This article presents a systematic mapping study classifying existing scientific contributions on synergies of Model-Driven Engineering (MDE), AI/Machine Learning (ML), and DevOps, with the overall objective of supporting the continuous development of Cyber-Physical Systems (CPSs). We collected papers from bibliographic sources and selected primary studies to analyze. Then, we characterized and classified the current state of the art, focusing on (1) main aspects already tackled at the intersection of at least two of the three studied areas, and (2) findings emerging from the analysis as a framework for potential future research, notably regarding the integration of the three studied areas. The results reveal that few approaches combine MDE, AI/ML, and DevOps for software and systems engineering. In contrast, several approaches have combined two of them, specifically MDE and DevOps. Approaches combining AI/ML with MDE or DevOps are also becoming more frequent and will most likely continue to progress in the future. These synergies cover a range of engineering activities, from requirements and design to monitoring, maintenance, and evolution. Open research challenges include advancing AI/ML, MDE, and DevOps integration, supporting scalable, data-oriented solutions, proposing new continuous engineering methods, and adapting DevOps practices to diverse systems. © 2026

Performance testing with the aim of generating an efficient and effective workload to identify performance issues is challenging. Many of the automated approaches mainly rely on analyzing system models, source code, or extracting the usage pattern of the system during the execution. However, such information and artifacts are not always available. Moreover, all the transactions within a generated workload do not impact the performance of the system the same way, a finely tuned workload could accomplish the test objective in an efficient way. Model-free reinforcement learning is widely used for finding the optimal behavior to accomplish an objective in many decision-making problems without relying on a model of the system. This paper proposes that if the optimal policy (way) for generating test workload to meet a test objective can be learned by a test agent, then efficient test automation would be possible without relying on system models or source code. We present a self-adaptive reinforcement learning-driven load testing agent, RELOAD, that learns the optimal policy for test workload generation and generates an effective workload efficiently to meet the test objective. Once the agent learns the optimal policy, it can reuse the learned policy in subsequent testing activities. Our experiments show that the proposed intelligent load test agent can accomplish the test objective with lower test cost compared to common load testing procedures, and results in higher test efficiency.

Test automation has been an acknowledged software engineering best practice for years. However, the topic involves more than the repeated execution of test cases that often comes first to mind. Simply running test cases using a unit testing framework is no longer enough for test automation to keep up with the ever-shorter release cycles driven by continuous deployment and technological innovations such as microservices and DevOps pipelines. Now test automation needs to rise to the next level by going beyond mere test execution.

Testing resource allocation is the problem of planning the assignment of resources to testing activities of software components so as to achieve a target goal under given constraints. Existing methods build on Software Reliability Growth Models (SRGMs), aiming at maximizing reliability given time/cost constraints, or at minimizing cost given quality/time constraints. We formulate it as a multi-objective debug-aware and robust optimization problem under uncertainty of data, advancing the stateof- the-art in the following ways. Multi-objective optimization produces a set of solutions, allowing to evaluate alternative tradeoffs among reliability, cost and release time. Debug awareness relaxes the traditional assumptions of SRGMs – in particular the very unrealistic immediate repair of detected faults – and incorporates the bug assignment activity. Robustness provides solutions valid in spite of a degree of uncertainty on input parameters. We show results with a real-world case study.

The importance of efficient software testing procedures is driven by an ever increasing system complexity as well as global competition. In the particular case of manual test cases at the system integration level, where thousands of test cases may be executed before release, time must be well spent in order to test the system as completely and as efficiently as possible. Automating a subset of the manual test cases, i.e, translating the manual instructions to automatically executable code, is one way of decreasing the test effort. It is further common that test cases exhibit similarities, which can be exploited through reuse when automating a test suite. In this paper, we investigate the potential for reducing test effort by ordering the test cases before such automation, given that we can reuse already automated parts of test cases. In our analysis, we investigate several approaches for prioritization in a case study at a large Swedish vehicular manufacturer. The study analyzes the effects with respect to test effort, on four projects with a total of 3919 integration test cases constituting 35,180 test steps, written in natural language. The results show that for the four projects considered, the difference in expected manual effort between the best and the worst order found is on average 12 percentage points. The results also show that our proposed prioritization method is nearly as good as more resource demanding meta-heuristic approaches at a fraction of the computational time. Based on our results, we conclude that the order of automation is important when the set of test cases contain similar steps (instructions) that cannot be removed, but are possible to reuse. More precisely, the order is important with respect to how quickly the manual test execution effort decreases for a set of test cases that are being automated. 

In testing, engineers want to run the most useful tests early (prioritization). When tests are run hundreds or thousands of times, minimizing a test set can result in significant savings (minimization). This paper proposes a new analysis technique to address both the minimal test set and the test case prioritization problems. This paper precisely defines the concept of mutant stubbornness, which is the basis for our analysis technique. We empirically compare our technique with other test case minimization and prioritization techniques in terms of the size of the minimized test sets and how quickly mutants are killed. We used seven C language subjects from the Siemens Repository, specifically the test sets and the killing matrices from a previous study. We used 30 different orders for each set and ran every technique 100 times over each set. Results show that our analysis technique performed significantly better than prior techniques for creating minimal test sets and was able to establish new bounds for all cases. Also, our analysis technique killed mutants as fast or faster than prior techniques. These results indicate that our mutant stubbornness technique constructs test sets that are both minimal in size, and prioritized effectively, as well or better than other techniques.

Mutation analysis has proven to be a strong technique for software testing. Unfortunately, it is also computationally expensive and researchers have therefore proposed several different approaches to reduce the effort. None of these reduction techniques however, focuses on non-functional properties. Given that our goal is to create a strong test suite for testing a certain non-functional property, which mutants should be used? In this paper, we introduce the concept of targeted mutation, which focuses mutation effort to those parts of the code where a change can make a difference with respect to the targeted non-functional property. We show how targeted mutation can be applied to derive efficient test suites for estimating the Worst-Case Execution Time (WCET). We use program slicing to direct the mutations to the parts of the code that are likely to have the strongest influence on execution time. Finally, we outline an experimental procedure for how to evaluate the technique.