Embedded Abnormal Sound Detection Using FFT and CNN-LSTM
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing abnormal sound detection methods for vehicles and mechanical equipment rely heavily on human judgment and complex deep learning systems, which are costly, network-dependent, and prone to failures due to high computing demands and network instability, especially when dealing with high-frequency sound data from air conditioners and compressors.
Innovation Solution
An offline embedded system for abnormal sound fault detection comprising a sound acquisition module, audio feature extraction module, and a neural network module with CNN, LSTM, and trigger decision layers, which processes audio signals locally and reduces network dependence, using a 48 KHz sampling frequency and FFT for accurate anomaly classification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning-based abnormal sound detection methods are used, then detection accuracy is improved, but system complexity and computing burden increase
Solution Approach 1:
The patent uses pre-trained deep learning models that have been copied and deployed on embedded devices. The complex model training is performed offline, and the trained model parameters are copied to the embedded end system, allowing accurate detection without requiring the embedded device to perform complex training computations
Solution Approach 2:
The patent performs model training and optimization in advance before deployment. The deep learning models are pre-trained on comprehensive datasets, and the trained models are then deployed to embedded devices for inference only, separating the complex training phase from the simple detection phase
2Measurement precision
If fault detection systems are deployed on server with network connection, then detection capability is improved, but network dependence increases and reliability decreases
Solution Approach 1:
The patent enables the embedded device to perform fault detection independently using locally deployed deep learning models. The system processes audio data locally without requiring network connection to server, making the system self-sufficient and eliminating network dependence while maintaining detection capability
Solution Approach 2:
The patent separates the detection function from the server infrastructure and embeds it directly in the edge device. This segmentation allows the detection system to operate autonomously at the edge without relying on centralized server resources or network connectivity
3Measurement precision
If high sampling frequency (48 KHz) is used for accurate sound detection, then detection precision is improved, but data transmission volume increases putting pressure on network
Solution Approach 1:
The patent extracts only the essential audio features needed for detection and processes them locally on the embedded device. By performing feature extraction and detection locally, the system eliminates the need to transmit large volumes of high-frequency audio data over the network, keeping only necessary data transmissions
4Ease of operation
If real-time audio is transmitted over network, then remote detection is enabled, but frame drop occurs causing detection failure
Solution Approach 1:
The patent enables the embedded device to perform complete audio processing and detection locally without network transmission. The system captures, processes, and analyzes audio data in real-time on-device, eliminating network transmission entirely and thus preventing frame drop while maintaining detection 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 provides reliable, cost-effective, and accurate offline anomaly detection with reduced network dependence, improved accuracy through LSTM and trigger decision layers, and stable performance using a digital microphone, avoiding frame drop and generalization errors.
Implementation Method 1
The sound acquisition module converts sound from a sound source to be detected into an audio digital signal
Implementation Method 2
The sound audio feature extraction module processes the audio digital signal in a frequency domain to obtain an audio frequency sample
Data Source
AI summary
System and method for offline embedded abnormal sound fault detection are disclosed, the system comprising a sound acquisition module, a sound audio feature extraction module, and a neural network module. The sound audio feature extraction module uses fast Fourier transform to process sample data in a frequency domain, and then inputs the sample data to the neural network module to complete anomaly classification. The neural network module comprises at least one CNN feature extraction layer, a long short-term memory (LSTM) layer, at least one fully connected and at least one classification layer, and a trigger decision layer. The number of network layers of the at least one CNN feature extraction layer is dynamically adjustable, a network structure of the at least one fully connected layer and the at least one classification layer is dynamically variable, and the trigger decision layer is configured to eliminate generalization errors generated by a neural network. The present disclosure works in an offline environment, has less dependence on the network, strong performance and reliable operation, and can adapt to changing working environment for anomaly diagnosis.