Quality enhancement has been on the minds of manufacturers for as long as the industry has been around. Given the complex and varied factors that affect production quality, manufacturers often spend a lot of time and resources and yet still find themselves struggling with efficiency and yield problems. Further compounding the problem is the difficulty involved in terms of industry knowledge and technique transfer that hinders quality improvement.
With the intelligent quality enhancement system developed by Delta Research Center (DRC) in partnership with IABG, manufacturers are able to address the anomalies that cannot be detected by conventional quality management methods, identify possible root causes of the issues through data-oriented diagnoses, and allow production line personnel to be more effective and efficient in troubleshooting. The system owes its origin to the need of the exploration of production test data. Delta's products are delivered to customers after meticulous testing. However, despite the tests, product returns still occur occasionally. Even with the introduction of the Statistical Process Control (SPC) technique, anomalies have not been eliminated entirely due to inappropriate specification settings.
In view of this, DRC looked to the use of unsupervised learning as a solution. In the case of not knowing the quality of a product, various dimensions are taken into account to allow data to follow the learning rules and users to observe the distribution of test data, or, the grouping of the data (such as Group A in red and Group B in gray in the graph below) from different aspects. When data are grouped, it usually means that variation or offset has occurred. The phenomenon can then be correlated with the 4M1E (Man, Machine, Material, Method and Environment) elements in the production process which are the key factors of production quality and possible sources of issues. After anomaly detection, root cause analysis serves as the core function module of the intelligent quality enhancement system. Building upon the system, DRC has been working on the development of an application module of solder paste inspection (SPI) data containing spatial information, feature extraction methods of SPI data, and multivariable “outlier analysis” module which allows users to add annotations on data to train outlier prediction models. With continuous interaction, the models can be constantly updated and finetuned to improve the accuracy of predictions as well as the quality of products.
The system provides insights to whether or not product test data are stable, main test items that are causing instability and the factors involved (the unstable state of Group A is higher than that of Group B), so that improvement strategies can be communicated and formulated with suppliers.
The success of this project were attributable to a number of reasons: correct goal setting, division of work based on specialization between DRC and IABG, and Delta’s strategic focus on AI. Well aware that quality was a pain point of the manufacturing industry, the project team set their goals in line with the practical needs of production lines. DRC and IABG each gave full play to their respective strengths. IAGB took advantage of its deep understanding of the production landscape and rich experience in data acquisition to assess the needs of manufacturing facilities, pain points and data acquisition status on the front line. IABG also implemented the Mapping System, which it jointly developed with DRC, to assist factories with the test data structuring and storage work, and tailored the user experience (UX) based on user needs, as well as trained the users of the intelligent quality enhancement system. Meanwhile, DRC’s data scientists, software development team and project management team focused on the development of the system and analysis modules.
Sean Chang, Senior Director of Smart Sensor & Meter Business Division of IABG, said “The project development spanned across Taoyuan, Taipei, Beijing and Wuhan. The seamless collaboration of IABG and DRC was what made the project such a success, with efficient and accurate communication constantly going between the two parties, from problem identification, goal setting, all the way to project execution. Moreover, the way the research and development work approached data—through grouping and root cause algorithms—made it possible to perform analyses based on data, allowing the intelligent quality enhancement system to be further expanded to other areas, not just on test data, but also on machine parameters or production process information, etc.”
Dr. Vincent Lo, Leader of DRC AI CoE (Center of Excellence) said “Good data quality is pivotal to valuable and meaningful analysis. The project team made efforts to collect complete data and properly manage the data, so when they loaded the data into analytical applications, they could ensure a smoother flow and obtain better results.”
Sean Chang further explained “The algorithm development team was engaged in the project requirement interview to match customer needs with algorithm functions, which not only helped improve development efficiency, but also sparked more ideas about how to best utilize various analysis dimensions.”
IABG and DRC worked hand in hand throughout the project, applying new data analysis techniques to detect variances in production quality and using relevant process data to decipher possible causes of anomalies. Since 2017, the intelligent quality enhancement system has been introduced to Wujiang plant 3, Wujiang plant 2, Dongguan plant 3 and Dongguan plant 5, etc. one after another. In addition, its outstanding performance has also earned multiple recognitions, including the Award of “New Business Model and New Business Process” in Four Major Awards of 11th “Delta Innovation Award” and the First Prize of the “First Delta Intelligent Manufacturing Technology Forum”.
The intelligent quality enhancement system has been introduced to Wujiang plant 3, Wujiang plant 2, Dongguan plant 3 and Dongguan plant 5.