Thermoacoustic stability prediction using classification algorithms

Abstract Predicting the occurrence of thermoacoustic instabilities is major interest in a variety engineering applications such as aircraft propulsion, power generation, and industrial heating. Predictive methodologies based on physical approach have been developed past decades, but moderate-to-high computational cost when exploring large number designs. In this study, stability prediction capabilities four well-established classification algorithms—the K -Nearest Neighbors, Decision Tree (DT), Random Forest (RF), Multilayer Perceptron (MLP) algorithms—are investigated. These algorithms are trained using an in-house physics-based low-order network model tool called OSCILOS. All able to predict which configurations thermoacoustically unstable with very high accuracy low runtime. Furthermore, frequency intervals containing modes for given configuration also accurately predicted multilabel classification. The RF algorithm correctly predicts overall finds all 99.6 98.3% configurations, respectively, makes it most accurate training examples available. For smaller sets, MLP becomes algorithm. DT found be slightly less accurate, can extremely quickly runs about million times faster than traditional tool. findings could used devise new generation combustor optimization tools that would run much existing codes while retaining similar accuracy.