Miller Magazine Issue: 127 July 2020

54 COVER STORY MILLER / JULY 2020 If we remember the historical process of industrial revolutions and try to establish a correlation between them and the mainly applied or suitable maintenance method corresponding to it, we can summarize the de- velopment as given in Figure 2 5 . Facilitating effect of Predictive Maintenance on Industry 4.0 Returning to Predictive Maintenance (PdM), we can see that its application can be also defined based on the data structure or the data collection and analysis method it uses. Predictive Maintenance (PdM) may be “rule based” which is actually “condition monitoring” where sensors collect data continuously and send alerts according to defined rules and thresholds. This would not meet the expectation if a factory or plant is desired to be smart. On the other hand, predictive maintenance may be “machine learning based” which relies on large sets of historical data or test data combined with special ma- chine learning algorithms 6 . It is possible in this way to run different scenarios and predict what will go wrong when, and generate alerts to point it out. A Machine Learning Process consists of the succes- sive data collection, feature extraction and reduction, model creation, model validation, and deployment steps [6] where the main goal is to run the created model and improve the system in relaying back discrepant be- haviours. So, the loop is improving the behaviours in- stead of just repeating them. In Industry 4.0 environments, machine learning is even considered as the core of PdM, and is advised to be integrated also to business operations in order to link estimations and action recommendations for a better decision making 7 . In the beginning of PdM applications, rule based maintenance might be a good choice to obtain quick business results and to give a stepping stone into ma- chine learning as it is viable both for plants having al- ready IoT, and also for those who have not yet designed their system, but, for an optimum application to realize smart manufacturing, rule based method may not be sufficient if IoT is not implemented, and as historical data will not be available or not continuous in this case. Among the “Industry 4.0” criteria, two key criteria in context with PdM are “technical assistance” and “de- Figure 3: Technologies Enabling PdM Source: Deloitte analysis. Deloitte University Press. http://dupress.deloitte.com