Prediction Of Remaining Strength Of Corroded Pipelines Based On Improved BP Algorithm And Genetic Algorithm

Compared to highway, railway and water transportation, pipeline is considered a moresecure and economical way for oil and gas transmission. But accidents of pipeline failureoften occur due to adverse of weather condition, corrosion of pipeline, destruction ofthird-party et al, causing the big lose in life safety, environment and economy. Corrosionfailure is one of the most common failure modes, which causes piping thinned and bearingpressure ability reduced. Long-distance pipeline residual strength prediction mainly focuseson whether corrosion defects can be allowed to exist under the operating pressure, and themaximum allowance of the existing defect sizes. Thus, the maximum pipeline carryingcapacity can be predicted. Therefore, pipeline maintenance and safety management can beproposed based on scientific evidence. There have been lots of research on this issue bothhome and abroad. A large number of experiments and researches have been carried on andsome methodologies have been developed and standardized in some specifications andnorms such as ASME B31G, Modified ASME B31G, DNV-99, Battelle and Shell-92.The similarities and differences among these standards and norms was discussed in thisthesis. By introducing the artificial neural network, the failure pressure of long-distance gaspipeline was predicted based on the nonlinear mapping function of artificial neural network.The effect of pipe diameter, pipe wall thickness, material yield strength, radial corrosionrate, longitudinal corrosion rate, defect length and pit depth on the pipeline failure wasanalyzed comprehensively. In order to illustrate the generality of neural network, thenetwork was trained using sample training set from six corroded pipelines with differentdiameters. The result showed that the neural network can be a more accurate andconvenient method to predict pipeline failure. Because the genetic algorithm is independenton the gradient information and it is a parallel random search optimization method, thegenetic algorithm which has the global search properties can be used to compensate for thelimitations of BP neural network to get the optimal BP neural network weights andthreshold. Then using the well-trained GA-BP network to predict failure pressure oflong-distance pipeline. The results showed that the GA-BP neural network is superior to theimproved BP neural network in meeting the engineering needs, and can be preferably usedto predict residual strength of the corroded pipeline.