Beyond the Error: Unveiling Uncertainties in Pressure Sensor Measurements

In every sensor lies inherent uncertainty. It is crucial to distinguish between error, accuracy, and uncertainty. Error reflects the discrepancy between an exact and measured value, a difference often elusive due to the inaccessibility of the ‘exact’ value. Uncertainty denotes the range of plausible values attributed to a measurement. For accuracy, understanding the nature of uncertainty and defining the confidence interval is essential. Statistically, broader confidence intervals imply greater uncertainty. For instance, for a normal distribution, 68.3%, 95.4%, and 99.7% confidence intervals are typically represented by σ1, σ2, and σ3, respectively. Many datasheets cite accuracy in terms of average values, with a variation roughly equivalent to σ1. Calibrating sensors involves many components, including reference pressure sensors and temperature sensors, power supplies, interface cards, and acquisition equipment. Each has its inherent uncertainties which cumulatively influence the tested sensor. It is not known but most of the uncertainty (50% to 80%) is attributed to the test bench used to calibrate/test the sensor. That is what we will show in this paper. There are several methodologies to evaluate uncertainties. Primarily, two main methods can be distinguished: • Type A: This approach focuses on repetitive measurements of a single component. It is thorough and often yields more precise uncertainties specific to that element. However, its detailed nature can be time-consuming. • Type B: This method leans on datasheets provided by manufacturers. It is faster, but can lead to broader uncertainties, especially if based on generalizations drawn from large component batches. When manufacturer data is lacking or insufficient, resorting to Type A becomes necessary. Regarding pressure sensors, their operation is defined by a linear equation linking the applied pressure with the output and supply voltage. Using a model to represent this relationship inherently introduces another layer of uncertainty. It is worth noting that no model can perfectly match every data point, hence certain deviations will always exist. The objective is to determine the sensor’s transfer function using the least squares method, which provides a clear understanding of its performance under varying conditions.