Vibration Fault Detection Using 2D-CNN Time-Frequency Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current mechanical device fault detection methods lack high accuracy due to difficulties in extracting fault features from vibration signals, especially in real-world environments with noise and varying conditions, failing to capture both transient and long-time deterioration features effectively.
Innovation Solution
A method and apparatus using a 2D-CNN neural network model that processes 3D time-frequency joint distribution cubes, performing dimensionality-reduced feature extraction to generate LST-FD joint distribution matrices, which are used to classify faults, capturing both short-time transient and long-time deterioration features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional vibration signal analysis methods are used for fault detection, then the detection process is simple, but the prediction accuracy is low and fault features cannot be extracted effectively in noisy environments
Solution Approach 1:
The patent transforms 1D vibration signals into 2D time-frequency joint distribution cubes by introducing the frequency domain dimension through STFT transformation. This dimensional expansion allows the CNN model to capture both transient and long-time deterioration features simultaneously, resolving the contradiction between detection accuracy and environmental noise interference
Solution Approach 2:
The patent introduces LST-FD joint distribution matrices as an intermediary representation between raw vibration signals and fault classification. These matrices serve as a bridge that preserves both time-domain transient features and frequency-domain characteristic features, enabling accurate fault detection without requiring complex multi-sensor systems
2Measurement precision
If only time domain statistical indicator extraction is used for fault feature extraction, then the method is simple, but short-time transient features and long-time decay features cannot be captured comprehensively
Solution Approach 1:
The patent applies STFT transformation to convert time-domain signals into time-frequency joint distribution cubes, adding the frequency dimension to the original time domain. This allows simultaneous capture of short-time transient features (through time localization) and long-time deterioration features (through frequency spectrum analysis), resolving the limitation of single-domain analysis
Solution Approach 2:
The patent segments the vibration signal into multiple time-frequency blocks that are processed independently by the CNN model. Each block captures local transient features, while the sequence of blocks preserves long-time deterioration trends. This segmentation approach enables comprehensive feature extraction without requiring excessively complex processing
Data Source
AI summary
A method and apparatus for determining a mechanical device fault are provided. The method includes: obtaining vibration data corresponding to a target test position of a device and a preset neural network model (S101); inputting the vibration data to the preset neural network model, and obtaining an output result of the preset neural network model (S102); determining a target label included in the output result, where the target label is either of a fault label and a non-fault label (S103); and determining, in a case that the target label is the fault label, that a fault occurs at the target test position; otherwise, determining that the fault does not occur at the target test position (S104). Therefore, the technical problem that a high-accuracy mechanical device fault detection means is lacked in the related art is resolved.