Context: The rapid prevalence and potential impact of big data analytics platforms have sparked an interest amongst different practitioners and academia. Manufacturing organisations are particularly well suited to benefit from data analytics platforms in their entire product lifecycle management for intelligent information processing, performing manufacturing activities, and creating value chains. This requires re-architecting their manufacturing legacy information systems to get integrated with contemporary data analytics platforms. A systematic re-architecting approach is required incorporating careful and thorough evaluation of goals for data analytics adoption. Furthermore, ameliorating the uncertainty of the impact the new big data architecture on system quality goals is needed to avoid cost blowout in implementation and testing phases. Objective: We propose an approach to reason about goals, obstacles, and to select suitable big data solution architecture that satisfy quality goal preferences and constraints of stakeholders at the presence of the decision outcome uncertainty. The approach will highlight situations that may impede the goals. They will be assessed and resolved to generate complete requirements of an architectural solution. Method: The approach employs goal-oriented modelling to identify obstacles causing quality goal failure and their corresponding resolution tactics. It combines fuzzy logic to explore uncertainties in solution architectures and to find an optimal set of architectural decisions for the big data enablement process of manufacturing systems. Result: The approach brings two innovations to the state of the art of big data analytics platform adoption in manufacturing systems: (i) A systematic goal-oriented modelling for exploring goals and obstacles in integrating manufacturing systems with data analytics platforms at the requirement level and (ii) A systematic analysis of the architectural decisions under uncertainty incorporating stakeholders’ preferences. The efficacy of the approach is illustrated with a scenario of reengineering a hyper-connected manufacturing collaboration system to a new big data architecture.

Introduction

Product lifecycle management is a data intensive process comprising market analysis, product design, development, manufacturing, distribution, post-sale, and recycling (Stark, 2015). The process involves a variety of voluminous data, e.g. customers’ comments on social media, product functions, product configuration, and failure incidences reported by installed sensors to monitor parameters of environment and products. Manufacturing organisations view such data as a valuable business asset to achieve good performance and to reduce cost in the product lifecycle. They also regularly seek to increase their productivity using new advanced information technologies that place further demand on their data processing storage requirements such as Internet of Thing (IoT) and radio-frequency identification (RFID) tags in their daily production. For example, Toyota automotive company equip cars with smart sensors and continuously collecting data about its locks, location, ignitions, and tyres which can be later used by the manufacturer assembly. Continuous product innovations lead to further product data generation coupled with a great diversity of types, sources, meaning, and format. Given its increasing volume and variety, manufacturing data is increasingly difficult to process using common manufacturing data platforms be they computer aided design (CAD), supply chain management (SCM) manufacturing execution system (MES), or enterprise resource planning (ERP). Indeed, the high volume, velocity, variety, veracity, and value adding data requirement all point to the need to complement manufacturing systems with big data platforms (McAfee, Brynjolfsson, & Davenport, 2012). New platforms such as Apache Hadoop, Google’s Dremel, or S4 are promising ways forward to address the abovementioned processing complexity (Lycett, 2013). They provide a support for capturing, processing, and visualising large volume of data sets that organisational systems may have collected over the years.