结论：任何测量结果都有不确定度，紧贴限值的结论要谨慎

检测报告上的数值看起来是确定的，但任何测量都有不确定度 ——同一个样品、同一个方法、同一个实验室，重复测量的结果也会有差异。

理解这一点有实际意义：

对判定的影响。 测量值紧贴限值时，考虑不确定度后可能落在限值两侧，判定就不那么确定。

对比较的影响。 两个数值的差异如果小于不确定度，这个差异可能没有实际意义。

对设计的影响。 设计时应当留出超过不确定度的余量。

对采购的影响。 两个产品的数据相差很小时，不足以说明性能差异。

不确定度的主要来源

来源 说明
测量设备 设备本身的精度与校准
方法 方法固有的分散性
样品 样品之间的差异
环境 温湿度等条件的波动
操作 操作者之间和操作次数之间的差异
试验装置 夹具、加载方式的差异
读数与记录 读数的分辨率和判读

在轮椅检测中，样品差异和试验装置的差异往往是较大的来源，因为轮椅是组装产品，各部件的公差累积起来差异不小；而试验装置中夹具的位置和加载点的细微差异会影响结果。

各类测量的不确定度特点

尺寸测量。 相对较小，主要来自测量工具和测量点的选择。

质量测量。 较小，来自称量设备。

力的测量。 中等，来自传感器精度、加载方式、以及摩擦。

稳定性（倾翻角）。 较大，受试验台水平度、样品摆放位置、加载点位置影响。

续航。 较大，受载重、路面、温度、电池状态、驾驶方式影响。

制动距离。 较大，受速度控制、路面、反应时间影响。

疲劳寿命。 很大，材料和制造的分散性使疲劳寿命的分散往往达到数倍。

「疲劳寿命的分散」需要特别注意。 同一批产品的疲劳寿命可能相差很大，所以单个样品的疲劳数据代表性有限，需要多个样品。

在判定中的作用

判定规则。 是按测量值直接判定，还是考虑不确定度后判定，这应当在方案中规定。

紧贴限值的情形。 测量值与限值的差距小于不确定度时，判定需要说明。

报告中的表述。 是否给出不确定度、以什么形式给出。

风险的分担。 不同的判定规则意味着生产方和使用方承担不同的风险。

「判定规则的选择」有实际影响。 严格的规则（测量值加不确定度仍要在限值内）对使用方更有利；宽松的规则对生产方更有利。规则应当事先明确而不是事后争论。

对设计的意义

留出余量。 设计的目标值与限值之间应当留出超过不确定度的距离。

关注分散大的项目。 疲劳、续航这类分散大的项目，余量应当更大。

批次差异的叠加。 测量不确定度加上批次差异，实际的分散更大。

内控限值。 企业内部的控制限值应当严于标准限值。

「内控限值严于标准限值」是合理的做法，因为标准限值是合格的底线，而实际生产有波动，卡着底线做必然出现不合格。

对采购比较的意义

小差异不说明问题。 两个产品的某个数据相差百分之几，可能在不确定度范围内。

关注明显的差异。 数量级或明显幅度的差异才有说服力。

关注余量而非合格与否。 距限值的距离比「合格」两个字有信息量。

关注数据的条件。 条件不同的数据不可比。

关注样本量。 单个样品的数据代表性有限。

「关注余量」是有经验的采购人员的做法 ——两个产品都合格，但一个的数据接近限值、另一个有明显余量，实际的可靠性差别是存在的。

降低不确定度的做法

设备的校准。 定期校准并使用在有效期内的设备。

方法的规范。 严格按方法执行，减少操作差异。

条件的控制。 控制环境条件。

重复测量。 多次测量取平均可以降低随机部分。

样品量。 更多样品能更好地反映真实分布。

人员的培训。 减少操作者之间的差异。

「重复测量」对随机误差有效，但对系统误差无效 ——系统误差需要通过校准和方法改进来消除。

不确定度与产品开发

研发阶段理解不确定度有实际价值：

方案比选时。 两个方案的数据差异小于不确定度时，不足以判断优劣。

改进效果的判断。 改进后数据的变化要超过不确定度才说明有效。

目标值的设定。 设计目标应当高于标准限值一定幅度。

验证的样本量。 分散大的项目需要更多样本。

「改进效果的判断」在实务中容易出偏差 ——改了之后测一次，数据好了一点就认为改进有效，而这个差异可能在不确定度范围内。

报告中的呈现

不确定度在报告中的呈现方式有几种：给出扩展不确定度；说明测量条件和样本量；给出多个样品的数据范围；以及说明判定规则。

「给出多个样品的数据」是直观的做法，读者能直接看到分散的程度。

常见的误解

「合格就是好」。 忽略了余量的差别。

「数据高就是好」。 忽略了条件和不确定度。

「测一次就够」。 忽略了分散性。

「不同报告可以直接比」。 忽略了条件和方法的差异。

这四项误解在采购和研发中都常见，而理解不确定度能避免它们。

与报告使用的关系

理解不确定度之后，看报告的方式会不同：关注实测值与限值的距离；关注样本量；关注测试条件；以及对小差异保持谨慎。

这几点加起来，就是从「看合格与否」转向「看数据的实际含义」，而后者对采购和研发都更有价值。

我们的做法

我们在报告中给出实测值，对于不确定度影响较大的项目，会说明测量条件和样本量，因为这些信息是读者判断数据可靠程度的依据。

对于紧贴限值的结果，我们会在沟通中说明情况，建议委托方考虑是否需要增加样本量或者改进产品留出余量 ——因为紧贴限值的产品在批量生产中出现不合格的概率不低。

疲劳类项目我们建议多个样品，因为疲劳寿命的分散大，单个样品的数据说服力有限。

如果你在解读检测数据时对不确定度有疑问，可以直接联系：132 4819 8029。检测能力见服务介绍，产品分类见手动轮椅检测与电动轮椅检测，案例见案例。

English version

Conclusion. Figures in a test report look definite, yet every measurement carries uncertainty: the same sample, the same method and the same laboratory produce slightly different results on repetition. Understanding this has practical consequences. For conformity decisions, a value close to the limit may fall either side once uncertainty is considered, making the verdict less definite. For comparison, a difference between two values smaller than the uncertainty may carry no practical meaning. For design, margin should exceed the uncertainty. And for procurement, small differences between products do not demonstrate a performance difference.

Principal sources of uncertainty. Measuring equipment contributes through its own accuracy and calibration. Method contributes through inherent scatter. Samples contribute through variation between units. Environment contributes through fluctuation in temperature, humidity and similar conditions. Operation contributes through differences between operators and between repetitions. Test fixtures contribute through variation in fixturing and loading. And reading and recording contribute through resolution and interpretation. In wheelchair testing, sample variation and fixture variation are often the larger contributors: wheelchairs are assemblies whose component tolerances accumulate, and small differences in fixture position and loading point affect results.

