Improving pipeline inspection with machine learning in Non-Destructive Testing

Author: Ing. Euclides Quiñonez, November 2, 2023.

Introduction

Non-destructive testing (NDT) has experienced a notable boom due to growing demand and multiple applications in the field of industrial asset integrity. The wide variety of uses of Non-Destructive Testing and the rigorous quality control regulations established have driven researchers to seek accurate, cost-effective and time-efficient approaches.

Currently, precise, economical and time-efficient methods are being developed, such as Artificial Intelligence (AI) algorithms and Machine Learning; which have proven to be effective numerical tools¹ that have enhanced the way in which inspections of industrial assets are carried out.

Pipelines are located in all parts of industrial facilities, occupying a significant part of space and resources, to transport fluids (water, chemicals, oil, gases, etc.) from one site to another within. In this context, it is important to guarantee the safety, efficiency and reliability of these assets, detecting defects, corrosion and other problems that may arise in time to avoid possible failures; therefore, the Non destructive essays (END) represent the most appropriate tools to locate discontinuities or damage in materials.

What is machine learning?

This tool is an artificial intelligence application that uses statistical techniques to allow computers to learn and make decisions without being explicitly programmed. It is based on the notion that computers can learn from data, detect patterns and make judgments with little help from humans.

The learning process is automated and improved based on the machines’ experiences throughout the process; which receive the data, which must be of good quality, and use different algorithms to create machine learning models to “train” the machines with this data. The choice of algorithm depends on the type of data available and the type of activity that needs to be automated.

Non-Destructive Testing in pipeline inspection

They are methods that allow the integrity of a pipe to be evaluated without damaging it. These methods include visual inspection, ultrasonic, radiographic, dye penetrants, magnetic particles, and thermography, among many others. These tests are necessary in detecting defects, corrosion and other problems without interrupting operations.

Machine learning in pipeline inspection

This tool is strengthening the way NDTs are performed. Machine learning algorithms can process large amounts of data collected during inspections and provide more accurate and faster results than traditional methods. This system is capable of processing images captured by modern sensors, they can detect a variety of data, such as images, readings made by the different NDT techniques to the pipes.

How does machine learning work?

The principle is to learn from data without a priori knowledge about physical laws and relationships between parameters. Machine learning models are generated using established algorithms to make subsequent predictions with unknown data. Their learning methods are the following:

• Supervised learning: means that there is a large set of labeled data available, where the label represents the answer that is assumed to be correct. The algorithm will be trained with that data, the correct training will be validated and, once the algorithm is trained, unknown data sets can be processed. Supervised learning covers both data with real-value labels, called regression problems, and cases with discrete labels, called classification problems. Typical tools are logistic regression, support vector machines or neural networks.

• Unsupervised learning: means the discovery of hidden regularities or anomalies in the data (for example, some unusual machine function or a network intrusion). Therefore, unsupervised learning does not require any annotated data. Typical tools are ensemble methods, clustering algorithms or principal component analysis.

• Reinforcement learning: The algorithm takes active steps and learns to optimize its actions to maximize some reward, i.e. the desired outcome, in a continuous development of itself. Such an algorithm can be used,to learn to play an interactive game, where the problem domain must,be explored to create the learning data set.

While all of these machine learning models offer potential benefits in the field of NDT, supervised learning models can be more easily integrated into this current framework. But these methods can be applied to an already available batch of data (batch learning) or to an incoming data stream (online learning).

Applications of machine learning in pipeline NDT

• Defect detection: Algorithms can identify defects or anomalies in pipelines by analyzing collected images and data. This allows for early and accurate detection of problems that could lead to costly failures.

• Predictive Maintenance: Models can predict when a pipeline failure is likely to occur based on historical data, operating conditions, and wear patterns. This allows you to predict when maintenance is needed on a pipeline, reducing the risks of failure, unplanned downtime and associated costs.

• Resource Optimization: By analyzing inspection data, this method can help companies allocate resources more efficiently, focusing on areas that require immediate attention.

• Reduction in human errors: Automating inspection with machine learning reduces reliance on human interpretation, reducing errors and increasing the reliability of results.

• History analysis: Data collected over time can be analyzed to identify trends and patterns that may be indicative of potential problems in the future.

Challenges in implementing machine learning

While this tool offers many advantages in pipeline inspection, its implementation is not without challenges. Some of the challenges include:

• Quality data collection: Data must be accurate and representative, this collection can be complicated in industrial environments. One of the first steps when facing Non-Destructive Testing in pipelines with machine learning is to have quality and appropriate data that serves as an example for the algorithms to learn from.
• Model training: Model building in this area requires a significant amount of labeled data and expertise in algorithm selection and tuning.

• Interpretation of results: It can be complex, and requires experience to make informed decisions based on model results.

Conclusions

Pipe inspection is a critical aspect in a variety of industries. Early detection of defects, optimization of resources and reduction of human errors are just some of the advantages that this technology offers.

Successful implementation of machine learning in pipeline inspection requires close collaboration between NDT experts, data engineers and professionals. In addition, high-quality data collection and training.

References

1. https://www.azorobotics.com/Article.aspx?ArticleID=470