These slides were presented at the NAFEMS World Congress 2025, held in Salzburg, Austria from May 19–22, 2025.

Abstract

There is a long history of the use of engineering simulation for design. This virtual approach is typically followed by physical prototyping, testing and refinement to reach a final design, followed in many cases by a physical testing regime to meet regulatory requirements. For many companies, the use of simulation tools has reduced the time and cost associated with getting new products to market due to the ability to explore multiple designs, and has reduced resource usage and improved product quality by enabling exploration of aspects of manufacturability and long term in-use performance. Companies are increasingly seeking to gain similar benefits beyond the design stage. Some companies whose products are subject to extensive regulatory testing requirements are seeking to provide evidence of compliance through a combination of simulation and testing. It is common in such industry sectors to have a 'œtesting pyramid', where the safety of a complex multi-component product is demonstrated by carrying out tests of materials, components, assemblies and complete systems, with the number of tests carried out decreasing as the complexity of the object under test increases. A 'œsmarter testing' approach would replace some of these physical tests with simulations and would feed information between the various tests to improve the validation of the simulation and the confidence in the evidence of safety. Some products cannot be fully tested via physical testing alone because they cause a risk to human safety. An example of current relevance is autonomous vehicles. The artificial intelligence (AI) that controls an autonomous vehicle is often trained on data obtained from human-controlled journeys of a vehicle with the sensor suite in operation, so that the AI is shown what safe driving looks like under typical conditions. However, many of the situations most likely to lead to an accident are not encountered in typical driving conditions, and could cause risk to life if recreated deliberately. A simulation can potentially recreate high-risk scenarios safely for both training and testing purposes. Some products can significantly improve lifetime prediction and understanding of real-world performance by linking models and data in a digital twin. This approach can lead to improved future design iterations and more effective maintenance plans as the understanding of the product improves. Application of this approach could support personalisation of devices such as medical protheses, where monitoring, adjustment and individualisation could significantly improve people'™s lives. In all of these applications it is important to note that the company developing the simulations are not the only people that need to have trust in the simulation results. That trust needs to be shared by regulators, end users, and in some cases the general public. These three seemingly distinct themes of activity are strongly linked, not least because they have the same need underpinning them: they need to combine validated models and measured data to make trustworthy predictions of real-world behaviour. This need can be answered most efficiently by a combination of activities in several technical areas, including data quality assessment, software interoperability, semantic technologies, model validation, and uncertainty quantification. The technology readiness level of these areas is varied, and the level of awareness and uptake of good practice of each technical area varies across sectors. This paper discusses the common features, and differences between, the fields of smart testing, virtual test environments, and digital twins. Starting from a consideration of commonality we will highlight areas where existing methods and expertise could be better exploited, and identify areas where further research and development of tools would accelerate successful application of trustworthy digital assurance approaches in industry.