Generalized inverse matrix - long short-term memory neural network data processing algorithm for multi-wavelength...

Area of Science:

• Physics
• Engineering
• Computer Science

Background:

• Multi-wavelength pyrometry (MWP) faces challenges with underdetermined equations (N equations, N+1 unknowns) in data processing.
• Traditional iterative optimization methods lack the accuracy and efficiency required for real-world measurements.
• Artificial intelligence (AI) presents a potential solution for overcoming these data processing limitations.

Purpose of the Study:

• To develop an advanced data processing algorithm for MWP that addresses accuracy and efficiency issues.
• To completely eliminate the influence of emissivity on temperature measurements.
• To enable real-time monitoring applications, particularly for high-emissivity objects.

Main Methods:

• A hybrid approach combining a Generalized Inverse Matrix (GIM) with a Long Short-Term Memory (LSTM) neural network is proposed.
• GIM is utilized for emissivity classification.
• The classified emissivity data is then processed by the LSTM network for temperature inversion.

Main Results:

• The GIM-LSTM algorithm demonstrated superior accuracy compared to traditional GIM-EPF and BP methods.
• The algorithm exhibits excellent anti-noise performance, maintaining a low relative error even with added random noise.
• Application to rocket engine exhaust plume data showed reduced average relative error compared to conventional methods.
• The GIM-LSTM algorithm achieves millisecond-level operational time, enabling real-time monitoring.

Conclusions:

• The GIM-LSTM algorithm provides a highly accurate and efficient solution for MWP data processing.
• This AI-based method effectively handles emissivity variations and noise, outperforming existing techniques.
• The algorithm's speed is crucial for real-time applications, such as monitoring rocket exhaust plumes.