Probabilistic Models to Evaluate Effectiveness of Steel Bridge Weld Fatigue Retrofitting by Peening

Abstract

The purpose of this study was to evaluate, with two probabilistic analytical models, the effectiveness of several alternative fatigue management strategies for steel bridge welds. The investigated strategies employed, in various combinations, magnetic particle inspection, gouging and rewelding, and postweld treatment by peening. The analytical models included a probabilistic strain-based fracture mechanics model and a Markov chain model. For comparing the results obtained with the two models, the fatigue life was divided into a small, fixed number of condition states based on crack depth, similar to those often used by bridge management systems to model deterioration due to other processes, such as corrosion and road surface wear. The probabilistic strain-based fracture mechanics model was verified first by comparison with design S–N curves and test data for untreated welds. Next, the verified model was used to determine the probability that untreated and treated welds would be in each condition state in a given year; the probabilities were then used to calibrate transition probabilities for a much simpler Markov chain fatigue model. Then both models were used to simulate a number of fatigue management strategies. From the results of these simulations, the performance of the different strategies was compared, and the accuracy of the simpler Markov chain fatigue model was evaluated. In general, peening was more effective if preceded by inspection of the weld. The Markov chain fatigue model did a reasonable job of predicting the general trends and relative effectiveness of the different investigated strategies.