Highway Weigh-In-Motion System Based On BP Neural Network And Its DSP Implementation

The development of communication and transportation industry has undoubtedly played an active role in the construction of national economy. However, the overloading of trucking vehicles, which is known as stealthy killers of the highway, remains incessant in despite of repeated prohibition. Harms created by overloading are of manifolds, accelerating road damage, increasing road maintenance cost, leading to substantial run off of taxation and toll fee, bringing about frequent occurrence of traffic accidents, seriously polluting the environment, speeding up vehicle wear-off etc. So that it becomes more and more important to reinforce control over road transportation. Weigh-in-Motion, i.e. weighing the vehicle on its moving state, with the characteristic of higher efficiency and time-saving, avoids the disturbance to the traffic in the weighing process, as compared with static weighing. Vehicle Weigh-in-Motion is the technical precondition to the modernized and scientific management of reinforcing overloading control and forcing overloading control regulations. And the study of Weigh-in-Motion system is of economic significance and social value in assurance of appropriate use of the roads. For the highway Weigh-in-Motion systems, the weighing accuracy of the vehicle moving is the most import standard specification. It indicates the technical level of the WIM systems. While the weighing signal processing of the actual Weigh-in-Motion is simple digital filter and no further signal processing technique. So the WIM system's accuracy is hardly improved. Because there are so many factors that affect the WIM system's accuracy, and there are not exact function relations among those factors. While neural network has its own special advantage of processing non-linear and complex problems.In a word, the design is described in 3 parts in this dissertation. In the first part, from improving the WIM system's accuracy, this paper introduced neural network, and mostly introduced the basic idea, calculated process, executed steps and existed problems of BP network, and put forward the improved method based on those problems. A BP network with one hide layer is designed, and on the basis of analyzing the signal characteristics of field experiment data for various vehicle types and different vehicle speed, the max value, the average value of the signal, the speed of vehicle and car type are selected the inputs of the network.The second part tells the theory and realization of hardware design. The author analyses the reason why DSP are selected. The core of the design is single chip microprocessor and DSP. The author explains each component of the hardware. The author uses single chip microprocessor which is cheap and universally applied to display and communicates to the host. It effectively shares the burden of DSP in this way, and the programming is easy. So the author explains the structure characteristics and adjacent hardware circuit design of single chip microprocessor and DSP in big space. And there must be A/D converting circuit to convert the analog signal to digital signal recognized by DSP and single chip microprocessor. The author uses 24-bit A/D converter AD7714 to ensure the precision. The digital displayer is used to display the data. Communicating circuit realizes the communication to the host in favor of data storing and inquiring.The third part explained in discusses the software design. Because BP study algorithm is very complex and need great math operation, sometimes this work is done on PC. This shortcoming restricts BP study algorithm applying on Weigh-in-Motion. This paper implemented successfully BP study algorithm of Weigh-in-Motion on DSP. Some significant attempt to applying neural network on Weigh-in-Motion was tried.