Published on 01-Nov-2023

Artificial Intelligence in NDT: Exploring the Untapped Potential

Artificial Intelligence in NDT: Exploring the Untapped Potential

Table Of Contents

  • Introduction
  • What is Artificial Intelligence (AI)?
  • There are several key components and approaches within AI
  • What is the current performance of NDT (without AI)?
  • How does Artificial Intelligence (AI) in NDT work?
  • Using Artificial Intelligence (AI) for NDT 4.0
  • How can Artificial Intelligence reduce human factors in NDT?
  • Applications of AI in NDT
  • Conclusion
  • Key Takeaways


Introduction

In the dynamic realm of Non-destructive Testing (NDT), precision and safety are crucial and, some of the key activities in NDT like inspecting a component for defects have proven difficult to automate in many cases.

But not anymore, since Artificial Intelligence (AI), the game-changer is revolutionizing NDT by seamlessly merging cutting-edge technology with tried-and-tested inspection methods.

AI is not just enhancing accuracy, but also paving the way for safer and more reliable processes in Non-destructive Testing.

What is Artificial Intelligence (AI)?

Artificial Intelligence (AI) is a multidisciplinary field of computer science that focuses on creating intelligent machines capable of performing tasks that typically require human intelligence.

It encompasses a wide range of NDT Methods and Techniques, algorithms, and approaches aimed at simulating and replicating human-like cognitive functions such as learning, problem-solving, reasoning, perception, and language understanding.

At its core, AI aims to develop systems that can process information, adapt to changing conditions, and make decisions based on data.

It's a field that has evolved over decades and is now experiencing rapid growth and innovation, driven by advances in computing power, data availability, and algorithm development.

There are several key components and approaches within AI


  • Machine Learning (ML):

Machine Learning is a subset of AI that focuses on developing algorithms that enable computers to learn from and make predictions or decisions based on data.

It consists of supervised learning, reinforcement learning, and unsupervised learning.

  • Neural Networks:

Inspired by the human brain, neural networks are a fundamental component of deep learning, a subfield of ML.

These interconnected layers of artificial neurons are used for tasks like image and speech recognition.

  • Natural Language Processing (NLP):

NLP is concerned with enabling computers to understand, interpret, and generate human language.

It's essential for applications like chatbots, language translation, and sentiment analysis.

  • Computer Vision:

Computer vision enables machines to interpret and understand visual information from the world, making it crucial for tasks such as image and video analysis, facial recognition, and autonomous vehicles.

  • Robotics:

AI-driven robots are designed to perform physical tasks in the real world, ranging from manufacturing and healthcare to autonomous drones and self-driving cars.

  • Expert Systems:

These AI systems emulate the decision-making abilities of a human expert in a specific domain, making them valuable for applications like medical diagnosis and financial analysis.

Artificial Intelligence (AI) is already making a significant impact in various industries, including AI for Healthcare, finance, transportation, and entertainment.

It's poised to drive further innovation, with the potential to revolutionize how we work, communicate, and live.

Its positive impacts can be channelized to various aspects of Non-destructive Testing as well. 

However, as AI continues to advance, it also raises important ethical and societal questions regarding privacy, bias, transparency, and the potential for job displacement.

Addressing these challenges is essential to ensure that AI benefits society as a whole.

What is the current performance of NDT (without AI)?

In the realm of Data Science and Artificial Intelligence, metrics play a crucial role in evaluating the effectiveness of a given method.

When dealing with binary decision problems, one of the most prominent metrics employed is the receiver-operator characteristic (ROC).

The ROC metric delineates four distinct states of prediction outcomes as given below:

  • True Positive:

A given sample has flaws and the NDT Methods detects it correctly and rejects it.

  • True Negative:

There exist no defects in the given sample and the NDT method doesn’t indicate any defect.

  • False Positive:

The given sample has no defects but the Non-destructive Testing method indicates the presence of a defect and rejects it.

  • False Negative:

The given item has defects but the NDT method could not detect it

Within the domain of Non-destructive Testing (NDT), the performance and accuracy of the finest NDT evaluators are estimated to achieve a maximum of 85.2% Probability of Detection (POD) with a 2.1% Probability of False Alarm (PFA).

However, it's noteworthy that the overall results tend to deteriorate in cases with a high false alarm rate.

How does Artificial Intelligence (AI) in NDT work?

Industry 4.0 represents the digital transformation of industrial production following the previous industrial revolutions.

Non-destructive Testing and Non-destructive Evaluation 4.0 (NDT & NDE) heavily relies on advancements in NDT/E equipment.

There exists a significant interplay between progress in Industry 4.0 and NDT 4.0, with Artificial Intelligence (AI) playing a pivotal role, particularly in image processing and analysis. 

