Researchers have developed a new machine learning-based technique that uses only spectral data to monitor laser temperature—no contact sensors required—opening the door to simpler, non-invasive diagnostics in complex laser environments.
Study: Towards Neural-Network-Based Optical Temperature Sensing of Semiconductor Membrane External Cavity Laser. Image Credit: NicoElNino/Shutterstock.com
Published in Advanced Sensor Research, the study outlines how neural networks can be trained to read subtle changes in a laser’s emitted light spectrum to accurately infer temperature. Traditional temperature monitoring typically relies on contact sensors, which can be invasive, impractical in high-temperature settings, or disruptive to laser system design.
This new approach offers a cleaner alternative by tapping into the intrinsic thermal signatures already present in the light lasers emit.
Background
Optical methods for sensing temperature aren’t new. Techniques based on fluorescence, phosphorescence, or material-specific spectral shifts have long been explored. But these often demand specialized components and tend to fall short in range, resolution, or environmental adaptability.
Machine learning—especially neural networks—has emerged as a promising way to interpret complex spectral data, particularly where traditional methods struggle due to noise or sparsity.
Neural networks are well-suited to pattern recognition, making them effective for identifying subtle spectral cues linked to temperature. While prior studies have shown success applying neural networks in areas like laser diagnostics and ultrafast photonics, their use in optical thermal sensing for laser systems remains relatively unexplored.
The Current Study
To test their concept, the researchers worked with two laser sources: a near-infrared diode laser operating around 800 nm and a membrane external cavity laser (MECSEL) emitting at 1070 nm. Emission spectra was collected using three compact spectrometers spanning visible and NIR ranges, offering resolution fine enough to capture temperature-induced spectral shifts.
Data collection involved carefully controlled temperature settings (15 to 25 °C), with multiple measurements taken at each point to ensure robustness against real-world noise. This rich dataset was used to train multi-layer feed-forward neural networks, fine-tuned through hyperparameter optimization to balance accuracy and model complexity.
The networks were designed to match the resolution and wavelength range of the spectrometers, and were trained to correlate spectral patterns with actual temperatures, verified using conventional sensors. The authors also suggest that transfer learning could help adapt this technique to other laser systems or temperature ranges by building on pretrained models.
Results and Discussion
The trained models demonstrated impressive accuracy, achieving sub-percent mean square error rates compared to standard sensors. This highlights the potential of combining spectral data with neural networks to perform precise thermal monitoring.
Several practical advantages stand out. First, the method enables remote, non-contact sensing—ideal for environments where physical access is limited or undesirable. Second, the models are computationally efficient, making real-time predictions possible even on low-power embedded systems. Third, the network design can be tailored to different performance or resource requirements, offering flexibility across various use cases.
Conclusion
By integrating neural networks with compact spectrometers, the researchers have created a lightweight, accurate system for monitoring laser gain medium temperatures without the need for additional hardware or intrusive setups.
The system’s scalability through transfer learning and its potential to sense other laser parameters via spectral analysis point to broader applications in photonic diagnostics. Ultimately, this work showcases how machine learning can streamline temperature monitoring in laser systems, improving their reliability, performance, and ease of maintenance across research, industrial, and commercial settings.
Journal Reference
Mannstadt J., & Rahimi-Iman A. (2025). Towards Neural-Network-Based Optical Temperature Sensing of Semiconductor Membrane External Cavity Laser. Advanced Sensor Research, e70009. DOI: 10.5281/zenodo.15262006, https://advanced.onlinelibrary.wiley.com/doi/10.1002/adsr.70009