An Adaptive Denoising Method for Industrial Big Data with Multi-indicator Fusion
Xuemei Jiang, Xiao Wang, Ping Lou, Xiaomei Zhang, Junwei Yan, Jiwei Hu · 2019
With the development of information technology and network technology, as well as sensing technology, more and more data are collected from the processing of producing and machining in industrial sites. Industrial data are usually with noises and errors because the interference of ambient and the fault of sensors. Data preprocessing is necessary for analyzing further. The local polynomial regression is usually used to data denoising because of its characters of simplicity and flexibility, but the smoothing parameter of this algorithm need to be manually set by trial and error. In this paper, an adaptive parameter optimization method is used to calibrate its parameter, that is, the multi-indicator fusion method is used to fuse the root mean square error (RMSE), signal-to-noise ratio (SNR) and smoothness (r) of the denoised signal into a composite evaluation indicator. The smoothing parameter corresponding to the minimum value of the indicator is the optimal smoothing parameter of the denoising algorithm. A data set which is collected from monitoring the temperature field and thermal error of a heavy-duty CNC machine tool is used to validate the method.