Learning factory

Process mining in production and logistics: data, variants and limits

Process mining reveals actual flows from system events. It becomes reliable only when case concepts, object relations and booking semantics are correct.

In brief

Case ID, activity and timestamp form the classic event log. Production and logistics often require an object-centric view of orders, materials, batches and deliveries.

What does process mining provide?

Process mining reconstructs observed process behaviour from event data and connects variants with time, quantity, resources and deviations. It can support discovery, conformance checking, performance analysis and operational monitoring.

The event log is a business model

A classic event log contains a case, activity and timestamp. Production and logistics also require a decision on what the case represents. Orders, materials, batches, handling units and deliveries interact. Forcing them into one case ID can create duplication or loss of context.

Process-mining chain with quality gates
  1. 01

    Question

    Decision and boundary

  2. 02

    Data model

    Cases, objects and events

  3. 03

    Quality

    Completeness and semantics

  4. 04

    Analysis

    Variants, time and rules

  5. 05

    Verification

    Cause and action

Data quality before process maps

Timestamp meaning, overwritten states, duplicates, cancellations, late manual postings, time zones and missing unrecorded cases are tested explicitly.

Limits of directly-follows graphs

A directly-follows graph shows observed adjacency. It can misrepresent concurrency and loops, and frequency filtering changes which behaviour remains visible. Van der Aalst explains these risks in A practitioner's guide to process mining: Limitations of the directly-follows graph.

Analysis levels

LevelQuestionOutput
DiscoveryWhat happened?variants and model
PerformanceWhere do time and WIP accumulate?waiting and bottleneck patterns
ConformanceHow does observed behaviour differ?deviations with context
EnhancementWhich attributes explain behaviour?segments and hypotheses

From correlation to action

A pattern is not yet a cause. Process owners verify whether a site, product or shift explains an issue or merely correlates with another factor. Measures receive target metrics and a before-after or comparison design.

Related: process-mining consulting, value stream analysis, event log and object-centric process mining.

FAQ

Frequently asked questions

Which three fields does a classic event log need?

A case or process instance, an activity and a timestamp. Resources, material, cost and other attributes extend the analysis.

Why is the case ID difficult in manufacturing?

Several objects such as orders, materials, batches and handling units interact; one ID can lose relations or multiply events.

Does a directly-follows graph show the complete process?

A directly-follows graph shows immediate adjacency and can misrepresent concurrency and behaviour removed by filters.

Does process mining automatically identify causes?

It identifies patterns. Causes require process knowledge, additional data and verification through controlled change or comparison.

Contact

Turn your challenge into a sound plan

Talk to us about your data basis, alternatives and the next robust decision.