Uncertainty by measurement type. Dimensional measurement is comparatively small, arising mainly from instruments and choice of measuring points. Mass measurement is small, arising from the weighing equipment. Force measurement is moderate, arising from transducer accuracy, loading method and friction. Stability, as a tipping angle, is larger, affected by platform levelness, sample placement and loading point. Range is larger, affected by load, surface, temperature, battery state and driving style. Braking distance is larger, affected by speed control, surface and reaction time. And fatigue life is very large, since material and manufacturing scatter often give fatigue lives differing several-fold. Fatigue scatter deserves particular attention: fatigue lives within one batch can differ greatly, so data from a single sample are of limited representativeness and several samples are needed.

Role in conformity decisions. The decision rule must state whether conformity is judged on the measured value directly or after considering uncertainty, and this belongs in the test plan. Values close to the limit require explanation where the gap is smaller than the uncertainty. Report presentation must decide whether and how uncertainty is stated. And risk sharing differs between rules, allocating risk differently between producer and user. The choice of rule has practical effect: a strict rule, requiring the measured value plus uncertainty to remain within the limit, favours the user, while a lenient rule favours the producer. Settle the rule in advance rather than arguing afterwards.

Implications for design. Leave margin, with the design target set further from the limit than the uncertainty. Attend to high-scatter items, since fatigue and range deserve greater margin. Consider accumulation, since measurement uncertainty plus batch variation gives a wider real spread. And set internal control limits tighter than standard limits. Tighter internal limits are sound practice: the standard limit is the threshold of conformity, production varies, and designing to the threshold guarantees eventual nonconformity.

Implications for procurement comparison. Small differences prove nothing, since a few per cent between two products may lie within the uncertainty. Attend to obvious differences, since only substantial differences are persuasive. Attend to margin rather than conformity, since distance from the limit carries more information than the word conforms. Attend to conditions, since data under different conditions are not comparable. And attend to sample size, since single-sample data are of limited representativeness. Attending to margin is what experienced buyers do: two products both conform, but one sits close to the limit while the other has clear margin, and their real reliability does differ.

Reducing uncertainty. Calibrate equipment and use it within its calibration period. Follow methods rigorously to reduce operational variation. Control environmental conditions. Repeat measurements and average, which reduces the random component. Increase sample numbers to reflect the true distribution better. And train personnel to reduce differences between operators. Repetition addresses random error but not systematic error, which requires calibration and method improvement.

How we handle it. We report measured values and, for items where uncertainty matters most, state the measurement conditions and sample size, since those let the reader judge how far the data can be relied upon. Where a result sits close to the limit we say so in discussion and suggest that the client consider increasing sample numbers or improving the product to create margin, since products sitting at the limit have a considerable probability of nonconformity in production. For fatigue items we suggest several samples, since fatigue scatter is wide and single-sample data carry limited weight.

If you have questions about uncertainty when interpreting test data, get in touch: +86 132 4819 8029.