Pipeline damage identification based on an optimized back-propagation neural network improved by whale optimization algorithm

With the advantages of high economy and large transportation capacity, pipeline transportation is commonly used in industrial production. Pipeline damage induced by various factors will result in changes of physical properties, further leading to changes of dynamic parameters such as natural frequency and vibration mode. Recently, as a new type of tool, artificial intelligence is widely used for pipeline damage identification. In this study, to promote the accuracy of pipeline damage identification, a novel method that employs the artificial neural network (ANN) and swarm intelligence algorithm is proposed. In detail, based on the original whale optimization algorithm (WOA), an improved WOA (IWOA) is presented in which an adaptive coefficient strategy and a stochastic optimal substitution strategy are introduced. Then, the IWOA and back-propagation neural network (BPNN) are hybridized into IWOA-BPNN. Subsequently, a damage location detector and a damage degree detector are established based on the proposed IWOA-BPNN. By taking a pipeline fixed at both ends and its curvature and displacement modes, the proposed damage identification method is verified to confirm its effectiveness and accuracy in different damage states. Experimental results demonstrate that the comprehensive performance of IWOA-BPNN is better than other compared models. The relative error of the predicted results obtained by IWOA-BPNN is less than 2.2% when evaluating the damage location and degree for 12 randomly selected test samples, indicating the superiority of the proposed method. The proposed method has broad application prospects in modern industries.