Building Vibration Anomaly Detection Using FFT and CNN Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Analyzing large datasets from building equipment for operational issues is time-consuming and costly, especially with limited analysts, and requires manual parsing which can be inefficient with extensive data sets.
Innovation Solution
A building management system utilizing machine learning models, specifically convolutional neural networks, to analyze vibration data sets by performing fast Fourier transforms and identifying abnormalities, thereby automating the detection of equipment faults and reducing the workload for human analysts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual analysis by human analysts is used, then operational problems can be detected, but the process becomes time-consuming and costly with limited analysts
Solution Approach 1:
The patent replaces the mechanical system of manual human analysis with an automated machine learning system. The processing circuit automatically performs Fast Fourier Transforms on vibration data sets and uses trained machine learning models to identify abnormalities, eliminating the need for manual parsing while maintaining detection accuracy.
Solution Approach 2:
The system enables self-service by allowing the building management system to automatically analyze its own vibration data without external human intervention. The machine learning models are trained on historical data and autonomously process new data sets, generating anomaly detections that can trigger automated alerts or corrective actions.
2Reliability
If manual parsing of extensive data sets is performed, then operational issues can be identified, but efficiency decreases with large volumes of data
Solution Approach 1:
The patent replaces manual data parsing with automated computational processing. The processing circuit systematically applies Fast Fourier Transforms and machine learning algorithms to vibration data sets, enabling rapid analysis of extensive data volumes while maintaining reliable fault detection capability.
Solution Approach 2:
The system transforms the vibration data from time-domain signals to frequency-domain representations through Fast Fourier Transforms. This parameter transformation enables the machine learning models to more effectively identify abnormal patterns in the frequency spectrum, improving detection capability while processing efficiency.
3Reliability
If human analysts are trained to analyze vibration data, then detection accuracy improves, but training costs and time increase
Solution Approach 1:
The patent replaces the need for trained human analysts with an automated machine learning system. The processing circuit executes predefined algorithms and machine learning models that have been trained on historical vibration data, achieving consistent detection accuracy without requiring human training time or ongoing analyst development.
Solution Approach 2:
The system creates a digital copy of expert analysis capability through machine learning models. These models are trained on labeled historical data representing expert judgments, effectively copying and encoding detection expertise into the algorithm, which can then be replicated indefinitely without additional training time or cost.
4Productivity
If automated machine learning analysis is implemented, then processing efficiency improves, but system complexity increases
Solution Approach 1:
The patent segments the analysis system into distinct functional modules: a processing circuit for data acquisition and Fast Fourier Transforms, machine learning models for pattern recognition, and a output interface for anomaly reporting. This segmentation enables efficient automated processing while managing complexity through modular design, where each component has a specific function.
Data Source
AI summary
A building management system includes building equipment operable to affect a variable state or condition of a building and a controller including a processing circuit. The processing circuit is configured to obtain a vibration data set related to vibrations of the building equipment and analyze the vibration data set by one or more machine learning models to generate a set of probabilities. The set of probabilities is related to a probability that the vibration data set is abnormal. The processing circuit is configured to identify the vibration data set as normal or abnormal based on the set of probabilities and initiate a corrective action responsive to identifying the vibration data set as abnormal.