Detecting Micro-Stops: External Sensors, Thresholds, and False Positives

Automatic micro-stop detection is the technical prerequisite for any serious program to reduce invisible production losses. Without exhaustive timestamped capture, no Pareto analysis is credible, no gain is measurable, and no project can be prioritized on numbers rather than intuition. This article is the technical guide to detection — sensor technology choice, threshold parameterization, false-positive and false-negative management, routine calibration.

It is written for methods engineers, instrumentation leads, Industry 4.0 project managers, and automation technicians who must specify, commission, and operate a micro-stop detection instrumentation. It assumes a basic familiarity with OEE concepts and the production-time hierarchy of the canonical OEE framework.

Why external sensors rather than PLC integration

The choice between direct PLC integration and wireless external sensors is the structuring technical decision of a micro-stop detection project. Both approaches are valid — relevance depends on the industrial context.

PLC integration via OPC-UA, Modbus, or proprietary protocol offers absolute precision: the PLC knows exactly when the machine is running and when it is stopped. The machine bit is the reference truth for that piece of equipment. Limitations are operational: PLC access locked by IT/OT, mandatory change control in pharma and food validated environments (typically six to twelve weeks of validation), frequent incompatibility with older PLCs (pre-2010) that lack standard network interfaces, and the need to adapt PLC code on certain architectures. Hidden integration cost in engineering hours is high.

External wireless sensors — vibration, current, optical, accelerometer — mount on the equipment surface with no PLC intervention, no code change, no change control, and no IT validation. Typical installation time is fifteen to thirty minutes per machine; total commissioning time on a five-station line is a half-day. Detection precision reaches two to five seconds depending on configuration, sufficient to capture 99 % of micro-stops. It is the approach with the best deployment-speed-to-precision ratio in 2026 on most industrial sites.

The case where PLC integration remains preferable is strict pharma manufacturing with ALCOA+ requirements, where the source data must come directly from the validated equipment. In all other cases — automotive, general food, aerospace, chemistry, metallurgy, mechanical engineering — external sensors are faster and equally exploitable.

The three families of external sensors for micro-stops

Sensor type selection depends on the physical signature of the machine state to be detected. Three families cover most industrial cases encountered on the field.

The current sensor measures the electrical current consumed by the machine via a clamp on the power supply. When the machine produces, current exceeds a threshold; when it stops, current drops below a low threshold. It is the most robust technology for equipment with stable electrical signatures: CNC machine tools, presses, motorized assembly lines. The typical false positive is the warm-up phase or auxiliary load (ventilation, hydraulics) that may maintain non-zero current while the machine is not producing — the low threshold must be calibrated above this residual current.

The vibration or accelerometer sensor detects the machine’s vibratory signature when running. When the machine produces, the accelerometer sees a characteristic vibration amplitude; when stopped, amplitude drops to a residual ambient level. It is the most versatile technology for rotating equipment: conveyors, mills, mixers, bottling lines. The typical false positive is external vibration transmitted through the floor or from neighboring machines — calibration consists of filtering out ambient noise.

The optical sensor detects piece or cycle passage via an infrared beam or photoelectric sensor. When pieces pass at regular cadence, the sensor counts; when flow stops, the counter stalls and a stop is detected after a delay expires. It is the most precise technology for packaging and wrapping lines where the product flow is directly measurable. The typical false positive is the extended pause between lots that can be interpreted as a stop — operator qualification resolves this case.

On most lines, two complementary sensors (typically current + optical, or vibration + optical) give coverage above 99 % of machine states, with robustness to individual sensor failures.

Threshold parameterization: the sensitivity/specificity trade-off

The detection threshold is the parameter that determines which event is counted as a micro-stop and which is ignored. Its calibration follows two complementary principles.

The signal threshold defines the amplitude above which the sensor considers the machine to be producing. It is expressed in the sensor’s unit: amperes for current, meters per second squared for vibration, Hz or pieces/minute for optical. Calibration rests on an experimental routine measurement over one to two weeks, during which the signature is recorded in production and in confirmed stop, then the threshold is set in the middle of the separating zone. On machines with stable signatures, this calibration holds for years without readjustment. On machines with multiple regimes (cadence changes by product), the threshold may require per-product adjustment.

The duration threshold defines from what duration a low signal counts as a stop. It filters out short transients that are not real stops — cycle time between two pieces, micro electrical fluctuations, occasional ambient vibrations. The typical value is two to five seconds depending on equipment type. The shorter the threshold, the more sensitive the detection (fewer false negatives) but the more false positives it generates; the longer it is, the more specific (fewer false positives) but the more it misses real micro-stops.

The sensitivity/specificity trade-off is the central technical decision of parameterization. The practical recommendation proven on 450 sites is to start with a duration threshold of five seconds, then progressively reduce it to three or two seconds once the team has confidence in the calibration. Initial over-detection is less harmful than under-detection — a false positive is corrected at qualification, a false negative disappears forever.

Managing false positives: filtering rules and operator qualification

False positives — events detected as stops but not actually so — are inevitable in any automatic instrumentation. Their acceptable rate is 5 to 10 % of detected volume; beyond that, the field team loses confidence in the measurement and abandons qualification.

