A Home Assistant custom component to monitor washing machines via smart sockets, learn power profiles, and estimate completion time using shape-correlation matching.
Caution
ELECTRICAL SAFETY WARNING: Using smart plugs such as Shelly or Sonoff with high-amperage appliances (washing machines, dryers, dishwashers) carries significant risk.
- Fire Hazard: Cheap or low-rated smart plugs may overheat, melt, or catch fire under sustained high loads (heating/drying phases).
- Rating Check: Ensure your smart plug is rated for the maximum peak power of your appliance (often >2500W). Standard 10A plugs may fail; 16A+ or hardwired modules are recommended.
- Use at Your Own Risk: The authors of this integration are not responsible for any electrical damage or fires caused by improper hardware usage. inspect your hardware regularly.
- Multi-Device Support: Track Washing Machines, Dryers, Dishwashers, Coffee Machines, or Electric Vehicles (EV) with device-type tagging.
- Smart Cycle Detection: Automatically detects starts/stops with Predictive End logic. Includes End Spike Protection for dishwashers to capture final pump-outs.
- Power Spike Filtering: Ignores brief boot spikes to prevent false starts.
- Shape-Correlation Matching: Uses
numpy.corrcoefwith Confidence Boosting to distinguish similar cycles. - Manual Training: You define your profiles (e.g., "Cotton", "Quick") once; the system learns to recognize them thereafter. Integration does not auto-create profiles.
- Smart Time Estimation: "Phase-aware" prediction detects variance (e.g., heating) and locks the countdown to prevent erratic jumps.
- Changeable Power Sensor: Switch plugs without losing history.
- Minimal Status Card: Optional custom Lovelace card.
- Manual Program Override: Select the correct program manually if detection is uncertain; the system learns from your input.
- Manual Profile Creation: Create profiles even without historical cycles by specifying a baseline duration (e.g., "Eco Mode - 3h").
- Ghost Cycle Suppression: Intelligent filtering with "Suspicious Window" (20 mins) prevents end-spikes from triggering duplicate cycles. Now Persistent: Remembers cycle history across HA restarts to prevent ghost detections after reboots.
- Robust vNext State Machine: Advanced filtering with
start_energyandend_energygates prevents false starts/ends. - Multi-Stage Matching Pipeline: Uses Fast Reject -> Core Similarity -> DTW-Lite tie-breaking for superior accuracy.
- Local Only: No cloud dependency, no external services. All data stays in your Home Assistant.
- Notifications: Integrated alerts for cycle start, finish, and pre-completion (e.g., 5 mins before finish).
- Self-Learning: Gradually adjusts expected durations based on your confirmed historical data.
- Realistic Variance: Handles natural cycle duration variations with configurable tolerance.
- Progress Tracking: Clear cycle progress indicator with automatic reset after unload.
- Auto-Maintenance: Nightly cleanup - removes broken profiles, merges fragmented cycles (Empty/New profiles are safely preserved).
- Export/Import: Full configuration backup/restore with all settings and profiles via JSON.
Designed for new users to get up and running quickly.
This integration is a default repository in HACS.
- Click the button below to open the repository in HACS:
- Click Download.
- Restart Home Assistant.
Alternatively, open HACS > Integrations > Explore & Download Repositories > Search for "WashData".
- Download the
custom_components/ha_washdatafolder from the latest release. - Copy it to your Home Assistant
custom_componentsdirectory. - Restart Home Assistant.
Follow these steps to get accurate results quickly.
- Go to Settings > Devices & Services > Add Integration > WashData.
- Name: Name your appliance (e.g., "Washing Machine").
- Device Type: Select the type (Washer, Dryer, etc.) - this sets smart defaults for the internal logic.
- Power Sensor: Select your smart plug's power entity (Watts). Note: The system is now optimized for polling intervals of 30-60 seconds (defaults adjusted automatically).
- Initial Profile (Optional): If you know your standard program (e.g. "Cotton"), create it now.
If you are using Zigbee2MQTT with smart plugs, ensure your device reporting is responsive enough for accurate matching:
- Reporting Intervals: Decrease the reporting intervals in Z2M (e.g., Min: 1s-10s, Max: 1200s).
- Power Threshold: Decrease the minimum updating threshold (e.g., from 5W to 1W or 2W) to ensure small power changes are captured promptly.
- Note: These changes may slightly increase Zigbee network traffic.
WashData does not come with pre-built profiles because every machine model is different. You must teach it what your cycles look like.
This gives you the cleanest data.
- Open the Dashbord Card.
- Select Record Cycle (Manual) from the program list.
- Start your machine.
- When finished, go to Manage Data & Profiles, find the recording, and create a profile.
If you prefer to just use it and label later:
- Run a Cycle: Use your machine as normal. WashData will track it as an "Unknown" cycle.
