Development of a recommendation system in PCBA repair

• Autor(a) principal: Cozinheiro, José Pedro Neves
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/34920
• topic: Big data; Bosch; Data analysis; Fault identification; PCBA repair; Recommendation system; Quality control

Industry 4.0 has promoted the digitalisation of industrial activities, representing a significant impact on process improvement and increased productivity. However, the systems that operate using these innovations present themselves generating quantities of information at an exacerbated velocity and in an unimaginable volume, big data. As a result, it becomes impractical and detrimental in many procedures for the analysis of the data set and decisions not to be taken by using the digital means available. The background process related to this dissertation takes place at Bosch Security Systems, where every day around twelve hundred printed circuit board assemblies (PCBA) are submitted for quality control tests from the production line. Each evaluation encompasses approximately four hundred measuring points at which the measured physical quantity is varying. Given the size of the referred around thirty-seven thousand test files with a total of approximately fourteen million eight hundred and eighty thousand measurements are generated each month. Therefore, assuming that for each type of fault identified in a file referring to a failed test there exists a repair method regarded as the most effective to fix the defect and guarantee the functionality levels of each board, the work proposal arises in the context of studying the possibility and practicability of developing a model of a recommendation system applied to the mentioned process. This has the objective of providing suggestions for repair methods, increasing both the rapidity of human decision making and the efficiency of the actions applied by the repairers. The proposed solution relies on being a recommendation system of hybrid category. At its base the essential tools used are the Python programming language, the Pandas library for data analysis and the algorithms based on gradient descent method for applying the technique of utility matrix completion, along with K-means clustering to consolidate the system in cases where data sparsity and cold start issues occur. Briefly, to reinforce the fact that the theme of this work should be seen with an innovative character, once currently this type of systems are present every day in people’s lives in ecommerce platforms, streaming, music, among many other areas, but not in industrial activities. Based on the dissertation and as a proposal for future work, if the operation of the developed model is seen as viable by the company’s responsibles in the short and long term, it can be implemented to the referred process as well as to others. In this way practical benefits will be brought that corroborate some of the previously mentioned goals of Industry 4.0.