Vibration-Based Human Activity Classification Using FFT and Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies struggle to accurately identify and categorize the actions causing impacts on a structure based on the vibrations generated, as they primarily focus on locating and measuring the force of impacts rather than classifying the actions.
Innovation Solution
A system that utilizes sensors to capture vibration signals, decomposes these signals into time and frequency domain components, and employs machine learning or artificial intelligence algorithms to classify the actions causing the vibrations, without the need for amplification or specific localization techniques.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional impact sensors and accelerometers are used to detect vibrations, then the force and location of impacts can be measured, but the ability to accurately classify and identify the specific action causing the vibration is insufficient
Solution Approach 1:
The patent segments the vibration signal into multiple frequency components using Fast Fourier Transform (FFT), dividing the complex signal into analyzable frequency bands. This segmentation allows extraction of specific features from different frequency ranges that correspond to different action types, enabling accurate classification while maintaining measurement precision.
Solution Approach 2:
The patent transforms the vibration analysis from traditional time-domain only measurement to include frequency-domain analysis. By adding the frequency dimension through spectral analysis, the system extracts additional features (frequency components, spectral centroid, bandwidth) that provide rich information for action classification without losing the original impact location and force data.
2Reliability
If amplification and specific localization techniques are employed to enhance vibration detection, then detection sensitivity improves, but system complexity increases
Solution Approach 1:
The patent replaces complex mechanical amplification systems with signal processing algorithms. Instead of using physical amplifiers and complex localization hardware, the system uses digital signal processing techniques (FFT, feature extraction, machine learning classifiers) to enhance detection sensitivity and identify action types directly from the raw vibration signals.
Solution Approach 2:
The patent enables the vibration sensor system to self-analyze and self-classify actions through embedded signal processing and machine learning algorithms. The system processes its own output signals through frequency analysis and feature extraction, eliminating the need for external amplification equipment or complex localization systems, thereby reducing overall device complexity while maintaining high reliability.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
The system effectively categorizes actions based on vibrations, providing a high degree of accuracy in identifying the specific action that caused the vibrations, even in scenarios where the location of impact is unknown.
Implementation Method 1
An accelerometer associated with the rigid body measures the rate of deceleration of the rigid body, as a function of time, in terms of a voltage signal waveform
Implementation Method 2
When the member vibrates after being impacted by the object, an oscillatory electrical signal is produced by a piezoelectric sensor
Data Source
AI summary
The current disclosure is directed to classifying/identifying an action based on the vibrations caused in/on/around a structure by the action and providing a confidence level association between the vibration and the action that caused the vibration.