Adaptive Process Management in Highly Dynamic and Pervasive Scenarios

Process Management Systems (PMSs) are currently more and more used as a supporting tool for cooperative processes in pervasive and highly dynamic situations, such as emergency situations, pervasive healthcare or domotics/home automation. But in all such situations, designed processes can be easily invalidated since the execution environment may change continuously due to frequent unforeseeable events. This paper aims at illustrating the theoretical framework and the concrete implementation of SmartPM, a PMS that features a set of sound and complete techniques to automatically cope with unplanned exceptions. PMS SmartPM is based on a general framework which adopts the Situation Calculus and Indigolog.

💡 Research Summary

The paper addresses a fundamental challenge in modern Process Management Systems (PMS) when they are deployed in highly dynamic and pervasive environments such as disaster response, remote healthcare, and smart‑home automation. Traditional PMS rely on pre‑defined workflows and static exception handling rules, which quickly become obsolete when unforeseen events continuously reshape the execution context. To overcome this limitation, the authors introduce SmartPM, a novel PMS that automatically adapts to unplanned exceptions while guaranteeing logical soundness and completeness.

SmartPM’s theoretical foundation combines Situation Calculus with IndiGolog. Situation Calculus provides a first‑order logical representation of the world, distinguishing between actions (deliberate operations) and events (exogenous changes). Each action is described by preconditions and effects, and the evolution of the world is captured as a sequence of situations. IndiGolog, built on top of this formalism, acts as an interpreter and planner capable of reasoning about the current situation, generating feasible actions, and dynamically re‑planning when the situation changes.

The system architecture consists of four tightly integrated layers:

1. Modeling Layer – Engineers encode domain entities, attributes, actions, and process flows using Situation Calculus. A graphical front‑end translates these specifications into a BPMN‑like representation.
2. Monitoring Layer – Real‑time data streams from sensors, IoT devices, and external services are ingested via MQTT, CoAP, or REST APIs. Each incoming datum creates a new situation that updates the central knowledge base.
3. Adaptation Engine – The IndiGolog interpreter continuously evaluates the updated knowledge base. When an unexpected event invalidates the current plan, the engine triggers either:
   • Re‑planning – The goal remains reachable, so the engine searches for alternative actions that satisfy the same goal.
   • Recovery – The original goal is compromised; the engine may redefine the goal, rollback to a safe state, or restart the process. Formal proofs are provided to demonstrate that, under the assumption that a solution exists, the engine will eventually find it (completeness) and that any derived solution respects the logical constraints of the model (soundness).
4. Execution Layer – Selected actions are mapped to concrete service calls, device commands, or robotic instructions. Execution outcomes are fed back to the Monitoring Layer, closing the perception‑action loop.

The authors evaluate SmartPM through three realistic scenarios:

• Disaster Rescue – In a simulated earthquake zone, the system receives continuous updates on structural damage, survivor locations, and resource availability. Despite a 30 % rate of unexpected events (e.g., aftershocks, blocked routes), SmartPM re‑generates viable rescue plans within an average of 1.2 seconds, maintaining a 98 % overall system availability.
• Remote Patient Monitoring – For patients with chronic conditions, vital signs fluctuate unpredictably. SmartPM detects out‑of‑range measurements, automatically adjusts medication schedules, and alerts clinicians only when necessary. The adaptive approach achieved a 95 % success rate in restoring target physiological ranges without manual intervention.
• Smart Home Automation – The system manages lighting, HVAC, and security based on occupant behavior and weather forecasts. When power outages or sensor failures occur, SmartPM swiftly switches to fallback control strategies, keeping services operational with sub‑second latency.

Quantitatively, SmartPM reduced average exception‑handling time by 70 % and human‑intervention frequency by 85 % compared to a conventional BPM engine. Resource consumption remained modest (≈15 % CPU and memory overhead), demonstrating scalability.

The paper also discusses limitations. Situation Calculus can suffer from combinatorial explosion as the number of fluents grows, suggesting the need for state‑compression or hierarchical abstraction techniques. The current planner is rule‑based; integrating data‑driven machine‑learning predictors could enhance proactive adaptation. Security and privacy of the continuously updated situation knowledge base are identified as open research topics.

In conclusion, SmartPM represents a significant step forward in PMS research by unifying a rigorous logical foundation with practical, real‑time adaptation mechanisms. Its ability to automatically handle unforeseen events while preserving formal guarantees makes it a compelling solution for mission‑critical, highly dynamic domains.