Deep-learning-based fault detection in building automation systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing rule-based fault detection systems in building automation are limited in scalability and provide vague diagnostic information, requiring predefined rules and manual configuration, which can lead to errors and are difficult to implement in complex systems.
Innovation Solution
The implementation of deep learning techniques, including training models with contaminated data using Huber loss and dropout regularization, and applying cumulative sum control chart analysis to identify faults, allowing for more accurate and adaptive fault detection without the need for fault-free training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If rule-based systems are used for fault detection, then fault detection capability is provided, but scalability is limited and diagnostic information is vague
Solution Approach 1:
The patent replaces rule-based systems with machine learning models that automatically learn fault patterns from data. Instead of manually configuring rules, the system uses algorithms to detect faults, enabling better scalability and adaptability while maintaining fault detection capability.
Solution Approach 2:
The machine learning models automatically learn and adapt to system patterns without manual intervention. The system self-configures by training on historical data, eliminating the need for manual rule configuration and improving scalability across different building automation systems.
2Reliability
If predefined rules are manually configured, then fault detection can be implemented, but the process is difficult and error-prone in complex systems
Solution Approach 1:
Manual rule configuration is replaced with automated machine learning model training. The system learns fault detection patterns automatically from data, eliminating manual configuration steps and reducing errors in complex system implementation.
Solution Approach 2:
The patent introduces data as an intermediary between the system and fault detection. Instead of directly configuring rules, the system learns from historical data, which mediates the detection process and simplifies implementation by removing manual configuration complexity.
3Loss of information
If complex rules are created for better diagnosis, then diagnostic information improves, but configuration complexity and error risk increase
Solution Approach 1:
Complex manual rule configuration is replaced with machine learning models that automatically learn diagnostic patterns. The models process complex relationships in data without requiring manual rule creation, improving diagnostic quality while reducing configuration complexity.
Solution Approach 2:
The patent changes the approach from discrete rules to continuous learning models. By using parameters learned from data rather than fixed rules, the system achieves better diagnostic information without the complexity of configuring multiple interdependent rules.
Data Source
AI summary
Methods, mediums, and systems include use of a system manger application in a data processing system for fault detection a building automation system using deep learning, to receive point data for a hardware being analyzed, where the received point data is contaminated data, train a deep learning model for the hardware being analyzed, generate predicted data based on the deep learning model, analyze the predicted data and the received point data, identify a fault in the hardware being analyzed according to the received point data and the predicted data, and produce a fault report according to the identified fault.