Traditional spectroscopic techniques require multiple lasers, adding complexity and cost. By integrating AI, we use a single laser for highly selective, simultaneous multi-species detection, making LAS more practical for real-time analysis.
Our research proposes four key methods to advance laser-based spectroscopic sensing:
Multi-Speciation Detection Using a Single Laser: We developed a methodology using a DFB-ICL near 3.3 µm, with DNNs for simultaneous detection of gases like methane and propane during high-temperature processes.
Deep Denoising Autoencoders: DDAEs are used to clean noisy spectra from shock tube experiments, uncovering hidden spectral features and improving data quality.
Unsupervised Blind Source Separation (BSS): We used autoencoders for hydrocarbon quantification without reference spectra, simplifying spectroscopic analysis by estimating species concentrations.
Three-Phase Sensing for Water Cut Measurement: A self-calibrating laser-based sensor measures water content in oil-water mixtures, using unsupervised learning for continuous self-calibration, eliminating the need for external calibration.
Impacts:
- Field Applicability: These AI-enhanced methods are designed to operate effectively in harsh conditions, providing accurate data even in the presence of noise and interference.
- Operational Efficiency: The three-phase self-calibrating sensor reduces operational costs by eliminating the need for manual recalibration, making it more practical for continuous use in the field.
- Enhanced Data Quality: Deep denoising autoencoders significantly improve the quality of data, allowing for precise, real-time monitoring, which is crucial for safety and optimization in combustion and industrial processes.
.png?sfvrsn=1b6bae1a_0)
(FASTER).png?sfvrsn=ac211bda_0)
