computational overhead while maintaining the ability to adapt to new data patterns through scheduled retraining cycles.
Solution Approach 2:
The patent incorporates feedback mechanisms that monitor data distribution changes and trigger feature extraction and model updating only when necessary. By detecting shifts in process patterns and activating computation only in response to meaningful changes, the system maintains adaptability while minimizing unnecessary computational overhead.
Data Source
AI summary
A computer implemented method of preparing process data for use in an artificial intelligence (AI) model includes collecting and storing raw data as episodic data for each episode of a process. An episode data generator assigns an episode identifier each set of episodic data. The raw data per episode is transformed into a standardized episodic data format that is usable by the AI model. Metrics are assigned to the episodic data and the episodic data is aggregated in an episode store. The data in the episode store is used by a feature extraction and learning module to extract and rank features.