Building Equipment Fault Diagnostics Using Multi-Modal Generative AI
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems face challenges in generating precise data for servicing building equipment due to limitations in existing AI and machine learning models, which often produce incorrect, imprecise, or irrelevant outputs, requiring manual adjustments and significant computational resources, and struggle to accurately process unstructured data from diverse sources.
Innovation Solution
The implementation of a method using generative artificial intelligence models, specifically generative adversarial networks, to process multi-modal data inputs from building equipment, including images, videos, audio, and time series data, to generate accurate and relevant service outputs, such as service plans and actions, by training data generators to augment missing data modes and update models based on field technician feedback.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing AI and machine learning models are used to process building equipment data, then computational resources can be utilized, but the output precision and relevance deteriorate due to incorrect or irrelevant results
Solution Approach 1:
The patent implements a feedback mechanism where field technician observations and corrections are fed back to retrain and refine the generative AI models. This continuous feedback loop improves output precision and reliability by learning from real-world corrections and adjusting model predictions accordingly.
Solution Approach 2:
The patent uses generative adversarial networks to create synthetic data copies that augment training datasets. These artificial analysis packages replicate real equipment data patterns, enabling more robust model training and improving prediction accuracy without requiring additional physical equipment or extensive real-world data collection.
2Measurement precision
If manual adjustments are made to correct AI model outputs, then output accuracy can be improved, but the time and operational complexity increase
Solution Approach 1:
The system implements self-service through automated model retraining using feedback from field technicians. Instead of requiring continuous manual adjustment of each output, the system automatically learns from corrections and improves future predictions autonomously, reducing ongoing manual intervention time while maintaining high accuracy.
Solution Approach 2:
The patent performs preliminary actions by pre-training models with extensive synthetic and real data before deployment. This preliminary training phase prepares the models to handle diverse equipment scenarios, reducing the need for manual adjustments during actual field operations and speeding up service response times.
3Loss of information
If diverse data sources are processed to improve service recommendations, then data completeness improves, but the difficulty of processing unstructured data increases
Solution Approach 1:
The patent introduces an intermediary layer of multi-modal data processing that unifies diverse data sources including images, videos, audio, and time series data. This intermediary processing layer converts unstructured data from various sources into a standardized format that the generative AI models can effectively analyze, maintaining data completeness while reducing processing difficulty.
4Measurement precision
If generative AI models are trained with more data to improve accuracy, then model performance improves, but computational resource requirements increase
Solution Approach 1:
The patent uses generative adversarial networks to create synthetic data copies that augment training datasets. This approach allows the model to be trained on diverse and extensive data without proportionally increasing physical data collection costs, improving model accuracy while managing computational resources efficiently through artificial data generation.
Data Source
AI summary
A method for servicing building equipment using generative artificial intelligence models includes receiving a multi-modal data input characterizing operation of the building equipment using multiple modes of data, associating related data portions from each mode of the multi-modal data input to form a set of original analysis packages, training at least one data generator to generate artificial analysis packages using the original analysis packages, using the at least one data generator to generate a set of artificial analysis packages, and adjusting an output model using the set of artificial analysis packages and the set of original analysis packages. The output model is configured to generate a service relevant multi-modal data output for use in servicing the building equipment.