Probability of Detection (POD) serves as a crucial metric in NDT, ensuring compliance with inspection quality standards.

Read on to understand how these technological advancements, driven by the industrial revolution's digitalization, enhance the Probability of Detection (POD) in NDT processes.

Probability of Detection (POD)


The Probability Of Detection (POD) is a crucial metric in assessing the accuracy of testing procedures, particularly in Non-destructive Testing (NDT).

This statistical tool plays a vital role in determining how effectively an inspection method can identify critical defects.

In the realm of NDT, human error is an unavoidable factor. Even the most skilled inspectors may make errors.

This is where the POD comes into play, bridging the gap between quantitative and qualitative parameters and evaluating the minimum flaw size that can be reliably detected.

POD = True Positive / (True Positive + False Positive)

POD is typically represented using a POD Curve, with the x-axis denoting the size of the defect, and the y-axis indicating the POD percentage.

The curve gradually approaches 100%, signifying a high probability of detection as the flaw size increases.

For instance, at a marked point on the curve, the POD might be 90% for a flaw size of 22mm.

The significance of POD extends across various industries, including the military, Aerospace Industry, construction, and pipelines. It serves multiple purposes:

  • Establishing Compliance:

POD helps ensure that inspection procedures meet the required standards for quality.

  • Input for Risk-Based Inspection (RBI):

In risk assessment, POD data informs decisions on inspection priorities.

  • Probabilistic Safety Assessment:

POD contributes valuable insights into the safety assessment of systems and structures.

  • Damage Tolerance Analysis:

It aids in evaluating the tolerance of materials and components to defects.

  • Scheduling Inspection Intervals:

POD influences the timing and frequency of inspections.

  • Construction and Part Acceptance Criteria:

In manufacturing and construction, it determines whether components meet acceptance criteria.

The '90/95' rule in NDT emphasizes achieving a 90% POD with 95% confidence.

This rule guides how consistently a testing method or system performs, with the confidence interval ensuring reliability.

Determining POD involves collecting data through various techniques within NDT, such as Ultrasonic Testing (UT), radiographic testing (RT), and visual testing (VT).

By conducting multiple experiments under controlled conditions, inspectors assess how many defects of a certain size are correctly identified.

This data is then used to create statistical models for the POD curve, employing distribution functions, parameter estimation, and log-logistic functions.

Thus POD is a fundamental tool in NDT, enhancing inspection accuracy, ensuring compliance, and contributing to risk assessment across a spectrum of industries.

Using Artificial Intelligence (AI) for NDT 4.0


Here’s how AI can contribute towards NDT 4.0 and revolutionize the process of detecting defects:

  • Image Enhancement and Denoising

Before an image is projected on a monitor or film, several environmental conditions may cause some portions of the image to be distorted or the entire image to be enveloped in noise.

Noise has many different origins, making rule-based enhancement impossible. 

Salt and pepper noise is an example of noise in which individual pixels are either completely black or completely white.

Gaussian noise, which is evenly dispersed over the image, is another example.

The signal-to-noise ratio (SNR), which is stated in numerous standards for Welding Inspection, is the accepted technique for assessing noise.

Using static filters (such as low/high pass) that are applied to the image as a convolution is the traditional method for denoising.

Since this method treats every aspect of the image equally, the fundamental drawback of these filters is the lack of spatial resolution. 

In this context, contemporary AI algorithms can analyze information locally, produce hundreds to thousands of filters, and combine them, each of which can focus on a particular aspect of a picture.

Manually performing this requires separating the image into several patches (for example, 1616 pixels), optimizing each view according to your knowledge and the patch's location in the image, and then putting everything back together.  

AI models like this find wide applications in photography to enhance images with low resolution, improve lighting, or along with filters.

Developing these AI algorithms for similar applications in digital radiography/radioscopy can bring tremendous improvements in NDT.

For instance, while using national standards, noise in weld images is a significant concern.

Consequently, denoising is essential to represent the flaws. This increases the human observer's detection rate.

  • Automatic Defect Recognition (ADR)

AI can directly detect flaws or defects in an image in addition to image augmentation and denoising.

Although this detection has not been standardized yet, it is nevertheless a tool that aids in faster and more accurate detection by human observers.

Additionally, the AI performs consistently and is unaffected by things like fatigue.

A supervised AI is employed for defect recognition.

Based on samples of photographs with their accurate labels that are provided, this AI is taught.

There are three main categories of AI in supervised vision. They are instance segmentation, object detection, and image classification.

Instance segmentation returns one label per image (defect or no defect).

This is suitable when there is only a single object per image and if the presence of this defect leads to the rejection of the entire sample.

A bounding box is drawn around a flaw using object detection.

This is beneficial when a part may have many different flaws but a human observer must determine whether the item is scrap or not.

