Landing AI, an industrial AI solutions provider uses AI-powered vision defect detection models to limit human error impact.
By Daniel Bibireata, Principal Engineer, Landing AI
For many years, manufacturing lines relied heavily on human intervention and inadequate technology to identify leaks and other product faults. Leak detection is an important step manufacturers must take to ensure product quality, but inadequate detection can cost companies time and money when defective products aren’t identified early in the process.
Recent developments in AI make it possible for companies to use automation to standardize and streamline quality checks, digitizing the leak detection process to ensure products ranging from home appliance components to medical devices are performing optimally before they leave the factory floor.
Challenges in bubble immersion leak detection
Historically, manufacturers have used an iterative mix of tools for detecting leaks, from helium testing to sensor-based techniques. Bubble immersion remains one of the oldest and most common techniques.
To perform leak detection using the bubble immersion technique, a product is immersed in a water-filled test chamber and pressurized. At this point, human operators visually inspect the product for gas bubbles that could indicate the product is improperly sealed, and therefore, defective.
While bubble immersion provides a simple method to detect leaks, the repetitive and cognitively demanding nature of the task compromises its accuracy over time. In fact, a study by Sandia National Laboratories found that human inspectors across various types of inspections may miss as many as 30% of defects.
Inspectors must receive specialized training to use equipment and manage leak detection processes, leading to additional human variables that produce inconsistent results. It’s common for less experienced operators to perform manual inspections and classify a good product as defective (or vice versa).
AI helps manufacturers prevent variability in quality assurance processes, increase productivity and reduce costs — it absorbs dull, taxing work and frees up teams to perform higher-value strategic initiatives.
This left side of the video shows traditional leak detection where the product is immersed in the water-filled test chamber and pressurized, and human operators inspect for bubbles. The right side depicts AI-powered and processed bubble immersion leak detection. The system feeds image data into an AI model trained with labeled data, then determines whether there is a defect and outlines the bubble in a red box.
The benefits of AI-enabled leak detection systems
AI-powered bubble immersion leak detection uses real-time video capture to monitor pressurized water tanks. The system feeds image data into an AI model trained with labeled data, then analyzes outputs to determine whether there is a defect.
As more data enters the model over time, your inspection process becomes more robust, quality improves and the elimination of human error ensures uniformity. Essentially, AI improves processes across plants because it is:
- Accurate and consistent: AI delivers more accurate and consistent inspection results and continues to improve over time as data grows more robust and models are deployed across additional plants. For example, AI significantly reduces the rate of “escapes” — defective products falsely identified as good — that are common in the manual inspection process.
- Efficient and scalable: With a significantly reduced need for human presence, AI-powered systems allow plants to perform inspection routines for longer hours without risking quality. The automated system enables manufacturers to scale production and speed the quality assurance process by replicating and deploying AI-operated inspection systems into new factories and plants.
- Cost-effective: An AI-powered system quickly alerts line operators about suboptimal operating conditions in leak detection tanks (e.g., murky water, a dirty tank or low water levels) so team members can respond quickly to systems and parts malfunctions. Solving for product defects quickly allows companies to optimize costs on the production line and post-production. In addition, reliable automated solutions enable manufacturers to redeploy their workers to more value-added tasks — such as fixing defects — and optimize costs in the long run.
Success in manufacturing hinges on your ability to consistently deliver high-quality products, while simultaneously optimizing costs. Manufacturers that adopt AI technology for fault detection will undoubtedly raise the bar on quality standards for the rest of the industry given the fact that AI-enabled leak detection technologies produce improvements in quality, safety and compliance. But just as importantly, these technologies create competitive advantage — and that’s a major win in today’s industry climate.
Daniel Bibireata is a Principal Engineer at Landing AI, an industrial AI company that provides enterprise-wide transformation programs and solutions with a focus on Computer Vision. Prior to Landing AI, Daniel was a Principal Engineer at Amazon where he worked on the Computer Vision technology behind the Amazon Go stores.