Our society is surrounded by technical systems that increasingly influence andtake over aspects of everyone’s everyday life. As we build on the correct functioning, the need for appropriate methodologies that support the development of reliable products persists. However, the development and in particular the qualification become challenging due to their increasing inherent complexity. Virtual testbeds and digital twins are modern key concepts in this context and offer a wide range of opportunities for simulation-based verification and validation processes with virtual prototypes, especially in cases where real tests are hard to perform. However, the validity of test results derived from virtual experiments depends on the quality of the underlying simulation models.Of course, the models themselves must be verified before. But in the contextof virtual testbeds and digital twins this becomes challenging since the corresponding models are usually complex themselves. Therefore, we presented an approach to establish structured verification and validation activities for digital twins and virtual testbeds, based on a modular configuration framework that supports the development of complex simulation models.
The growing relevance of artificial intelligence (AI) for technical systems offers significant potential for the realization and operation of autonomous systems in complex and potentially unknown environments. However, unlike classical solution approaches, the functionality of an AI system cannot be verified analytically, which is why data-driven approaches such as scenario-based testing are used. With the increasing complexity of the required functionality of the AI-based system, the quantity, and quality of the data needed for development and validation also increase. To meet this demand, data generated synthetically using simulation is increasingly being used. Compared to the acquisition of real-world reference data, simulation offers the major advantage that it can be configured to test specific scenarios of interest. This paper presents an architecture for the systematic generation of virtual test scenarios to establish synthetically generated test data as an integral part of the development and validation process for AI systems. Key aspects of this architecture are the consistent use of digital twins as virtual 1-to-1 replicas and a simulation infrastructure that enables the generation of training and validation data for AI-based systems in appropriate quantity, quality, and time. In particular, this paper focuses on the application of the architecture in the context of two use cases from different application domains.
Support Vector Machine (SVM) is currently an efficient classification technique due to its ability to capture nonlinearities in diagnostic systems, but it does not reveal the knowledge learnt during training. It is important to understand of how a decision is reached in the machine learning technology, such as bioinformatics. On the other hand, a decision tree has good comprehensibility; the process of converting such incomprehensible models into an understandable model is often regarded as rule extraction. In this paper we proposed an approach for extracting rules from SVM for microarray dataset by combining the merits of both the SVM and decision tree. The proposed approach consists of three steps; the SVM-CHI-SQUARE is employed to reduce the feature set. Dataset with reduced features is used to obtain SVM model and synthetic data is generated. Classification and Regression Tree (CART) is used to generate Rules as the Last phase. We use breast masses dataset from UCI repository where comprehensibility is a key requirement. From the result of the experiment as the reduced feature dataset is used, the proposed approach extracts smaller length rules, thereby improving the comprehensibility of the system. We obtained accuracy of 93.53%, sensitivity of 89.58%, specificity of 96.70%, and training time of 3.195 seconds. A comparative analysis is carried out done with other algorithms.
Experimentable Digital Twins are capable of combining different simulation domains on a system level. This has been shown for a multitude of simulation domains, e.g., rigid body dynamics, control, sensors, kinematics, etc., and application scenarios, e.g., automotive, space, and industrial engineering. In our work, we investigate how to include structural loads into an Experimentable Digital Twin while maintaining computational efficiency and interoperability on a system level. We combine rigid body dynamics with the transfer matrix method to simulate forces and stresses. We show our approach for statically determinate beam structures in a simulation on a system level and validate it experimentally and numerically with static and dynamic example problems. The results show a strong agreement in these comparisons, confirming the accuracy and reliability of our method. For practical applications, we see force and stress simulation using the transfer matrix method as an additional tool to facilitate simulation-based engineering in the early stages of structural design processes, e.g., when dealing with uncertain loading conditions and operational complexity on a system level.
1 September 2021 in International Journal of Advances in Applied Sciences
Photovoltaic (PV) power has emerged as the most attractive resource in form of a clean and green energy. However, one major challenge associated with PV interfacing is its intermittent output characteristic which varies dramatically with the operating conditions. Thus designing an effective maximum power point tracking (MPPT) algorithm is a key aspect for optimizing PV system performance. Numerous MPPT algorithms have been proposed earlier having their own specific advantages. However, these are found to have two major limitations which have to be essentially addressed. Firstly, they become ineffective in the dynamic conditions where there is rapid change in environmental parameters like insolation and temperature. Secondly, they fail to discriminate between global and local peaks under partial shading conditions. Therefore, to achieve a reliable and efficient system operation, this paper presents an enhanced model-based (MB) algorithm that overcomes both these deficiencies. Based on new governing equations and precised estimation technique, it predetermines the MPP analytically. First simulated results are obtained where it is tested for dynamic variations of all the three parameters. Then the experimental validation is carried out on a 2 KW installed panel where real time data is recorded through CR1000 data logger and environmental parameters are sensed with elements like pyranometer and humidity sensor. A large number of experimental results are obtained for tracked MPP in the dynamic conditions, which are then summarized in tabular forms. These are finally plotted and compared with simulated results to illustrate the effectiveness of proposed MB algorithm.