Operationalization of a Machine Learning and Fuzzy Inference-Based Defect Prediction Case Study in a Holonic Manufacturing System

摘要

Industry 4.0 capabilities have enabled manufacturers to collect and analyze smart manufacturing data across a broad array of diverse domains including but not limited to scheduling, production, maintenance, process, and quality. This development necessarily proceeds in a logical sequence by which first the organization develops the capability to capture and store this data and, at best concurrently but frequently lagging, develops and refines the competencies to analyze and effectively utilize it. This research presents an applied case study in surface mount technology (SMT) manufacture of printed circuit board (PCB) assemblies. Parametric data captured at the solder paste inspection (SPI) station is analyzed with machine learning models to identify patterns and relationships that can be harnessed to preempt electrical defects at downline inspection stations. This project is enabled by the recent conclusion of an Industrial Internet of Things (IIoT) capability enhancement at the manufacturing facility from which the data is drawn and is the logical next step in achieving value from the newly-available smart manufacturing data. The operationalization of this analysis is contextualized within the product-resource-order-staff architecture (PROSA) of a Holonic Manufacturing Systems (HMS). A Trigger Holon is nested between the Resource Holarchy and Product Holarchy that, at scheduling, distributes implementation instructions for the defect-prediction model. The Defect Prediction Holon is containerized within the Product Holarchy and provides instructions for corrective action when the model flags a record as exhibiting increased probability of a downline electrical defect.