Object-centric process mining

What is object-centric process mining?

Object-centric process mining assigns an event to one or more business objects. Relationships between orders, materials, batches, handling units and deliveries remain available rather than being forced into one case ID.

Why one case ID can fail

One order can have many items, one delivery can combine orders and one material can move through batches and load carriers. Treating every object as a separate case duplicates shared events; selecting only one object removes context.

Object-centric event model

Events have an activity and time and reference the involved typed objects. Object relationships and changes remain traceable. Analysis can follow one object type or study interaction across types.

Industrial use cases

• Order-to-cash across order, item, delivery and invoice
• Production across order, material, batch and resource
• Intralogistics across mission, handling unit and vehicle
• Maintenance across asset, notification, order and spare part

Object identities, relationships and timestamp semantics still require business validation. OCPM is used where it answers a defined question better than a classic event log.

Convergence and divergence — the two errors with names

What goes wrong when everything is forced into a single case ID has two established names. Van der Aalst and Berti identify them in 2020 as the two problems of flattening:

• Convergence. One event belongs to several cases. Flattening copies it once per case, so it appears several times in the analysis although it happened once. Frequencies and effort are systematically overstated.
• Divergence. Within one case, several instances of the same activity occur — ten items in an order, say, when the order is chosen as the object type. The model then shows loops and sequences that the process does not have.

Neither effect raises an error. Both produce a plausible-looking model with wrong numbers, which is why the object type has to be decided on business grounds before the analysis, not after seeing the result.

Sources

van der Aalst, W. (2019): Object-Centric Process Mining: Dealing with Divergence and Convergence in Event Data. In: Ölveczky, P.; Salaün, G. (eds.): Software Engineering and Formal Methods, LNCS 11724, Springer. Further: van der Aalst, W.; Berti, A. (2020): Discovering Object-Centric Petri Nets. Fundamenta Informaticae 175(1–4), pp. 1–40.

Applying Object-centric process mining in a project

Object-centric process mining is relevant when orders, materials, batches, handling units and deliveries must remain connected in one analysis.

Before applying Object-centric process mining, define the objective, system boundary and decision to be supported. The distinction from adjacent methods and systems is equally important: which processes are included, which interfaces remain outside the scope and which metrics indicate an improvement? This prevents a term from becoming a label and avoids local optimisation that creates new problems elsewhere.

A reliable assessment of Object-centric process mining combines current-state data with documented assumptions. Sources, reference periods, units and exceptions need to be transparent. Alternatives or measures can then be compared using consistent criteria. Depending on the task, these include performance and cost as well as space, inventory, ergonomics, quality, feasibility, risk and expandability.

The output from applying Object-centric process mining should support a concrete decision or a verifiable next step. Ownership, a target value and a review date make the expected effect measurable. To transfer the concept to a real assignment, it can be combined with the relevant consulting, planning, optimisation and digital capture services.