中文摘要：

为正确评估管道的使用寿命和安全状态,需要对管道缺陷进行准确的定量分析。提出一种基于泛化能力优化的在线学习径向基函数（RBF）神经网络,给出具体的算法步骤,采用自适应学习机制训练网络,并利用Ansoft Maxwell 3D建立仿真缺陷数据作为样本进行测试。结果表明：该网络训练效率高,泛化能力好,显著增强了样本适应能力。该方法有助于量化具有不同形态的缺陷,为管道的安全评估提供依据。

英文摘要：

To properly evaluate the service lifetime and safety of pipeline,the defects should be quantitatively analyzed. An online learning radial-basis function（ RBF） neural network based on optimized generalization ability was proposed. The detailed procedures of the iterative algorithm were introduced and self-adaptive mechanism to train the network was used. Then,simulation defect data from Ansoft Maxwell 3D was used as sample for testing. The test result proves that the neural network system has high efficiency,good generalization capability which can greatly enhance the ability to adapt to the sample. It helps to quantify the different forms of defects and provids reliable basis for the security evaluation of pipelines.