PSO优化多尺度一维卷积神经网络的风机基础螺栓松动诊断

为在非经验指导下获取多尺度一维卷积神经网络中卷积核数目和尺度最优参数,实现风机基础螺栓松动智能诊断,提出粒子群优化(particle swarm optimization, PSO)多尺度一维卷积神经网络的风机基础螺栓松动诊断方法。首先,获取风机一维原始振动信号,划分训练集与验证集;然后,将多尺度一维卷积神经网络中卷积核数目和尺度作为PSO的粒子,以验证精度作为适应度值,根据适应度值更新粒子速度和位置,经训练后获得最优卷积核数目和尺度参数下的多尺度一维卷积神经网络;最后,输入测试样本,得到风机基础螺栓松动诊断结果。在稳定转速和升降速下进行风机基础螺栓松动诊断试验,结果表明,PSO优化多尺度一维卷积神经网络的风机基础螺栓松动诊断方法可在非经验指导下获取最优参数,可从一维原始信号中提取出有效松动特征,具备良好的松动诊断效果。

PSO optimized multi-scale one-dimensional convolutional neural network for fan foundation bolt looseness diagnosis

In order to obtain the number and scale of convolution kernels in multi-scale one-dimensional convolution neural network under non-empirical guidance, and realize intelligent looseness diagnosis of fan foundation bolt, a bolt looseness diagnosis method for fan foundation based on multi-scale one-dimensional convolution neural network optimized by the particle swarm optimization(PSO) was proposed. Firstly, the one-dimensional original vibration signal of the fan was collected and divided into the training set and validation set. Then, the number and scale of the convolution kernel in the multi-scale one-dimensional convolutional neural network were taken as the particles of the PSO. Verification accuracy is set as fitness value, and the particle speed and position are updated according to the fitness value. After training, a multi-scale one-dimensional convolution neural network with optimal number of convolution kernels and scale parameters was obtained. Finally, Test samples were input to obtain diagnostic results of the fan foundation bolt looseness. The method was applied to a bolt loosening diagnosis experiment of the fan foundation under stable speed and rising and falling speed. The experimental results show that the multi-scale one-dimensional convolutional neural network optimized by PSO can obtain optimal parameters under non-empirical guidance, can extract effective looseness features from the one-dimensional original signal, and has a good looseness diagnostic effect.