Development of an information discovery system to achieve process optimization

Abstract

Quality is now acting in the essential role of helping manufacturers to survive in the competitive industrial market, thus ensuring the final products fulfill the required standard while keeping manufacturing costs and time to a minimum. The traditional inspection-oriented quality control methods can only provide limited information as to why defective products are produced. Furthermore, as this information is so limited, how to correct the fault cannot be identified. In an actual manufacturing environment, projects and productions are changed from time to time. Due to the enormous complexity of many processes and the large number of influencing parameters, conventional approaches to modeling and optimization are no longer sufficient. Because of this, a systemized and well-designed approach was formulated, that performs the desired task of problem identification based on a specific process. Two techniques were employed. The first, a data warehousing technique called On-Line Analytical Processing (OLAP) converts complex data into useful analyzed information. The second is an Artificial Intelligence (AI) techniques including decision tree classification and Artificial Neural Networks (ANNs) that extrapolates probable outcomes based on available patterns of events. These were both deployed together to perform data drilling and analysis which are essential procedures for identifying factors of quality discrepancy in terms of dimensional accuracy and dynamic performance. The proposed integrated approach of Neural-OLAP was built as the Information Discovery System (IDS) which made use of the AI and experience of engineers in identifying foreseeable failure modes of a process or a series of distributed processes and planning for its elimination. This intelligent integrated system aims to process huge amount of production data from multi-processes in multi-manufacturing sites by using the object-oriented database design and multi-dimension data cube structure. Quality prediction can be achieved for performing manufacture feasibility evaluation, identifying potential quality problems and providing guidance for improvement through the decision tree-based ANNs, thereby achieving continual and real-time process enhancement. With this IDS, essential support for users who wish to identify the cause and source of problems can be provided so that immediate action can be taken for rectification. The system prototype was implemented in a factory, which manufactures magnetic heads for Hard Disk Drives (HDDs), in order to validate the workability of the proposed methodology and the feasibility of the adoption of IDS. An application of the mass production environment of the slider fabrication process, Reactive Ion Etching (RIE), was studied. Substantial improvements in terms of the product quality defect enhancement and process optimization through the corresponding machine settings prediction can be achieved. There is a significant decrease in total quality costs. The frequency of rework and scrap are also reduced resulting in better customer satisfaction, which proves the effectiveness of the proposed methodology.