Tractian · Innovation Case 2026

DIAL.
Tractian OEE Terminal.

Unplanned downtime costs U.S. manufacturers $50B annually. Most of it goes unlabeled. DIAL is a purpose built operator terminal that captures stop reason codes at the machine in under 5 seconds, gloves on, WiFi optional. Every attributed event becomes structured input for Tractian's OEE engine and a labeled training sample for predictive maintenance models.

Try the interactive demo
DIAL terminal render
Built for the floor

Everything a machine stop needs.
Nothing it doesn't.

DIAL tilt
100M+ shaft revolutions
Avg. log time
5sec
Dial. Confirm. Done.
Enclosure
IP65
Dust-tight. Splash-proof.
Buttons
1M cycle life · Gold-plated contacts
Offline memory
72h
NVS flash queue
DIAL top-down
Connectivity
Offline-first
WiFi + BLE when available.
NVS queue when it's not.
Dial top
256 PPR · Hall effect encoder
Auto-commit
30sec
UNP-TIM timeout
Mount options
Magnetic
VESA 75mm
Wall
Field-replaceable surround · TPE overmold
Integration
Tractian-
native
Every stop feeds directly into the Tractian OEE engine as a labeled data point.
The problem

The machine stopped.
Nobody logged why.

A CNC machine goes down. The operator has 5 seconds to log a stop reason before walking back to their station. They are wearing thick Kevlar gloves soaked in cutting fluid and metallic dust. The WiFi in this corner of the plant has been down for 40 minutes. Nothing gets logged. The OEE report records "unplanned downtime" with no root cause attribution. The same failure repeats next shift.

C15-second input window

Operator attention resets the moment the machine restarts. Any interaction exceeding 5 seconds is consistently abandoned, making input speed a hard design constraint, not a preference.

C2Kevlar gloves and process contamination

Capacitive touchscreens fail under gloved input. Thin membrane buttons corrode under cutting fluid exposure. Exposed mechanical components accumulate metallic particulate and seize. Input hardware must be designed for the environment from the ground up.

C3Intermittent connectivity

WiFi outages of 1 to 2 hours are common in heavy manufacturing environments. Cloud dependent solutions lose data silently during these windows. The operator logged the event. The infrastructure did not preserve it.

C4Plant specific stop taxonomies

No two facilities use the same reason code structure. The terminal must support customer defined hierarchies configurable remotely, one hardware SKU deployable across any operation without field modification.

C5OEE data integrity

An unattributed stop event does not just represent missing information. It corrupts the Availability component of OEE and removes the signal needed to identify recurring failure patterns at the maintenance level.

The solution

DIAL, a dedicated stop reason terminal.

Not a repurposed tablet. Not a general purpose HMI panel with downtime logging bolted on. A single function device engineered around the operator's input constraints.

Two inputs. Under 5 seconds.

A sealed magnetic rotary encoder for navigation and one oversized confirm button for commitment. The interaction requires no authentication, no mode switching, and no screen management. Two deliberate gestures from stop event to logged record.

Engineered for process environments.

IP65 rated enclosure with a non contacting magnetic encoder. No mechanical wear surfaces, no particulate ingress path, rated for mud, dust, grease, and cutting fluid. IP67 pushbuttons at both input points. Rotary shaft seal at the encoder penetration.

Modular hardware architecture.

The electronics core is a self contained unit. The rubber surround functions as a replaceable protective shell, covering edges and back while leaving the face fully accessible. Field damage to the surround does not require electronics replacement.

Offline first data integrity.

Every stop event is committed to NVS flash storage before any network operation is initiated. WiFi connectivity is a synchronization channel, not an operational dependency. The local queue drains to Tractian's MQTT broker in chronological order on reconnection.

Interactive demo

Try it.

Scroll the dial or use arrow keys. OK confirms. BACK steps back. Space wakes the device.

CNC-04 · CELL B · SHIFT 2
SYNC
Device state

Turn the dial to navigate stop categories and reason codes. Every logged event syncs to the Tractian OEE engine.

IDLEawaiting
Logged events
No events logged yet
scroll turn dial
↑↓ navigate
Enter confirm
Esc back
Space wake device
The interaction

Three steps. Under 5 seconds.

The two level stop taxonomy, category then reason code, limits visible options to eight at any point in the flow, keeping selection time predictable regardless of how large the customer's full code set grows.

01Stop event detected

State: IDLE → SELECT_GROUP. DIAL registers the stop event via machine signal or vibration input. The display activates and presents the top level category list without requiring any operator initiated input. The 30 second timeout window begins immediately.

