Neural Network Waveform Signal Processing for Noise Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing waveform signal processing systems struggle to accurately detect elastic waves in industrial equipment and structures due to noise interference, leading to decreased evaluation accuracy and the inability to acquire desired signals in noisy environments.
Innovation Solution
A waveform signal processing system employing a neural network that generates time-series data for noise and signal components, using a learner to update parameters based on a loss function that minimizes the difference between synthetic and observed waveforms, and a target signal extractor to isolate the desired signal, effectively improving signal-to-noise ratio and maintaining the integrity of the elastic wave signal.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If noise filtering is performed using frequency domain methods (low-pass filter or band-pass filter), then the signal-to-noise ratio is improved, but the frequency domains of noise and elastic wave overlap making it difficult to achieve effective filtering
Solution Approach 1:
The patent replaces traditional mechanical frequency domain filtering methods with a deep learning-based neural network approach. The neural network is trained to automatically distinguish and separate noise from elastic wave signals in the time domain, eliminating the need for frequency domain separation and manual filter design. This substitution enables effective noise filtering even when frequency domains overlap.
Solution Approach 2:
The patent changes the approach from fixed frequency domain parameters to adaptive time-series parameter processing. The neural network dynamically adjusts its filtering behavior based on learned patterns in the input signals, allowing it to adapt to varying noise characteristics and signal conditions without requiring explicit frequency domain knowledge or manual parameter tuning.
2Measurement precision
If high-sensitivity sensors and high amplification are used to detect weak elastic wave signals, then the signal detection capability is improved, but the system becomes vulnerable to noise interference
Solution Approach 1:
The patent converts the harmful noise that inevitably enters the high-sensitivity detection system into a learnable pattern. By training the neural network on data containing both signal and noise, the system learns to identify and separate the useful elastic wave signals from the harmful noise components. The noise, rather than being a limiting factor, becomes part of the training data that enables the network to robustly distinguish signal from noise.
Solution Approach 2:
The neural network acts as an intermediary processing layer between the high-sensitivity sensor and the final signal interpretation. This intermediary learns the complex relationship between noisy sensor outputs and underlying physical signals, effectively decoupling the high amplification (which increases both signal and noise) from the final measurement quality. The network filters out noise while preserving the amplified signal features.
3Measurement precision
If traditional frequency domain filtering is applied to reduce noise, then the signal-to-noise ratio increases, but the shape of the signal may be changed decreasing evaluation accuracy
Solution Approach 1:
The patent replaces mechanical frequency domain filtering with a learned time-domain processing approach. The neural network is trained to preserve the temporal and waveform characteristics of elastic waves while removing noise, avoiding the signal distortion that occurs with traditional filters. This substitution maintains signal integrity because the network learns from labeled examples what the true signal waveform should look like.
Solution Approach 2:
The training process incorporates feedback through the loss function that compares the network's output with ground truth signals. This feedback mechanism ensures that the network learns to preserve important signal characteristics while removing noise. The continuous refinement through backpropagation adjusts the network parameters to minimize waveform distortion while maximizing noise removal effectiveness.
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 removes noise while preserving the waveform characteristics of elastic waves, enhancing the signal-to-noise ratio by up to 6 dB and improving evaluation accuracy even in high-noise environments, allowing for precise detection of structural deterioration.
Implementation Method 1
an AE sensor using a piezoelectric element
Data Source
AI summary
According to one embodiment, a waveform signal processing system according to an embodiment includes a neural network, a learner, and an extractor. The neural network is configured to generate at least first time-series data on noise and second time-series data on a signal other than noise on the basis of input random noise. The learner is configured to update parameters of the neural network on the basis of a loss function including a main limitation term of which a value becomes lower as synthetic time-series data obtained by adding the first time-series data and the second time-series data generated by the neural network and an observed time-series waveform including noise become more similar to each other. The extractor is configured to extract at least one of the first time-series data and the second time-series data generated by the neural network as a target signal on the basis of the parameters updated by the learner.