Process Mining in Logistics - A Need for Multi-Dimensional Analytics

29-10-2018, by Dirk Fahland

This post is part of the 'Insights' obtained in the Process Mining in Logistics project between TU Eindhoven and Vanderlande where we develop process mining technology for logistics systems.

Logistics systems form a network of processes, cases, and objects that interact with each other in a multi-dimensional fashion: Multiple cases synchronize at batching and de-batching steps, multiple objects may queue in high-load situations over multiple processing steps, and the routing decisions for each object depend not only on its case, but also on the load in other system parts. The aim of the Process Mining in Logistics project is to provide fact-driven process analytics for such multi-dimensional dynamics.

Process mining has become a mature technique for fact-driven process analytics. However, logistics processes are very different from the “information handling” businesses processes for which existing process mining techniques have been built. In the following, I explain how logistics processes are different and which challenges they raise for process mining technology.

  • In logistics processes, materials have to be moved and processed in synchronization with “information handling” business processes (such as order fulfillment), and the position and state of each material throughout the entire process is central to the analysis. As a result, logistics processes cover 1000's of process steps that have to be considered in the analysis. Even a focused analysis on a specific system part may still involve 100's of steps - in contrast to at most a few dozen steps considered in business processes. Analytics on process models and process maps of this scale becomes intractable with existing process mining technology.
  • Complex logistics processes are supported by modern material handling solutions such as Vanderlande's who take over the task of transporting, managing, and processing materials of a 10,000s of cases at the same time. Many business processes see such volumes only over the course of a month or a year. With cases having 100's or 1000's of events, event data quickly reaches a million events on a single day - in a single system - which places very high demands on scalability.
  • While in information handling processes, cases are isolated from each other, in logistics systems, two material items cannot occupy the same position on, for example, a conveyer belt, sorter, storage unit, or scanner. As a result, different cases immediately influence each other: For example, if a bag could not be processed fast enough in one step in an airport baggage handling system, then all other bags (also those for other flights) behind would queue up, potentially leading to a cascade of delays or a system halt. Inter-case dependencies and dynamics are dominant in logistics processes while they hardly occur in information handling processes - and very limited analysis techniques exist to date.
  • Material handling systems are very versatile and support - just like information systems - dozens and more process variants. But because of the physical inter-case dynamics in the system, the different variants cannot be filtered out and analyzed in isolation. Logistics processes have to be analyzed as a whole considering all cases and variants together.
  • Because of the high degree of automation offered by material handling solutions, logistics processes suffer much less from uncontrolled, yet persistent deviations or performance problems than business processes. The processes are designed upfront to continuously deliver steady, reliable performance within the system boundaries. When, however, the system needs to operate beyond its boundaries - because of delays in flight schedules or incoming deliveries, deviations and performance problems may turn severe. Once resolved, the system returns to normal operation. Process deviations and performance problems in well-designed logistics processes are inherently rare and transient but severe and only occur for a very short period of time, rendering existing steady-state analysis techniques inapplicable.

All these characteristics together, makes identifying and predicting process deviations and performance problems notoriously difficult as they simply either disappear or are not addressed in dashboards or existing process mining tools.

Check our Project Insights for how we solve these challenges in the “Process Mining in Logistics” project.