KR-20260065689-A - DISASTER PREVENTION AND PREDICTION SYSTEM USING MULTIMODAL TECHNIQUES USING NPU-BASED GENERATIVE ON-DEVICE AI

Abstract

The present invention relates to a disaster prevention and prediction system using a multimodal technique utilizing NPU-based generative on-device AI, and is characterized by comprising: a data collection unit that collects process data from preset sensors provided on the mold fixed axis and moving axis in an injection molding process; a monitoring unit that monitors in real time the molding condition data stored in advance and the process data in the injection molding process; a correlation evaluation unit that evaluates the correlation by multivariately analyzing the process data and the actual defect rate; a defect prediction unit that predicts the probability of defect occurrence by applying the correlation analyzed by the correlation evaluation unit to a preset algorithm; a molding condition adjustment unit that adjusts the molding conditions in response to the probability of defect occurrence predicted by the defect prediction unit; and a disaster prediction unit that predicts the occurrence of a disaster using the data collected by the data collection unit.

Inventors

• 정양근
• 설동철

Assignees

• 신라정보기술(주)

Dates

Publication Date

20260511

Application Date

20241101

Claims (5)

1. A data collection unit that collects process data from preset sensors provided on the mold fixed axis and moving axis in an injection molding process; A monitoring unit that monitors in real time the molding condition data and process data stored in advance in the above injection molding process; A disaster prediction unit that predicts the possibility of a disaster using process data collected by the above-mentioned data collection unit, previously stored disaster occurrence records, and molding conditions; A correlation evaluation unit that evaluates the correlation between the above process data and actual defect rate through multivariate analysis; A defect prediction unit that predicts the possibility of defect occurrence by applying the correlation analyzed by the above-mentioned correlation evaluation unit to a preset algorithm; A molding condition adjustment unit that adjusts molding conditions in response to the possibility of defect occurrence predicted by the defect prediction unit above; and A disaster prevention and prediction system using a multimodal technique utilizing NPU-based generative on-device AI including
2. In paragraph 1, The above monitoring unit is, The process data collected by the above data collection unit is refined and noise is removed, and The difference between the above-mentioned stored molding condition data and process data is monitored in real time, If the difference between the above process data and the above stored molding condition data exceeds a preset ratio, at least one of the following operations is performed: displaying a notification, sending a notification to a preset user terminal, and sending a process stop signal to the injection molding machine. A disaster prevention and prediction system using a multimodal technique utilizing NPU-based generative on-device AI, characterized by real-time visualization of the defect rate predicted by the defect prediction unit and automatic transmission to the preset user terminal.
3. In paragraph 1, The above correlation evaluation unit, Visualize the correlation between the above real-time process data and the above actual defect rate, and A multimodal disaster prevention and prediction system utilizing NPU-based generative on-device AI characterized by quantifying the association between key variables and actual defect rates using principal component analysis and regression analysis.
4. In paragraph 1, The above defect prediction unit is, A disaster prevention and prediction system using a multimodal technique utilizing NPU-based generative on-device AI, characterized by calculating the probability of defect occurrence (Y) based on process data collected in real time using a learned regression model according to the following [Equation 1]. [Mathematical Formula 1] (Here, β₀ is the constant term (intercept), which is the default value affecting the probability of defect occurrence; β₁ to β₄ are the regression coefficients representing the influence of each variable on the defect rate; X₁ is the mold temperature; X₂ is the cylinder temperature; X₃ is the ambient temperature; X₄ is the ambient humidity; and ε represents the error term, which is other variation factors that the model cannot explain.)
5. In paragraph 4, The above-mentioned molding condition adjustment unit is, If the probability of the above defects exceeds a pre-stored threshold, the molding conditions are automatically adjusted, and Collect process data corresponding to adjusted molding conditions in real time to re-predict the probability of defects, but A disaster prevention and prediction system using a multimodal technique utilizing NPU-based generative on-device AI, characterized by repeating a feedback loop to adjust the molding conditions until the probability of a re-predicted defect occurs is calculated to be below the previously stored threshold when the re-predicted probability of defect occurrence exceeds the previously stored threshold again.

Description

Disaster Prevention and Prediction System Using Multimodal Techniques with NPU-Based Generative On-Device AI The present invention relates to a system for predicting molding defects and accidents in products manufactured in an injection molding process, and more specifically, to a multimodal method accident prevention and prediction system utilizing a Neural Processing Unit (NPU)-based generative on-device AI that digitizes the molding conditions of an injection molding machine and predicts defects and accidents in molded products through AI-based data relationship analysis. Recently, product quality control and process optimization have emerged as critical challenges in the manufacturing sector. Traditionally, quality management in manufacturing processes has been conducted by utilizing methods such as Statistical Process Control (SPC) and Design of Experiments (DoE) to detect variations within the process and thereby prevent quality issues in advance. However, these traditional quality control techniques require significant time and economic effort to process and analyze large amounts of data in complex process environments, and there are currently limitations in the accuracy of predictions. Meanwhile, existing quality management has been operated by manually analyzing and processing data to predict potential quality defects within the process. However, due to the continuous advancement of manufacturing technology and equipment, the volume and complexity of data required in the manufacturing industry are steadily increasing, leading to the emergence of a need for technologies capable of analyzing and predicting this data in real time. Consequently, research is currently underway on new technologies that automate quality management and enable more sophisticated quality prediction. Among these, Artificial Intelligence (AI) and Machine Learning (ML) technologies are attracting attention as important technologies for solving these problems. In particular, AI technology utilizes machine learning algorithms to learn patterns from process data and predict potential quality issues based on this, thereby contributing to lowering process defect rates and improving productivity. Furthermore, by analyzing large volumes of process data in real time and predicting quality defects in advance, it provides higher accuracy and efficiency than existing quality management methods. Furthermore, deep learning technology demonstrates excellent performance in analyzing unstructured data such as images, sound, and sensor data, enabling the analysis of various data generated during manufacturing processes. Additionally, quality prediction utilizing deep learning is accelerating the smartification of the manufacturing industry, as it enables real-time detection of product appearance defects or predicts the condition of equipment to facilitate proactive maintenance. Furthermore, with the advancement of Smart Factory and Internet of Things (IoT) technologies, an environment has been established to collect and analyze large amounts of data generated in manufacturing processes in real time. By collecting various process data from sensors in real time via IoT technology and applying it to AI-based quality prediction solutions, it has become possible to simultaneously improve process efficiency and quality. However, despite the introduction of these AI and machine learning-based quality prediction technologies, limitations in the accuracy and effectiveness of prediction models have become apparent due to process complexity and data diversity. Consequently, there is a growing need for technologies that improve the performance of prediction algorithms and perform more sophisticated quality predictions based on real-time fluctuating process data. Furthermore, since the molding process is fraught with numerous risk factors such as high temperature and high pressure, there is a need for accident prevention and prediction technology that ensures safety in preparation for accidents and maintains process continuity. This is achieved by integrating and analyzing sensing data collected from process equipment and environmental measurement sensors to minimize worker safety issues, providing alerts in the event of an accident through multimodal AI analysis, risk prediction, and warning notifications, and improving accident prediction accuracy through continuous learning of accident cases. This technology facilitates worker safety, efficient process operation, and data-driven process improvement. Therefore, the present invention requires research on a disaster prevention and prediction system using a multimodal technique based on NPU-based generative on-device AI, which optimizes quality management by analyzing manufacturing process data in real time and predicting quality defects in advance using artificial intelligence-based quality prediction technology, while simultaneously preventing human casualties caused by equipment accidents and maximizing m