概率矩阵分解在数控机床振动预测中的应用

Application of Probabilistic Matrix Factorization in Predicting of Vibration Trend of Numerical Control Machine

数控机床的振动会造成机械设备的损坏,对数控机床振动趋势的预测能够有效的提高振动故障诊断的准确率,由于存在数据的稀疏性和冷启动问题,导致现有的振动趋势预测方法无法对数控机床的振动趋势进行准确预测,利用矩阵分解思想,提出一种使用概率矩阵分解的数控机床振动趋势预测方法,在矩阵分解中使用利用贝叶斯准则和logistic函数计算概率,表示潜在因子之间的非线性关系,通过实验表明该方法能克服现有预测方法的不足,有效解决了稀疏性和冷启动问题,提高了振动预测的准确率。

Mechanical equipment will be d^d by vibration of numerical control machine. The accurate rate of vibration fault diagnosis can be improved by predicting of vibration trend of numerical control machine. Because of the data sparse problem and cold start, the existing methods of predicting of vibration trend have low accurate rate in predicting of vibration trend of numerical control machine. The thinking of matrix factorization is introduced to predicting of vibration trend. A method of predicting of vibration trend that use probabilistic matrix factorization is proposed. Bayesian criteria and logistic function are used in matrix factorization and capture non-linear relationships between latent factors. An example shows that the new method overcomes the defects of the existing methods, the data sparse problem and cold start are solved, and the predicting of vibration trend is improved.