Probabilistic Diagnosis of Aviation Engine with Invertible Neural Network

Invertible neural network (INN) is a promising technique for the reasoning and causal inference of degraded system performance. While generating forward predictions from given inputs to the system response, INN enables the inverse process without much extra cost. The inverse process of INN predicts the possible input parameters for the specified system response qualitatively. For the purpose of fault diagnosis for aviation engine fuel efficiency, accurate predictions from the inverse process are required. Moreover, INN predictions lack effective uncertainty quantification for regression tasks, which increases the challenges of decision making. This paper proposes a probabilistic diagnosis framework with INN. The dependency of INN on large dataset has been addressed through generating training data through Gaussian Processes while an uncertainty prediction for the INN forward process has been designed. The representational capability of INN for posterior distributions has been enhanced through probabilistic inference. Optimum engineering setups have been explored with specialized post-process approaches. The proposed method has been applied to the diagnosis of aviation engine fuel efficiency and demonstrated better accuracy over the standard INN in both forward and inverse processes.