Use of thermodynamic steady-state data for the analysis of turbofan engines: towards automatic fault detection and analysis

As part of this research, the thermodynamic behavior of turbofan engines has been under investigation on the basis of synthesized, steady-state data in order to detect anomalous behavior. The paper focuses on failure detection, which constitutes a crucial component of the broader process of diagnosis. Subsequent stages will expand the diagnosis capabilities by implementing the analysis of the underlying causes of failures. A dataset consisting of thermodynamic data has been generated with a gas turbine performance simulation program, considering both component and measurement failures to replicate real-world conditions. Based on this simulated dataset, a clustering methodology has been developed to categorize and identify patterns. Utilizing synthetic data allows for a gradual escalation in complexity, enabling a systematic enhancement of clustering techniques to address increasing challenges effectively. In addition this approach enables the investigation of reliable limits for the detection of anomalies and the differentiation from normal operation. To carry out this investigation, new modules have been integrated into an existing software system, firstly to improve analytical capabilities and secondly to enable seamless integration of automated fault detection and analysis. The findings indicate that detecting various failure scenarios by identifying deviations from the nominal engine state is feasible through the use of clustering algorithms, even in datasets containing noise. This approach not only improves failure detection, but also lays the groundwork for failure analysis of the underlying causes which will be part of future work. As such, this research contributes to the advancement of automated methods for diagnosing engine performance based on typical thermodynamic engine measurements.