- Label It: After the cycle finishes:
- Go to Manage Data & Profiles (via the Configure button or the Tile Card).
- Find the recent "Unknown" cycle.
- Click Create Profile, name it (e.g., "Cotton 40"), and Save.
- Repeat: Do this for your 2-3 most common programs.
Once profiles are created, WashData starts matching new cycles automatically.
- Feedback: If a match is found but confidence is moderate, you may get a "Verify Cycle" notification.
- Refinement: Go to Configure > Learning Feedbacks to Confirm or Correct the detection.
- Self-Improving: Confirming a cycle helps the system refine its duration models.
If "Auto-Detect" isn't working perfectly, use Advanced Settings to tune the logic for your specific machine.
📊 Click here for a Visual Guide to these settings - Graphs explaining what the numbers actually do.
| Problem | Likely Cause | Solution |
|---|---|---|
| Cycle Starts Too Early | Smart plug reports brief power spikes during boot/standby. | Increase Start Energy Threshold (e.g., to 2 Wh). This forces the machine to consume actual energy before stating "Running". |
| Cycle Ends Too Early | Machine pauses (soaking) or has long low-power intervals. | Increase Off Delay. Give it more time (e.g. 5 mins) to wait before deciding the cycle is truly finished. |
| False "Ghost" Cycles | High-power usage at the very end (e.g. anti-crease or pump-out) triggers a new start. | Increase Minimum Off Gap (e.g. to 120s). Forces a mandatory cooldown period between cycles. |
| "Unknown" Matches | Your profiles are too strict or variance is high. | Increase Duration Tolerance (e.g. 0.25). Allows ±25% duration difference when matching. |
| Notifications Too Late | You want to know before it finishes. | Set Notify Before End Minutes (e.g. 5). Get an alert 5 minutes before the estimated finish time. |
| Persistent 'Running' State | Integration stays locked to a long profile after a short cycle diverged. | Adjust Matching Stability Thresholds in const.py. Divergence detection now automatically reverts to "Detecting..." if confidence drops below 60% of the peak score. |
Pro Tip: Use the Apply Suggestions button in Settings. It analyzes your history and calculates the perfect text-book values for your specific machine.
- 📗 IMPLEMENTATION.md - Deep dive into NumPy matching, State Machine logic, and Learning algorithms.
- 🧪 TESTING.md - How to test with the virtual socket.
📸 Screenshots (Click to expand)
All your WashData-monitored appliances appear as devices with sensors and controls.
The central hub for managing your appliance - access all features from here.
Configure power sensor, device type, off delay, and notification preferences.
Fine-tune detection thresholds, matching parameters, and timeout values for your specific appliance.
View, create, edit, or delete learned power profiles for different wash programs.
Browse cycle history, label unknown cycles, merge fragments, or delete bad data.
Confirm or correct the system's profile matches to improve learning accuracy.
Run database cleanup, repair corrupted data, and export/import configurations.
sensor.<name>_state: Current Status (Idle, Running, Detecting...).sensor.<name>_program: Best-matched Profile Name.sensor.<name>_time_remaining: Smart countdown (locks during high variance phases).sensor.<name>_total_duration: Total predicted duration (Elapsed + Remaining). Ideal fortimer-bar-card.sensor.<name>_cycle_progress: 0-100% (Resets to 0% after unload timeout).binary_sensor.<name>_running: Simple On/Off state.switch.<name>_auto_maintenance: toggle nightly database cleanup.
Most management is now done via the Interactive UI (Configure > Manage Data), but services are available for automation:
ha_washdata.export_config: Full JSON backup/restore.ha_washdata.import_config: Import a JSON backup, restoring all custom thresholds and profiles.
ha_washdata.record_start / record_stop:
Manually start/stop a recording. Useful for automations (e.g. triggering from a physical button or separate sensor).
service: ha_washdata.record_start
data:
device_id: "washer_device_id"ha_washdata.label_cycle: Assign profile to history programmatically.
🇬🇧 English • 🇨🇿 Čeština • 🇩🇰 Dansk • 🇩🇪 Deutsch • 🇬🇷 Ελληνικά • 🇪🇸 Español • 🇪🇪 Eesti • 🇫🇮 Suomi • 🇫🇷 Français • 🇭🇷 Hrvatski • 🇭🇺 Magyar • 🇮🇹 Italiano • 🇯🇵 日本語 • 🇱🇹 Lietuvių • 🇱🇻 Latviešu • 🇳🇴 Norsk • 🇳🇱 Nederlands • 🇧🇪 Nederlands (BE) • 🇵🇱 Polski • 🇵🇹 Português • 🇷🇴 Română • 🇸🇰 Slovenčina • 🇸🇮 Slovenščina • 🇷🇸 Srpski • 🇸🇪 Svenska • 🇺🇦 Українська • 🇨🇳 简体中文
Non-commercial use only. See LICENSE file.
