The AI-ready NDT infrastructure Guide.

For years the hardware and software tools in non-destructive testing (NDT) have grown and added step by step. With the rise of Artificial Intelligence (AI) and the whole technology stack of NDT 4.0 new tools and approaches are needed to use the full potential of a company’s data. This article deals with AI-ready NDT infrastructure consisting of hardware and software, its benefits and requirements needed to use AI’s full potential.

Inhalt

What is an AI-ready NDT infrastructure?

An AI-ready NDT infrastructure is an ecosystem designed to support the integration of AI and other modern technologies into NDT processes. It seamlessly combines acquisition hardware, data storage, evaluation software, and advanced algorithms to create an intelligent, adaptable, and efficient inspection environment.

What are the components of an AI-ready infrastructure?

AI-ready NDT Digital Infrastructure
  • Acquisition Hardware: High-quality data acquisition hardware (sensors, scanners, detectors and cameras etc.) ensures that accurate, consistent, and detailed inspection data is captured. AI-ready acquisition systems store data digitally and pre-process and structure it for further handling. Data should be losslessly stored so that an AI can easily be trained and  understand the data.

  • Revision-safe data storage / PACS: Effective storage solutions are crucial. AI-driven NDT systems require extensive datasets for training, validation, and evaluation. Often cloud and decentralized storage systems are used to be scalable, secure, and fast, ensuring that data can be accessed and processed as needed. In non-destructive testing it is also important to store the data in a revision-safe manner. Often a Picture Archiving and Communication System (PACS) is used in this context. To support the latest AI and data frameworks we suggest to use an newer, encapsulated data-format on top of DICOM / DICONDE (supported by second generation evaluation tools like the sentin EXPLORER), because this standard was introduced in the 90s without AI in mind.

    Learn more here: What is a PACS in NDT?

  • Evaluation / Viewer Software: To become AI-ready the latest evaluation / viewer software like the sentin EXPLORER should be used. These not only have AI workflows integrated but also support working with data to build an AI model. The first generation of digital evaluation software may not be suitable for preparing and storing data with AI in mind. When „burning in“ image information / findings on top of an x-ray image for example, the information beneath is lost and cannot be used by an AI. The second generation of tools, on the other hand, store the data losslessly. These tools boost evaluation with AI models e.g. for text-recognition, defect recognition or fraud detection, and automatically generate reports as well as support the dataset cleaning and working with AI models (more on that later).

  • IMS / ERP Integration: Integration with Inspection Management Systems (IMS) and Enterprise Resource Planning (ERP) solutions ensures that the whole inspection workflow is considered when working with AI. It will get inputs from the other connected systems (e.g. label an order as done, when all inspection reports have been generated) and provide insights that can be leveraged across multiple departments and processes.

  • AI / Algorithms: At the heart of an AI-ready NDT infrastructure are the advanced AI models and algorithms. These use historical data to learn patterns, detect anomalies, and even predict failures before they happen (e.g. predictive maintenance). Continuous retraining of these models ensures they stay accurate and relevant. The applications of AI are so diverse they range from super-resolution algorithms reducing irradtion time in CT, over defect recognition in ultrasonic testing, to fraud detection and pattern recognition in the order data.

    Learn more here: How does Artificial Intelligence (AI) in NDT work?

What are the benefits of an AI-ready NDT infrastructure?

Benefits AI-ready NDT Infrastructure

Implementing an AI-ready NDT infrastructure can transform operations in numerous ways:

  • Unlocking AI Potential: Setting up an AI-ready infrastructure allows companies to work with data-driven insights and become future-proof.

  • Time and Cost Savings: Improving data acquisition, analysis, and reporting saves significant time. The reduction in manual processes also lowers operational costs, as AI-driven systems can work around the clock with minimal human intervention and higher precision.

  • Automated Workflows: AI-ready systems enable automated workflows, eliminating bottlenecks and streamlining tasks from acquisition to evaluation. This makes complex NDT processes easier to manage and reduces downtime.

  • Increased Transparency and Insights: AI systems provide deeper insights into testing processes and results. These systems can uncover hidden patterns and trends that humans might overlook, resulting in a more transparent and data-driven approach to NDT.

  • Maintainability and Scalability: AI-ready NDT infrastructures are designed to be easy to maintain and scale. As new testing requirements emerge, the system can evolve, and is ready for new datasets, hardware, and algorithms without major changes.

  • Revision-Proof Systems: The new systems store and track all changes and updates, making it easier to review historical data and ensure compliance with regulatory standards, while using top-notch technology to boost the inspection efficiency.

What tasks does an AI-enabled infrastructure assist with?

Tasks AI-ready NDT Infrastructure

For an AI-ready NDT infrastructure has to support several tasks. These include the normal inspection tasks and new tasks that are arising from the work with AI.

Regular inspection tasks:

These tasks are part of the normal inspection workflows. The inspection infrastructure should support them to not interrupt the regular processes and ensure the company can work as before.

