Research On Fault Diagnosis Of Sensors For Bridge Structural Health Monitoring System

Having experienced a period with large scale of construction, Chinese bridge engineering gradually turns into a new period when the main task is maintenance of the existing bridges, for which structural health monitoring system can provide important data support. The lifetime of a bridge is sometimes more than a century while that of the sensors is usually no more than decades. This fact together with the natural damage such as corrosion and lightning will probably cause malfunction of the sensors. Wrong data from these faulted sensors is likely to mislead further structural safety evaluation and cause wrong alarm. Therefore, it's necessary to detect the faulted sensors before the structural safety evaluation. This thesis proposed a self-detecting method for the faulted sensors based on generalized likelihood ratio test. The method to classify different types of malfunction is proposed based on classification theory of relevance vector machine. These methods are used on a real set of data from a bridge to check their applicability, and are proved to be new useful methods that ensure the structural health monitoring system to work well. The main content and conclusions for this research are listed below:(1)A self-detecting method for the faulted sensors based on generalized likelihood ratio test is proposed, which can quickly detect the malfunction of sensors and map position of faulted sensors. Numerical case of simply supported beam proves that the proposed method can work properly no matter there is structural damage or not. If there is structural damage while all the sensors are working well, this method will reach a conclusion of malfunction for all sensors, which can be used as a signal of structural damage.(2) Classification method for sensor malfunction type is proposed based on classification theory of relevance vector machine and method of binary tree. Classification parameters are suggested at the same time according to the relevance between predicted and real signals from sensors. Numerical case of simply supported beam proves that 5 common types of malfunction in acceleration sensors can be classified properly with 4 relevance vector machines classifier.(3)Applicability of the malfunction detection and classification method for sensors mentioned above is checked with the numerical case of a continuous rigid frame bridge, and the results show the method can work properly in detecting different types of malfunction.(4)The method proposed is applied to a real set of data from a structural health monitoring system for acceleration sensors on a cable-stayed bridge. The result shows that all the sensors work properly, which agrees with the result from manual inspection. This again proves the applicability of the method proposed.