Generative AI forDefect Detection and Root Cause Analysis

What if manufacturing companies could pinpoint the exact cause of a defect the moment it occurs, preventing costly production delays and ensuring top-notch quality? Generative

artificial intelligence (genAI) makes this possible by revolutionizing how manufacturers approach quality control and defect monitoring. With genAI, manufacturing companies can conduct continuous defect monitoring and root cause analysis (RCA) for real-time quality control to ensure a more efficient and reliable production process.

You’ve likely encountered several case studies where AI-powered vision systems are used for defect detection and quality control. To clarify the differences between vision AI and genAI in such applications, it’s important to realize some key differences. Vision AI excels at processing visual data and identifying patterns indicative of defects, while genAI brings a higher level of intelligence to the process through its learning capabilities, contextual understanding and natural language processing abilities.

Differentiating vision AI vs genAI-based defect detection

Here’s how the two processes work:

Vision AI-based defect detection combines traditional image processing techniques and machine learning to examine images and videos and detect anomalies or flaws during production. Vision AI’s advantages include speed, elimination of human error and scalability. Limitations include poor adaptability to new defect types and performance that can be affected by lighting and environmental factors. It may also struggle to identify subtle defects.
GenAI-based defect detection analyzes data streams from smart cameras, Internet of Things (IoT) devices and other sources. It then intelligently filters and prioritizes information to reduce noise, spot deviations from learned patterns and improve defect detection efficiency. GenAI can learn from historical data, real-time feedback and changing conditions to identify new, emerging defect patterns. It also recognizes relationships and underlying factors for more insightful RCA.

Three key differences exist between genAI and vision AI. First, genAI models learn from new data and adapt their behavior. Second, genAI looks at product type, the manufacturing process and other factors to better understand the context of a defect. Finally, genAI processes and understands natural language queries, encouraging more intuitive interaction with the system.

GenAI’s impact on quality control and defect monitoring

With the help of genAI, manufacturers can identify patterns and correlations in vast amounts of data in a fraction of the time. For example, genAI can continuously monitor the production line in steel manufacturing to detect and analyze defects in hot-rolled steel strips. It then identifies the root causes of the problems and recommends preventive measures to avoid future issues. By integrating genAI into the manufacturing process, companies can save time and free up staff to work on other essential tasks.

McKinsey and Company expects genAI “to herald a new age of efficiency in operations. In manufacturing and supply chain alone, it could reduce expenses by up to half a trillion dollars.” Most of those savings would come from using genAI in RCA “to predict failures and reduce defects, and draft easy-to-follow dynamic work instructions.” GenAI is highly capable of identifying a wide variety of defects across the manufacturing sector, including surface flaws on metal sheets, inconsistencies in engines, issues in heavy machinery, and defects in automobile parts.

Rajat Jain, HCL Tech — Rajat Jain, HCLTech

Companies in every sector are realizing the benefits of genAI’s quality-control and defect-monitoring capabilities. For instance, when minor alignment issues in BMW’s battery pack assembly process produced costly defects and assembly line disruptions, the company created a digital twin of the assembly line and integrated genAI for RCA. In this setup, the digital twin collected data from every component, including robotic arms, conveyor belts and alignment sensors. Then, when a misalignment occurred, the AI analyzed every preceding step, identifying patterns that pointed to root causes like equipment calibration or environmental shifts. By running simulations in the digital twin, engineers could test adjustments without disrupting the production line. This approach reduced alignment-related issues by 30% and improved production line efficiency.

McKinsey and Company also highlighted how Citic Pacific Special Steel, a steel manufacturer in China, leverages genAI-based RCA to predict the inner workings of its blast furnaces during production. McKinsey noted that this integration enabled the company to optimize process parameters in real-time, increasing “throughput by 15% while reducing energy consumption by 11%.”

Training data for genAI

It’s vital for manufacturers considering the implementation of genAI-based RCA to overcome two common challenges that frequently arise: guaranteeing the availability of
diverse and high-quality training data and ensuring the system’s adaptability to evolving production processes.

The keys to addressing these challenges are to collaborate with AI experts when developing robust training datasets and implement continuous learning mechanisms into the genAI system to better adapt to production process changes. Once manufacturing companies address the potential challenges, they can integrate genAI into their existing quality assurance processes using the following five steps:

Step 1. Conduct a thorough assessment of existing quality assurance processes.
Step 2. Identify areas where genAI can add real value.
Step 3. Create the genAI system and train it with relevant, up-to-date datasets.
Step 4. Integrate genAI into existing production and quality control systems.
Step 5. Provide training for operators and quality assurance personnel.

Manufacturers can ensure the ongoing accuracy, reliability and ethical use of genAI by taking these additional steps:

Step 1. Continuously update and train the genAI system with new data.
Step 2. Implement robust validation and testing procedures.
Step 3. Adhere to ethical guidelines and standards for AI use. For instance, a company could establish a dedicated team to stay current on industry ethical guidelines and ensure that the manufacturer’s AI system continues to meet those guidelines.

When measuring the success of real-time quality control via genAI-based RCA, manufacturing companies can utilize a variety of key performance indicators (KPIs), including defect detection rate, reduction in production downtime, cost savings from reduced waste and rework, and improvement in product quality and customer satisfaction.

For example, a machine parts manufacturer might acquire information to make better business decisions by tracking the reduction in defective parts per million (PPM) as a key metric.

Rajat Jain is a senior director and sales leader for HCLTech, a global technology services company that provides digital, engineering, cloud and AI technologies for businesses.

See more content from Automation World on applying AI to improve quality control: