A Reference Model for Predictive Maintenance Model Development

Deterioration modeling plays a pivotal role in various industries, enabling predictive maintenance strategies and cost-effective resource allocation. Furthermore, with an escalating influx of data across multiple domains, opportunities for predictive analysis continue to expand. The development of wear models becomes a more and more complex process especially when acquiring and handling large amounts of customer data of different products and stakeholders since additional topics such as data privacy have to be included. Therefore, the development process causes high efforts in time, capacity and coordination between the different disciplines such as domain experts, data scientists and IT-administration. This paper presents a reference model for predictive maintenance model development. Based on adopted best practice approaches from the industrial production context, the reference model is structured around the four phases of the CRISP-DM model. Altogether it encompasses 48 defined steps. Covering the whole life cycle, beginning with component identification, use case description and culminating in model deployment and maintenance, each step is meticulously crafted to ensure quality and speed up the predictive maintenance model development. By adhering to this systematic approach, wear models for components can be developed with confidence, with lower effort and costs and mitigating uncertainties. The reference model is validated in the automotive industry since there already exist large amounts of fleet data of different products and customer. By providing a reliable and systematic approach to predictive maintenance model development, this reference model empowers stakeholders to optimize maintenance schedules, reduce downtime, and enhance overall operational efficiency.