Hierarchical Machine Learning for Industrial Plant Signal Compression
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Industrial plants face challenges in training machine learning models due to high-dimensional input data, which leads to overfitting and requires either manual feature selection or large amounts of data, making it difficult to effectively predict events or monitor processes.
Innovation Solution
A hierarchical machine learning method that uses a topology model with a tree-like structure to group data signals and units, reducing dimensionality by learning representations at each level and training models on compressed or reconstructed data, thereby avoiding overfitting and reducing the need for extensive data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If manual feature selection and engineering are used, then model complexity is reduced, but productivity and automation are decreased
Solution Approach 1:
The patent segments the feature selection process into hierarchical levels (plant level, unit level, asset level, component level), where each level automatically generates features for the next lower level. This segmentation resolves the contradiction by maintaining automation across all levels while keeping each individual level's complexity manageable through localized feature generation.
Solution Approach 2:
The system performs preliminary feature generation and selection at each hierarchical level before final model training. By pre-processing data and generating features automatically at upper levels (plant, unit, asset), the system reduces the complexity burden on lower levels while maintaining high automation throughout the entire process.
2Measurement precision
If large amounts of data are used for training, then measurement precision is improved, but loss of time and productivity are worsened
Solution Approach 1:
The patent extracts and removes redundant information by generating compact feature representations at each hierarchical level. Instead of using all raw sensor data directly, the system extracts essential features (e.g., statistical summaries, trends, anomalies) at plant, unit, asset, and component levels, significantly reducing the data volume required for training while maintaining or improving prediction accuracy.
Solution Approach 2:
The system transforms the data from the raw sensor dimension to hierarchical aggregation dimensions. By organizing data across multiple hierarchical levels (plant → unit → asset → component) and temporal dimensions (historical trends, rolling windows), the system reduces the effective dimensionality of training data while preserving critical information for accurate predictions.
3Device complexity
If hierarchical representation is used, then device complexity is reduced, but loss of information is increased
Solution Approach 1:
The patent implements feedback mechanisms where feature generation at each hierarchical level is continuously refined based on performance metrics from lower levels. The system uses feedback loops to adjust feature extraction parameters, ensuring that hierarchical aggregation preserves critical signal details while reducing complexity. This feedback-driven approach maintains information fidelity throughout the hierarchical processing chain.
Solution Approach 2:
The system dynamically changes parameters at each hierarchical level (e.g., aggregation windows, feature types, transformation methods) to optimize the balance between complexity reduction and information preservation. By adapting parameters based on the specific characteristics of data at each level (plant-level macro trends vs. component-level micro variations), the system maintains information fidelity while achieving dimensional reduction.
Data Source
AI summary
A method of hierarchical machine learning includes receiving a topology model having information on hierarchical relations between components of the industrial plant, determining a representation hierarchy comprising a plurality of levels, wherein each representation on a higher level represents a group of representations on a lower level, wherein the representations comprise a machine learning model, and training an output machine learning model using the determined hierarchical representations.