System and Method for Estimating Coffee Grinder Burr Wear from Acoustic, Motor Current, Grind Time, and Extraction Telemetry

Abstract

Disclosed is a system and method that estimates how worn a coffee grinder's burrs are without opening the grinder, sieving the grounds, or waiting for the coffee to turn bad. The burrs are the cutting teeth that fracture roasted beans into particles, and they dull slowly, over months in a home and over hundreds of kilograms in a cafe. Nobody watches them dull. The first symptom is usually a shot that will not dial in: the barista grinds finer, the shot still runs erratically, the cup tastes sour and bitter at the same time, and the burrs get blamed only after the beans, the machine, and the barista have all been blamed first. This disclosure replaces that guessing with a wear index built from signals the grinding and brewing process already produces. It reads grind time per gram at a fixed setting, motor current and torque ripple during grinding, the sound of the grind itself, how far the dial setting has drifted to hold a fixed recipe, and how shot time and late-shot flow behave at that recipe. No single channel is trusted alone, because every channel has a confound that mimics wear. The channels are fused, corrected for bean, dose, and setting, and separated from the three great mimics: new-bean seasoning, burr misalignment, and a dirty or retention-clogged grind path. The output is a wear score on a fixed scale, an estimated remaining throughput in kilograms before the score crosses a replacement threshold, and a plain statement of which channel drove the verdict. A cafe gets a burr order before the busy weekend, not after it. A home user gets a date, not a shrug.

Field of the Invention

This invention relates to predictive maintenance of food preparation equipment, specifically to estimating cutting-tool wear in coffee grinders from acoustic, electrical, timing, and brewing telemetry, and to separating true burr wear from bean variation, burr seasoning, burr misalignment, and grind-path retention, for replacement scheduling in homes, cafes, and grinder fleets.

Background

A burr grinder does not crush beans the way a blade grinder does. It shears them between two toothed surfaces, flat rings or a cone inside a ring, whose gap sets the target particle size. Sharp teeth cut. Dull teeth tear. That one-word difference is the whole story of burr wear, because tearing makes two kinds of wrong particle at once: fines, the microscopic dust that over-extracts into bitterness and clogs a filter bed, and boulders, the oversized chunks that under-extract into sourness. A worn burr set therefore does not simply grind coarser. It grinds wider. The distribution stretches at both ends, and the cup carries both defects in the same sip.

The published evidence for that widening is consistent, even though the numbers come from different labs and different burrs. In a worn-burr simulation measured with a particle analyzer, worn burrs at identical gap and speed produced nearly twice as many fines as new burrs, with a wider and less stable distribution overall, and significant wear appeared after roughly 1,500 kilograms of throughput in commercial use (LeBrew worn burr guide). Scott Rao's account, summarized in a 2026 alignment guide, agrees on the mechanism and adds the second tail: as burrs dull, they make both more fines and more boulders, and the net result is slower flow and less extraction (JayArr burr alignment guide). The same guide reports maker burr-life ratings that differ by more than a factor of two, 300 kilograms for one home grinder's burrs and more than 750 kilograms for another's, which is itself a finding. A calendar or a kilogram counter cannot schedule replacement across burr models. The wear has to be measured on the grinder in front of you.

Three mimics make that measurement hard, and each one has fooled experienced baristas. The first is seasoning. New burrs change during their first kilograms of use, grinding coarser at first and then more evenly, with one maker recommending about 10 pounds (4.5 kilograms) to stabilize and one measured case taking 24 pounds to settle. Any wear detector that treats the first week as wear will condemn a healthy new burr set. The second is misalignment. Burrs that are not parallel, concentric, or seated flat produce a wider distribution, session-to-session inconsistency, and channeling at fine settings, exactly like wear, and Barista Hustle's 2017 account puts the sensitivity at about 0.05 millimeters of misalignment. Replacing burrs in a tilted carrier changes nothing, which is why misalignment must be tested separately rather than folded into a wear score. The third is the grind path itself: retained grounds, static cling, and a dirty chute change dose and flow without touching the burr edge at all, and cleaning resets them. A system that cannot tell these four states apart, wear, seasoning, misalignment, and retention, will be right by accident and wrong at exactly the moments a replacement decision costs money.

