With new advances in Cyber-Physical Systems (CPS) and Internet of Things (IoT), more and more discrete software controllers interact with continuous physical systems. Workflow models are a classical approach to define controllers. However, the effect of the associated actions that are activated by executing the workflow may not spontaneously be realized but have to be realized over time. Generally, behavioral model elements such as activities in workflow languages are displayed mostly as black box, meaning that it is not possible to trace variable changes over time in most of the classical modeling approaches. In this paper, we introduce an envisioned architecture to cope with this challenge.
Workflow Model
Transformation Messaging Middleware
SensoLroL Vogagslsue
Stream
Event Pattern
Query M a tsceh TSDB
A First Architectural Style for Continuous Behaviour Mining
Consider a system where sensors continuously send data to a controller, which sends
signals to the actuators based on these data according to a predefined workflow. This
workflow is modeled at design-time and defines the expected execution order for the
system. At runtime, it is important to trace and monitor whether the system behaves
as intended or not. A promising way to achieve this requirement are event pattern
queries [
At runtime, each component within the system communicates via a messaging system middleware. The middleware can be also used to transfer sensor values to produce logs of a system in a time series database (TSDB). A single log contains the following fields: (i) timestamp, the actual time; (ii) sensor, the name of the sensor; and (iii) value, the received sensor value. By recording the sensor data in a TSDB, it is possible to illustrate variable changes of components continuously. Beside the recording of the data, the value streams can be analyzed for detecting the defined activities of the WM. Thereby, the value streams are filtered by an event pattern query which may be derived from the WM. By this query, one can continuously sample the runs of the system, e.g., to detect process duration changes over time (i.e., time between matches becomes longer or shorter). The query responses produce a stream of matched actions, the socalled “action stream”. Now, it is possible to map the logs of the received sensor values and the query matches together in the TSDB to produce condensed figures of continuous variable value shapes within activities. With the help of the presented architectural style, complex behavior of the system can be continuously monitored.
Outline for future work. Based on the logs of the sensors and the logs of the query matches, it is possible to analyze specific slices of time series. For instance, hidden activities may be detected where variable changes are observed between two defined activities, i.e., they are not described in the WM.