Pipeline failure prediction with dynamic model constructed using a hybrid modeling approach

  • Yongxiang Wang

Student thesis: Master's Thesis


The objective of this report is to give an overview of a master's project that aims to predict vibration at any location along the span of a pipeline using a model built based on experimental data, and subsequently utilize this model to predict pipeline failure. The hybrid approach involves the construction of an initial model from experimental impact test data of a pipe section. The initial model is subsequently converted into state-space format and a linear-quadratic Gaussian (LQG) is formulated. The optimal LQG is tuned to minimize the error between measured acceleration at certain locations on the pipe span and their corresponding estimates. The uniqueness of this approach is that minimal feedback measurement is required from the physical system to produce reliable vibration estimates at any location on the pipe span without the need to place a sensor at that location. And this approach can be utilized to predict pipeline anomalies from changes in the estimated vibration signal. The advantage of this approach is that, it can be utilized for pipes or structures, which are inaccessible for measurements except for a few locations where sensors can be placed. Moreover, changes in the physical system require no adjustment to the initial model since those changes will affect the measurement and hence, LQG will adjust estimates to minimize the error, implying implicit adjustment of the initial model. The proposed method can be employed as a practical damage detection method of underground pipelines and buried structures. A detailed literature survey is carried out to fully understand the system identification method, hybrid and, modeling approaches and compares it to existing ones for validation. Experimental facility is designed and constructed to acquire experimental data from a pipe using MATLAB and dSPACE data acquisition system. Accelerometers are utilized as feedback sensors. The validity of the proposed approach has been examined experimentally. Two factorial experiment is designed to study the effect of three factors namely mass, position and direction, on the natural frequencies of the pipe. Analysis of variance (ANOVA) is carried out to determine the significance of the effect of these factors on pipe natural frequencies. Two methods (F-statistics and normal probability plot) are adopted to analyze the relationship between factors and natural frequencies. The relationships between effects and change of natural frequencies are established. Finally, the relation equations are fitted to provide a model for failure prediction.
Date of AwardDec 2016
Original languageAmerican English
SupervisorAmeen Hussein El-Sinawi (Supervisor)


  • Applied sciences
  • Damage detection
  • Dynamic model
  • System identification
  • Mechanical engineering
  • 0548:Mechanical engineering

Cite this