Existing practice does not measure wear at all. Home users replace burrs every one to two years on advice, cafes replace them on a throughput schedule from the maker, and everyone else waits for the coffee to complain. Sieve analysis measures particle distribution directly but needs lab equipment, a sample protocol, and a barista willing to do it. Visual inspection needs disassembly and a trained eye for edge rounding and coating loss. Motor overload trips and rising noise are late warnings, not early ones. Meanwhile the signals that reveal wear early are already being generated every morning: the grinder takes longer to produce a dose, its motor draws more current doing it, it sounds different doing it, the dial creeps finer to hold the same shot, and the shot itself gets less repeatable. What the art lacks is a method that fuses those signals, corrects them for bean and recipe, and separates wear from its three mimics with a stated uncertainty. That is what this disclosure provides.

Detailed Description

1. Inputs and System Architecture

The preferred embodiment is phone-only and installs no hardware in the grinder. A smartphone microphone records the grind, the phone's clock times the grind and the shot, a kitchen scale weighs the dose, and the user logs the dial setting, the bean, and the shot result (dose in, beverage weight out, shot time). A second embodiment adds a smart plug or an inline current sensor on the grinder's mains lead, which reports RMS current, power, and, where sampling allows, torque ripple during the grind. A third embodiment is grinder-embedded: a current sensor, a microphone, and a rotation sensor inside the grinder, with throughput counted in kilograms by the grinder itself. A fourth embodiment is a fleet aggregator for a cafe group or a roaster: anonymized wear curves per burr model across many grinders, used to set model-specific replacement thresholds and to spot a bad batch of burrs before any single site can see the pattern. In all embodiments only derived features leave the premises: grind times, current summaries, acoustic summaries, and shot statistics, never audio recordings of the room.

Every measurement is tagged with the variables that move it. Dose weight in grams, dial setting in clicks or microns where the grinder reports them, bean identity and roast level, grinder temperature class (cold start versus warmed up), and cleaning state (days since the burr chamber and chute were last cleaned). A measurement without its tags is not used for wear. It is logged as untagged and excluded from the index, because an untagged measurement is how a bean change becomes a false wear alarm.

2. Baseline Learning and Seasoning Exclusion

Before any wear verdict, the system learns the grinder's healthy behavior. The baseline is taken after seasoning, not at unboxing. Seasoning is detected as an early transient: grind time per gram and shot time drift in one direction over the first kilograms and then stabilize. The system fits that transient and discards it, or, where the grinder is already seasoned when monitoring starts, it takes the baseline from the first two stable weeks at a reference recipe and flags the baseline as "seasoning unknown" with a wider uncertainty band until stability is demonstrated. A reference recipe is fixed for baseline purposes: one dose, one dial setting, one bean or one roast class, and one brewing target, for example 18 grams in and 36 grams out for espresso. Other recipes are still logged, but they are normalized to the reference before they enter the index (section 5).

The baseline stores five fingerprints per reference recipe: grind time per gram, RMS current during the loaded portion of the grind, acoustic spectral centroid and high-frequency band share, dial setting, and shot time mean and standard deviation over at least ten shots. These are the healthy values. Wear is defined strictly as drift away from this baseline under matched conditions, never as an absolute number, because two healthy grinders of different models differ more from each other than a new burr set differs from a worn one of the same model.

3. The Five Wear Channels

Grind-time drift is the simplest channel and the most easily misread. As teeth dull, the grinder must work longer to fracture the same dose, so seconds per gram rise at a fixed setting. The system measures grind time from first bean fracture to motor stop, not from button press, because hopper feed delay is not burr work. It corrects for dose, because a larger dose takes longer for reasons that have nothing to do with wear, and for roast class, because dense light roasts resist fracture more than dark roasts. A rise of 15 percent or more at matched conditions, sustained over at least five grinds, is a wear vote, not a verdict. One slow grind is a bean. Five slow grinds are a trend.

