Data model-based sensor fault diagnosis algorithm for closed-loop control systems

•A linear model is dynamically established based on the gradient change of the control variables from the sensor data, and a baseline model tracker is designed.•The engine baseline model is trained and updated using historical data from healthy sensors.•Reasonable diagnostic thresholds are calculate...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Measurement : journal of the International Measurement Confederation Ročník 246; s. 116715
Hlavní autoři: Han, Xinhao, Zhou, Xin, Lu, Feng, Huang, Jinquan
Médium: Journal Article
Jazyk:angličtina
Vydáno: Elsevier Ltd 31.03.2025
Témata:
Adaptive threshold setting
Data model
Neural network
Nonlinear tracker
Real-time sensor fault diagnosis
Turboprop engine
Nonlinear tracker
Data model
Turboprop engine
Neural network
Real-time sensor fault diagnosis
Adaptive threshold setting
ISSN:0263-2241
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:•A linear model is dynamically established based on the gradient change of the control variables from the sensor data, and a baseline model tracker is designed.•The engine baseline model is trained and updated using historical data from healthy sensors.•Reasonable diagnostic thresholds are calculated using the characteristics of historical data from healthy sensors.•The performance of four sensor fault threshold-setting methods is compared under various sensor fault modes. The varying operational parameters and random noise make it difficult to determine the fault diagnosis thresholds for engine sensors under different working conditions. Therefore, an adaptive threshold-based fault diagnosis method for aeroengine sensors is proposed. A multivariable control system based on the MFAC method is established for the aeroengine. The OS-ELM algorithm employs historical sensor data to train and update the engine baseline model. MFAC dynamically establishes a linear model based on the pseudo-gradient change of control variables from the current sensor data and designs a baseline model tracker to calculate reasonable diagnostic thresholds based on historical sensor data characteristics, thereby improving the efficiency of threshold calculation and diagnostic accuracy. The experimental results validate that this method improves the fault detection rate by at least 30% while ensuring a low false alarm rate, reduces the minimum detectable fault magnitude by 39%, and keeps the fault detection time within 0.2 s.
ISSN:0263-2241
DOI:10.1016/j.measurement.2025.116715