An Enhanced Deep Extreme Learning Machine for Integrated Modular Avionics Health State Estimation

Abstract

Integrated modular avionics (IMA) is one of the most advanced systems whose performance deeply impact on the security of civil aircraft. In order to enhance the safety and reliability of aircraft, the health state of the IMA must be estimated accurately. Since IMA is a real-time system, the estimation algorithm should have fast learning speed to satisfy the real-time requirement. In this paper, an enhanced deep extreme learning machine is developed to estimate the health states of IMA. First, the enhanced deep extreme learning machine is built in a novel fashion by using a dropout technique and extreme learning machine autoencoder. Second, multiple-enhanced deep extreme learning machines with different activation functions are employed to estimate the health states, simultaneously. Finally, a synthesis strategy is designed to combine all the results of different enhanced deep extreme learning machines. In such a manner, the robust and accurate estimation results can be obtained. In order to collect the data under different health states, a performance degradation model of IMA is built by the intermittent faults. The proposed method is applied to health state estimation, and the results confirm that the proposed method can present a superior estimation to the conventional methods.