Перейти к основному содержимому
Unlisted page
This page is unlisted. Search engines will not index it, and only users having a direct link can access it.

Elegoo-Obico Failure Detection API

The APIs documented on this page are designed for Elegoo partners to detect print failures using Obico's AI-powered failure detection system.

Authentication​

Authentication is performed using device credentials passed as form data parameters:

  • serial_no: The device serial number registered in the system
  • access_token: The access token associated with the device

These credentials must be included in the POST request along with other parameters.

подсказка

Use the Elegoo-Obico Access Token API to manage device credentials before using the failure detection API.

Endpoint​

  • https://elegoo-app.obico.io/. Production endpoint. Please use this endpoint unless instructed by the Obico team differently.
  • https://elegoo-app-stg.obico.io/. Staging endpoint. Please don't use unless instructed by the Obico team.
  • https://elegoo-cn-app.elegoo.com.cn. Production endpoint within China.
  • https://elegoo-cn-app-stg.elegoo.com.cn. Staging endpoint within China.

POST /ent/partners/api/elegoo/predict/​

Request​

This POST request should be sent as multipart/form-data format.

Form parameters​

  • serial_no: The device serial number. Required for authentication.
  • access_token: The access token for the device. Required for authentication.
  • print_id: A id that can uniquely identify the print within the printer it belongs. Max 256 characters.
  • img: Snapshot from the webcam for failure detection. In JPEG format.
  • fd_gen: Which generation of the failure detection AI model scores the snapshot. 1 (default): the original model. 2: the next-generation model. Optional. See Model generations below.

Response​

Status code: 200​

API request was processed successfully.

Body​

{
  "result": {
    "p": 0,
    "temporal_stats": {
      "ewm_mean": 0,
      "rolling_mean_short": 0,
      "rolling_mean_long": 0,
      "prediction_num": 0,
      "prediction_num_lifetime": 0
    },
    "detections": [
        [0.541085422039032, [422.7984619140625, 236.30227661132812, 61.9364013671875, 74.49552917480469]],
        [0.43781569600105286, [426.05596923828125, 264.619140625, 42.386478424072266, 4.73854064941406]],
        [0.2545202076435089, [423.3209533691406, 238.6829071044922, 113.47953796386719, 135.73854064941406]],
        [0.20370429754257202, [456.3966369628906, 236.23785400390625, 39.029632568359375, 67.34481811523438]]
      ]
  }
}
  • p: A number between 0 and 1.0. 0 means no failure is detected. 1 means the maximum confidence on predicting a print failure.
  • temporal_stats: The temporal stats that may be useful in determining if a failure has actually occurred. These stats are important for smoothening the noises in failure detection. See the tip below for details.
    • ewm_mean: Exponentially weighted mean for p. EWM window span = 12.
    • rolling_mean_short: Short-term rolling mean for p. Rolling window span = 310. This rolling mean is reset to 0 when a new print starts.
    • rolling_mean_long: Long-term rolling mean for p. Rolling window span = 7200. This rolling mean is accumulated over the lifetime of the printer.
    • prediction_num: The number of predictions for the current print so far.
    • prediction_num_lifetime: The number of predictions for the life-time of the printer.
    • detections: A list of tuples. Each tuple is [confidence, [xc, yc, w, h]].
      • confidence: Range: [0, 1], where 0 means not failure and 1.0 means the maximum confidence on predicting a print failure.
      • [xc, yc, w, h]: Rectangle of the detected area. xc and yc are the X and Y coordinates of the center of the rectangle. w and h are the width and hight of the rectangle.
подсказка

It's a good practice to use the temporal stats to smoothen out the noises in failure detection. Otherwise there may be excessive amount of false alarms.

In Obico open-source server, the way these temporal stats are used can be simplistically described as below:

  • If ewm_mean - rolling_mean_long < 0.36: no failure.
  • Else if ewm_mean - rolling_mean_long > 0.99: failure.
  • Else if ewm_mean - rolling_mean_long > 0.78: maybe failure.
  • Else if ewm_mean > (rolling_mean_short - rolling_mean_long) * 3.8: maybe failure.
подсказка

All these "magic numbers", such as the rolling window sizes, or thresholds such as 0.36 or 0.78, should be considered as hyper-parameters. You are highly recommended to go through the hyper-parameters tuning process to find the optimal values for them.

Status code: 400​

API request was NOT processed successfully for other reasons, such as missing required parameters.

Body​

{
  "error": "Detailed error message"
}

Examples of error messages:

  • "Missing or invalid image"
  • "print_id is required"
  • "fd_gen must be 1 or 2"

Status code: 401​

Authentication failed. This can occur when:

  • Missing serial_no or access_token
  • Invalid credentials (including expired access tokens)

Body​

{
  "error": "serial_no and access_token are required"
}

or

{
  "error": "Invalid credentials"
}

Status code: 429​

API request was NOT processed successfully because of rate throttling.

Body​

{
  "error": "You are running too hot! Take it easy buddy..."
}

Model generations​

The fd_gen parameter selects which AI model scores the snapshot. Both generations return the same response shape, but the two models produce different score distributions, so the temporal stats must be interpreted with different hyper-parameters.

fd_gen=1 (default): original model​

Use the hyper-parameters in the tips above.

fd_gen=2: next-generation model​

The next-generation model is more accurate, and its scores are smaller in absolute terms. detections contains at most the 8 highest-confidence detections in the frame.

In Obico's own server, the decision for the next-generation model can be simplistically described as below:

  • EWM window span = 13.
  • Short-term rolling window span = 380.
  • The first 30 predictions of a print are never treated as a failure.
  • If ewm_mean - rolling_mean_long < 0.10: no failure.
  • Else if ewm_mean - rolling_mean_long > 0.15: failure.
  • Else if ewm_mean - rolling_mean_long > (rolling_mean_short - rolling_mean_long) * 2.8: failure.

As with the original model, treat these values as starting points for your own hyper-parameter tuning.

предупреждение

Use the same fd_gen value for all requests from a device. Do not switch a device between generations.

подсказка

Please coordinate with the Obico team before switching a fleet to fd_gen=2, so that the next-generation model is provisioned on the endpoint you use.

Usage Example​

curl -X POST https://elegoo-app.obico.io/ent/partners/api/elegoo/predict/ \
  -F "serial_no=ELEGOO_DEVICE_001" \
  -F "access_token=your_access_token_here" \
  -F "print_id=print_456" \
  -F "img=@/path/to/snapshot.jpg"

With the next-generation model:

curl -X POST https://elegoo-app.obico.io/ent/partners/api/elegoo/predict/ \
  -F "serial_no=ELEGOO_DEVICE_001" \
  -F "access_token=your_access_token_here" \
  -F "print_id=print_456" \
  -F "fd_gen=2" \
  -F "img=@/path/to/snapshot.jpg"
подсказка

Make sure to register your device credentials using the Elegoo-Obico Access Token API before calling the failure detection API.