Motor current is the channel that sees effort directly. Dull teeth tear rather than cut, so the motor draws more RMS current for the same dose and shows larger torque ripple at the rotation frequency as beans jam and release. The system reports current only from the loaded portion of the grind, excluding the no-load spin-up and the empty tail, and it normalizes by dose and by grinder temperature class, since a cold motor draws differently from a warm one. A sustained RMS rise with rising ripple is a wear vote. A current spike on one grind is a stone, a foreign object, or a jam, and it is flagged as an incident, not as wear.

Acoustics is the channel that hears the cutting mechanism itself. Sharp teeth produce short, high-frequency fracture transients as beans crack cleanly. Dull teeth produce longer, lower-frequency crushing events, more rumble, less snap. The system summarizes each grind with three features: spectral centroid, the energy-weighted mean frequency; high-frequency band share, the fraction of energy above roughly 4 kilohertz; and burst kurtosis, how spiky versus smeared the fracture events are in time. Wear moves all three in the same direction: centroid down, high-frequency share down, kurtosis down. The microphone need not be calibrated in absolute decibels, because the features are ratios and shapes, not levels, and the baseline is taken on the same phone in the same kitchen. A phone change re-baselines the acoustic channel only, not the whole index.

Dial drift is the channel the barista already feels but never logs. To hold a fixed shot time as burrs dull, the barista grinds finer, step by step, week by week. The system logs the dial setting required to hit the reference shot target and reports its cumulative movement from baseline in clicks or microns. Dial drift is powerful evidence because it is a human-in-the-loop measurement: the barista is continuously re-calibrating the grinder against the cup, and the drift of that calibration is a wear signal even when no instrument is attached. It is also the channel most contaminated by bean changes, since a new bean dials in at a different setting for innocent reasons, so dial drift is only counted when the bean is held constant or when the drift persists across a bean change.

Extraction drift is the channel that closes the loop in the cup. Worn burrs widen the distribution, and a wider distribution shows up in the shot before it shows up in taste. Shot time at a fixed dose and target weight becomes more variable from shot to shot, and late-shot flow falls relative to early-shot flow as fines migrate and clog the puck. The system tracks shot time standard deviation over a rolling window of ten shots and the late-to-early flow ratio where the machine reports flow or where beverage weight over time is logged. Rising variance with falling late flow is a wear vote. Rising variance alone, with stable late flow, points at puck preparation or machine temperature, not at burrs, and the system says so.

4. The Wear Index and Replacement Threshold

The five channels are combined into a single wear index W on a 0 to 1 scale, where 0 is the seasoned baseline and 1 is a burr set that cannot hold the reference recipe at any dial setting. Each channel contributes a normalized drift, its current value minus its baseline divided by a channel scale set from the fleet distribution for that burr model or, where no fleet data exists, from a default scale published with the system. The channels are weighted: grind time and dial drift carry the most weight because they are the most robust, acoustics and current carry medium weight, and extraction variance carries weight only when late flow also moves, since variance alone is noisy. The weights and the scales are shown to the user with the score. A black-box score that cannot be audited will not be trusted at the moment it recommends spending money, and it should not be.

Two thresholds govern action. At W of 0.5, the system issues a watch notice: wear is measurable, the burrs are not yet failing, and the next order window is named. At W of 0.75, the system issues a replace notice: the burrs are measurably degrading the cup, remaining throughput is estimated from the recent slope of W against kilograms ground, and a swap is scheduled. The remaining-throughput estimate is reported as a range, never a point, with the range widening when the slope is estimated from few kilograms or across a bean change. A threshold crossed on a single channel alone does not trigger a replace notice. At least three channels must agree, or two channels plus a dial drift of at least half the grinder's usable fine range, because a single-channel verdict is how a new bean becomes a new burr order.

