![인사이트 이미지]()
With People
Driving Manufacturing Innovation Through AX (AI Transformation):
Driving Manufacturing Innovation Through AX (AI Transformation):LG Innotek Enhances Quality and Customer Trust with AI Vision Inspection(Left) Byungwook Kim (Right) Heechul YangIntroAX (AI Transformation) is emerging as a driving force behind industrial reinvention. Across industries, it is redefining how businesses operate and setting new standards of competitiveness. In finance, AI enables personalized services and advanced risk management through the analysis of massive datasets. In healthcare, it accelerates drug discovery and improves the diagnostic accuracy. In mobility, it is reshaping transportation itself with autonomous driving and traffic optimization.Manufacturing is no exception to this paradigm shift. From product development and supply chain management to production and quality assurance, AX is reshaping the entire value chain. In quality inspection in particular, AI is actively applied to tackle key manufacturing challenges such as minimizing defect rates, increasing productivity, and reducing resource consumption.LG Innotek is at the forefront of this shift. To understand how AX drives innovation on the production floor—and how it has contributed to the company’s global position as the world’s No. 1 camera module company—we spoke with Heechul Yang and Byungwook Kim of the AI Inspection Technology Team.Q. Please introduce the AI Inspection Technology Team and describe its role within LG Innotek.The AI Inspection Technology Team specializes in inspection AI and automation to drive manufacturing innovation at LG Innotek. The team develops company-wide inspection solutions and applies them directly in mass production settings. We undertake projects such as improving decision accuracy through deep learning, automating and enhancing visual inspections, and utilizing AI to analyze inspection data. These initiatives allow us to continuously optimize processes, increasing both product quality and operational efficiency on the production floor.Q. What prompted LG Innotek to adopt AX in its manufacturing operations?LG Innotek adopted AX primarily to strengthen both production efficiency and quality competitiveness. Manufacturing is inherently complex, involving numerous interdependent processes. While each process operates independently, a robust framework for oversight, monitoring, and control is essential to achieve stable and optimized operations. This is particularly true for large-scale quality inspection processes, where optimizing the overall operation of the production line and advancing system capabilities are critical tasks for improving efficiency.To tackle these challenges, LG Innotek has been advancing its systems with proprietary technologies. Today, every stage of the manufacturing process is digitized, allowing AI to learn from the data and help identify potential pitfalls in each process that might otherwise go unnoticed by human operators. Previously, defective products could only be identified after completing the entire production process, which typically took more than two days depending on the production stage, leaving no option but to discard them. Today, however, with AI-enabled early detection and predictive capabilities, defects can be identified in less than an hour. As a result, we have been able to minimize discarded components and significantly reduce unnecessary resource waste.Q. Could you elaborate on one of the AX applications you mentioned, specifically AI vision inspection?Vision inspection refers to the process of examining product exteriors during manufacturing to identify defects. Traditionally, this inspection was carried out manually, but recent trends have shifted toward automation using cameras and sensors, enhanced further with AI for greater precision. Across manufacturing, AI is now widely applied to various processes to lower defect rates and elevate overall product quality. Essentially, AI vision inspection enables machines to replicate human visual perception and judgement. Where humans once relied solely on sight, cameras and sensors now capture detailed data, which AI algorithms analyze with accuracy comparable to human cognition. LG Innotek’s AI vision inspection solution takes this a step further by integrating these processes into a cohesive system—functioning much like the coordinated operation of the human body—to ensure precise, efficient, and reliable inspection across the production line.Q. Before implementing AI vision inspection, what were the main limitations or challenges of the traditional inspection methods?The primary challenge in traditional inspection was the discrepancies in results. Take camera modules, for example: though small, they have complex, three-dimensional structures that require careful visual scrutiny. Even then, inspection outcomes could vary from day to day depending on an inspector’s condition, sensory acuity, and experience. Even highly skilled personnel found it difficult to deliver perfectly consistent assessments.In fact, in our own experiments, when inspectors were asked to distinguish between good and defective products using the same samples, human accuracy remained at around 99%—even in relatively straightforward cases. The remaining 1% typically involved details that were difficult to catch with human subjectivity and perception. Of course, there are advantages to manual inspection, such as being able to adjust angles while holding the product directly. However, because no two inspectors share exactly the same standards or sensory judgment, maintaining consistent quality remained a challenge. To overcome this, we developed a method of training AI with images captured under diverse lighting conditions. This has enabled us to achieve accuracy levels beyond human capability, effectively covering that critical remaining 1% in quality determination.Q. What was the biggest challenge in developing AI vision inspection, and how did you overcome it?Within the unit processes of manufacturing, inspection requires the largest workforce. Even with automation, products initially flagged as defective still had to be manually re-checked under microscopes or magnifiers. Our goal was to make this secondary inspection fully possible with AI—while