基于数据特征抽取技术的多模态异常监测

Feature extraction technique based on multi-modal data anomaly detection

由于设备磨损,过程负荷等因素的变化,生产过程会出现多个模态,并且稳定模态之间会经历一个缓慢变化的过渡模态。然而现有多模态处理方法没有对过渡模态的动态特性进行有效的过程监测。鉴于此,本文采用微分几何特征抽取的方式提取过渡模态的动态特征,主要是通位置、斜率、曲率等几何特性刻画过渡模态的动态曲线特征,建立起过渡模态基于滚动球的异常检测模型,进而进行过渡模态的异常监测。仿真结果验证了本文所采用算法的有效性。

Due to the changes of the abrasion of equipment, the load of process and so on, production process will appear multiple modes, and the production process will experience a slowly changing transition mode from a stable operation mode to another one. However, there were not effective multi-mode processing methods for the dynamic characteristics of the transition mode. Given this, this article adopts the method of differential geometric feature extraction to extract the dynamic characteristics of transition mode, mainly use the geometric elements such as location, slope, curvature to describe the dynamic curve characteristics of the transition mode, establish the anomaly detection model based on the rolling balls for fault detection of transition mode. Simulation result shows the effectiveness of the algorithm.