# Crowdsourcing Framework for Security Testing and Verification of Industrial Cyber-Physical Systems

## TL;DR

This paper introduces a crowdsourced testing platform to improve the security of industrial cyber-physical systems by enabling efficient and scalable security verification.

## Contribution

A novel distributed framework for testing and verifying ICPS security with support for asynchronous execution and LLM-assisted control.

## Key findings

- The proposed framework improves testing and verification speed by approximately 2.6 times compared to Apache JMeter.
- The platform supports configurable delay characteristics and multithreaded parallel execution for realistic ICPS environments.
- Experimental validations demonstrate the effectiveness of the dynamic test-task management and asynchronous testing mechanisms.

## Abstract

With the widespread deployment of Industrial Cyber-Physical Systems (ICPS), their inherent vulnerabilities have increasingly exposed them to sophisticated cybersecurity threats. Although existing protective mechanisms can block attacks at runtime, the risk of defense failure remains. To proactively evaluate and harden ICPS security, we design a distributed crowdsourced testing platform tailored to the four-layer cloud ICPS architecture—spanning the workshop, factory, enterprise, and external network layers. Building on this architecture, we develop a Distributed Input–Output Testing and Verification Framework (DIOTVF) that models ICPS as systems with spatially separated injection and observation points, and supports controllable communication delays and multithreaded parallel execution. The framework incorporates a dynamic test–task management model, an asynchronous concurrent testing mechanism, and an optional LLM-assisted thread controller, enabling efficient scheduling of large testing workloads under asynchronous network conditions. We implement the proposed framework in a prototype platform and deploy it on a virtualized ICPS testbed with configurable delay characteristics. Through a series of experimental validations, we demonstrate that the proposed framework can improve testing and verification speed by approximately 2.6 times compared to Apache JMeter.