5. Correction for Bean, Dose, Setting, and Temperature

Correction is what separates this disclosure from a timer with a chart. Bean density and roast level shift grind time, current, and acoustics together, so the system maintains a bean correction table: for each roast class, a multiplicative factor for grind time and current learned from the fleet or from the grinder's own history when the same bean returns. Dose is corrected by seconds per gram rather than total seconds, and by current per gram over the loaded portion. Dial setting is corrected by expressing every grind at its setting's expected particle size, using the grinder's setting-to-gap map where the maker publishes one and a learned map otherwise, so that a grind at a finer setting is not mistaken for a slower grinder. Temperature is handled by class, cold start versus warmed up, with separate baselines where the grinder shows a consistent warm-up effect. Humidity and static are logged as confounds for retention (section 7) rather than corrected away, because their effect runs through dose loss, not through cutting.

6. Separating Wear from Misalignment

Misalignment is tested, not assumed away. The system runs two checks that wear cannot fake. The first is a no-load chirp test: with the burrs closed to near contact and the grinder running empty, a well-aligned set produces an even, deep chirp across the rotation, while a tilted set produces a once-per-revolution chirp that comes and goes. The phone microphone records the test, and the system measures the once-per-revolution modulation depth. The second is a loaded modulation check: during a normal grind, a tilted carrier produces amplitude modulation at the rotation frequency in both current and sound, because the gap varies as the burr turns. Wear produces no such periodic modulation. It produces a slow, uniform change across the whole rotation. If either check shows strong once-per-revolution modulation, the system reports suspected misalignment, suspends the wear verdict, and recommends a seating and cleaning check before any burr order. If the modulation persists after reseating, the carrier itself is suspect, and the report says that in plain words rather than selling burrs into a tilted carrier.

7. Separating Wear from Retention, Fouling, and Damage

Retention and fouling are the reversible mimics. Grounds left in the chute and chamber go stale, absorb moisture, and change the next dose, while oily buildup changes flow without dulling any edge. The system handles them by state, not by statistics: after a cleaning, the dose, grind time, and shot variance are re-measured and compared against the pre-cleaning values, and any improvement is attributed to cleaning and removed from the wear trend. A step improvement after cleaning is evidence against wear, and the system lowers W accordingly rather than letting a dirty grinder accumulate a false wear debt. Damage is handled as an incident class of its own: a sudden current spike, a sudden acoustic change, or a sudden grind-time jump on a single grind, with no trend before it, is flagged as a possible foreign object or chipped tooth, and the user is told to inspect before grinding again. Wear is slow. Damage is sudden. The time course is the diagnosis.

8. Worked Example: The Cafe Grinder That Stopped Holding Its Recipe

Consider a cafe grinder with a flat burr set in the 1,000 to 1,500 kilogram maker class, seasoned and baselined at installation. At baseline, an 18 gram dose grinds in 5.9 seconds at setting 12, draws its reference RMS current, and produces a 36 gram shot in 27 seconds with a shot-to-shot standard deviation of 0.9 seconds. The grinder serves a stable menu, one bean at a time, and throughput is logged by dose count. At 620 kilograms cumulative, the picture has changed on every channel at once. The same 18 gram dose takes 7.4 seconds, a rise of 25 percent. RMS current over the loaded portion is up 14 percent with visibly larger torque ripple. The acoustic centroid is down 22 percent and the high-frequency band share has fallen by a third, the sound of tearing replacing the sound of cutting. The dial setting needed to hold the 36 gram, 27 second target has moved from 12 to 8, four steps finer, and it keeps moving. Shot time standard deviation has grown from 0.9 seconds to 2.4 seconds, and late-shot flow is down 18 percent relative to early-shot flow. The chirp test shows no once-per-revolution modulation, cleaning produced no step improvement, and the trend survived a bean change, so the three mimics are excluded. The fused wear index reads 0.78, past the 0.75 replace threshold. That it crosses the threshold at 620 kilograms, ahead of its class rating, is the point of measuring wear rather than counting kilograms: a rating describes a class average, and this unit wore faster than average, with a remaining-throughput estimate of roughly 40 to 120 kilograms at the recent wear slope. The cafe orders burrs that week and swaps them on the next quiet morning, before the busy weekend, instead of discovering the problem through a weekend of erratic shots. That is the entire value proposition in one paragraph: a planned swap instead of a lost weekend.

