基于DA改进深度置信网络的滚动轴承故障诊断

Fault Diagnosis of Rolling Bearing Based on DA Improved Depth Confidence Network

针对滚动轴承结构及其运行环境的复杂性和信号特征参数信息难以被提取的问题,提出一种由蜻蜓算法(DA)优化改进的深度置信网络方法(DA-IDBN)并应用于轴承故障诊断中.采用DA对神经网络结构进行优化设计,确定最优结构.应用最小批量随机梯度下降法对每一个自适应受限波尔茨曼机(RBM)进行预训练,并采用BP神经网络反向微调DBN的权重参数.通过实验验证了笔者所提方法的有效性,并取得了更好的分类效果.

In view of the complexity of rolling bearing structure and its operating environment and the difficulty of extracting signal characteristic parameter information,a depth confidence network method(DA-IDBN)optimized and improved by dragonfly algorithm(DA)is proposed and applied to bearing fault diagnosis.DA is used to optimize the neural network structure and determine the optimal structure.The minimum batch random gradient descent method is used to pre-train each adaptive restricted Boltzmann machine(RBM),and BP neural network is used to fine-tune the weight parameters of DBN in reverse.The effectiveness of the proposed method is verified by experiments,and a better classification effect is achieved.