Building Equipment Fault Prediction with Multi-Device Data Augmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing fault detection systems in building management systems often rely on robust historical data that is not always available, limiting their effectiveness in predicting equipment faults and malfunctions.
Innovation Solution
A system that augments data sets for fault prediction models using supplemental data from multiple devices, clusters devices based on characteristics, and trains per-device or global models to improve fault prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If fault prediction models are trained using only local device data, then device-specific accuracy is improved, but insufficient data availability deteriorates model effectiveness
Solution Approach 1:
The patent combines data from multiple devices of the same type to create augmented training datasets. When a target device has insufficient local data, the system retrieves and integrates data from other devices, merging them into a comprehensive training set that enables effective model training while maintaining device-specific prediction accuracy.
Solution Approach 2:
The patent creates a universal data pool that serves multiple devices of the same type. A single augmented dataset can be used to train models for any device within a cluster, making the data resource multi-functional and applicable across multiple devices, thereby solving the data scarcity problem for individual devices.
2Quantity of substance
If data is augmented using data from multiple devices, then data availability is improved, but data heterogeneity worsens training effectiveness
Solution Approach 1:
The patent applies local quality by selecting and weighting data from source devices based on their similarity to the target device. Devices with more similar operational characteristics contribute more heavily to the augmented dataset, ensuring that the combined data maintains consistency relevant to the target device while still expanding data availability.
Solution Approach 2:
The patent transforms heterogeneous data into a standardized format by adjusting parameters such as temporal resolution, feature normalization, and operational condition alignment. This parameter transformation ensures that data from different devices can be effectively combined while maintaining training effectiveness.
3Adaptability or versatility
If clustering is used to group similar devices, then data reusability is improved, but system complexity increases
Solution Approach 1:
The patent performs device clustering and data pool creation in advance, before actual fault prediction is needed. By pre-organizing devices into clusters and pre-augmenting data pools for each cluster, the system reduces real-time computational complexity while maintaining high data reusability when making predictions.
Data Source
AI summary
A predictive modeling and control system for building equipment assesses whether a data set from a first device of building equipment is sufficient to train a prediction model for the first device. In response to a determination that the data set from the first device is insufficient to train the prediction model for the first device, the system generates a ranking of a plurality of additional devices of building equipment based on similarities between the first device and the plurality of additional devices, augments the data set with supplemental data from one or more of the plurality of additional devices in an order based on the ranking until the augmented data set is sufficient to train the prediction model, and trains the prediction model for the first device using the augmented data set. The system influences operations of the first device using the prediction model.