Research On The Method For Truck Scale’s Eccentric Error Compensation And Its Weighing Fusion

As an important branch of weighing apparatus, the truck scale has many advantages such as wide weighing range, fast speed measurement and easily controlled by computer which is widely applied to storage, trade, transport, communication, industry and mine. The accuracy of truck scale’s weighing results needs to improve which affected by eccentric error and linearity error, the truck scale’s weighing data difficult to obtain and weighing system in small sample state. In order to solve these disadvantages, under the supports of National Natural Science Foundation Project "The study of large weighing apparatus mechanism of eccentric error and the method of multisensor weighing fusion", this thesis carry out research about the method of truck scale weighing error compensation:a method for truck scale’s weighing fusion based on prior knowledge with a partial derivative constraint and Lagrange multiplier neural network(PD-LMNN) is proposed, which improving neural network’s generalization ability in small sample state, thus reduce the truck scale’s weighing error; through a test of truck scale experiment platform which is built by single-chip MSP430F449 as the information processing core prove the effectiveness of this method.The thesis mainly for the following works:Firstly, The research status and development trends of truck scale is discussed, the composition of truck scale and their principle are introduced, points out the lack of truck scale weighing error compensation and the focus point of this paper; Secondly, the truck weighing error compensation model of BP neural network is built, training by traditional method which the neural network is only trained depend on using data sample(DINN), and point out the insufficient of this method when the sample is small; by studying the truck scale input-output function partial derivative, the constraint condition of a neural network is constructed with the truck scale’s prior knowledge, this proposed method can improve the neural network’s generalization ability when the training samples are lacking, the comparative simulation experiment results show the superiority of PD-LMNN. Thirdly, by using 24bit high-precision A/D converter CS5532 as signal acquisition unit and single-chip MSP430F449 as the information processing core building a scale range of 250kg, accuracy of 0.2kg truck scale experiment platform, the hardware circuit and flow chart of software design is given. Finally, according to the Non-Automatic Scale General Verification Regulations, the performance test results of truck scale experiment platform which using PD-LMNN method is given, including eccentric error, repeatability error, indication error and discrimination.The test have shown that under the condition of laboratory, the truck scale experiment platform’s eccentric error, repeatability error, indication error and discrimination are better than the III level scale of Chinese National Standards "JJG555-1996 Non-Automatic Scale General Verification Regulations".