Machine learning-based predictive maintenance: empirical insights of challenges and countermeasures
通过德尔菲法识别并排序了工业中采用机器学习进行预测性维护的关键挑战，同时基于从业者洞察提出了有效对策，发现培训等挑战被文献忽视，而某些对策可应对多个挑战。
Predictive Maintenance (PdM) has gained attention to reduce production-related costs and downtime, with Machine Learning (ML) emerging as a prominent technique. However, ML benefits are often achieved using laboratory or reference datasets. These may differ from real-world industrial data, raising doubts about ML applicability in real-world settings. This work addresses this issue, showing that ML adoption for PdM in industry is low. Furthermore, using a Delphi study, key challenges hindering ML adoption are identified and prioritised. Interestingly, some relevant challenges (e.g. the need for training employees) are overlooked by the literature. Furthermore, to boost PdM adoption, we identified and prioritized potential countermeasures based on practitioner insights. It emerged that some countermeasures can tackle multiple challenges (e.g. training programs). Our findings benefit both scholars and practitioners. Scholars may focus on relevant challenges to facilitate ML adoption for PdM. Practitioners are provided with a set of effective countermeasures to cope with relevant challenges.