Company News

In December 2023, the Ministry of Industry and Information Technology issued the "Implementation Opinions on the Manufacturing Excellence Quality Project," emphasizing that quality is the lifeline of manufacturing and promoting the shift from quantity expansion to quality improvement is a practical need for high-quality development in the new era. In this process, the level of intelligence in manufacturing plants continues to increase. For example, in the field of industrial product inspection, engineers have turned AI into "new workers" to help factories solve challenging problems.

Although manufacturing has long used data collection devices like industrial cameras for product inspection, the collected data still required manual identification and judgment, resulting in low efficiency and difficulty in ensuring quality stability. Breakthroughs in machine learning algorithms have provided a possibility: training AI models based on accumulated data to autonomously judge production line inspection data, thereby overcoming the inefficiency and high error rates of manual judgment. Thus, AI+Machine Vision has opened a door for industrial intelligence. What is machine vision? According to the definition by the Automated Imaging Association (AIA): Machine Vision is a combination of hardware and software applied in industrial and non-industrial fields, providing operational guidance for equipment based on captured and processed images. Machine vision can be divided into two main parts: imaging and image processing analysis. Imaging relies on hardware components of the machine vision system, such as light sources, light source controllers, lenses, and cameras; image processing analysis is conducted through the visual control system based on imaging. The core functions of a machine vision system include: recognition, measurement, positioning, and inspection. The difficulty of these functions increases progressively, making production inspection one of the areas that best showcases the "capabilities" of machine vision. Currently, machine vision technology is widely used in consumer electronics, automotive manufacturing, semiconductors, photovoltaics, and other fields, with a continuously expanding market size.

AI Visual Inspection Originating from Semiconductor Manufacturing ScenariosIn 2018, Getech was incubated by TCL, bringing with it the digital capabilities TCL accumulated in solving specific problems, and based on this, developed a series of related products and solutions. Deep industry know-how became Getech's natural advantage. Unlike some AI companies that seek scenarios and customers with technology, Getech was born from specific scenarios, giving it a strong innate understanding of scenarios and customer needs. Getech's machine vision inspection solution—Tianshu AI Visual Inspection System—is a small branch of its industrial intelligence solutions, originating from TCL Huaxing's semiconductor panel production inspection. Each key process in panel production requires AOI (Automatic Optical Inspection) equipment to capture images and identify related defects. Initially, defect classification was done manually; only after classification could the next steps be determined. The entire process involves over a hundred steps, requiring extensive manual labor for each key process. With the rise of machine vision technology, Getech collaborated with TCL Huaxing to develop an AI visual inspection system, using artificial intelligence technology for image recognition and classification. After implementation, the system significantly improved inspection efficiency, replacing 80-90% of inspection personnel and processing nearly three million images per day. It also enhanced inspection accuracy by effectively avoiding issues like human fatigue and cognitive differences between individuals. Subsequently, as technology matured, Getech's machine vision solutions gradually expanded beyond the semiconductor industry and have now successfully empowered 22细分 industries, including photovoltaics, 3C electronics, home appliances, petrochemicals, and aerospace.

How Do AI Models Empower Manufacturing?The process of AI empowering machine vision inspection has evolved from small models to large models, and then back to "small models." Initially, Getech combined its deep industry data积累 and understanding of specific industrial scenarios and inspection indicators with machine learning algorithms to build small models for specific inspection scenarios. Small models are designed to closely match specific business scenarios and needs, emphasizing targeted applicability and high adaptability. Through focused and customized development, small models can meet industry-specific needs while improving operational efficiency and decision quality. However, small models risk overfitting, where the model becomes too focused on details and noise in the training data, performing poorly on new, unseen data. This necessitates data normalization and standardization to achieve optimal data distribution and比例. Model development and tuning require professionals familiar with both AI technology and industry-specific knowledge, which many customer companies lack. In 2022, breakthroughs in large model technology offered hope for solving these issues. Compared to small models, large models have stronger compatibility and stability. Large models are not limited by data complexity, do not require deep scenario understanding, only need sufficient data for training, and require minimal parameter tuning or architecture design. They can handle various types of data more easily without overfitting. In short, the emergence of large models significantly lowers the barrier to model implementation.


Getech, which has long pondered how to make machine vision systems better and easier to implement, actively responded to the large model trend, training and developing its own large models based on extensive existing data and small models. However, directly deploying large models on the client side posed new challenges, such as high resource consumption for推理 needs in real-time industrial scenarios and cost pressures难以 meeting the requirements for continuous rapid inspection and seamless integration with production processes. Getech's solution was to first train large models internally and then perform "model slimming" for specific scenarios. This approach leverages large models for learning and feature extraction, enabling more efficient training of small models with significantly reduced data requirements, making model implementation more convenient and feasible. Additionally, to improve product deliverability, the Tianshu AI Visual Inspection System developed visualization features, simplifying the model development process into intuitive drag-and-drop operations, allowing even non-AI expert IT personnel to easily develop and optimize AI models. Customers can choose from a series of preset sub-models and algorithms, combining them through simple operations to build models that meet specific needs without delving into complex algorithmic details. This way, many customer companies can benefit from the latest machine vision technology without additional huge human resource costs. How much impact does AI have on the industrial intelligence process? Beyond machine vision, industrial intelligence applications can effectively improve product quality through precise control, intelligent analysis, automated production, quality traceability, and other means, leading the digital transformation and upgrading of manufacturing. Enhancing quality with "intelligence," leading the new era of industrial intelligence. On the journey to achieving new industrialization, Getech will continue to leverage its technological advantages, committed to transforming more integrated software-hardware digital application成果 into practical productivity, injecting continuous new momentum into high-quality development, and exploring the infinite possibilities of industrial intelligence.