Effective Maintenance of Components in T700 Engine Using Backpropagation Neural Network

Predicting the exact time of failure for aircraft components is critical as a failure may cause a fatal accident, have a high cost, and waste a large amount of time. Accurate prediction will help reduce the occurrence of unexpected failures and ensure safe flights. Thus, we propose a model for predicting the lifetime and failure of components, which uses the modified Delphi method and a backpropagation neural network (BPNN). To select the significant factors that affect the lifetime, a questionnaire survey on experts was first carried out. As a result, 17 factors were defined, and through a second survey, the following seven factors were selected from the criteria of average scores and standard deviations: operation hours after installation, the resistance of the thermocouple assembly, and the ohm values obtained from a hydraulic machinery control unit linear displacement sensor, a power turbine speed sensor, a torque and overspeed sensor, an overspeed leakage solenoid valve, and the torque motor of the hydraulic control unit. The training data were obtained from maintenance data using various sensors of the electronic control unit (ECU) of an engine (T700) of a helicopter in Taiwan collected during 2011-2013. By using Alyuda NeuroIntelligence software, the relationship between the input and output data (predicted time to component failure) was found and used in the prediction model. The coefficients of relevance and model fitting were 0.999 and 0.997, respectively, and the average prediction accuracy of 15 data sets calculated from the mean absolute percentage error (MAPE) was 92.45%. This result confirmed that the new BPNN model predicted the time of component failure effectively. The validated prediction ability of the BPNN model provides a reference for the maintenance management of various aircraft components and an effective maintenance strategy.