AI for Visual Inspection & Non-destructive testing

What is non-destructive testing ?

Non-destructive testing ¹ (NDT) designates the technics used to evaluate a component or a system without causing any damage. This is a very interesting technique, as it does not alter the system being evaluated, allowing it to be used in a production line without disrupting its efficiency.

These innovative techniques are key to maintaining safety, reliability, and efficiency in industries such as aerospace, construction, manufacturing, and energy. By employing methods like radiographic testing, ultrasonic testing, and electromagnetic testing, NDT professionals can detect flaws, prevent failures, and ensure the longevity of critical assets.

Why use AI in non-destructive testing ?

Although non-destructive testing techniques and methods have been used in industry for a long time, the arrival of artificial intelligence (AI) has brought a new dimension to non-destructive testing. Indeed, it is possible to exploit AI models to enhance level of automation of non-destructive testing in order to benefit from their advantages in terms of speed, efficiency and cost.

AI, through computer vision and deep learning algorithms, can detect tiny defects and anomalies that might escape the human eye might miss. Used in parallel of human inspection to enhance level of quality or in order to limit human validation on important tasks, this reduces the risk of human errors and improves the quality of inspections, which is particularly important in industries where safety and quality are paramount.

What are the challenges to deploy AI for non-destructive testing ?

Deploying an AI system that ensures confidence in the quality of information provided is a significant challenge. Achieving a high level of trust in AI driven non-destructive testing involves rigorous testing, validation, and continuous improvement processes. This ensures that the AI systems are reliable and trustworthy in real-world applications. AI can enhance the accuracy and reliability of visual inspections and non-destructive testing, but achieving this requires overcoming several hurdles.

Keeping on trust on quality

While AI algorithms can be trained to recognize specific patterns and anomalies with high accuracy, ensuring that these algorithms are robust and generalizable across different production lines is a complex task. Moreover, ensuring the integrity and accuracy of the data fed into these systems is crucial and challenging, as it involves addressing issues like data bias, noise, and completeness.

Adaptable to production change and adapted to new environment

To be efficient AI systems might be able to adapt the environment to take into account new context and data discovered during inspection. Operation such as adding a new material, a new product is essential in order to generate scallability and operation efficiency of such system. These AI systems can be configured to validate data and provide detailed reports, enhancing confidence in the results.

Empower the operational team for improvement decision

AI does not replace human workers but assists them by taking over repetitive and tedious tasks. Visual inspections and non-destructive testing can be mentally exhausting, especially when they require sustained attention over long periods. AI can analyze images and data with consistent precision, thereby reducing the cognitive load on operators and minimizing the risk of errors due to fatigue. This allows operators to focus on more complex and higher value-added tasks, thereby improving their job satisfaction and efficiency.

Moreover, in dangerous or hard-to-reach environments, the use of AI for visual inspections and NDT can enhance worker safety. Drones equipped with cameras and sensors can inspect high-risk areas without endangering human personnel, thereby reducing workplace accidents.

Human operators can provide real-time feedback on the results of inspections performed by AI. For example, if an operator identifies a defect that the AI missed, this information can be used to adjust and improve the detection algorithms. This feedback loop allows the AI to become increasingly precise and reliable over time.

How does it work ?

The integration of artificial intelligence (AI) into visual inspection and non-destructive testing (NDT) processes follows a rigorous protocol to ensure its effectiveness and safety.

First, the AI is configured to operate within a predefined design domain that specifies the conditions and limits within which the AI is authorized to operate. This includes the types of defects to detect, the working environments, lighting conditions, and other relevant parameters.

Before becoming fully operational, the AI starts by operating in “shadow mode.” In this mode, the AI observes the responses and actions of human operators and proposes its own responses without applying them. This allows for a comparison of the AI’s results with those of the operators and an evaluation of its accuracy and reliability. The data collected during this phase is used to refine and improve the AI’s algorithms.

Once operational, the AI is constantly monitored to ensure it functions correctly. Monitoring systems allow for the rapid detection of any abnormal behavior or deviation from expected performance. Monitoring data is analyzed to identify potential biases and deviations well before their impact becomes significant. This proactive monitoring allows for quick corrective actions and maintains the reliability of the AI.

AI Observability understandable for operational teams associated to incremental improvement are key to AI deployment success in industry

The successful deployment of AI in industrial operations hinges on systems that provide observability of AI for operational teams, coupled with incremental improvements :

• Observability ensures that teams can monitor, understand, and trust AI-driven processes, facilitating better decision-making and quick identification of issues.
• Incremental improvements allow for continuous refinement and adaptation of AI systems, ensuring they evolve with operational needs and technological advancements.

Together, these elements create a robust framework for integrating AI into industrial operations, enhancing efficiency, reliability, and overall performance.

Finally, the integration of AI in visual inspection and non-destructive testing offers significant advantages in terms of complementarity with operational teams, reduction of cognitive load, improvement of safety, reduction of errors, continuous training, and predictive maintenance. By adopting these technologies ethically, companies can not only improve the quality of their products but also optimize the working conditions of their employees, resulting in increased competitiveness in the market and a safer, more satisfying work environment.