Rule-based diagnosis method for business processes

OData support
Gönczy László
Department of Measurement and Information Systems

The implementation of software systems is often based on a design captured by business processes. This has the advantage that a domain expert can easily understand the operation of the system. In such systems, some important questions can be raised; i) how to prove that the specification meets the stakeholders' (users, operators, authorities, etc..) expectations, ii) how to prove that the operation of the system meets the specifications, iii) how to evaluate the system in terms of IT and business performance and dependability. In addition, activities in the processes can be based on functionalities provided by complex systems where internal faults and data errors can prohibit the desired functionality. Rule-based diagnosis can help to answer these questions. However, after many runs or long operational time, huge amount of data can be produced. In this case the analysis may become difficult therefore an effective tool is needed, which supports the experts' work.

For this problem a possible solution could be an environment where the business expert has the possibility to perform analysis by writing rules in a domain

specific language generated from the process model. These will be evaluated on events generated by executed processes.

In the maintenance and development of such rulebases it's worth to investigate the activation of rules. The frequent execution of particular rule may indicate a design flow (e.g., the cause of this effect can be a bad condition formulation in the rule, or maybe it's worth to split the rule for multiple smaller cases) or it can be caused by a faulty component of the system.

From the result of executed rules I create a date format, which is able to use for exploratory data analysis. In my thesis I present example for investigations

on my system.

An other way to support analytics work is calculating different statistic values on processes. In my thesis I will present how I implemented this.

After evaluating rules we have quite huge amount of data. I use them to create an illustrative visualisation of processes. (For example the failed processes

have intuitive, red color.) For this visualization I had to trace back statistic datas to the primary model.


Please sign in to download the files of this thesis.