Recommendation function for smart data analytics toolbox to support semantic merging of middle-of-life data streams Fatima-Zahra Abou Eddahab, Imre Horváth Department of Sustainable Design Engineering Delft University of Technology Delft, The Netherlands {f.aboueddahab, i.horvath}@tudelft.nl Abstract—Continuous enhancements of connected products make them able to generate and communicate a huge amounts of middle-of-life data streams to their producers. This affordance also creates a challenge for current data analytics tools unable to keep up with the heterogeneous nature and characteristics of these type of data. Accordingly, a function able to combine data from multiple data streams and analyze them as one source of information is definitely needed in a next-generation data analytics toolbox to support product enhancements by designers. As a result of a recent Ph.D. project, this paper presents the conceptualization and the implementation of a novel function of merging middle-of- life data streams. The implemented computational mechanism (i) acquires middle-of-life data streams, (ii) pre-processes them individually, (iii) merges information from the concerned streams, (iv) derives recommendation based on the merged information, and (v) send a recommendation as a message to the designer. The performance of the computational implementation was tested in an application case of data steaming and management to white goods designers for enhancing a connected washing machine. From a computational point of view, the testing proved that the set of proprietary algorithms designed for the realization of computational merging, together with the existing ones taken from the literature, were able to efficiently perform the subtasks. The advantages of merges were: (i) it provides more information than the one obtained by processing sensors’ data individually, (ii) it reflects the condition of the product with a higher fidelity, (iii) it communicates information about the product while it is in use by the customer, (iv) it reduces the sensors analyses time and effort, and (v) it provides recommendation as an action plan concerning the product at hand. The outcomes of this study will be used in a follow up research to develop a comprehensive smart data analytics toolbox to support product designers in product innovation. Keywords—data analytics; middle-of-life data; data merging; semantic interpretation; product designers; white goods I. INTRODUCTION During the last decades, data merging has become a rapidly evolving topic in various application fields [1]. It is defined as a synthesis of information provided by multiple data sources. Its objective is to establish a relatively consistent and complete description through a more complete and accurate set of information [2]. The need for data merging, especially for semantic fusion of middle-of-life data streams (MoLD-Ss) was reported by product designers in an investigation that we conducted to determine the needs, satisfaction and expectations of white goods designers concerning a next-generation smart data analytics toolbox (SDATB) [3]. Designers concretely required (i) semantic interpretation of data analytics outputs, as well as (ii) merging different data streams from different sources. To explore the need in a multi-disciplinary manner, we combined five implicative theories based on the principles of the axiomatic theory fusion (ATF) approach [4]. The specific theories considered for fusing were dedicated to (i) professional needs of product designers, (ii) advanced technological enablers, (iii) evolution of data analytics, (iv) combined creative problem solving and decision-making, and (v) functional and structural interoperability of enablers. The process of the ATF consisted of five main stages, (i) selecting theories based on their usefulness as source theories, (ii) axiomatic discretization of component theories which consists of semantic discretization of theories, and arrangement and composition of axioms and postulates structures, (iii) semantic and visual capturing of relationships that is done in three steps: creation of relationship network, matrix representation and rearrangement, and deriving propositions in a given context, (iv) actual fusion of the component theories that is done in three steps: syntactic processing and merging of component theories, deriving propositions based on units of resultant theory, and transferring propositions into a narrative description, finally (v) validation of the new theory in the context of the planned application. In this sense The results of the ATF confirmed the relevance of the hypothesis that semantic merging of MoLD-Ss and offering recommendation to designers based on combined data streams is a computational function that needs to be provided a SDATB. According to published works, implementation of a multi- sensor-based data merging approach (i) improves the probability of proper detection, (ii) extends the spatial and temporal coverage, (iii) reduces ambiguity, (iv) enhance systems reliability, and (v) increases system robustness [5]. However, computational is a complicated task, especially when semantic merging is targeted. This is a so-called high-level merging and is difficult to realize for two reasons [6]. First, inferring semantic knowledge needs the transformation of a low-level data/information into higher-level ones, which typically suffers from information deficit. Second, understanding semantics by a system requires the ability of (i) working towards a set purpose, 978-1-7281-6999-6/20/$31.00 ©2020 IEEE Authorized licensed use limited to: TU Delft Library. Downloaded on December 20,2023 at 15:19:37 UTC from IEEE Xplore. Restrictions apply.