  1. Data Acquisition: Capturing high-quality, consistent data from a variety of NDT methods (ultrasonic, radiographic, electromagnetic, etc.) using advanced sensors and acquisition devices.
  2. Data Storage: Providing reliable and scalable storage solutions capable of handling vast amounts of data with the ability to quickly retrieve and process information for real-time decision-making.
  3. Evaluation / Interpretation: Utilizing AI and ML algorithms to analyze data, identify defects, and detect anomalies with greater accuracy and speed compared to traditional methods.
  4. Reporting: The generation of reports and insights should be automated, clear, and customizable to meet the needs of different stakeholders, from field technicians to upper management.
UT Pipeline
Example: Data Acquisition of an Ultrasonic Scan (UT)
Lineprofile
Example: Evaluation of a Double Wire Image Quality Indicator
Export to report
Example: Writing an inspection report of a weld (xray) with porosities
AI related tasks:
 

These tasks arise from working with AI. They are things that data scientists do regularly, but they are new to the world of NDT. Not every inspector needs to understand them or do them regularly, but an organisation’s success in applying AI will depend on good tools and managers who understand these processes.

  1. Dataset Creation, Labeling, and Cleaning: AI models require clean, well-labeled datasets for effective training. The infrastructure should support automatic dataset creation, accurate labeling, and effective cleaning processes to eliminate noise / outliers and improve data quality.
  2. (Re-) training: The process of “teaching” an AI model what to do is called training. It requires special hardware (like GPUs) and software tools. You can use a prior state and existing training pipeline with new data to re-train and improve the model’s performance or teach it new things. This is necessary if a new inspection challenge arises (e.g. new parts that are inspected or data points / images that look significantly different from what you have had before).
  3. Validation: Before rolling out an AI model and use it in a real world environment you want to be sure it performs as expected e.g. having a certain accuracy. This process is called validation. Modern tools will allow you to work with the model e.g. on a validation data set and show you statistics and insights for you to make informed decisions whether you want to use a new model, its prior version or train a completly new one.
  4. Deployment and Monitoring: The infrastructure should be capable of continuously supporting you in retraining and validating AI models as new data becomes available. For using it in an productive environment it must also deploy updated models efficiently (e.g. to make it available for all inspectors) and monitor their performance to ensure accuracy and reliability. It should be possible to see how a model is currently doing and deploy an update to a server with just a few clicks.
Polygon Annotation
Example: Dataset creationg and labeling of a porosity
Model Performance
Example: Insights of an AI model loss convergence (the lower the better)
Update Model
Example: Deployment of an AI model to another remote system

What are the requirements of an AI-ready NDT infrastructure?

Requirements AI-ready NDT Infrastructure

For the infrastructure to support AI effectively, certain requirements must be met:

  • Interoperability: The system must be able to integrate with existing hardware and software platforms, supporting different NDT methods and ensuring seamless data exchange between systems. Using a mix of industry standard and modern interfaces is the key to success here. An example can be the DICONDE standard combined with RESTful web interfaces.

  • Workflow Awareness: The infrastructure should enhance existing workflows by making them AI- and data-centric, enabling smoother collaboration between human experts and AI algorithms and tools. Besides enhanced existing workflows there also will be new workflows that emerge from using AI like Critical Image/Item Detection (CID). It is therefore important to have a holistic overview of existing and new workflows.

  • Modern Interfaces and Data Formats: To create usable datasets and facilitate real-time analysis, the system should support modern data formats and interfaces, allowing for flexible interaction and integration with other digital systems. The DICONDE standard and PACS are commonly used, but to support AI-centric workflows new interfaces and data formats e.g. based on readable formats like JSON are the key to success.

  • Maintainability and Scalability: An AI-ready infrastructure must be both easy to maintain and scalable, ensuring that it can adapt to evolving industry requirements and increasing volumes of data without significant reconfiguration or downtime. NDT can learn from past years of cloud and web development which made huge advances with container and micro-service architectures.

How to get an AI-ready NDT infrastructure?

After describing benefits and requirements of AI-ready infrastructure solutions we now want to show you some applications of AI and the tools that are commonly used.

We support you from data acquisition to report generation
with AI and automation solutions.
Your AI-ready NDT Infrastructure

Software Products:

Wall

sentin EXPLORER - The imaging / viewer software with Artificial Intelligence. More...

sentin asset collector heroshot

Asset Collector - The Smartphone App for industrial device & type plate capture. More...

Services:

  • Analyzing the status quo and potentials of analog and digital workflows
  • AI dataset acquisition, annotation and cleaning
  • AI model training & validation
  • Development of digital automation & assistance tools as well as customized workflows
  • External interface implementation
  • User-Experience & User-Interface Optimization
  • Software or cloud integration, updates and monitoring

More…

AI & Automation Tools:

  • Wall Thickness Measurement / Corrosion Detector [X-Ray]

  • Weld Defect Detector (in-field, manual welds) & Standard Validation [X-Ray]

  • Weld Defect Detector & Validator (inline, manufacturing) [X-Ray]

  • CT & X-Ray Image Quality Enhancement / Irradition time reduction

  • OCR / Text Recognition [X-Ray & RGB]

  • Inline Surface Inspection [X-Ray, RGB]

  • Aerial Drone Inspection [RGB]

  • Inspection Data Anonymization e.g. personal data [RGB]

Hardware & Services with partners:

  • Inspection Hardware setup
  • Digitization of analog X-Ray archives
  • Setup of digital archives like DICONDE-compliant PACS
  • And more to make your inspection faster and more reliable…
BW Plus logo

In collaboration with

Found or missing something?
Preivew PACS NDT EN

What is a PACS in NDT?

It is no longer possible to imagine non-destructive testing (NDT) without PACS. What has been standard in medical technology for years (under the keyword –

Weiterlesen »