maintaining uncompromised accuracy in defect detection. Achieving that balance was the most demanding aspect of this project. As customer demand for higher volumes and improved quality continued to grow, the number of inspection criteria also expanded. Under conventional methods, it became increasingly difficult to meet these requirements. While automation improved defect detection, scaling inspection capacity to keep up with production volumes required a far more efficient solution.We approached the challenge by developing our vision inspection solution step by step. In the early phase of automation, multiple cameras and light sources were used to inspect products. To improve both efficiency and accuracy, we later reduced the system to a single camera, using mirrors to reflect images so that all 20 surfaces of four camera modules could be examined simultaneously. We then went a step further by designing a dome-shaped structure that allowed lights to be positioned at 0, 45, and 90 degrees. This not only optimized lighting angles but also reduced the need for additional equipment space and management points.Through a series of refinements, we were able to achieve a high-efficiency solution that enhanced inspection accuracy while reducing the number of cameras and overall equipment size. As a result, productivity increased by about 40 percent, and equipment space requirements were cut by roughly 75 percent.Q. What technological approaches and strategies have you employed to advance AI performance?While dome-light based automated inspection marked a breakthrough in the physical process, the greater challenge lay in elevating the “intelligence” required for precise defect detection. As inspection criteria increasingly include elements that are difficult to discern with the naked eye—such as foreign substances or contamination—as well as defects that inevitably involve human perception and subjectivity, the proactive application of AI technology has become indispensable.Enhancing AI performance begins with effective data training. To accelerate the learning process, we generated virtual defect data to train AI models at scale. By leveraging this approach, along with big-data-driven ‘Active learning,’ we achieved a defect detection accuracy of 99.95 percent for targeted defects. In practice, this has translated into zero defect leakage for customers—achieving an operational standard that is virtually flawless.Cross-organizational collaboration also played a pivotal role. By combining the AI core algorithms developed by the LG AI Research Institute with the domain knowledge of our inspection team, we established a strategy that enabled AI to make judgments akin to human reasoning. This integration of AI expertise and field experiences allowed LG Innotek to build a distinctive AI solution while continuously improving its performance.Our own team is also evolving to stay agile in the face of rapid changes in the manufacturing environment. Seasoned professionals with deep technical knowledge are working alongside newer members with high proficiency and familiarity with AI, creating synergies that make the team stronger and more adaptable.Q. From a customer perspective, what values or changes can be expected from the expansion of AX applications such as AI vision inspection?The most immediate and tangible value for customers is an apparent improvement in product quality. At LG Innotek, we have applied AI technologies capable of delivering exceptional inspection accuracy even with limited data across our entire quality inspection process. This approach has reduced the likelihood of defective products reaching customers to virtually zero, while also providing transparent process data that strengthens trust and confidence in our manufacturing.In the past, human involvement in inspection made it difficult to identify precisely why product quality varied. Limited data meant there was little visibility into which factors influenced outcomes. With the introduction of AX, however, we can now systematically trace and understand the root causes behind quality difference.This advancement has also enabled us to detect and address potential large-scale defects in advance, delivering direct benefits to customers in meeting delivery schedules and managing lead times more effectively. From the perspective of end-users, it further minimizes so-called “progressive defects”—issues that may go undetected during conventional testing but emerge during actual use—ensuring longer-lasting, more reliable products.Beyond product quality, the optimization of manufacturing processes reduces unnecessary resource consumption and maximizes the efficiency of equipment and energy usage, thereby lowering carbon emissions. This not only enhances productivity but also contributes meaningfully to ESG commitments, underscoring the broader value of AX in driving both operational excellence and sustainable growth.Q. What changes do you expect AX to bring to the manufacturing industry in the future, and what is LG Innotek’s vision for this transformation?In manufacturing, the core value that is continuously demanded is ultimately quality. Quality is not simply about maintaining a stable level—it must ensure both consistency and growth. We see AI as the definitive technology capable of meeting these demands, enabling manufacturers to respond proactively to diverse inspection variances and rapidly evolving market conditions.Currently, we are working toward an “AI-agent” structure in which the training, validation, and evaluation of data occur autonomously without direct human intervention. When humans are needed only for final verification, AX is expected to accelerate far beyond its current pace.Ultimately, LG Innotek aims to realize an Autonomous Factory, in which AI can independently make decisions, take action, self-correct, and predict defects before they occur. To achieve this, our AI Inspection Technology Team is focused on advancing solutions that inspect more precisely, across broader areas, and at higher speeds. Through these innovations, we aim to resolve issues that even our customers may not yet recognize, ensuring that both clients and end-users experience unparalleled quality. LG Innotek will continue to push the boundaries of innovation without pause.
