摘要
信号处理和机器学习是故障诊断过程中的关键技术,针对机械关键零部件的传统诊断技术,提出了一种采用字典学习和AdaBoost算法的信号诊断方法。该方法基于原始振动信号驱动训练数据,通过K-SVD和OMP算法更新字典并对其在字典空间稀疏表示,筛选重构所得增强信号时、频域特征,采用集成算法在AdaBoost神经网络分类器中实现振动信号的诊断。研究表明,采用字典学习和AdaBoost算法的信号诊断方法自适应强,能准确提取信号本质特征,诊断精度高,优于传统诊断技术。
Signal processing and machine learning are the key technologies in fault diagnosis. For the traditional technology, this paper proposes a signal diagnosis method with dictionary learning and AdaBoost algorithm. By using the K-SVD and OMP algorithm to update the dictionary, the training data is driven by the original vibration signal, represented in the dictionary space sparsely. Then, after screening the features of enhanced signal in time domain and frequency domain, the vibration signal is diagnosed in the AdaBoost neural network classifier. Results of research show that the signal diagnosis method with dictionary learning and AdaBoost algorithm is adaptive and accurate, and it is superior to the traditional diagnosis technology.
出处
《机械设计与制造》
北大核心
2017年第S1期113-116,共4页
Machinery Design & Manufacture
基金
国家自然科学基金(51275426)
关键词
故障诊断
字典学习
神经网络
集成算法
Fault Diagnosis
Dictionary Learning
Neural Network
Integrated Algorithm
