Neurocle’s AI Auto Deep Learning Drives a New Era in Automotive Vision Inspection

[Youngsun Ha] Artificial intelligence (AI) is taking the lead in automotive quality inspection, a field once dominated by human and Rule-based systems. The growing sophistication of Deep Learning algorithms is making it possible to automate and refine inspections across complex manufacturing environments.

Every automobile passes through thousands of components and dozens of processes before completion, each demanding exceptional precision and systematic quality control. The production process typically consists of six major stages: pressing, body assembly, painting, trim assembly, engine machining, and final inspection.

In the pressing process, steel sheets are molded into frames and body panels using dies. The body assembly process joins structural components through welding, while painting coats the vehicle’s surface to enhance appearance and durability.

During trim assembly, key parts such as engines, wiring, and piping are installed. In the engine machining stage, critical parts like cylinder heads are cast, processed, and checked for defects. Finally, the final inspection ensures that the completed vehicle meets all required quality standards.

Neurocle has applied its proprietary “Auto Deep Learning technology” throughout the entire automotive production line, enabling precise detection of irregular and non-standard defects that traditional visual or Rule-based inspections often fail to catch.

In the body welding process, cracks or missing welds are detected in real time. During the sealing stage, cameras attached to robotic arms immediately inspect the sealant’s application width, position, and continuity. In the assembly stage, the AI automatically analyzes the alignment and fitting between components, identifying even minute deviations invisible to human eyes.

By implementing Deep Learning vision inspection for each production step, manufacturers can now perform more efficient, consistent, and intelligent quality management across the entire assembly line.

The sealing process, positioned between body assembly and painting, is essential to vehicle integrity. It prevents water, dust, and air intrusion, protects painted surfaces, and absorbs corrosion, noise, and vibration. If the sealant is unevenly applied or incomplete, it can lead to reduced waterproofing, structural weakness, or noise issues making accurate inspection and control indispensable.

At Automotive OEM Company A, robotic arms apply sealant along the vehicle’s joints as the body moves down a conveyor line. However, variations in the sealant’s thickness and placement often resulted in inconsistent sealing quality.

The company’s previous Rule-based inspection system struggled to handle these irregularities. Because sealant patterns are non-uniform, the system frequently produced false detections or missed defects, and its fixed position cameras couldn’t provide immediate feedback or correction. As a result, defects were only discovered during later inspection stages, causing rework delays, lower productivity, and increased costs.

To address these challenges, Neurocle implemented two innovative solutions.

First, a real-time integrated inspection system was introduced. By merging sealing and inspection into a single process, cameras mounted on robotic arms capture images during sealant application and instantly transmit them to an analysis system. This enables real-time quality monitoring and immediate feedback during production. The approach dramatically reduced both delayed inspection issues and the frequency of false or missed detections.

Next, Neurocle applied Deep Learning inspection model. By defining a Region of Interest (ROI), the system focuses exclusively on areas requiring inspection, eliminating unnecessary computations and improving overall accuracy. Using a Segmentation model, the AI precisely extracts sealant regions and automatically flags defects when variations in sealant width, discontinuity, or positioning exceed acceptable thresholds.

Neurocle’s Auto Deep Learning algorithm automatically optimizes model structures and hyperparameters, achieving high accuracy even with limited data.

Since adopting Neurocle’s solution, Company A has reduced both false detection and missed detection rates by over 70%. The accuracy of sealant inspection has improved dramatically, and sealant related quality issues have been virtually eliminated. By combining sealant application and inspection in one process, total production time decreased, overall productivity increased, and worker burden was reduced resulting in a smarter, more systematic approach to quality management.

This case is widely recognized as a breakthrough in applying Deep Learning to automotive sealing, showcasing how AI inspection can overcome traditional limitations and set new standards for precision and efficiency in smart manufacturing.

Beyond the sealing process, Neurocle continues to innovate in AI-based vision inspection even under conditions of limited or imbalanced defect data. In tire X-ray inspection, for instance, defect occurrences are rare, and defect types vary widely from air bubbles and internal layer separations to foreign material contamination. This makes it difficult to secure consistent, representative training data.

To overcome this, Neurocle utilizes a “Synthetic Defect Generator(GAN) model” to generate realistic synthetic defect images. These generated dataset enable stable model training even when real defect data are scarce, allowing AI models to detect irregular and rare defects with high precision. This significantly improves both defect detection rates and production efficiency across the manufacturing line.

In addition, Neurocle provides “Unsupervised models” that can train exclusively on normal data, as well as “AI-based Labeling Tools” that maintain accuracy and consistency while minimizing manual effort. These technologies make it possible to achieve effective quality control even in data scarce or imbalanced environments.

Recently, Neurocle launched “Neuro-T version 4.5”, which increases model inference speed by up to 28%, enabling faster real-time defect detection on production lines and preventing productivity losses caused by inspection delays.

Neurocle is a Seoul-based Korean company specializing in AI Deep Learning vision software, founded in 2019. With the mission of “Making Deep Learning Vision Technology More Accessible”, Neurocle provides vision inspection solutions that apply Deep Learning to computer vision. The company has now expanded to over 30 countries worldwide, strengthening its presence in the global market.