An integrated manufacturing platform, the law of sciencetechnology,industrial ecosystem - smart productionintelligent precision manufacturing with digital decision, AI modeling, big data governancekernel technologies
Summary This project aims to apply big data and AI and other analysis technologies to develop a bottom-up and complete smart manufacturing flexible decision-making framework, starting from the perspective of factory operation, and gradually go to the factory level and even the supply chain level to improve the resilience under the uncertain risk and decision-making quality. In the early stage of this project, the Advanced Process Control System (Advanced Process Control, APC) and the Advanced Planning & Scheduling System (Advanced Planning & Scheduling System, APS) were developed respectively for the quality and production capacity of the factory operation end, and then from different decision-making units it can be divided into two decision-making units: capacity planning and demand planning, including local optimization modules for customer demand fulfillment and capacity utilization, and integrated system integration for each level of enterprise planning. Finally, the data between different decision-making units such as APS, APC, capacity planning requirements, and inventory management planning were integrated to build the factory's own war room, so that decision-makers can more comprehensively grasp the factory's operation overview, and at the same time go further. The highlights of the project can be divided into three directions: digital decision-making, smart production, and total resource management, which are explained in order as follows:

1. Digital decision-making: The advanced process control system dynamically avoids, warns, manages, or compensates abnormal conditions in production in real time, avoiding the loss of production capacity and yield rate.
In the industry-university project "Engineering data analysis to build a solder resist opening (SRO) size improvement module", through engineering data and historical data analysis, a data analysis module was constructed for the printed circuit board solder resist opening size, including data health check and data exploration analysis, virtual metrology, rolling update mechanism and model confidence score evaluation, single batch full measurement data analysis. A proposed SRO key feature extraction and virtual metrology model can be applied to the prediction and improvement of the size of the solder resist opening, which can replace the retesting of the production line and save the cost of retesting.
This project also establishes a rolling forecast model under three material numbers, and updates and verifies the model with weekly production batches and model confidence scores. For one kind of SRO size, the ratio of the error less than 0.005 is 81.07%, and the ratio of the average error of the other size is less than 0.01 is 96.49%. Using this virtual measurement model to replace production line retesting can save RMB 1,020,000 a year. In addition, the variation of production line machines can be detected through model prediction errors, which can be used to find problem machines and conduct inspections or repairs in advance. The results of this project have been imported into the application of Zhen-Ding Technology.

2. Smart production: Advanced planning and scheduling system considers demand and production capacity uncertainty to develop materials, inventory, and production capacity modules.
The industry-university program "Decision Evaluation Model for Dedicated SMT Process Line and Co-line Production Allocation" assists relevant departments to make the best SMT process dedicated line and co-line production allocation decision, and provides optimized product demand and capacity combination configuration, which can be used in different Under the demand situation and considering the existing resources, conduct benefit analysis. By providing optimized configuration of product demand and capacity combination, the following are achieved: (1) Effective use of capacity resources, calculation of the line distribution of the old material number and the new material number combination, reducing the number of line body configurations to reduce possible cost expenditures, real-time To meet customer needs; (2) Assist decision makers to quickly and correctly perform numerical calculations of a large number of combinatorial optimization; (3) Systematize the company's decision-making process and methods, and reduce the time cost of re-analyzing similar decision-making problems.
This project first provides a unified format of data specifications, standardizes the data that was originally processed manually, and does not need to manually consolidate data to achieve descriptive analysis, reducing time costs and the possibility of human error. Then communicate with the company to understand the needs and limitations of the production system, discuss the goal-based construction, write an algorithm based on the capacity allocation constraints, and establish a production decision-making evaluation model for dedicated lines and common lines. After the system reads the input report data, it first checks the data, deletes the data that does not conform to the format, and generates a report to the user. Then the core calculation is performed, and the improved genetic algorithm is used to obtain an approximate optimal solution that meets the production capacity allocation limit work order and line body allocation results.
In the validity verification, we found that the optimal combination of maximizing the capacity utilization rate and adding local search to improve the algorithm efficiency can obtain the optimal work order and line body combination in a short period of time. The algorithm can obtain the best solution under actual data and conditions, and to verify the reproducibility and robustness of the algorithm, we design different data combinations and conditions, and finally the average capacity utilization rate can be increased to 94%. The results of this project have been imported into Zhen-Ding Technology for use.

3. Total resource management: Integrate all aspects of the enterprise planning level, build the factory's own war room, and enable decision makers to have a more comprehensive overview of factory operations
The industry-university project "Advanced Planning and Scheduling System for Combining Assembly and Test Packages at the Front and Back of SMT", develops a complete scheduling system from the front and back of SMT to assembly, testing, and packaging through heuristic algorithms, sorting out and simplifying the current multi-departmental processes including the weekly scheduling and detailed daily scheduling process to maximize resource utilization and personnel operation flexibility. Relevant visual charts and reports were constructed to provide multi-departmental decision-making basis and flexible scheduling suggestions. This project proposes a system connection method to more effectively consider and deal with problems that may occur at the practical level. Using real data for empirical verification, among the 10 scheduling data between 2022.06.20 ~ 2022.07.04, the proposed method is better than Benchmark's 3.2 in terms of the performance of the total number of delayed work orders, with an average delay of 0.3 , can reduce the total number of delayed work orders by 91%, and the average execution time of the proposed method is 3 minutes and 36 seconds, successfully demonstrating the effectiveness of the proposed method. The results of this project have been imported into Garmin for use.

The industry-university project "AI system solution for intelligent supplier selection in the aerospace industry" can comprehensively evaluate the various conditions of various suppliers in the past, such as: price, quality, or the degree of cooperation with special needs of customers, and then through machine learning models, systematically calculate and provide forecasted best decision-making suggestions to provide more effective and accurate choices for high-level decision makers.
This project integrates the buyer's point of view and extracts domain knowledge to quantify. It uses procurement targets and supplier categories as classification divisions and establishes systematic supplier selection rules with different target hierarchy structures to reduce the unstable quality of supplier selection. Objective data and quantified subjective information were used to establish a scoring system to efficiently rank business sources and reduce manual work time. After obtaining all supplier rating and ranking decision-making suggestions, the buyer makes the final decision and earns a procurement record. By reducing the difference in quality of supplier selection caused by different experiences and supervisors’ feelings, speeding up the evaluation of supplier selection and providing a systematic decision-making process, and providing priority for supplier selection suggestions, can provide indicators for learning from long-term excellent suppliers and improve supplier quality.
According to the multi-attribute decision-making evaluation results, it assists enterprises to verify the validity of the model, evaluate the rationality and suitability of business selection, and cooperate with the user unit to import. The framework of the research method adopts the UNISON decision framework combined with the hierarchical analysis method to evaluate and select suppliers, and further explores historical data and extracts information with data science analysis methods to improve the quality of decision-making. The research results of this project have been imported into Aerospace Industrial Development Corporation (Han-Xiang) for use.

The research results of this project have been published in several representative intelligent manufacturing journals, including International Journal of Logistics Research and Applications, Journal of Intelligent Manufacturing, Computers & Industrial Engineering, IEEE Transactions on Semiconductor Manufacturing, etc.
Keyword Digital decision-making Smart production Total resource management AI modeling Big data governanace
Research Project An integrated manufacturing platform, the law of sciencetechnology,industrial ecosystem - smart productionintelligent precision manufacturing with digital decision, AI modeling, big data governancekernel technologies
Research Team Led by PI:Prof. Chen-Fu Chien, National Tsing Hua University
More like this
Provide the latest information of AI research centers and applied industries