How Closely Does a Digital Twin Match Reality?

A digital twin makes an ambitious promise: investigate a representation on a computer and learn something useful about its real-world counterpart. The word “twin” suggests a close match, but it leaves an essential question unanswered. Close enough for what? A model that helps explain how a machine works may need different evidence from one used to decide whether that machine can safely continue operating. The purpose of the model matters as much as its appearance.

数字孪生提出了一个雄心勃勃的承诺：研究计算机中的一个表征，就能了解与之对应的现实对象，并获得有用的信息。“孪生”一词暗示两者高度相似，却没有回答一个关键问题：对什么用途而言足够相似？一个帮助解释机器如何运作的模型，与一个用来判断机器能否继续安全运行的模型，可能需要不同的证据。模型的用途与其外观同样重要。

In its announcement published on 14 February 2025, the US National Institute of Standards and Technology, or NIST, introduced a report on digital twins, security and trust. Its broad account includes electronic representations of physical objects, such as buildings, and non-physical entities, such as processes. The report discusses existing modelling capabilities alongside a wider vision for the technology. It should therefore be read as an examination of capabilities and challenges, rather than proof that every proposed application already works.

美国国家标准与技术研究院（简称NIST）在2025年2月14日发布的公告中，介绍了一份关于数字孪生、安全与信任的报告。其广义描述包括建筑等物理对象的电子表征，也包括流程等非物理实体的电子表征。报告既讨论现有的建模能力，也展望这项技术更广泛的前景。因此，应把它视为对能力与挑战的考察，而不能把它当作每一种拟议应用都已有效运行的证明。

IBM’s educational explanation emphasises digital twins that use real-time data to reflect physical objects or systems. Such information allows a representation to follow a particular object as its condition changes. A simulation can then explore what might happen under different conditions. These activities are related but serve different purposes: monitoring asks what is happening, while prediction asks what could happen. An attractive display alone cannot establish that either answer is accurate.

IBM的科普说明强调利用实时数据来反映物理对象或系统的数字孪生。这些信息让表征能够随着某个具体对象的状态变化而变化。随后，模拟可以探索不同条件下可能发生的情况。这些活动彼此相关，却有不同的用途：监测询问正在发生什么，预测则询问可能发生什么。仅凭一个好看的显示界面，无法确定这两种问题的答案是否准确。

Consider a hypothetical building manager investigating an unusually warm meeting room. A digital twin might combine a model of the building with temperature measurements and information about its cooling equipment. The manager could simulate a change in operating hours before trying it in the building. This would provide a way to compare possible actions without immediately disturbing the occupants. It would not, by itself, demonstrate that the predicted temperature or energy use would occur.

以一个假设情境为例：一位楼宇管理人员正在调查一间异常温暖的会议室。数字孪生可以把建筑模型、温度测量值和制冷设备的信息结合起来。管理人员可以先模拟调整运行时段的效果，再在建筑中实际尝试。这样就能在不立即打扰使用者的情况下比较可能采取的措施。但它本身并不能证明，预测中的温度或能源消耗一定会出现。

The same example reveals several opportunities for error. A sensor might be poorly positioned, so its reading would describe one corner better than the whole room. Information about occupancy might be outdated. The model might assume that a door stays closed when people regularly leave it open. These are hypothetical problems, but they illustrate why collecting more data and understanding a system are different tasks. Measurements need context, and assumptions need to be made visible.

同一个例子也揭示了几种出错的可能。传感器的位置可能不合适，因此其读数更能反映某个角落，而不是整个房间。有关使用人数的信息可能已经过时。模型可能假定门始终关闭，而人们实际上经常把门敞开。这些都是假设问题，但它们说明，收集更多数据与理解一个系统是两项不同的任务。测量需要背景信息，假设需要清楚地呈现出来。

NIST’s report identifies time, environment and functional equivalence among its trust considerations. Put simply, users need evidence that the representation behaves sufficiently like the thing it represents. For our hypothetical room, that could mean comparing predicted temperatures with observations under the same conditions. Agreement during an ordinary working day would be useful evidence, yet it would leave questions about an exceptionally hot afternoon. Testing should reflect the decision being considered and the conditions in which it will be made.

NIST的报告把时间、环境和功能等效性列为信任方面的考量。简单来说，使用者需要证据，证明表征的行为与其所代表对象的行为足够相似。对于我们假设中的房间，这可能意味着将相同条件下的预测温度与观测结果进行比较。普通工作日中两者一致，会是有用的证据，但对于异常炎热的下午，仍然存在疑问。测试应当对应正在考虑的决策，以及作出该决策时的条件。

There is also a problem of shared meaning. NIST’s announcement stresses interoperable definitions, tools and standards. Suppose, again hypothetically, that two building systems exchange temperature values. Sending a number successfully would achieve little if one system interpreted it in degrees Celsius and the other in Fahrenheit. Useful exchange requires agreement about meaning as well as format. Standards can help different components work together, but they do not automatically make the underlying measurements correct.

此外还有共同理解信息含义的问题。NIST的公告强调可互操作的定义、工具和标准。再设想一个假设情况：两个建筑系统交换温度数值。如果一个系统按摄氏度理解，另一个系统按华氏度理解，那么成功传送一个数字也没有多少用处。有用的信息交换既需要格式一致，也需要对含义达成一致。标准能够帮助不同组件协同工作，但不会自动让底层测量变得正确。

A concrete environmental illustration appears in ESA’s video Digital Twin Hydrology. It describes combining satellite observations, ground observations and a model to derive information such as river discharge rate. ESA explains that this combined resource can support water management and help identify places and times of risk. Here, the relationship between observation and modelling is central: observations help improve the representation, while the model helps turn separate measurements into information about a larger system.

欧洲航天局的视频《水文数字孪生》提供了一个具体的环境领域示例。它介绍了如何结合卫星观测、地面观测与模型，推导河流流量等信息。欧洲航天局解释说，这种整合后的资源能够支持水资源管理，并帮助识别存在风险的地点和时段。在这里，观测与建模之间的关系至关重要：观测帮助改进表征，模型则帮助将分散的测量转化为有关更大系统的信息。

The most revealing demonstration of a digital twin may therefore be a comparison between its output and independent observations. Where do they agree? Where do they diverge, and why? For the hypothetical building manager, an unexplained mismatch would be a reason to investigate the sensor, the assumptions or the building itself. A useful twin makes those questions easier to pursue. Its credibility grows through evidence about a specific task, rather than through the resemblance implied by its name.

因此，最能说明数字孪生能力的展示，可能是将其输出与独立观测进行比较。两者在哪里一致？在哪里出现偏差，原因是什么？对于假设中的楼宇管理人员，无法解释的不一致会成为调查传感器、模型假设或建筑本身的理由。一个有用的孪生体能让这些问题更容易得到追查。它的可信度来自针对具体任务的证据，而不是名称所暗示的相似性。