02Category selection

State: SELECT_GROUP → SELECT_CODE. The operator rotates the encoder to scroll through stop categories: Tooling, Material, Machine, Process, External. Each detent advances one position. Input is reliable under heavy gloves, contaminated hands, or indirect contact. OK confirms.

03Reason code selection and commit

State: SELECT_CODE → LOGGED. The operator scrolls to the applicable reason code and confirms with OK. The event is written to local flash before any network transmission is attempted. The display presents a logged confirmation and returns to idle within 2 seconds.

If the operator does not complete input within 30 seconds, DIAL automatically commits a UNP-TIM record to local storage. The downtime window and timestamp are preserved for OEE calculation and flagged for supervisor review. No stop event is ever silently discarded.

Hardware specification

Component decisions.

Each component was selected for a specific engineering reason. Click any component to see the rationale and specifications.

01
MCU
ESP32-S3
02
Rotary encoder
Non-contacting magnetic
03
Input buttons
Sealed panel mount
04
Display
Transflective LCD
05
Enclosure
Structural plastic + TPE
06
Shaft seal
Rotary lip seal
ESP32-S3 Microcontroller
The ESP32-S3 runs FreeRTOS across both cores, which allows encoder polling, display refresh, and network communication to operate as independent tasks without blocking each other. Native WiFi and Bluetooth eliminates an external radio module. The NVS flash subsystem provides reliable queue storage without requiring a file system layer. A Raspberry Pi introduces a 20 to 40 second Linux boot time and depends on an SD card, a documented failure point in high vibration environments. An Arduino Uno has 2KB of SRAM, insufficient for MQTT, display buffer, and offline queue simultaneously.
PlatformESP32-S3-WROOM-1
CoresDual core, 240MHz
RTOSFreeRTOS
WirelessWiFi 802.11n + BLE 5.0
StorageNVS flash, ~72h queue
Idle current<10μA deep sleep
Data pipeline

Offline first. Then cloud.

Local persistence is not a fallback. It is the primary write path. Cloud synchronization operates as a background process whenever connectivity is available.

Local commit
On confirmation, DIAL writes the full event record to NVS flash in under 10ms. The event is preserved regardless of network state.
machine_id · timestamp · reason_code · operator_id
Queue accumulation
Events accumulate in an ordered FIFO structure with approximately 72 hours of capacity, covering the longest documented WiFi outage windows in CNC environments.
MQTT sync
When connectivity restores, the queue drains to the Tractian broker in chronological order. Each event is acknowledged before being marked flushed locally.
OTA configuration
Customer stop taxonomies are pushed from the Tractian cloud on the next sync window. Updated code trees are active within minutes. No physical access required.
OEE attribution
Each attributed stop enables accurate Availability calculation, root cause Pareto analysis, and cross facility failure pattern identification in the Tractian platform.
How the protocol works

MQTT was chosen over HTTP because it maintains a persistent broker connection, making reconnection after an outage fast and the queue drain immediate. Delivery is set to QoS 1, meaning the broker must acknowledge each event before the device removes it from the local queue. A duplicate event is recoverable. A lost event is not.

Encoder input is debounced at 50ms to prevent phantom ticks from machine vibration. Button input at 20ms. The 30 second timeout auto-commits a UNP-TIM record so no stop window is ever unaccounted for.

OEE + AI impact

The device isn't the product.
The labeled data is.

OEE impact
~40%
Of unplanned stops have a recurring, fixable root cause, invisible without reason code attribution.
Data quality gap
±15%
Variance between manual OEE reporting and automated sensor based calculation.
Industry baseline
55–60%
Typical global manufacturing OEE. World class benchmark is 85%.
Annual cost
$50B
US manufacturers lose to unplanned downtime. 70% rely on manual data collection.

Tractian's sensors capture the physical signature of every failure. What they cannot provide is operator attributed causation. DIAL closes that gap. The result is a labeled dataset that makes every Tractian predictive model more accurate over time.

01
Sensor detects anomaly. Vibration, current deviation, or thermal event. Timestamp recorded.
02
Machine stops. DIAL activates. Operator selects stop reason within 5 seconds.
03
Platform correlates sensor anomaly with operator attributed reason code. Labeled record created.
04
Labeled signatures accumulate across machines, cells, shifts, and facilities.
05
Predictive alert issued ahead of the next qualifying anomaly. Downtime avoided.