Three complementary mechanisms keep the false-positive rate under control.

Rule-based filtering eliminates patterns known to generate false positives: planned meal breaks, shift changes, inter-lot intervals. A rule “stop between 12:00 and 12:45 = meal break (not counted)” filters meal breaks automatically without operator intervention. The rule list enriches progressively over weeks.

Operator qualification is the ultimate safety net. Each detected stop is presented to the operator on the qualification terminal, who can either assign a stoppage cause or mark it as “not relevant” (false positive or planned non-productive event). Typical operator action time is under five seconds per event. Stable qualification rate sits between 85 and 95 % after four weeks of routine.

Automatic pattern learning is the third layer, used by the most advanced solutions. Operator qualification history progressively trains a model that pre-suggests the likely cause and identifies recurring false positives for automatic filtering. On sites that have accumulated six months of qualified data, operator time savings reach 40 to 60 % versus purely manual qualification.

Routine calibration: what must be checked periodically

A sensor calibrated once is not a reliable sensor indefinitely. Calibration maintenance is an operational cost to anticipate in any enduring micro-stop program. Three periodic checks structure the routine.

The monthly coherence check compares the operating time declared by the sensor with an independent measurement: piece counter from quality control, MES declaration, or simply visual observation over a full shift. A gap above 3 % between the two measurements signals either sensor drift, threshold drift, or a change in machine behavior. The cause is traced and the corrective action documented.

The semi-annual threshold recalibration repeats the initial experimental measurement. Production conditions evolve (new products, equipment aging, process parameter changes) and initially well-calibrated thresholds may become inadequate. Recalibration takes one to two hours per line and preserves long-term detection quality.

The annual instrumentation audit covers the hardware: physical state of sensors, wireless signal quality, battery autonomy, fixture integrity, hardware drift. On autonomous wireless sensors, typical battery life is three to five years depending on sampling frequency. The audit anticipates replacements and avoids data loss.

Total annual calibration routine cost for a five-station line is typically one day of methods-engineer time and one day of instrumentation-technician time. Low relative to the OEE gains generated.

Data architecture: from sensor to decision

The complete data architecture of a micro-stop program comprises four layers that must work together without rupture for the value chain to be exploitable.

The sensor layer captures the raw physical signal at high frequency (1 to 100 Hz depending on technology). This raw data is not fully stored — it is processed locally by the sensor electronics to produce a “machine stopped” or “machine producing” event with second-level timestamp.

The communication layer transmits events to a local gateway or directly to the cloud via LoRa, NB-IoT, industrial Wi-Fi, or Ethernet depending on configuration. Typical event-arrival latency in the central system is five to thirty seconds. This latency does not affect OEE calculation — it only affects real-time alerting reactivity on long stops.

The qualification layer presents events to the operator or shift supervisor via a touch terminal (industrial tablet, smartphone, shop-floor fixed display). The interface displays the count of stops pending qualification and allows event labeling in a few seconds. It is the system’s human interface.

The analysis layer aggregates qualified events by period, shift, line, product and computes the indicators: OEE, Availability/Performance/Quality decomposition, Pareto top causes, trends. It is the layer consumed by plant managers, continuous improvement engineers, and operations leadership.

A rupture between two layers — for example, a terminal too slow for qualification or an unsuitable analytic visualization — is enough to compromise the entire chain. This is why integrated end-to-end solutions are operationally more reliable than architectures composed of disparate bricks.

Selection criteria for a detection solution in May 2026

The micro-stop detection market in 2026 counts dozens of active solutions globally, ranging from generalists (MES integration) to specialists (autonomous IoT sensors). Six criteria allow rational comparison of offerings during a consultation.

• Total time to service from first sensor mount to first exploitable routine data. Reference: 48 to 72 hours for an external-sensor solution, six to twelve weeks for a PLC integration with change control.
• Detection precision measured in seconds versus reference measurement (piece counter or PLC bit). Reference: 2-5 seconds for external sensors, 1 second for PLC integration.
• Stoppage coverage rate measured as the share of real stoppages effectively captured. Reference: above 99 % for performant solutions.
• Routine false-positive rate after four weeks of calibration. Reference: below 10 %, ideally under 5 %.
• Operator effort per event for qualification. Reference: under 5 seconds per event.
• Total annual cost including hardware, software, maintenance, and support. To compare against expected OEE gains (typically 6 to 12 points over 12 months).

Return on investment is typically achieved in three to nine months on lines of medium or high criticality in May 2026.

Detect and eliminate your micro-stops — 48 hours, no PLC modificationWireless external sensors installed in under 30 minutes per machine. Automatic micro-stop detection from 2 seconds. Free 48-hour POC on site, 450+ factories deployed across 30 countries.Request an OEE demo

External references

Internet of Things — Wikipedia · AFNOR — NF E60-182 · SME — Society of Manufacturing Engineers · SEMI Standards (E10)

Related TeepTrak reading: Micro-Stops on the Production Line: How to Detect and Eliminate Them · Eliminating Micro-Stops: Pareto, Top 3 Root Causes, and Targeted SMED Project · How to Calculate OEE: Formula, Method, and Worked Example