Asset Classification via Anchor-Based Neural Network Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning algorithms in Operational Technology (OT) and Building Management Systems (BMS) struggle with classifying assets due to unstructured data and the lack of training data for certain asset types, limiting their ability to handle partial positive or negative labels effectively.
Innovation Solution
The implementation of an anchor-based machine learning neural network that uses mined triplets with anchor, positive, and negative inputs to generate vector representations, allowing for the determination of asset types based on similarity thresholds, rather than binary classifications, and incorporating time series data and metadata for more accurate categorization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional machine learning algorithms are used for asset classification, then binary ground truth labels can be processed, but the system cannot handle unstructured data effectively and fails when training data for certain asset types is missing
Solution Approach 1:
The patent transforms the classification problem from binary labels to continuous similarity scores by changing the output parameter from discrete categories to continuous vector distances. This allows the system to handle unstructured data and asset types without predefined training labels by measuring similarity to anchor assets rather than matching against fixed classes.
Solution Approach 2:
The patent introduces a new dimensional approach by embedding assets into a vector space where classification is performed through distance measurement rather than traditional category matching. This dimensional transformation enables the system to process unstructured data and handle novel asset types by comparing them in the continuous vector space.
2Measurement precision
If manual asset classification is performed, then accurate categorization can be achieved, but substantial manual labor is required
Solution Approach 1:
The system enables automated self-service classification by computing similarity scores between assets and anchor assets automatically. The machine learning model performs categorization without human intervention by measuring vector distances and comparing against threshold values, eliminating manual labor while maintaining accuracy.
Solution Approach 2:
The patent replaces manual classification mechanics with automated computational mechanics. Instead of human operators examining asset metadata and assigning categories, the system uses machine learning models to compute vector representations and automatically determine asset types through similarity threshold comparisons.
3Device complexity
If traditional machine learning with binary labels is used, then simple classification can be performed, but partial positive or negative labels cannot be represented
Solution Approach 1:
The patent changes the label representation from binary discrete values to continuous similarity scores. This parameter transformation preserves partial information by representing degrees of similarity rather than forcing assets into rigid binary categories, allowing the system to capture nuanced relationships between assets.
Data Source
AI summary
A method for classifying assets in a facility includes receiving item point metadata and time series data associated with an asset, applying a machine learning neural network to the item point metadata and time series data to determine a similarity between metadata associated with the asset and metadata associated with a known asset type, and determining that the asset is of the asset type based on the similarity satisfying a predetermined threshold. The neural network comprises anchor-based learning trained on a plurality of mined triplets including an anchor input, a positive input, and a negative input. The neural network is configured to generate vector representations associated with each of the anchor input, the positive input, and the negative input. A loss function applied to the vector representations is configured to differentiate between the vector representation associated with the positive input and the vector representation associated with the negative input.