汽车衡偏载误差补偿与称重融合方法研究

发布时间：2018-01-30 15:57

  本文关键词： 汽车衡 称重误差补偿 神经网络 约束条件 先验知识　出处：《湖南师范大学》2015年硕士论文　论文类型：学位论文

【摘要】：汽车衡作为衡器的重要分支,具有称重范围广、测量速度快、便于控制计算等优点,广泛应用于仓储贸易、交通运输、工矿企业等部门。现有汽车衡受到偏载误差与线性度误差的影响,称重结果准确度有待提高；同时,汽车衡称重数据获取不易,称重系统处于小样本状态。针对这些缺点,在国家自然科学基金项目“大型衡器偏载误差机理与多传感器称重融合方法研究”的支持下,本文开展汽车衡称重误差补偿方法研究：利用汽车衡先验知识,构建一种基于偏导数约束与Lagrange乘子法神经网络(PD-LMNN)优化的称重融合方法,提高小样本下神经网络的泛化能力,从而减少汽车衡的称重误差；建立以单片机MSP430F449为信息处理核心的汽车衡实验平台,通过实验平台测试,验证了这种方法的有效性。本文主要进行以下工作：首先,介绍了汽车衡基本情况及发展趋势、汽车衡的构成及工作原理,指出了现有汽车衡称重误差补偿的不足,阐述本文工作的重点；其次,构建了BP神经网络的汽车衡称重误差补偿模型,通过传统的利用数据样本训练算法(DINN),对该模型进行训练,指出了这种方法在小样本情况下的不足；通过研究汽车衡输入-输出函数偏导数,并以此作为先验知识,构建有约束条件的神经网络,利用Lagrange乘子法构造增广拉格朗日函数作为神经网络的目标函数,弥补了传统神经网络因训练样本不足导致的泛化能力差的问题,通过两种算法对比仿真实验,验证了PD-LMNN算法的优越性；再次,以单片机MSP430F449为信息处理核心、24bit模／数转换器CS5532为称重数据采集单元,搭建了最大量程为250kg、分度值为0.2kg的汽车衡称重实验平台,给出了硬件电路与软件设计流程图；最后,根据非自动秤通用检定规程,对采用PD-LMNN方法的汽车衡称重实验平台进行了偏载误差、重复性误差、示值误差和鉴别力进行现场测试,给出了测试结果。测试表明,在实验室条件下,该汽车衡称重实验平台的偏载误差、重复性误差、示值误差和鉴别力指标均达到了国家标准《JJG 555-1996非自动秤通用检定规程》Ⅲ级秤要求。
[Abstract]:As an important branch of weighing instrument, automobile scale has the advantages of wide weighing range, fast measuring speed, easy to control calculation and so on. It is widely used in warehousing, trade, transportation and so on. The existing automobile scale is affected by bias error and linearity error, and the accuracy of weighing result needs to be improved. At the same time, the vehicle weighing data is not easy to obtain, the weighing system is in a small sample state. Supported by the project of National Natural Science Foundation "Research on the Mechanism of bias error and Multi-sensor weighing Fusion method of large weighing instrument", this paper carries out the research on the compensation method of weighing error of automobile scale: using the prior knowledge of automobile weighing scale. A weighing fusion method based on partial derivative constraint and Lagrange multiplier neural network (PD-LMNN) optimization is proposed to improve the generalization ability of neural networks with small samples. In order to reduce the weighing error of the vehicle scale; A vehicle scale experiment platform with MSP430F449 as the core of information processing is established, and the validity of this method is verified by the test platform. The main work of this paper is as follows: first. This paper introduces the basic situation and development trend of the automobile scale, the composition and working principle of the vehicle scale, points out the deficiency of the existing vehicle weighing error compensation, and expounds the emphases of the work in this paper. Secondly, a BP neural network model of vehicle weighing error compensation is constructed, which is trained by the traditional data sample training algorithm. The shortcomings of this method in the case of small samples are pointed out. By studying the partial derivative of the input-output function of the vehicle scale and taking it as a priori knowledge, a constrained neural network is constructed. The Lagrange multiplier method is used to construct the augmented Lagrangian function as the objective function of the neural network, which makes up for the poor generalization ability of the traditional neural network caused by the lack of training samples. The superiority of PD-LMNN algorithm is verified by comparing the two algorithms with simulation experiments. Thirdly, a 24bit A / D converter (CS5532) is used as a weighing data acquisition unit with MSP430F449 as the core of information processing. The maximum measurement range is 250kg. The design flow chart of hardware circuit and software is given in this paper. Finally, according to the general verification regulation of non-automatic scale, the bias error, repeatability error, indication error and discriminant force of the vehicle weighing experiment platform based on PD-LMNN method are tested on the spot. The test results show that, under the laboratory conditions, the bias error and repeatability error of the vehicle weighing experimental platform are obtained. Both the indication error and the discriminant index meet the requirements of the national standard < JJG 555-1996 general verification regulation of non-automatic scale > class 鈪，

本文编号：1476623