Fault Diagnosis via Neural Ordinary Differential Equations

Implementation of model-based fault diagnosis systems can be a difficult task due to the complex dynamics most systems, an appealing alternative avoiding modeling is use machine learning-based techniques for which implementation more affordable nowadays. However, latter approach often requires extensive data processing. In this paper, hybrid using recent developments in neural ordinary differential equations proposed. This enables us combine natural deep learning technique with estimated model system, making training simpler and efficient. For evaluation methodology, nonlinear benchmark system used by simulation faults actuators, sensors, process. Simulation results show that proposed methodology less processing comparison conventional approaches since data-set directly taken from measurements inputs. Furthermore, essay only structural approximation plant; no advanced required. also alleviate some pitfalls data-series, such as complicated augmentation methodologies necessity big amounts data.