Studies on leak detection in process pipelines using artificial neural networks/machine learning

Abstract

In process industries, fluids are commonly transported through pipelines. Leakages in pipelines are common and are mostly due to external factors [1]. These leakages can cause hazardous disasters and loss of lives if not monitored regularly. This can be witnessed by several gas leak accidents that took place in Bhopal, Visakhapatnam, etc. It is an environmental, health and economic issue to be addressed without fail. Safe transportation of fluids in a pipeline can be achieved by accurate leak detection and leak prevention in pipelines. Identifying these leaks with less human intervention, without false alarm rates, and accurate leak location identification is the need of the hour.

To detect the presence of leaks in pipelines, several conventional leak detection techniques are available. There are various methods for leak detection in pipelines, ranging from manual investigation to improved imaging through satellite. Based on various working principles and ideas, numerous methods for the detection of leaks are reported. From the literature, conventional techniques for the detection of leaks can be mainly classified into hardware-based, visual, and software-based methods. Among those, hardware-based and visual methods are frequently employed in the process industries. But there are drawbacks to these methods as well. Therefore, researchers have focused more on software-based methods because due to simple and reliable operation. Under software-based methods using machine learning algorithms that are a data-driven strategy for detection and localization of leaks is becoming popular because of its learning capabilities. Different machine learning algorithms like Artificial Neural Network (ANN), Neuro-Fuzzy Approach, and Support Vector Machines (SVM) are used for leak detection in process pipelines [2-4].

In this review, several recent conventional methods for the detection of leaks and machine learning-based leak detection methods are described. An attempt is made to identify machine learning-based leak detection techniques along with the common methodology followed for building a leak detection system, as shown in fig. 1. In one of the recent works leak diagnosis using combined ANN networks with cascade forward back propagation model to locate and measure leak using pressure and flow rate measurements at the pipe ends is carried out [2]. Therefore, the application of data science and machine learning algorithms for detecting leaks in pipelines in a process plant can help in handling and dealing with issues quite effectively and swiftly and thereby minimizing the wastage of time and increasing the efficiency of a process plant. Automation in leak detection based on machine learning models creates a less human intervention, fast, reliable, accurate, and economical solution. The future approach is to build a leak detection system using Machine learning algorithms to apply the same for real-time analysis in process industries, which has not been accomplished over the years.