Defining PSD stability

A grinder can produce a very good PSD on day one but doesn't necessarily maintain that same distribution after many hours or many kilograms of coffee. So “particle-size stability” should be understood as a time-based data series. Indicators worth tracking include the distribution's median, the fine fraction, distribution width, and changes in the tail. A single average value usually isn't enough to detect a small but meaningful shift affecting the brewing process.

Sources that can shift PSD

Many mechanisms can cause PSD to shift. A worn grinding mechanism can change the cutting-zone geometry; alignment can change due to heat or mechanical stress; buildup in the exit path can change how material flows; and operating conditions can change between batches. Research on monitoring grinder wear using precision sieving and laser diffraction shows PSD can be used as a signal source for tracking equipment condition. However, that signal only has meaning when the measurement method is stable enough to distinguish equipment drift from measurement error.

Tracking over time

A good tracking design should choose a control sample or condition and repeat the grind at defined milestones, such as by the mass of coffee processed through the machine. There's no need to assume “every 100 kg the grinding mechanism must be replaced”; the maintenance threshold should come from the specific equipment's own data. The same amount of PSD change can be normal on one machine but significant on another. What matters is maintaining the same bean type, roast level, sample temperature, settings, mass, and measurement method when building the baseline.

Distinguishing equipment drift from fluctuation

Another challenge is distinguishing random fluctuation from equipment drift. With only two measurements, a small change can easily be over-interpreted. Control samples, repeated measurements, and an accepted range of variation help establish the natural baseline of fluctuation. When many consecutive points shift in the same direction, the evidence for equipment drift is stronger. Conversely, if data fluctuates around a stable level without a trend, it shouldn't yet be called wear.

PSD as a maintenance signal

PSD also needs to be read by structure, not just by a single number. A 2025 study on PSD modeling for grind consistency emphasizes the relationship between particle-size distribution and extraction quality. In practice, the fine fraction and distribution tail can change while the median stays almost unchanged. If the goal is controlling grounds repeatability, the entire distribution curve, or at least a data set sufficient to describe its shape, should be stored.

When PSD starts to become a maintenance signal, the decision shouldn't be based on one unusual measurement. It needs to be reconfirmed with repeated measurement, checking cleaning and alignment condition, ruling out raw-material change, before attributing it to wear. This is how PSD data shifts from a snapshot at one point in time into an equipment-monitoring system. The greatest value lies not in precisely predicting the exact day a grinding mechanism needs replacing, but in the ability to detect change early enough to investigate the cause before it becomes an obvious deviation in the brewing process.

Tracking PSD over time should also be tied to the mass of coffee ground rather than just calendar date. A machine may run for a few hours but process very different masses on two different days, and grinding-mechanism wear can relate directly to the amount of material that has passed through the grinding zone. When both operating hours and mass are recorded, the data can show whether a drift signal relates to time or to actual load. This is useful information when setting an equipment-inspection schedule.

If drift is detected, the next step isn't necessarily replacing the grinding mechanism right away. You can check alignment, cleanliness, raw-material source, and thermal conditions first. A PSD series collected under the same procedure will help assess the effectiveness of each intervention. If PSD returns to baseline after cleaning, the cause may lie in grounds buildup; if it remains off after alignment and the trend keeps increasing, the wear hypothesis becomes stronger.

When tracking a piece of equipment long-term, raw-material conditions need to be managed as part of the measurement system. If the bean source or roast level changes, PSD can shift without it being the grinding mechanism's fault. A control sample or reference raw material can help separate the two effects. When the control sample still gives a stable PSD but the product changes, the cause may lie in raw material; if both drift together, the equipment deserves investigation.

A maintenance threshold shouldn't be set purely from another study's data. Grinding-mechanism characteristics, material, geometry, and machine load differ and will create different aging curves. External research is useful for choosing the method and measured variables, while the decision threshold must be confirmed on the specific equipment itself. This is an important principle when moving from research into quality control.

When establishing a baseline, it's worth storing at least one data set detailed enough to revisit if the trend changes. An average chart alone doesn't show which part of the PSD has shifted. Distribution data, grinding conditions, and machine history help distinguish a real change from error. This is especially useful when several machines are being tracked at once and a unified way of reading the data is needed.

References

• Khongphinitbunjong et al. (2025), Particle Size Distribution Model for Optimizing Coffee Grind Consistency, Journal of Food Process Engineering 48, e70129, DOI 10.1111/jfpe.70129.
• Wear Monitoring of Coffee Grind-On-Demand Grinders Using Precision Sieving and Laser Diffraction.
• The Craft and Science of Coffee (2017), The Grind—Particles and Particularities.
• Huang et al. (2025), Optimization of Coffee Grinder Quality Characteristics under a Sensor-driven Framework, Sensors and Materials 37, 4677–4689, DOI 10.18494/SAM5901.