基于数据挖掘的电子皮带秤皮带跑偏检测

发布时间：2018-07-31 08:48

【摘要】：为提高电子皮带秤连续累计称量精度,针对严重影响精度的电子皮带秤跑偏,采用对皮带秤现有原始传感器的数据挖掘实现跑偏量实时在线检测,以取代传统硬件检测设备。引入流形学习和深度学习,分别提出了基于局部切空间排列(local tangent space alignment,LTSA)+广义回归神经网络(generalized regression neural networks,GRNN)和基于连续深度置信网络(continuous deep belief networks,CDBN)的在线跑偏特征提取模型,再结合极限学习机(extreme learning machine,ELM)以跑偏特征为模型输入进行跑偏量预测。最后通过试验对该文提出的在线跑偏量预测模型的性能进行了验证:LTSA+GRNN+ELM平均跑偏预测精度为93.33%,平均每组预测时间38.29 ms;CDBN+ELM预测精度则高达98.61%,平均每组预测时间1.47 ms。二者预测精度和实时性皆表明能取代传统硬件检测装置,为皮带跑偏检测提供了一种方法,为进一步的皮带秤在线精度补偿和故障预测提供了必要依据。
[Abstract]:In order to improve the continuously accumulative weighing accuracy of the electronic belt weigher, aiming at the deviation of the electronic belt scale which seriously affects the accuracy, the real-time on-line measurement of the deviation of the belt scale is realized by the data mining of the original sensor of the belt scale, which replaces the traditional hardware testing equipment. In this paper, manifold learning and depth learning are introduced. Based on the generalized regression neural network (generalized regression neural) and the continuous depth confidence network (continuous deep belief), we present the online bias feature extraction models based on the locally tangent space permutation (local tangent space alignance) and the continuous depth confidence network (continuous deep belief), respectively. Combined with the extreme learning machine (extreme learning machine), the deviation is predicted by the model input of the runaway feature. Finally, the performance of the on-line deviation prediction model proposed in this paper is verified by experiments. The average deviation prediction accuracy of GRNN ELM is 93.33, and the average prediction time of each group is 38.29 msBNs. The prediction accuracy of CDBNs is 98.61 and the average prediction time of each group is 1.47ms. Both the prediction accuracy and the real-time performance show that they can replace the traditional hardware detecting device, which provides a method for belt deviation detection, and provides a necessary basis for further on-line precision compensation and fault prediction of belt scale.
【作者单位】：农业部南京农业机械化研究所;南京理工大学机械工程学院;南京三埃工控股份有限公司;
【基金】：科技型中小企业技术创新基金(13C26213202062)
【分类号】：TH715.195;TP311.13

【相似文献】