Big data analytics based fault prediction for shop floor scheduling

The current task scheduling mainly concerns the availability of machining resources, rather than the potential errors after scheduling. To minimise such errors in advance, this paper presents a big data analytics based fault prediction approach for shop floor scheduling. Within the context, machining tasks, machining resources, and machining processes are represented by data attributes. Based on the available data on the shop floor, the potential fault/error patterns, referring to machining errors, machine faults and maintenance states, are mined for unsuitable scheduling arrangements before machining as well as upcoming errors during machining. Comparing the data-represented tasks with the mined error patterns, their similarities or differences are calculated. Based on the calculated similarities, the fault probabilities of the scheduled tasks or the current machining tasks can be obtained, and they provide a reference of decision making for scheduling and rescheduling the tasks. By rescheduling high-risk tasks carefully, the potential errors can be avoided. In this paper, the architecture of the approach consisting of three steps in three levels is proposed. Furthermore, big data are considered in three levels, i.e. local data, local network data and cloud data. In order to implement this idea, several key techniques are illustrated in detail, e.g. data attribute, data cleansing, data integration of databases in different levels, and big data analytic algorithms. Finally, a simplified case study is described to show the prediction process of the proposed method.