Analysis of Volvo IT’s Closed Problem Management Processes By Using Process Mining Software ProM and Disco

In this study, the event logs for closed problems which was created by Volvo IT Belgium for the Business Process Intelligence Challenge in 2013 are evaluated by using Process Mining software. One of these software (ProM) is an open source program and the other (Disco) is a commercial product. Commercial application is available as a full version with academic license. Observed event logs record consist of 6660 units. This case study described the logic of the process of mining programs and issues in problem solving stage of the Volvo Company are illustrated. In this analysis, ping-pong behavior in the processes, the product which has the most frequency and the employee who has the most duration for the processes have been revealed


Introduction
Process mining is the new area of Computer Science which provides some tools and techniques to extract useful information from event logs in order to discover, monitor and improve business processes.In this manner process mining requires all event logs for a process and an event log requires a structure which is shown in Figure 1 (Kumaraguru, 2013, Van der Aalst et al., 2004).Process Mining is the concretization of process intelligence using event logs as a starting point and process intelligence is the combination of business intelligence and business process mining which is shown in Figure 2 (Aufaure andZimányi, 2013, van der Aalst et al., 2012).Process mining provides the missing link between on the one hand process model anlaysis and data-oriented analysis and on the other hand performance and coformance which is shown in Figure 3 (van der Aalst, 2014).

Literature Review
According to structure an event originates a case i.e. process instance (e.g.buy an ecommerce product) which combines an activity or task (e.g.login to website) with a timestamp (e.g.time at login) by an originator (a person who makes task) (van der Aalst et al., 2005).After the collecting of event logs, process mining is used to discover a model for example by constructing a Petri net (Petri, 1962) which can recreate observed process (van der Aalst et al., 2005).Then process mining performs a conformance checking if modeled process is compatible with observed model (van der Aalst et al., 2007).Finally, it extends the model onto a new initial model ( Van der Aalst et al., 2003).Three stages of process mining are shown in figure 4 (Günther, 2009).Process mining data can be obtained from various software such as Enterprise Resource Planning systems, Business Process Management systems, Product Data Management systems, Electronic Health Record systems or any database systems which stores event logs for process.Discovery starts with a model creation by using any process model such as Petri net (Petri, 1962) from event logs analysis.This model is actually a control-flow system and it may explain the steps of a process as well as organization and perspective (Van Der Aalst, 2011).Conformance makes a comparison between the analytical model and event logs (Van der Aalst et al., 2012).Extension is the improvements of a model by using extracted information from event logs (Burattin, 2015).

Aim of This Study
In industrial or organizational psychology, proactive behavior is to act and take control in advance of a future situation rather than being reactive.In this purpose process mining is the major point for being proactive in the context of business intelligence.Therefore, we learned process mining software and evaluated a real life example (it is event logs of Volvo IT closed problems in this study) and criticized the results.

Tools
Alphanumeric Journal Volume 4, Issue 2, 2016 We used ProM version 6.5.1 and Disco version 1.9.5 and a spreadsheet application (Excel version 2016) for the process mining.
ProM (which is short for Process Mining framework) is an Open Source framework for process mining algorithms written in JAVA (Aalst et al., 2009).Disco (which is short for Discover your process) is a commercial application for process mining but can be used as full version with an academic license (Rozinat and Günther, 2012).

Data
In this study, a real-life event log data is used which is provided by Volvo IT Belgium and published in Business Process Intelligence 2013 as a challenge (Volvo, 2013)

Analysis and Findings
After importing data in Disco according to those assignments, process analysis created a map for process frequency which is shown in figure 5.The numbers in figure 5 refer to case frequencies and arrows show the direction of process with the thickness according to frequency.The process from 'In Progress//Accepted' to 'Closed//Completed' has the highest density with the 1266 cases.According to performance graph we gather the total, median, mean, maximum and minimum durations by using combo box selection.From this graph, it is understood that there is the longest duration from 'Awaiting Assignment//Queued' to 'In Progress//Accepted' which makes a bottleneck in the overall process.
Statistical information of processes can be seen on the 'Statistics' tab which is shown in figure 7.This section gives an overall information about the process data with the activities, resources, status and product ordering to frequencies.We can gather the longest or shortest activity or status, who used how much time for their process and frequencies of each product.We can also filter the data by using filter option which is on the left bottom corner of the software according to timeframe, variations, performance, endpoints, attributes and endpoints.According to statistics there are 6660 events within 1487 cases with the 7 activities.In the table below the graph events in the cases can be seen with the start and finish time as well as duration.From this observation we learned the Carolyn used the longest time for process.Also the product PROD97 has the longest frequency.
Finally, we can observe all events as variants from 'Cases' tab.This section separates all cases into variants and variants provide a simple and sequential view on the process which is shown in figure 8.
From the created process map in figure 5, we can observe a loop between a problem 'In Progress//Accepted' and 'Assigned//Accepted' dominantly.However, we cannot observe how a single case operates through the process or how many cases loop twice or more.In order to understand this situations, we must check the variants.

Figure 7. Cases into variants tab
There are 327 variants and 1487 cases in total which is shown in figure 8.For example, we can understand that there is a loop between 'In Progress//Accepted' and 'Assigned//Accepted' for the resource Minnie related to PROD191.
Disco also provides various export options for data to be able evaluate in another program.In process mining most important point to create a standard data format for event logs.So far, this standard has been provided by using MXML format but it has some restrictions.In order to solve these problems a new format defined named XES.It is an XML-based event format and its name come from eXtensible Event Stream (Gunther and Verbeek, 2014).
After exporting VINST data from Disco as XES file format it is imported into ProM software.ProM has several algorithms for process mining that Disco cannot provide.
For example, we can create a timeline of cases and observe them by using Gantt chart which is shown in figure 9.  Support teams, organizations, and function divisions which have responsibility to most of the ping pong behaviors are given in table 3 (Kang et al., 2013).
From the observation we obtained the organizations for 'Wait//Accepted' problems which is given in table 4.

Figure 1 .
Figure 1.The structure of an event log Reference: Kumaraguru, P. V. (2013).Machine learning approach for model discovery and process enhancement using process mining techniques Ph.D. Thesis, Dr. M.G.R. Page: 82

Figure 4 .Figure 5 .
Figure 4. Process frequency map in Disco software

Figure 6 .
Figure 6.Statistics of process

Figure 8 .
Figure 8. ProM timeline graph of process . Data contains the problem management log for closed problems.Data has several attributes and first 10 records is shown in table 1.
Through this analysis 'Problem Number' attribute is used for Case ID, 'Problem Date+Time' attribute is used for Timestamp, 'Problem Status and Sub Status' attributes are used for Activity, 'Product' attribute is used for other and 'Owner' attribute (which is problem involved action owner first name) is used for Resource.

Table 2 .
Ping-Pong Behavior of a ProblemFor this problem's action owners were changed through three person involved with the same product in some same activities.The person Per has queued the problem two times for the awaiting assignment.Also problem's support team were changed from E_10 to C_6 and it was changed to E_10 again thus causing the involved person changing as well as problem duration.

Table 3 .
Responsibilities of the most of the ping pong behaviors Reference: Kang, C. J., Y. S. Kang, Y. S. Lee, S. Noh, H. C. Kim, W. C. Lim, J. Kim and R. Hong (2013).Process Mining-based Understanding and Analysis of Volvo IT's Incident and Problem Management Processes.BPIC@ BPM.Page 13.