Fault Diagnosis Of Wireless Sensor Node Used For Railway Monitoring

Wireless sensor network node integrates wireless communication technology, embedded technology, signal processing technology and other advanced technologies so that it has small size, low price, low power consumption, multiple functionalities and many other advantages. While the sensor node is of limited resources and works in complex bad work environment for a long time, which results that in practical applications, the sensor node is the weakest link in wireless sensor networks. Due to great temperature difference, intense electromagnetic interference and strong vibration impact, sensor nodes which are applied to rail state monitoring are easily to go wrong. So this paper carries out related fault diagnosis research and experiments from the below three aspects:off-line fault diagnosis, online fault diagnosis and the optimization design of sensor nodes.In view of the sensor nodes off-line fault diagnosis this paper studies on a fault detection method based on vibration signal. In the method a test system is set up to provide stable vibration signal. The system consists of a control unit, an exciter, a simply supported beam structure and monitoring software. The control unit outputs control signals and drives the exciter via a power amplifier. The exciter excites the beam and makes it produce deformation. Then the vibration signals will be transmitted to sensor nodes via the sensors installed on the beam. Through the analysis of the time-frequency characteristics of collecting data, fault node can be effectively found.In view of sensor nodes on-line fault diagnosis, this paper adopt a soft fault diagnosis method based on the comparison of sensory data and a hard fault diagnosis method based on rough sets theory and BP neural network:as to soft fault diagnosis, gateway receives node’s sensory data and transmit it to the task management terminal. The management terminal calls the present and last acquisition data. Then fault node can be judged through comparing the change amount of base value and the correlation coefficient of power spectrum with the prescribed threshold; as to hard fault diagnosis, according to the node failure types (failure of sensor module, failure of communication module, failure of acquisition module, failure of power supply module) and their corresponding fault symptoms, the task management terminal set up a fault decision table. Then the rough sets theory is applied to simplify the decision table and then use the reduction to train a build-up BP neural network. Node failure can be located by inputting test sample to the trained network model.In addition, to conduct fault prevention and fault repairing, improved designs are achieved. The monitoring of multiple working voltage signals of the sensor node has been realized through the built-in 12 AD converter in STM32 so that we can analyze the problems existing in the power supply module. DS18B20 sensors are used to achieve the node working temperature in order to analyze the relationship between temperature and work performance. Some problems such that the node cannot join in the network successfully, the program code is tampered with by unknown reason, the program cannot jump from one to another successfully has been solved through the optimization of the program. Also a new way has been put forward to upgrade the bootloader remotely avoiding the troubles caused by traditional way of JTAG or ISP.