Combining Machine Learning, Embedded Sensor Networks and Additive Burner Design for Combustor Structural Health Monitoring: GT2024-127925

The combination of novel technologies such as machine learning, embedded sensor networks and additive burner design can significantly improve structural health monitoring compared to state-of-the-art methods using thresholding, pattern recognition and alarm levels. An aircraft combustor that 'feels' its effective operating conditions in real-time and accurately assesses its adaptive maintenance needs is key to meeting tomorrow's more demanding requirements specific to hydrogen-fuelled aircraft engines. The combination of these three technologies is being addressed in the crystAIr project.

An additively manufactured, highly sensitised burner (of the Recursive Sequential Combustion type, where avoidance of flashback is crucial) is used as an experimental combustion test case. In this part the fuel is propane.

One goal is to use AI to predict flashback from fast pressure sensor data as early as possible. To obtain reliable ground truth, a model is trained with signals representing the normal combustion process under desired conditions. These processes are recorded on the instrumented setup (with fast pressure sensors, temperature sensors and differential pressure sensors monitoring the flow in the burner), combined with a photodiode observing the flame and other operational data. All this optimises an auto-encoder model that learns how to reconstruct the sensor data describing the normal combustion process. A series of flash-back data is then presented where the large deviations in the reconstruction error indicate and pinpoint the disturbance in the combustion process. These events in the data are then used in the next stage to train a classifier and later an early predictor of flashback events based on the fast-pressure transducer data alone.

The paper describes the concept, the burner, the measurement chain and the AI architecture. It comments on best practice in terms of measurement location and the relevance of these, and reports about the first results.