优化神经网络算法在航空发动机故障诊断中的应用研究

Analysis on the Application of Optimal Neural Network Algorithm in Aero-engine Fault Diagnosis

神经网络的机器学习、深度学习具有较强的特征分类能力,但权值和初值的选取影响着故障诊断的精度。遗传算法、粒子群算法及后续的优化算法容易陷入局部最优,表现出自身的局限性。大数据挖掘与基于优化算法的深度学习神经网络成为目前主流的研究方向。该文在研究航空发动机故障诊断的基础上,总结了航空发动机的故障类型、故障诊断数据来源,简单介绍了几种典型优化算法和诊断网络,分析了优化算法对浅层神经网络优化方式,研究了深度神经网络在相邻领域的应用及其成效。根据神经网络的收敛速度和诊断精度判定神经网络的优劣,并提出基于混合自适应粒子群算法与卷积神经网络的航空发动机故障诊断方法。

Neural network machine learning and deep learning have strong feature classification capabilities,but the selection of weights and initial values affects the accuracy of fault diagnosis.Genetic algorithm,particle swarm algorithm and subsequent optimization algorithms show their own limitations because they are easy to fall into the local optimum.Big data mining and deep learning neural networks based on optimization algorithms have become the current mainstream research directions.Based on the study of aero-engine fault diagnosis,this paper summarizes the fault types and fault diagnosis data sources of aero-engines,and briefly introduces several typical optimization algorithms and diagnosis networks.The optimization algorithm to the shallow neural network optimization method was analyzed,and the application and effectiveness of the deep neural network in adjacent fields were studied.According to the convergence speed and diagnosis accuracy of the neural network,the pros and cons of the neural network are judged,and an aeroengine fault diagnosis method based on hybrid adaptive particle swarm algorithm and convolutional neural network is proposed.