Process Model Discovery from Sensor Event Data

Abstract

Virtually all techniques, developed in the area of process mining, assume the input event data to be discrete, and, at a relatively high level (i.e., close to the business-level). However, in many cases, the event data generated during the execution of a process is at a much lower level of abstraction, e.g., sensor data. Hence, in this paper, we present a novel technique that allows us to translate sensor data into higher-level, discrete event data, thus enabling existing process mining techniques to work on data tracked at a sensory level. Our technique discretises the observed sensor data into activities by applying unsupervised learning in the form of clustering. Furthermore, we refine the observed sequences by deducing imperative sub-models for the observed discretised data, i.e., allowing us to identify concurrency and interleaving within the data. We evaluated the approach by comparing the obtained model quality for several clustering techniques on a publicly available data-set in a smart home scenario. Our results show that applying our framework combined with a clustering technique yields results on data that otherwise would not be suitable for process discovery.

Method to Identify Process Activities by Visualizing Sensor Events

“…4) [2]. It ranges from only knowing the characteristics of individual sensors (as discussed, e. g., in [8]) to being able to associate a concrete activity and process instance with a given sensor event (as discussed, e. g., in [13]). Our basic assumption is that a WfMS does not always exist to coordinate and monitor process executions in CPS [15].…”

“…Koschmider et al provide a framework to discover processes from sensor data. Accordingly, we focus on "Activity Discovery", and "Event Abstraction", where we relate events to the start or completion of process activities [8]. Going from sensor data to data suitable for process mining poses challenges regarding event extraction, abstraction, and event correlation [6,2].…”

“…When considering sensor data, the challenge of mapping fine-grained sensor data to more abstract process activities becomes more pronounced [6,24]. In literature [24], various approaches exist to bridge this abstraction gap using, e.g., Complex Event Processing (CEP) [15,18] or machine learning (ML) [4,20,8].…”

Uncertain Case Identifiers in Process Mining: A User Study of the Event-Case Correlation Problem on Click Data

“…Their results however show that this method is eluded by edge cases. Other prominent sources of sequential event data without case attribution are IoT sensors: Janssen et al [6] address the problem of obtaining process cases from sequential sensor event data by splitting the long traces according to an application-dependent fixed length, to find the optimal sub-trace length such that, after splitting, each case contains only a single activity. One major limitation of this approach that the authors mention is the use of only a single constant length for all of the different activities, which may have varying lengths.…”

Discovering Data Models from Event Logs

“…While, the extraction of such event logs is usually performed using information systems as a source, which in the case of digital twin represents the software systems accompanying the physical object. Another option to extract such log is sensor data [112,135], which are defined as a rich source of information related to the physical object [67]. An event log extracted from the sensor data can provide insights regarding the activities of people, machines, and how they behave in a specific environment.…”