9. Figures Description

• Figure 1: System block diagram showing the grinder, the phone or embedded sensors, the scale, the espresso machine or brewer, the bean and recipe tagging module, the five wear channels, the correction module, the mimic separation module, and the wear index and scheduling output.
• Figure 2: Particle distribution curves for a sharp burr set and a worn burr set at the same gap, showing the worn curve widened at both ends with more fines and more boulders.
• Figure 3: Wear channel trends against cumulative throughput: grind time per gram rising, dial setting drifting finer, acoustic centroid falling, current rising, and shot time variance rising, with the seasoning transient marked and excluded at the left edge.
• Figure 4: Confound separation diagram: seasoning as an early transient, misalignment as once-per-revolution modulation, retention as a step improvement after cleaning, damage as a sudden jump, and wear as a slow monotonic drift.
• Figure 5: Worked example dashboard: baseline versus current values for all five channels, the fused wear index at 0.78 against the 0.5 watch and 0.75 replace thresholds, and the remaining-throughput range.
• Figure 6: Fleet view for a cafe group: wear curves for many grinders of the same burr model plotted against throughput, showing model-specific wear rates and an outlier grinder flagged for misalignment rather than wear.

Claims

1. A system for estimating coffee grinder burr wear, comprising: a sensor set reporting at least one of grind audio, motor current, grind time, and brewing telemetry; and a processor configured to compute a baseline fingerprint for a reference recipe after burr seasoning, measure drift from the baseline under matched dose, setting, and bean conditions, and convert the drift into a burr wear index.
2. The system of claim 1, wherein the processor computes grind time per gram at a fixed dial setting and corrects the grind time for dose and roast class, and wherein a sustained rise in corrected grind time per gram contributes to the wear index.
3. The system of claim 1, wherein the processor measures RMS motor current and torque ripple over the loaded portion of a grind, excluding spin-up and empty tail intervals, and wherein a sustained rise in loaded RMS current with rising torque ripple contributes to the wear index.
4. The system of claim 1, wherein the processor computes acoustic features of a grind comprising spectral centroid, high-frequency band share, and burst kurtosis, and wherein a falling spectral centroid with falling high-frequency share and falling kurtosis contributes to the wear index as a crushing signature distinct from clean cutting.
5. The system of claim 1, wherein the processor logs the dial setting required to hold a fixed brewing target and reports cumulative dial drift from baseline in clicks or microns, and wherein dial drift persisting across a bean change contributes to the wear index.
6. The system of claim 1, wherein the processor tracks brewing telemetry at a fixed recipe comprising shot time mean and standard deviation and late-shot flow relative to early-shot flow, and wherein rising shot time variance combined with falling late-shot flow contributes to the wear index.
7. The system of claim 1, wherein the processor combines at least three wear channels into a single wear index on a fixed scale, issues a watch notice at a first threshold and a replace notice at a second higher threshold, and estimates remaining throughput in kilograms from the recent slope of the wear index against cumulative throughput.
8. The system of claim 1, wherein the processor detects burr seasoning as an early transient in grind time or shot time over the first kilograms of use, excludes the transient from the wear trend, and establishes the baseline only after stabilization.
9. The system of claim 1, wherein the processor separates burr misalignment from burr wear by measuring once-per-revolution modulation in a no-load chirp test or in loaded current or audio, and wherein strong once-per-revolution modulation suspends the wear verdict and reports suspected misalignment.
10. The system of claim 1, wherein the processor separates retention and fouling from burr wear by comparing measurements before and after cleaning, attributing step improvements after cleaning to the grind path rather than to the burr edge, and adjusting the wear index accordingly.
11. The system of claim 1, wherein the processor classifies a sudden single-grind change in current, acoustics, or grind time with no preceding trend as a damage incident distinct from wear, and reports an inspection prompt rather than a wear verdict.
12. The system of claim 1, implemented as a phone-only system using a smartphone microphone, clock, and logged dose and shot data with no grinder hardware modification, or as a grinder-embedded system, or as a fleet aggregator producing model-specific wear curves and replacement thresholds across many grinders.

