The Concept of Functional-Productiveness for Modelling Reliability in Energy-Based Maintenance Domain

Abstract

Applying the Energy-Based Maintenance (EBM) policy within manufacturing companies, the theoretical probability of non-random deteriorating failures is relatively low. However, poor industrial maintenance practices and market intelligence have been reported. Nonetheless, although various maintenance practices, including CBM (Condition-Based Maintenance) and PdM (Predictive Maintenance) concepts, are applied within manufacturing sectors, results show that performance differs with decision-making and policy-making in all layers of abstraction. The reasons for such propositions are provoked by three main pillars of evidence, namely (1) state-of-the-projects, (2) state-of-the-literature, and (3) state-of-the-practice. The author of the thesis uses this evidence as an apparatus for justifying the lack of maintenance impact and achievement in the industrial “4th Wave”. A specific in-detail description of the protocol is given for each given pillar. The main setbacks and lack of achievement are seen in decision-making since most engineers and scientists rely on static data-driven approaches. Utilising the PdM approach seems to exhibit difficulties in switching from a static to a dynamic data-driven approach. The setbacks are seen through the poor decision-making of top management. Hence, the outdated maintenance CBM frameworks that manufacturers rely upon fall short of providing long-term effects, especially in upcoming sustainable manufacturing. Encompassing sustainable manufacturing as one of the key enabling technologies (KET) and sustainability indicator(s) as a condition monitoring (CM) tool(s) that rely on energy and environmental dynamics, maintenance decision-making (MDM) differs between traditional PdM and EBM policy. The benefit of conventional CM tools (e.g. vibrational and acoustic) is that such waste energy can be transformed into indicators at every decision-making layer (strategical, tactical, operational). Therefore, it can be used both for diagnostic and prognostic purposes to boost decision-making and optimise maintenance asset support, or it can be used as an overall planning optimizer in determining appropriate decision-making maintenance policy. However, ongoing “sustainable maintenance” and PdM maintenance research strategies face barriers in delineating functionality thresholds by still relying on static control indicators for triggering decision-making instead of dynamic indicators. The author of the thesis is set to propose the functional-productiveness (FP) concept as a quantitative estimate in delineating functional from non-functional labels. Secondly, using machine learning (supervised and unsupervised) algorithms for binary classification, the goal is to determine the healthy from the non-healthy state by relying upon functional-productiveness markers (FPMs). These markers are extracted from classification hypothesis space by variable importance; as such can be used for establishing the reliability of systems and contributing to maintenance decision-making. Using a practical hydraulic control system of a rubber mixing machine, it was possible to establish high classification accuracy between healthy and non-healthy states. The author used: Gaussian naïve Bayes (GNB), Artificial Neural Networks (ANN), Logistic Regression (LR), Classification and Regression Tree (CART), and k-Nearest Neighbour (kNN) for classification, where ANN resulted in the highest classification accuracy (95%) given unseen data.