SUPPORTING COMPANIES IN THEIR DIGITAL TRANSITION TO SMART MANUFACTURING SYSTEMS

Industry 4.0 (I4.0), which is considered to be the fourth industrial revolution, is now a well-established concept. However, it presents significant challenges for several manufacturing sectors. Luxury fashion is a key example, showing two blocking factors. First, the need to keep not only design, but also production data in-house, to protect industrial secrets, prevents the export of production or design data to external cloud services. Second, the need to keep using specially tailored legacy machines, which are not I4.0 compliant by design, leads to the lack of automated data pipelines capable of seamlessly connecting industrial machines to Business Intelligence (BI) applications. In fact, existing integration solutions often require invasive interventions on production lines, potentially disrupting operational continuity or exposing sensitive processes and intellectual property. This drawback acts as a barrier to adoption, especially in sectors where preserving traditional workflows and data confidentiality is critical. Despite these challenges, several factors facilitate this transition. For instance many companies operate with legacy machinery, which is already capable of generating and transferring industrial production data. Furthermore, most companies have on-premises servers, often pre-configured for data warehouse implementation. An additional incentive comes from government funding, which provides financial support to businesses embarking on the digital transformation journey. This thesis proposes a non-invasive architectural solution to support the digital transition of companies in high-quality production sectors towards smart manufacturing systems. It integrates an innovative data pipeline based on machine learning, capable of automatically recognizing sensors from unknown machinery. The proposed solution is intended to be non-invasive, as no change in the core production machines, and no IT intervention is required for the core business development of textile companies, in order to preserve intellectual property and avoid interfering with the normal workflow of textile operators. The main goal is to design and develop a novel data pipeline architecture to support the digital transition of textile companies seamlessly and adaptively. The research addresses the following research questions: RQ1: Which hardware / software technologies and tools are necessary to implement a software architecture for the transition to I4.0? RQ2: How can such an architecture be designed to be non-invasive, minimum human industrial configuration, seamless integration of new machines? RQ3: How to design a data pipeline which enables self configuration (i.e. how to design effective Machine Learning (ML) pipelines)? RQ4: How can such an architecture serve to build intelligent manufacturing systems with adherence to a set of I4.0 rules, standards, regulations, and guidelines that govern the textile sector? The proposed conceptual architecture is constructed and validated incrementally with real installations and data from Italian luxury fashion manufacturing companies. The architecture is divided into three layers: Physical, comprising legacy industrial machinery equipped with sensors, actuators, and PLCs; Edge Node, the core component that includes a message broker and an intelligent data filter for data processing and transformation; and Cloud On-Premises, which leverages the existing IT infrastructure in companies for production monitoring and management via Manufacturing Execution System (MES) or Decisional Support System (DSS). A key component is the Smart Interface (SI), an SBC (Single-Board Computer) device that extracts data from PLCs and publishes it to a message broker. The Intelligent Data Filter, based on ML techniques, is responsible for filtering, categorizing, and recognizing new machinery connected to the company network. From a pure methodological point of view, the ML-based semantic recognizer is based on a novel ML pipeline for solving a peculiar time series classification problem involving mutual exclusion. We also explore the algorithmic optimization of specific logistics systems, such as the sorting of hanging garments, which are triggered by our architecture, demonstrating how such an approach can further enhance overall company efficiency in their digital transition toward I4.0. Some modules of our architecture have already been provided to over 50 clients in the high-quality textile sector in Italy, managing the data flow of more than 580 IoT devices, corresponding to tailored industrial machinery. Thanks to this infrastructure, these companies have obtained competitive advantages, as well as the necessary certifications to access tax incentives related to the digital transition.