Machine Fault Detection Using Neural Signal Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing fault detection methods in rotating machinery are challenged by indirect sensor measurements, noise interference, and the need for precise bearing geometry knowledge, with machine learning approaches being opaque and difficult to interpret.
Innovation Solution
A method combining envelope spectrum analysis with an unsupervised neural network to enhance fault detection by filtering noise and emphasizing impulsive components in vibration signals, using a cost function to optimize the neural network's performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning algorithms are used for fault detection, then detection accuracy is improved, but interpretability and transparency deteriorate
Solution Approach 1:
The patent introduces an intermediary step between raw sensor data and fault detection: a generative model creates a synthesized fault signal that preserves the characteristics of actual faults while removing confounding factors. This intermediary representation maintains interpretability by allowing domain experts to understand and verify the synthesized fault patterns against known fault signatures, while still achieving high detection accuracy through the power of generative modeling.
2Measurement precision
If envelope spectrum analysis is applied to raw vibration signals, then fault frequencies can be identified, but noise and other vibrations mask the fault signals
Solution Approach 1:
The patent extracts the essential fault characteristics from the raw vibration signal by using a generative model to synthesize a cleaned fault signal. This extraction process removes noise and confounding vibrations while preserving the characteristic fault frequencies and patterns, enabling clear identification of fault frequencies through subsequent envelope spectrum analysis without the masking effects of background noise.
3Measurement precision
If precise bearing geometry knowledge is required for fault detection, then detection accuracy is improved, but the method becomes less adaptable to different bearing configurations
Solution Approach 1:
The patent creates a synthetic copy of the fault signal through generative modeling, where the model learns the characteristic patterns of faults from training data and generates realistic fault signatures. This copied fault signal preserves the essential diagnostic information needed for accurate damage classification while being independent of specific bearing geometry details, allowing the method to adapt to different bearing configurations without requiring precise geometric knowledge for each case.
Data Source
AI summary
In a method for detecting a fault in a machine, a vibration signal recorded by one or more sensors sensing a vibration of the machine is received. A true fault signal is determined by applying to the vibration signal a neural network which is an unsupervised neural network, and an envelope spectrum analysis is applied to the true fault signal to detect the fault. The unsupervised neural network is designed to determine the true fault signal such that an impulsive component in the true fault signal is maximized, and is designed to determine the true fault signal such that a difference between the true fault signal and the vibration signal is minimized.