Bridge Damage Detection using Moving Force Identification

Visual inspections are the primary method today for ascertaining the condition of bridges. These methods incur problems such as human objectivity and inconsistency between inspectors. Some methods of bridge damage detection... [ view full abstract ]

Visual inspections are the primary method today for ascertaining the condition of bridges. These methods incur problems such as human objectivity and inconsistency between inspectors. Some methods of bridge damage detection make use of the relationship between changes in stiffness or mass and changes in first natural frequency of the bridge. Mode shapes also have been used but are more difficult to infer from measurements.
Moving Force Identification (MFI) is an algorithm that calculates the applied axle forces due to a passing vehicle. It has been found that a small amount of damage (loss of stiffness) in a bridge changes the calculated force history quite significantly. A damage indicator may then be inferred from this change. MFI applied to health monitoring was first proposed by OBrien et al. (2014) and an instrumented vehicle was proposed for drive-by inspections.
The governing equations of the MFI algorithm are ill-conditioned, necessitating the introduction of a regularisation technique. First order Tikhonov regularisation is used in the MFI algorithm being investigated in this paper. This requires an appropriate smoothing parameter to be found which is commonly acquired through the use of an L-Curve. The L-Curve is a log-log plot of the norm of the regularised solution versus the norm of the least squares solution for a range of smoothing parameters.
It is recommended in literature to use the area of maximum curvature of the L-Curve as the basis of finding the optimal smoothing parameter. For the case of using MFI to calculate the axle forces on a healthy bridge, this may very well be the optimal basis for choosing the smoothing parameter. This paper investigates the sensitivity of a damage indicator to the choice of smoothing parameter. The sensitivity corresponding to the point of maximum curvature is of particular interest as it indicates if the point of maximum curvature is suitable for damage detection purposes. Preliminary results show that the damage indicator magnitude is sensitive to the choice of smoothing parameter. Furthermore, the point at which the damage indicator is least sensitive to the smoothing parameter doesn’t correspond to the region of maximum curvature on the L-Curve.