Graph Transformation System (GTS) can formally specify the behavioral aspects of complex systems through graph-based contracts. Test suite generation under normal conditions from GTS specifications is a task well-suited to evolutionary algorithms such as Genetic and Particle Swarm Optimization (PSO) metaheuristics. However, testing the vulnerabilities of a system under unexpected events such as invalid inputs is essential. Furthermore, the mentioned global search algorithms tend to make big jumps in the system’s state-space that are not concentrated on particular test goals. In this paper, we extend the HGAPSO approach into a cost-aware Memetic Algorithm (MA) by making small local changes through a proposed local search operator to optimize coverage score and testing costs. Moreover, we test GTS specifications not only under normal events but also under unexpected situations. So, three coverage-based testing strategies are investigated, including normal testing, robustness testing, and a hybrid strategy. The effectiveness of the proposed test generation algorithm and the testing strategies are evaluated through a type of mutation analysis at the model-level. Our experimental results show that (1) the hybrid testing strategy outperforms normal and robustness testing strategies in terms of fault-detection capability, (2) the robustness testing is the most cost-efficient strategy, and (3) the proposed MA with the hybrid testing strategy outperforms the state-of-the-art global search algorithms.

中文翻译：

---

使用模因算法对基于契约的软件模型进行稳健性测试

图转换系统（GTS）可以通过基于图的合约正式指定复杂系统的行为方面。根据 GTS 规范在正常条件下生成测试套件是一项非常适合进化算法的任务，例如遗传和粒子群优化 (PSO) 元启发式算法。但是，在无效输入等意外事件下测试系统的漏洞是必不可少的。此外，所提到的全局搜索算法往往会在系统状态空间中进行大跳跃，而不是集中在特定的测试目标上。在本文中，我们将 HGAPSO 方法扩展为成本感知模因算法 (MA)，方法是通过建议的本地搜索算子进行小的本地更改，以优化覆盖率分数和测试成本。而且，我们不仅在正常事件下而且在意外情况下测试 GTS 规范。因此，研究了三种基于覆盖的测试策略，包括正常测试、健壮性测试和混合策略。所提出的测试生成算法和测试策略的有效性是通过模型级别的一种变异分析来评估的。我们的实验结果表明：（1）混合测试策略在故障检测能力方面优于正常和鲁棒性测试策略，（2）鲁棒性测试是最具成本效益的策略，以及（3）提出的 MA 与混合测试策略优于最先进的全局搜索算法。和混合策略。所提出的测试生成算法和测试策略的有效性是通过模型级别的一种变异分析来评估的。我们的实验结果表明：（1）混合测试策略在故障检测能力方面优于正常和鲁棒性测试策略，（2）鲁棒性测试是最具成本效益的策略，以及（3）提出的 MA 与混合测试策略优于最先进的全局搜索算法。和混合策略。所提出的测试生成算法和测试策略的有效性是通过模型级别的一种变异分析来评估的。我们的实验结果表明，（1）混合测试策略在故障检测能力方面优于正常和鲁棒性测试策略，（2）鲁棒性测试是最具成本效益的策略，以及（3）所提出的 MA 与混合测试策略优于最先进的全局搜索算法。