Thursday 27 March 2025
In recent years, artificial intelligence has made tremendous strides in various fields, including engineering and physics. Researchers have been working on developing new methods that can help us better understand complex systems and make accurate predictions about their behavior.
One such method is called sparse identification of nonlinear dynamics (SINDy), which uses machine learning algorithms to identify the underlying physical laws governing a system’s behavior from limited data. This approach has shown great promise in fields such as fluid dynamics, optics, and even biology.
Now, researchers have applied SINDy to a new domain: turboshaft engines, which power many modern aircraft and helicopters. Developing accurate models of these engines is crucial for designing more efficient and sustainable aircraft, but it’s a challenging task due to the complexity of the engine’s internal workings.
To tackle this problem, the researchers used a dataset of real flight tests from Leonardo’s AW189 helicopter prototypes. They trained two types of machine learning models: feedforward neural networks (FFNNs) and long short-term memory (LSTM) networks. These models were designed to predict the engine torque based on various input variables such as fuel flow, engine speed, and environmental conditions.
The results showed that the LSTM model performed significantly better than the FFNN model, accurately predicting the engine’s behavior over a wide range of flight conditions. However, even the LSTM model struggled with certain types of maneuvers, such as those involving high-frequency oscillations in the engine torque.
To improve the accuracy of the models, the researchers turned to SINDy, which uses a different approach to identify the underlying physical laws governing the system’s behavior. By analyzing the dataset and looking for patterns, SINDy was able to develop a simple yet accurate model that accurately predicted the engine’s behavior even in complex scenarios.
The key to SINDy’s success lies in its ability to identify the most important variables and relationships between them, allowing it to distill the underlying physics of the system down to a few simple equations. This approach not only improves the accuracy of the models but also provides valuable insights into the engine’s internal workings.
In addition to improving our understanding of turboshaft engines, SINDy has broader implications for many fields where complex systems are involved. By applying this method to different domains, researchers can develop more accurate and reliable models that can help us better understand and predict the behavior of complex systems.
Cite this article: “AI-Powered Models Revolutionize Turboshaft Engine Predictions”, The Science Archive, 2025.
Artificial Intelligence, Machine Learning, Sparse Identification Of Nonlinear Dynamics, Sindy, Turboshaft Engines, Fluid Dynamics, Optics, Biology, Feedforward Neural Networks, Long Short-Term Memory Networks.
