ion; we could mine data about business transactions instead of low level events and/or activities. The business transaction process visualizations were easier to understand for stakeholders than owcharts. Next to this, we determined the optimal inventory level for guaranteeing a certain service level (using Lean Six Sigma), making it possible for organizations to make a tradeo between item price and service level.
Recently we have come across several organizations that face di culties getting the right material to the right place at the right moment for carrying out preventive and corrective maintenance. Preventive maintenance deals with inspecting the current state of a machine, detecting potential problems and cleaning/replacing items before defects occur. It is scheduled after a certain xed Copyright c by the papers authors. Copying permitted for private and academic purposes. In: Aveiro et al. (Eds.): Proceedings of the EEWC Forum 2017, Antwerp, Belgium, 09-May-2017 to 11-May-2017, published at http://ceur-ws.org period or after a certain amount of usage (e.g. working hours of a factory machine or vehicle mileage). Corrective maintenance deals with xing the machine after a defect occurred. The demand for items is less predictable for the second type of maintenance (and quick delivery is even more important). Inventory can be stored at di erent locations having di erent lead times for transportation of the item to the machine or vice versa, e.g.: { the project location (for instance a construction area or an o shore location), { the mechanics workplace (for instance a garage or a hangar), { a local warehouse, { a central warehouse, { warehouse of the supplier.
Organizations must nd a balance between costs of inventory and provided service level to the mechanic who requires the items. Too little inventory may lead to unnecessary downtime of machines and mechanics waiting instead of working, too much inventory leads to high storage costs, less money available for other business activities, and higher risk of an item becoming obsolete, damaged or stolen.
The task we were faced with at several organizations was to nd bottlenecks
in logistics processes that lead to unnecessary downtime of machines. This task
included determining which items should be kept in inventory to provide an
optimal service level to mechanics for preventive and corrective maintenance.
We extracted data from ERP systems (Infor, SAP, and tailor made systems),
combined this data with data from other enterprise applications and converted
them to a structure suitable for process mining [
This type of process mining, however, does not deal with the semantics of
the individual activities/events. We combined process mining with the DEMO
methodology (based on the -theory [
Though process mining can be used to analyze item movement between different locations, it does not deal with optimal inventory levels. To nd such an optimal inventory level, we had to make additions to our process mining factory. We introduced ideas from Lean Six Sigma on this topic and used a continuous review model (inventory can be ordered at any moment). We determined for each item type when new items should be ordered and how many items should be ordered by calculating the inventory reorder point and the optimum order quantity. The inventory reorder point is the level of inventory at which the inventory should be replenished to make sure a certain service level can be guaranteed. It is calculated as follows: d
D n = L
+ [ (S)], where: { L is the lead time in days, { d is the annual demand for the item, { D is the number of working days in a year, { is the demand standard deviation (per lead time), { S is the required service level, { is the inverse of the standard normal distribution.
As we can see a higher variation in demand leads to a higher inventory level. In our cases we assumed we cannot in uence the demand or the variation in demand (though this might be possible by analyzing the maintenance process!). The service level is a process requirement which is, in general, dependent on the price of the item. In some situations (very expensive or rarely used items) only one item is ordered each time. In other cases we want to know how many items we should order. This optimum order quantity can be calculated as follows: Q = r 2od ui
, where: { o is the order cost, { d is the annual demand, { u is the unit cost, { i is the interest rate or carrying cost.
We made custom visualizations (see Fig. 1) in our process mining factory for presenting logistics processes. These comprised more coarse-grained business transactions instead of ne-grained activities/events. This enabled us to show the process to domain experts in a way that re ect their way of thinking (di erent actors communicating about services they o er to each other).
These visualizations show the following metrics per business transaction type: { total number successfully executed (only coordination steps from basic transaction pattern), { total number failed (includes decline, reject or cancellation acts), { average duration of the transaction (when item is delivered from inventory or backorder), { median duration of the transaction (when item is delivered from inventory or backorder), { the survival curve (how many cases are still `in the transaction' after a certain period).
Because DEMO processes have a tree structure we can compare the metrics of an individual business transaction to those of the complete process (which is the root transaction!). The bars of the metrics in the di erent business transactions are therefore made relative to the values of the metrics of the root transaction. We can now easily see which business transactions take up most time and which business transactions fail frequently. We can show this visualization for all types of material, but of course we can also slice it for a speci c type of material. When we do this we can view an additional visualization that shows inventory related information as depicted in Fig. 2.
In this paper we presented a way to combine fully automated process discovery with making manual annotations about the semantics of individual activities/events using the DEMO methodology. We did not only focus on analyzing the logistics process itself, but also on determining the optimal inventory level to guarantee a certain service level. We see some ways of further improving our approach. First, we would like to make a distinction between the di erent reasons why a business transaction may `fail' (currently we only distinguish between succeeded and failed). Is it because of a decline, reject, or cancellation? This can give the organization a better understanding on why things go wrong and what to do about it. A second improvement is to nd a way of dealing with declined, rejected and canceled orders in determining the inventory reorder point. At the moment we exclude them from this calculation, but this is not in all cases the best way to deal with them. A third improvement is to automatically mine relations between business transactions and to see where the order does not go as planned.
Want to know more about data science or process mining with our process mining factory (www.processminingfactory.com)? Email me or call us at +31 30 227 04 13.
Looking forward to hearing from you!