【摘要】 针对原始振动信号不可避免的包含多余噪声问题。提出一种基于稀疏滤波(sparse filtering, SF)和长短期记忆网络(long and short term memory network, LSTM)相结合的旋转机械故障诊断模型，该模型利用快速傅立叶变换将原始时域信号转换成频域信号，再通过SF提取低维故障特征，并将其输入到LSTM堆叠分类器中识别旋转机械故障状态。用轴承和齿轮振动信号为例开展试验研究，并与Softmax、深度神经网络(deep neural networks, DNN)、支持向量机(support vector machine, SVM)、降噪自编码器(denoising auto-encoder, DAE)等方法进行试验对比，结果表明所提方法不仅在噪声环境下具有更高的准确率和鲁棒性，而且针对数据不平衡集的诊断也能达到98%以上的准确率。更多还原

【Abstract】 Here, aiming at the problem of original vibration signals inevitably containing redundant noise, a rotating machinery fault diagnosis model based on combination of sparse filtering(SF) and long-short term memory network(LSTMN) was proposed.In this model, fast Fourier transform was used to convert original time domain signals into frequency domain ones, and then low-dimensional fault features were extracted with SF, they were input into LSTMNstack classifier to identify fault states of rotating machinery.Vibration signals of bearings and gears were taken as examples to conducttest studies, and the results were compared with those obtained using softmax, deep neural networks(DNNs), support vector machine(SVM), denoising auto encoder(DAE) and other methods. The results showed that the proposed method can not only have higher accuracy and robustness in noisy environment, its diagnosis fordata unbalance sets can also reach a correctnessrate of more than 98%.更多还原