Prior Art References

1. LeBrew, "How Worn Coffee Burrs Impact Coffee Grinding and When to Replace Them": worn burr simulation at identical gap and speed produced nearly twice as many fines with a wider, less stable distribution; significant wear after approximately 1,500 kilograms; commercial replacement often after 1,000 to 1,500 kilograms
2. JayArr, "How to Align Coffee Grinder Burrs: The Marker Test, and When Not To": Barista Hustle 2017 alignment sensitivity of about 0.05 millimeters; Rao on dull burrs producing more fines and more boulders with slower flow and less extraction; maker burr-life ratings of 300 kilograms and more than 750 kilograms for different home grinders; seasoning of about 10 pounds to stabilize in one maker case and 24 pounds in one measured case
3. Hendon, Colonna-Dashwood, and co-authors, "Systematically Improving Espresso: Insights from Mathematical Modeling and Experiment," Matter, 2020: fines and uneven beds produce channeling, with parts of the puck over-extracted and parts under-extracted
4. Scott Rao, Espresso Extraction: Measurement and Analysis: burr condition, distribution width, and extraction uniformity as determinants of shot consistency
5. Standard motor current signature analysis: changes in mechanical load appear in RMS current and in torque ripple at the rotation frequency, and loaded intervals must be separated from no-load intervals
6. Standard machining tool-wear monitoring: as a cutting edge dulls, cutting gives way to rubbing and ploughing, acoustic emission shifts to lower frequencies, and cutting force and motor load rise for the same material removal rate
7. Standard espresso brewing practice: a fixed dose-in, beverage-weight-out, and shot-time recipe is the control against which grinder changes are judged, and shot-to-shot repeatability is a primary quality measure
8. Barista Hustle, "A Wide Thought Distribution About Grinding," 2017, via the JayArr guide above: grinding burrs need to be parallel and concentrically and radially aligned to work properly

Implementation Notes

Build notes for whoever implements this. Ship the baseline before the score. A wear index without a seasoned baseline is a random number with a decimal point, and the first user who gets a replace notice on a week-old grinder will never trust the second notice. Require the seasoning check, show the baseline values next to every current value, and refuse a verdict when the bean, dose, or setting tags are missing, reporting "insufficient matched data" instead of a number. Do not ship a single-channel version. Grind time alone will blame every new bean, acoustics alone will blame every new kitchen, and shot variance alone will blame every new barista. The three-channel agreement rule is not a nicety. It is the product. Calibrate the channel scales per burr model from fleet data as soon as the fleet exists, because a 300 kilogram burr and a 1,500 kilogram burr wear on different clocks and a single global scale will either nag one owner or neglect the other. Present the remaining-throughput estimate as a range with its slope, and say which channel drove the verdict in the notice itself, so the owner can check it with their own eyes: the dial really has moved four steps, the grind really does take longer, the shot really is less repeatable. On a damage incident, tell the user to stop and inspect before the next grind. A chipped tooth or a foreign object is not a wear trend, and grinding through it turns a burr order into a grinder repair. Finally, never present the wear index as a taste verdict. The system measures cutting condition, not flavor, and the notice should say that the burrs are measurably worn and the recipe no longer holds, leaving the question of whether the cup still pleases its owner exactly where it belongs, with the owner.