Abstract: With the increasingly fierce competition in the market,the small batch production mode has become an important production method.Aiming at the characteristics of small batch production and the problem that a large number of pre-experiments are required in existing research,a state monitoring method of machining process based on recurrence analysis and machine learning was proposed.The power signal of a small number of trials processing of different workpieces was collected through processing experiments;the preprocessed power data was input into the deep belief network for training,and the trained workpiece recognition model was obtained by optimizing genetic algorithm;the recurrence analysis and iterative self-organizing data analysis were performed to obtain the state monitoring model.Finally,the case study verified the effectiveness of the condition monitoring method,in which the workpiece identification accuracy was 99.3%,and the condition monitoring accuracy was 98%.

Key words: small batch, recurrence analysis, deep belief network, iterative self-organizing data analysis, state monitoring

摘要： 随着市场竞争的日益激烈,小批量生产模式成为了一种重要的生产方式。针对小批量生产的特性和现有研究需要进行大量预先实验的问题,提出一种基于递归分析和机器学习的小批量生产方式下机械加工过程状态监测方法。首先通过加工实验采集不同工件的少量试加工的功率信号;其次将预处理后的功率数据输入深度信念网络进行训练,并通过遗传算法优化,得到训练好的工件识别模型;接着进行递归分析以及迭代自组织数据分析,得到状态监测模型。最后案例研究验证了状态监测方法的有效性,其中工件识别精度为99.3%,状态监测准确率为98%。

关键词: 小批量, 递归分析, 深度信念网络, 迭代自组织数据分析, 状态监测