Advances in weigh-in-motion with pattern recognition and prediction of fatigue life of highway bridges

The weigh-in-motion and response (WIM + R) system was developed to study both the loading and response of highway bridges. The two main objectives of the present study were: (1) to demonstrate the advantages of using the WIM + R system to evaluate the fatigue life of existing bridges which were not designed for fatigue with the current AASHTO specifications, and (2) to improve the WIM + R system by using pattern recognition in the analysis of the data, which would remove the need for laying tapeswitches on the pavement, therefore facilitating the field operations.;Four steel girder bridges were instrumented to obtain strain data at fatigue critical details, and at sections of maximum strain to compute the gross vehicle weight (GVW) of each truck. Two were simple spans, and two continuous spans.;A detailed description of the conventional WIM method reveals its limitations. Three new methods related to the WIM process were developed. The first replaces 2-D analysis with finite element analysis in the computation of the axle weights. The second computes influence tracks from strain data collected during the event of a test truck at normal traffic speed. The third optimizes the distribution of GVW to the individual axles by shifting the position of the truck.;A comparative study of three of the four alternatives suggested by AASHTO showed that the fatigue life computed with direct measurements of the stress ranges were greater than those computed with the simplified approaches. The effect of secondary cycles was negligible for the four bridges. The damage equivalent secondary cycle factor for fatigue was defined.;The applicability of three pattern recognition methods for WIM was investigated. The dynamic time warping, hidden Markov model, and feed forward neural network methods can classify trucks with the measured strain patterns. The improved WIM system could be used to survey the truck traffic while monitoring fatigue critical details. The neural network method is the most promising of the three methods. The velocity, axle spacings and weights, number of axles, lane travelled, and any other information could be output from the network by inputting only the strain data.