[Paper] Automating Execution and Verification of BPMN+DMN Business Processes

Source: arXiv - 2512.15214v1

Overview

The paper presents BDTransTest, a prototype that automatically turns a BPMN + DMN business process into executable Java code, generates a systematic test suite, and measures how thoroughly the tests cover the process model. By moving beyond mere syntax checks, the authors give designers a way to catch semantic bugs early—without relying on opaque, vendor‑specific simulation tools.

Key Contributions

• End‑to‑end translation from a BPMN + DMN model to a native Java program, preserving both workflow logic and decision tables.
• Automated test‑plan synthesis that explores the input space (with optional user‑guided disambiguation) to produce concrete execution scenarios.
• Coverage analysis that reports node‑ and edge‑level coverage of the original BPMN diagram, helping designers see which parts have been exercised.
• Experimental validation on a benchmark set of real‑world BPMN + DMN processes taken from the literature, demonstrating feasibility and scalability.

Methodology

1. Model Parsing – The tool reads the BPMN XML and the associated DMN decision tables, building an internal graph representation where BPMN activities are nodes and sequence flows are edges.
2. Java Code Generation – Each BPMN task becomes a Java method; DMN tables are compiled into decision‑logic classes using the open‑source dmn‑engine library. The generated program respects the original control‑flow semantics (gateways, parallel joins, etc.).
3. Test‑Plan Synthesis –
   • The input domain of each process variable is inferred from DMN column constraints.
   • A combinatorial strategy (pairwise or full‑factorial, selectable by the user) creates concrete input tuples.
   • When the domain is ambiguous (e.g., overlapping rules), the tool prompts the designer to pick a disambiguation rule.
4. Execution & Monitoring – The generated Java program is run against each test case. A lightweight instrumentation layer records which BPMN nodes and edges are traversed.
5. Coverage Reporting – The tool aggregates the execution traces and produces a visual overlay on the original BPMN diagram, highlighting uncovered parts.

Results & Findings

• Correctness – For all 12 benchmark processes, the generated Java code reproduced the expected outcomes of the original proprietary simulators.
• Coverage – Using pairwise test generation, the average node coverage reached 92 % and edge coverage 87 %; full‑factorial testing pushed both metrics above 98 % at the cost of longer execution time.
• Performance – Translation time was under 2 seconds for models with ≤ 150 elements; test‑plan synthesis scaled linearly with the number of input variables.
• Usability – Designers only needed to intervene in 3 out of 45 decision tables to resolve ambiguous input ranges, indicating that the automated domain inference works well in practice.

Practical Implications

• Early Fault Detection – Teams can integrate BDTransTest into CI pipelines, automatically flagging semantic errors (e.g., dead‑ends, unreachable tasks, contradictory decision rules) before deployment.
• Vendor‑Neutral Testing – Because the generated code is plain Java, organizations are no longer locked into proprietary BPMN simulators for regression testing.
• Compliance & Auditing – Coverage reports provide concrete evidence that critical business paths have been exercised, simplifying regulatory audits.
• Rapid Prototyping – Developers can prototype a BPMN + DMN workflow, generate executable code instantly, and iterate on the model without manual scripting.

Limitations & Future Work

• Scalability of Full‑Factorial Tests – Exhaustive input combinations become infeasible for processes with many variables; the authors suggest integrating smarter combinatorial techniques (e.g., adaptive sampling).
• Dynamic Data Sources – The current prototype assumes static input domains; handling external services or runtime‑generated data would require runtime stubbing or mock frameworks.
• User Interface – The tool is command‑line driven; a graphical front‑end for visualizing coverage overlays directly within BPMN editors is planned.
• Formal Guarantees – While empirical coverage is high, the authors acknowledge the need for formal proofs that the generated Java semantics are bisimilar to the BPMN + DMN specification.

Bottom line: BDTransTest bridges the gap between business‑process modeling and automated software testing, giving developers a practical, open‑source pathway to verify that their BPMN + DMN workflows behave as intended.

Authors

• Giuseppe Della Penna
• Igor Melatti

Paper Information

• arXiv ID: 2512.15214v1
• Categories: cs.SE
• Published: December 17, 2025
• PDF: Download PDF