Instance segmentation, predicts the precise polygon of a flaw. This approach is ideal for quantifying defects even though it requires the maximum data.

For instance, several standards for weld images require you to measure defects and make sure they are smaller than a predetermined threshold.

For detected faults, those geometric coherences can be estimated and samples can be automatically rejected.

How can Artificial Intelligence reduce human factors in NDT?

NDT 4.0 is a result of Industry 4.0's use of a broad range of cutting-edge technologies that improve present NDT Techniques.

One of the developing technologies of the fourth industrial revolution is augmented reality.

Other emerging technologies include cloud computing and storage, artificial intelligence, robotics, big data, blockchain, and robotics.

Any manufacturing process will eventually have defects. A sufficiently high probability of detection (POD) of known flaws necessitates manual inspection and sophisticated statistical techniques like Six Sigma.

When testing welds, such as in chemical plants where weld seams are manually inspected using X-rays, this is particularly time-consuming.

AI can reduce human mistakes and aid inspectors in detecting flaws.

Different individuals form different opinions after viewing the same facts.

Human variables like exhaustion may cause even one person's thoughts to diverge.

Non-destructive Testing (NDT) is a task that demands a lot of concentration. When defining the POD, elements like the inspector's focus and the inspection system's effectiveness should be considered.

Among the most significant human elements influencing the assessment process are the evaluator's personality, previous work experience, training, and exhaustion.

However, employing AI helps to avoid the mistakes due to these variables.

The complete system of tools, protocols, and inspectors must be reliable for an NDT process to be reliable.

Reliability is increased through technology that can improve perception and minimize the impact of human variables on POD.

Artificial intelligence can be useful in this situation. 

By automatically labeling potential flaws using Automatic Defect Recognition (ADR), AI enhances the inspector's perception.

This lessens the impact of human factors on POD, such as weariness, motivation, or attentiveness.

Additionally, AI offers an unbiased second opinion, which may assist less experienced inspectors and guarantee that they are categorising faults consistently.

The image quality is an issue that affects how inspectors perceive the POD.

The inspector can concentrate on the actual interpretation process rather than manually selecting filters if the contrast can be intelligently optimised and the noise in photos decreases.

The inspector does not need to manually select the appropriate filters if the entire image presents all information in an optimised manner, picking the suitable improvement for the right area.

AI can enhance image quality as discussed above. As a result, inspectors with less experience can pay closer attention to the image's details and identify fewer flaws.  

Applications of AI in NDT


AI has a wide variety of applications in NDT. Here are some of them:

  • Automated Defect Detection

One of the primary applications of AI in NDT is automated defect detection.

Traditionally, human inspectors would visually inspect materials for defects.

However, this process is time-consuming and can be prone to human error.

AI algorithms, on the other hand, can be trained to recognize and classify defects with high accuracy. 

For example, in Radiography Testing, AI can analyze X-ray or CT scan images to automatically identify and characterize defects like cracks, inclusions, or voids.

This not only speeds up the inspection process but also ensures a consistent and unbiased evaluation.

  • Image Recognition and Analysis

AI excels at image recognition, which is helpful in Non-destructive Testing. Various techniques like radiography, ultrasonics, and Visual Inspection generate a plethora of images.

Analyzing these images for anomalies or defects requires a high level of precision, especially in complex or noisy environments. 

AI algorithms can be trained on vast datasets to recognize patterns associated with defects.

For instance, in Ultrasonic Testing (UT), AI can analyze waveforms to detect and characterize flaws or material properties.

This not only improves the accuracy of inspections but also allows for a more detailed analysis of subtle defects.

  • Predictive Maintenance

Predictive maintenance is a critical aspect of asset management in many industries.

By analyzing data from NDT inspections over time, AI algorithms can predict when maintenance or repairs will be needed.

This enables organizations to plan maintenance activities more efficiently, reducing downtime and avoiding costly emergency repairs.

For example, in the aviation industry, AI can analyze data from NDT inspections of aircraft components to predict when they are likely to reach the end of their operational life, allowing for timely replacements.

  • Ultrasonic Testing (UT)

Ultrasonic Testing (UT) is a widely used NDT technique for detecting flaws in materials using high-frequency sound waves.

AI has significantly enhanced UT inspections by automating the analysis of ultrasonic waveforms.

AI algorithms can process the complex data generated during UT inspections to detect and characterize defects like cracks, voids, or material properties.

This not only increases the speed of inspections but also improves the accuracy of defect sizing and classification.

  • Radiographic Testing (RT)

Radiography Testing involves using X-rays or gamma rays to create images of the internal structure of a material or component.

AI has proven to be invaluable in RT by assisting in the identification and classification of defects in radiographic images. 

By training AI algorithms on a diverse range of radiographic images, they can learn to recognize indications of defects, such as cracks, inclusions, or voids.

