Modeling gas transmission pipes with artificial neural networks in order to detect their defects

Dissertation for Master's Degree in Mechatronics Engineering

Persian abstract

The aim of this project is to introduce a new approach for troubleshooting gas transmission pipelines using mechanical waves, which is much cheaper and easier than other methods. who are currently working. These lines are usually located in harsh environmental conditions and far away and in long distances, and the use of systems that can report the defects and leaks of this pipe in real time is vital. A delay in assessing a damaged pipeline can potentially have a direct impact as a result of fire hazards from pipeline ruptures or sudden shutdowns of critical care facilities. The powerful health monitoring system of a pipeline includes the reduction of operating and maintenance costs. The presented method includes modeling a 2-inch pipe piece 50 meters long in Abaqus 6.121 software, creating 15 holes with a radius of one millimeter at three-meter intervals, capturing the vibrations (acceleration) of the pipe in a healthy state and in a defective state, transferring acceleration data to the frequency domain, using statistical techniques, creating a radial basis neural network and using this network for the presence and location of defects.

Key words: gas pipe, abacus, variance, radial basis function networks, pipe fault diagnosis

Chapter 1-Introduction

Pipeline structures sensitive to corrosion and exhaustion. The damage detection system for pipeline structures can reduce operating and maintenance costs.

This project is a new approach for diagnosing gas transmission pipelines using mechanical waves, which is much cheaper and easier than the methods of ultrasound and fiber optics that are currently employed. The gauge can be diagnosed with very good accuracy and resolution at a very low cost.

The research questions that we must answer in this thesis are:

1) What is the maximum length of the pipeline that can be detected?

2) Bandwidth What is the appropriate effective value?

3) Can the location of the defect be detected using the acceleration signal?

Reference to the chapters of the thesis:

Chapter one: Introduction - Chapter two: Existing methods in pipeline troubleshooting - Chapter three: Steps of the project - Chapter four: Construction of the finite element model From the piece of pipe and model validation - Chapter 5: Pipeline modeling - Chapter 6: Troubleshooting - Chapter 7: Results - Chapter 8: Conclusion

First, the goals and specifications of each stage are briefly stated, and then the details of each stage are presented in the chapters of the thesis, so that it can be used independently by researchers in the future.

1-1-Stage 1: Modal Analysis Three-meter pipe

The modal analysis test was performed on a three-meter piece of a two-inch pipe used in urban lines, as a result of which the natural frequencies and mode shapes of the pipe were extracted in the range below two kHz. This experiment was conducted in Sharif University of Technology and its details are available in the thesis chapters. Although this design is mostly aimed at intercity lines, the choice of two-inch pipe has a specific reason: new modes appear in vibrations with an increase in the length-to-diameter ratio, in fact, the higher the length-to-diameter ratio in the simulation, the more natural frequencies appear in the vibration behavior and the results are more realistic. Due to limited financial resources, it was not possible for us to test pipes longer than three meters. With the maximum length fixed, the only way to get the most accurate lab results was to use a smaller diameter.It should be noted that the purpose of the first two stages of this project is to validate (a) the finite element method, (b) the proposed software for implementing this method, and (c) the type of proposed elements, and the test results are not directly used in troubleshooting, so the use of a two-inch pipe, even if the purpose is to troubleshoot larger pipes, does not cause any problems.  At the end of the first stage, the results have been compared with the simulation results from the second stage, which confirms the accuracy of the model built in the second stage. Finite elements have been subjected to vibration analysis. This stage itself includes several steps, including pipe modeling with solid and loose (separate) elements, assigning material and mechanical characteristics (such as modulus of elasticity and Poisson's ratio) and analysis. The results of the modeling with the above-mentioned two different elements are also presented in the thesis.

1-3- The third stage: vibration analysis

Increasing the length of the pipe to 50 meters and setting the pipe to be similar to the real state without considering the effect of soil and welding in the simulation space, then determining the location of the force application so that its effect on the length can be attention should be visible and obtaining the maximum amplitudes of stimulation so that an impact function does not damage the tube. Creating 15 holes with a diameter of 2 mm in different longitudinal positions of the tube and applying force on the perforated models and the healthy model for 0.1 seconds and recording the determined signal. style="direction: rtl;">transferring the obtained information to the frequency field by windowing and fast Fourier transformation and calculating the average absolute value of the acceleration every 30 Hz and obtaining the difference of the signal in the state of the healthy and defective pipe and statistical analysis of the data of the previous stage and finally learning the data to the radial basis neural network.

2-1-Methods in pipeline troubleshooting

Gas pipeline troubleshooting plays an important role in gas transportation both in terms of safety and economy. The two main categories for diagnosing defects are:

2-1-1-Category 1

In this category, the entire pipe must be examined for troubleshooting, so either the detection device must be moved along the entire length of the pipe or this device must be installed along the entire pipe. These include the use of light or sound sensors to find leaks [1]. Other examples are the injection of flammable chemicals and the use of flame detectors along the pipeline [2], the simultaneous use of electromagnetic sources and detector sensors [3]. Another method is to use a special robot called "pig" (These pigs are actually robots that move on the pipe. These robots are generally used to detect defects on the pipe such as stress corrosion cracks, etc. In most cases, this expensive device is used to inspect gas pipelines that are not located underground.) In order to perform this method, the pipeline needs to be out of service [4]. Another example for this category is the installation of optical fiber along the entire length of the pipe [5]. All these methods are time-consuming and/or expensive.

2-1-2-the second category

In this category, to diagnose gas pipelines, some variables need to be measured at limited points of the pipeline. There are two methods in this category, the first method: fault detection based on monitoring changes in fluid properties (for example, flow rate and pressure) [6, 7]. The second method of troubleshooting is performed using ultrasound waves [8]. In the first method, it is done by using a set of solving non-linear equations that describe the dynamics of the flow (for example, through linearization [9] or discretization of non-linear equations [10]), which is used to predict the flow speed or pressure in the presence/absence of defects.