Gas Sensing Device With Attention-Weighted Temporal Signals
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing gas sensing technologies face challenges in achieving a high accuracy, selectivity, robustness, and resource-efficient implementation for detecting target gases in gas mixtures, particularly due to sensitivity degradation and temporal dependencies.
Innovation Solution
A neural network with an attention layer is employed to weight the contributions of measurement signal samples, leveraging temporal dependencies for improved gas concentration estimation, using an attention-based architecture that exploits long-term dependencies and reduces processing latency and memory footprint.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If recurrent neural networks are used to exploit temporal dependencies, then measurement precision is improved, but device complexity and computational requirements increase
Solution Approach 1:
The patent segments the temporal signal processing into fixed-size windows with overlapping segments. Each window is processed independently by the neural network, breaking down the complex temporal dependency problem into manageable chunks while maintaining temporal context through the overlap between segments.
Solution Approach 2:
The patent applies preliminary signal processing steps including normalization and feature extraction before feeding data to the neural network. The measurement signal is pre-processed to enhance relevant features and reduce noise, allowing the network to focus computational resources on the most informative aspects of the signal.
2Measurement precision
If more measurement signal samples are processed, then measurement precision is improved, but loss of time increases
Solution Approach 1:
The patent implements periodic processing by dividing the continuous measurement signal into discrete time windows that are processed at regular intervals. This periodic approach allows the system to balance between using enough samples for accurate estimation and maintaining real-time processing capabilities through controlled update frequencies.
Solution Approach 2:
The patent performs preliminary feature extraction and signal normalization on incoming data streams before full neural network processing. This pre-processing prepares the data in advance, reducing the computational burden during the actual estimation phase and enabling faster processing of larger sample sets.
3Device complexity
If fixed input size is used for neural network, then device complexity is reduced, but adaptability worsens
Solution Approach 1:
The patent introduces dynamic window size selection that adapts to the characteristics of the measurement signal. The system can adjust the time window length and overlap parameters based on signal variability and processing requirements, allowing the fixed-architecture network to handle variable-length inputs effectively while maintaining implementation simplicity.
Solution Approach 2:
The patent segments variable-length measurement signals into fixed-size windows that match the neural network input requirements. This segmentation approach with configurable window parameters provides flexibility in handling different signal durations and sampling rates while maintaining a consistent fixed-size input format for the network.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A gas sensing device for sensing a target gas in a gas mixture, the gas sensing device comprises: a measurement module configured for obtaining a measurement signal, the measurement signal being responsive to a concentration of the target gas in the gas mixture; and a processing module configured for: determining, for each of a sequence of samples of the measurement signal, a set of features, the features representing respective characteristics of the measurement signal; and using a neural network for determining an estimation of the concentration of the target gas based on the sets of features determined for the samples of the sequence, wherein the neural network comprises an attention layer to weight respective contributions of the samples to the estimation.