Measurement repeatability

A PSD measurement can look very nice but still not be reliable enough for quality control if repeatability is poor. Two sources of variation need distinguishing: variation created by the grinder and variation created by the measurement method itself. If sampling, dispersion, or data processing produces a different result each time, the operator won't know whether a change comes from the coffee or from the measurement. So measurement repeatability needs to be established before using PSD as a decision-making signal.

Factors that must be fixed

Factors needing to be fixed include sample mass, sampling method, measurement time, equipment, dispersion method, environmental conditions, and data-processing method. For sieving, opening size, shaking time, and sample loading need to stay consistent. For laser diffraction or similar methods, dispersion state and processing algorithm can also affect the result. The more complex a procedure, the more clearly its steps need to be documented so later measurements can be reproduced.

Number of repeats and the goal

There's no single number of repeats that applies to every measurement. If the goal is quickly assessing a process, the number of repeats needs to be enough to estimate baseline variation. If the goal is establishing control limits, more data is needed and the equipment's actual level of variation must be considered. What matters is not choosing a number of repeats just because it's convenient. You need to know how small a change you need to distinguish to determine whether the measurement design has enough power to detect it.

Statistical indicators

Basic statistical indicators like standard deviation and coefficient of variation can describe dispersion but aren't always sufficient. A repeatability interval can be useful for knowing how large a difference between two results from the same condition can be. If PSD data has multiple characteristics, you can track median, fine fraction, and distribution width separately. The processing method should match the control question rather than running many indicators and picking whichever looks “nicest.”

Standards for use in quality control

Research on ground-coffee consistency shows PSD relates to extraction behavior, while research on espresso fines shows the very small particle fraction can affect extraction kinetics. So a measurement that only repeats well at the median but overlooks the fine fraction may not be sufficient for the control goal. This also explains why you need to define in advance which variable actually matters for the brewing process being controlled.

A measurement should only become control data once repeatability has been demonstrated under real operating conditions and an acceptable limit has been defined in advance. After that, any change exceeding the limit becomes a signal worth investigating. This procedure helps avoid changing the machine or raw material just because of one unusual measurement. PSD has the most value when it's reliable, repeatable data tied to a specific decision, not just a nice-looking chart in a report.

You can test repeatability by running the same sample through the entire measurement procedure multiple times without changing conditions. Then, if the system requires it, test reproducibility when the operator, measurement day, or equipment changes. These two concepts aren't the same. A method with good repeatability within one session but that differs a lot between days may still not be suitable as long-term quality-control data.

Once baseline variation is known, control limits need to be set based on data, not chosen arbitrarily. If limits are too narrow, the system will generate many false alarms; if too wide, important changes may be missed. This is why PSD measurement needs to be assessed alongside the control goal. A method “good enough” always means good enough for a specific decision.

Besides repeatability, the method's stability over time needs to be checked. Measuring equipment can be recalibrated, software updated, or sample-handling changed. If historical data is used to set limits, these changes need to be recorded. A method with the same name but a procedure that has actually changed shouldn't be treated as one fully homogeneous data series.

When PSD is used for control, reporting results should keep both measurement conditions and method version. If only a single number is stored, it becomes very hard to investigate when there's a deviation. Storing distribution data or raw data, when feasible, allows rechecking the fine fraction and tail later. This is the foundation of traceability in a measurement system.

Besides repeatability within a single measurement, you can check the method's stability across days or operators. If results change significantly when the operator changes, the operating procedure needs further standardization. Once the method is stable enough, the data can then be used to build control limits. The ultimate goal isn't producing a nice number but creating a measurement capable of correctly detecting the change that matters.

References

• Khongphinitbunjong et al. (2025), Journal of Food Process Engineering 48, e70129, DOI 10.1111/jfpe.70129.
• Smrke et al. (2024), The role of fines in espresso extraction dynamics, Scientific Reports 14, 5612, DOI 10.1038/s41598-024-55831-x.
• Huang et al. (2025), Sensors and Materials 37, 4677–4689, DOI 10.18494/SAM5901.
• Wear Monitoring of Coffee Grind-On-Demand Grinders Using Precision Sieving and Laser Diffraction.
• A measurement method is considered suitable when its error is smaller than the level of change the system needs to detect. If production only needs to catch large changes, a simple method may be enough. If very tight control is needed, investment in a lower-variation method is required. So the question shouldn't be whether a measurement is absolutely accurate; it should be whether it can sufficiently distinguish the states that need distinguishing.