下一前沿：人工智能驱动世界的工程仿真

McKinsey DirectThe next horizon: Engineering simulation As AI and its uses advance, the scope of engineering simulation ischanging, redesigning the way engineering is done and moving the Engineering simulation (ES)has long been a powerful catalyst for better, faster, and more cost-effective product development. Its benefits are well established: Virtual testing reduces the needfor physical prototypes and accelerates time-to-market. It enables teams to explore designs morethoroughly and resolve issues earlier in the development cycle. In industries characterized by Over the past decade, artificial intelligence (AI), specifically engineering AI (including gen AI) havebegun reshaping industrial systems more broadly (see sidebar “What is engineering AI?”). Automotivemanufacturers already deploy AI-based predictive maintenance to extend service intervals andoptimize repair interventions based on vehicle sensor data. Aerospace players are using engineering AI What is engineering AI? As AI advances and broadens the use cases for it, new terms arise, and old ones expand. Here is a briefdescription of some of the terms we use in this article. Artificial intelligence (AI).A broad class of computational techniques that enable machines toperform tasks typically requiring human intelligence such as pattern recognition, prediction, reasoning, Engineering AI.Used across the product life cycle to accelerate analysis, uncover hidden relationshipsin data, augment expert judgment, and generate novel engineering solutions. Machine learning (ML).A subset of AI in which algorithms learn statistical relationships from datato make predictions or decisions without being explicitly programmed for each task. ML underpinsmany AI applications in engineering simulation (ES), including response surface modeling, uncertaintyquantification, anomaly detection, and real-time model calibration. There is no one-size-fits-all ML Generative AI (gen AI).A subset of AI focused on models that generate new content such as text,code, images, or structured data, based on learned patterns from large datasets. In simulationcontexts, gen AI can automate model setup, propose design alternatives, generate surrogate models, Agentic AI (AI agents).AI systems designed to operate with a degree of autonomy, pursuing goals byplanning, acting, observing outcomes, and adapting over time. In ES, AI agents can orchestrate multi-step workflows, such as iterating designs, running simulations, interpreting results, and triggering next This AI-driven transformation is underway in ES. Engineering AI can dramatically accelerate simulation byaugmenting physics-based numerical solvers with deep-learning surrogates and advanced probabilisticmethods. It can auto-generate and optimize design candidates at a pace impossible for human engineersalone. Early adopters of AI simulation tools are already capturing significant benefits. Companies havereported reductions in time to market of up to 50 percent using deep-learning surrogates to accelerate These disruptions are profound, because they change both the workflows that engineers follow and theunderlying systems used to build and run simulations. To better understand the implications of AI-enabledsimulation, McKinsey, in partnership with the International Association for the Engineering Modelling, Analysis and Simulation Community, or NAFEMS, has conducted a series of in-depth interviews withusers, providers, and industry experts.1This work builds upon previous rounds of collaborative researchbetween McKinsey and NAFEMS. In 2023, we looked at the impact of market and technology shifts on Our 2025 research enabled us to test multiple hypotheses about the speed, direction, and ultimatedestination of the AI-driven transformation of ES. From an initial list of around 60, we have identified 11topics that will shape the evolution of ES over the next five to ten years (see sidebar “Our methodology”).Taken together, these insights suggest a future in which ES becomes more intelligent, more automated, Our methodology The topics described in this article emerged from a series of 24 structured interviews with engineeringsimulation (ES) experts, conducted in early 2025. Interviewees came from across the ES value The selected interviewees were people in senior technical or managerial roles, including CTOs, heads of The interview questions covered a range of forward-looking topics. We asked interviewees to outline theirpersonal vision for the evolution of ES over the next decade. We asked them for their views on major trends The initial interview process generated more than 60 different hypotheses about the future of ES.Combining those insights allowed us to synthesize the results into 11 key topics that offer a consensus AI across the ES stack ES ecosystems can be viewed as a layered stack of interdependent components, and the hypothesesemerging from our research cover each layer in that stack (exhibit). At its base is