基于小波包字典优化的旋转机械振动信号压缩感知重构方法

Compressed Sensing reconstruction forrotating machinery vibration signals based on the wavelet packet dictionary optimization

采用工业无线传感器网络的机械状态监测系统需要进行复杂的数据压缩和高精度的重构,而传感器网络节点资源受限,针对这一问题提出基于小波包字典优化的旋转机械振动信号压缩感知重构方法。该方法结合小波包多分辨率分析及K-SVD字典训练方法,提出了小波包字典优化方法代替传统的正交基字典稀疏表示方法,提高稀疏度。根据旋转机械振动信号自身特征,提出用块稀疏贝叶斯学习最大期望值算法,代替传统仅依赖于稀疏假设的算法实现信号重构。实际轴承振动信号仿真结果表明,该方法相对于传统的压缩感知方法重构性能明显提高。

Machinery condition monitoring systems using industrial wireless sensor networks need complicated data compression and high-precisionreconstruction,however, thereare some limitations in the node resources of wireless sensor networks. A compressed sensing reconstruction method for rotating machinery vibration signals was proposed based on the wavelet packet dictionary optimization. Combining the multiresolution analysis with the K-SVD dictionary training, the wavelet packet dictionary optimization was introduced to replace the traditional sparse transformation method based on the orthogonal basis dictionary for improving the signal sparseness. According to the rotating machinery vibration signal characteristics, a block sparse Bayesian learning framework was put forward in which the expectation-maximization method was applied instead of the common reconstruction algorithms only based on the sparsity assumption. The experimentalresults show the proposed method has better reconstruction performance than traditional methods.