This reduces the reliance on human interpretation and ensures a more consistent evaluation of radiographic images.

  • Eddy Current Testing (ECT)

Eddy Current Testing is a technique used to detect surface and near-surface defects in conductive materials.

It works on the principle of electromagnetic induction. AI can enhance ECT inspections by analyzing the complex data generated during the process.

By training AI algorithms on a variety of ECT signals corresponding to different defect types and conditions, they can learn to identify and characterize defects accurately.

This leads to more reliable and efficient ECT inspections.

  • Magnetic Particle Testing (MPT) and Dye Penetrant Testing (DPT)

Magnetic Particle Testing and DPT are surface inspection techniques used to detect surface-breaking defects.

AI can be employed to improve the analysis of images captured during MPT and DPT inspections.

By training AI algorithms to recognize indications of defects in images, inspectors can benefit from automated defect detection, reducing the time and effort required for manual interpretation.

  • Acoustic Emission Testing (AET)

Acoustic Emission Testing involves detecting and analyzing acoustic signals generated by the deformation or fracture of a material.

It is particularly useful for monitoring the structural integrity of materials under load.

AI can process and analyze these acoustic signals to identify potential defects or anomalies.

By learning from a dataset of acoustic emissions associated with different defect types and conditions, AI algorithms can provide early warnings of potential issues, enabling proactive maintenance.

  • Guided Wave Testing (GWT)

GWT is a technique used for inspecting pipelines and other structures over long distances.

It involves sending guided waves through the structure and analyzing the reflected signals to detect defects or anomalies.

AI can enhance GWT inspections by processing and interpreting the signals.

By training AI algorithms on a diverse range of GWT data, they can learn to detect and locate defects accurately, even in complex structures with multiple interfaces.

  • Heat Exchanger Tube Inspection

Heat exchangers are critical components in many industries, and the integrity of their tubes is of paramount importance.

Eddy current testing is commonly used for inspecting heat exchanger tubes.

AI can improve this process by automating the analysis of eddy current signals.

By training AI algorithms on a diverse dataset of signals corresponding to different defect types and tube conditions, they can learn to accurately detect and classify defects, ensuring the reliability of heat exchangers.

  • Corrosion Detection

Corrosion is a common issue in many industries, particularly in environments with aggressive chemicals or exposure to the elements.

AI can assist in the detection and monitoring of corrosion levels.

By analyzing data from various NDT Methods, including ultrasonics, radiography, and visual inspection, AI algorithms can identify indications of corrosion, allowing for timely intervention and maintenance.

  • Integration with Robotics

The integration of AI with robotics is a significant advancement in NDT.

AI-powered robotic systems can perform inspections in hazardous or hard-to-reach environments, reducing the risk to human inspectors.

These robots can navigate complex structures, capture inspection data, and even perform some repairs autonomously.

By leveraging AI, these robotic systems can enhance the efficiency and safety of NDT inspections.

  • Data Analysis and Reporting

With the increasing volume of data generated during NDT inspections, the need for efficient data analysis and reporting has become critical.

AI-powered systems can process and analyze large volumes of NDT data, extracting valuable insights and generating comprehensive reports.

This not only saves time but also ensures that inspectors have access to the most relevant information for decision-making.

Conclusion

Artificial Intelligence is propelling Non-destructive Testing into a new era of NDT 4.0.

The integration of Artificial Intelligence in Non-destructive Testing represents a significant leap forward in ensuring the safety, reliability, and efficiency of inspections across various industries.

AI has a wide range of applications ranging from ultrasonic testing to corrosion detection.

From automated defect detection, predictive maintenance, and advanced signal analysis to processing complex data and images, it is transforming the way NDT is conducted. 

Key Takeaways

  • Artificial Intelligence (AI) is a multidisciplinary field of computer science that focuses on creating intelligent machines capable of performing tasks that typically require human intelligence.
  • When dealing with binary decision problems, one of the most prominent metrics employed is the receiver-operator characteristic (ROC). 
  • The ROC metric delineates four distinct states of prediction outcomes.
  • The Probability Of Detection (POD) is a crucial metric in assessing the accuracy of testing procedures. POD = True Positive / (True Positive + False Positive)
  • Image enhancement, denoising, and automatic defect recognition are the different methods through which AI can contribute to NDT 4.0.
  • Artificial Intelligence can reduce the human factors in NDT.
  • The applications of AI in NDT include predictive maintenance, ultrasonic testing, heat exchanger tube inspection etc.

References:

1. Sentin

2. ASNT Pulse

3. IMechE



Tree PNG back

Companies

Tree PNG back

Articles

Webinars

Webinars

Tree PNG back

Jobs

Application Notes

News

onestop ndt application

NDT Trends #83 is out!!
Grab your FREE copy now.