基于残差连接的Conv1D-BiGRU动态称重模型  

Conv1d-BiGRU Dynamic Weighing Model Based on Residual Connection

作　　者：史柏迪 庄曙东[1,2] 蔡鸣 江志伟 SHI Bai-di;ZHUANG Shu-dong;CAI Ming;JIANG Zhi-wei(School of Mechanical Engineering,Hohai University,Changzhou 213022,China;Jiangsu Key Laboratory of Precision Instruments,Nanjing University of Aeronautics and Astronautics,Nanjing 213009,china;METTLER TOLEDO Measurement Technology Limited,Changzhou 213022,china)

机构地区：[1]河海大学机电工程学院,江苏常州213022 [2]南京航空航天大学,江苏省精密仪器重点实验室,江苏南京213009 [3]梅特勒·托利多测量技术有限公司,江苏常州213022

出　　处：《仪表技术与传感器》2021年第9期110-115,共6页Instrument Technique and Sensor

基　　金：江苏省高校实验室研究会立项资助研究课题(GS2019YB18);江苏省精密仪器与微细制造技术重点实验室开放基金(CZ520007812);中央高校基本科研项目(2018B44614)。

摘　　要：针对现有深度学习模型在高速重载下无法有效提取动态秤压力传感器高频时序特征从而精确输出货物质量的问题,提出基于残差连接的Conv1D-BiGRU模型。使用振动传感器检测秤体的三轴加速度信号,经皮尔逊特征相关性检测,发现动态测重时传感器压力与振动信号相关系数为0.94,呈现高度相关,可将信号混合后作为特征输入模型。将压力与加速度信号经一维残差卷积网络处理后输入BiGRU模型,模型训练完成后,在载重40.4784 kg、带速135 m/min时,最大误差err max为0.052、平均绝对误差mae为0.029、方差std为0.019,性能良好,且各项指标均较DNN、BiLSTM、BiRNN模型提升显著。Aiming at the problem that the existing deep learning model can not effectively extract the high-frequency time series characteristics of dynamic scale pressure sensor under high-speed and heavy load,so as to accurately output the weight of goods,a Conv1d-BIRGU model based on residual connection was proposed.The vibration sensor was used to detect the three-axis acceleration signal of the scale body.The correlation coefficient between the sensor pressure and the vibration signal was 0.94 showing a high correlation when the dynamic weight measurement was detected by Pearson characteristic correlation.The signal is mixed as the feature input model.The pressure and acceleration signals were processed by one-dimensional residual convolution network and input into BiGRU model.After the training of the model,the performance of err max(0.052),mae(0.029)and std(0.019)were good under 40.4784 kg load and 135 m/min belt speed,and each index was significantly improved compared with DNN,BiLSTM and BiRNN models.

关 键 词：一维卷积网络 双向门控循环单元 时间序列分析 动态称重 传感器融合 

分 类 号：TH932[机械工程]