Predictive maintenance for industry 4.0, a holistic approach to performing predictive maintenance as a service.

Abstract

In modern collaborative industry, the machine equipment involved has rapidly increased. Many of the involved machines are complex and can only work in good maintenance conditions. Any failure of this equipment and related tools can easily lead to unintended disruption. Due to the collaborative nature of the manufacturing systems, one machine failure could result in undesired downtimes beyond single production lines and add costs to the value-added processes of the partner enterprises in the entire value chain. Industry 4.0 provides a concept of the interoperation of data, processes, and services within one enterprise as well as interoperation among different partner organizations. This increases dependencies and potential for failure related costs. There is, however, a lack of work that focusses on predictive maintenance services in the context of Industry 4.0 supported architecture and standards. This thesis looks at how data-driven predictive maintenance under existing Industry 4.0 concepts, architecture, and platforms can be supported. A flexible predictive maintenance case is used to design the predictive maintenance modules that fit within the industry standard Reference Architectural Model Industrie 4.0 (RAMI 4.0) model. Beyond looking at predictive maintenance for a specific manufacturing type, the research further looks at predictive maintenance as a service as well as forming a virtual factory specialized in supporting predictive maintenance. Adopting the design science research methodology, the dissertation designs Industry 4.0 Predictive Maintenance Architecture, algorithms of predictive maintenance modules for estimating RUL (Remaining Useful Life) and maintenance scheduling modules for supporting multiple machines/components. The design of architecture and algorithms are implemented within the leading FIWARE platform. The results are verified in terms of performance. The modular predictive model achieves higher accuracy and lower RMSE score at over 19% than comparator methods. The predictive maintenance service enabled by designed algorithms of predictive model and maintenance service scheduling can offer over 30% for optimal cost and 10% for downtime impact to the manufacturing network.