WISE-PaaS/AFS
  • Service Provider:Advantech
  • Last Update: 3/13/2020 6:31:12 PM
AFS is designed to serve as multi-model training and multi-model deployment services.
{{vm.prdAction.Point}} up
Perpetual
Monthly Subscription
{{vm.prdAction.Point}} up
Monthly Subscription
{{vm.prdAction.Point}} up
Monthly Subscription
(Trial Available)
{{vm.prdAction.Point}} up
Monthly Subscription / Redeem
Limited
Advantech professional can help find the best solution for your business.
  • Service Details
    Introduction

    WISE-PaaS/AFS is designed to serve as an AI framework service for scale up the AIoT solution. AFS offers five major benefits:

    1. 1. “Compute” resource management: CPU, GPU, and memory resources.
    2. 2. “Data” fusion from multiple sources: OT data, IT data, and contextual data.
    3. 3. “Algorithms” For model development and management, AI framework integration, and model repository and version control.
    4. 4. “Mass deployment at edge” Remote model deployment to Window, Linux, Docker, and EdgeX Foundry devices.
    5. 5. “Life cycle management” Auto model training and deployment, re-training and re-deployment, all triggered by a task scheduler.
    AFS_1.png

    Feature Highlights
    1. 1. Workspace:
      1. ● Online Code IDE: By using the Jupyter Notebook, you can quickly develop many kinds of analytic models based on Python 3. The Jupyter Notebook supports machine learning and deep learning frameworks, such as Scikit-Learn, Pytorch, Keras and TensorFlow.
      2. ● Upload Code: If you want to program offline, you can upload your code package and docker image during runtime.
    2. 2. Task Management:
      1. ● Training Tasks: Allows you to easily create a schedule for running training tasks automatically.
      2. ● Model Deployment Tasks: Allows you to easily create model deployment schedules and define deployment rules such as the best model and model performance.
      3. ● Hyperparameter Tuning: Allows you to find optimal hyperparameters for a learning algorithm and optimize the target variable that you have specified.
    3. 3. Catalog: Provides many prebuild analytic modules for you to subscribe to and start training tasks immediately.
    4. 4. Model Board: Visualizes models’ learning outcomes, helping you assess the models.
    5. 5. Inference Engine:
      1. ● Edge Inference: Allows you to quickly deploy models to edge devices without having to conduct on-site installation.
      2. ● Cloud Inference: Allows to you push the inference engine app via RestfulAPI to your WISE-PaaS/EnSaaS cluster, workspace, or namespace.
    6. 6. AFS SDK:
      1. ● Model SDK: Allows you to easily upload your training models to the AFS model repository.
      2. ● Data Source SDK: Allows you to easily input, query, and update your training data in WISE-PaaS PostgeSQL, MongoDB, InfluxDB, Blob Service, and WISE-PaaS/APM.
    7. 7. Three domain-specific AI models on the AFS catalog
      1. ● Prognostic and health management, PHM for smart equipment
      2. ● Process quality prediction analysis, PQA for iFactory
      3. ● Automatic optical inspection, AOI for iFactory


    Pricing details

    Unit: WISE-Point

    Basic monthly fee for the AFS instance

    AFS_pricing01.png

    Hourly fees for using AFS computing resources

    (Pay As You Go: Charged as soon as you start to use Jupyter Notebook, Training Task, Hyperparameter Task and Image Labeling).

    AFS_pricing02.png

  • Pricing Details
    {{category.PlanCategoryName}}
Introduction

WISE-PaaS/AFS is designed to serve as an AI framework service for scale up the AIoT solution. AFS offers five major benefits:

  1. 1. “Compute” resource management: CPU, GPU, and memory resources.
  2. 2. “Data” fusion from multiple sources: OT data, IT data, and contextual data.
  3. 3. “Algorithms” For model development and management, AI framework integration, and model repository and version control.
  4. 4. “Mass deployment at edge” Remote model deployment to Window, Linux, Docker, and EdgeX Foundry devices.
  5. 5. “Life cycle management” Auto model training and deployment, re-training and re-deployment, all triggered by a task scheduler.
AFS_1.png

Feature Highlights
  1. 1. Workspace:
    1. ● Online Code IDE: By using the Jupyter Notebook, you can quickly develop many kinds of analytic models based on Python 3. The Jupyter Notebook supports machine learning and deep learning frameworks, such as Scikit-Learn, Pytorch, Keras and TensorFlow.
    2. ● Upload Code: If you want to program offline, you can upload your code package and docker image during runtime.
  2. 2. Task Management:
    1. ● Training Tasks: Allows you to easily create a schedule for running training tasks automatically.
    2. ● Model Deployment Tasks: Allows you to easily create model deployment schedules and define deployment rules such as the best model and model performance.
    3. ● Hyperparameter Tuning: Allows you to find optimal hyperparameters for a learning algorithm and optimize the target variable that you have specified.
  3. 3. Catalog: Provides many prebuild analytic modules for you to subscribe to and start training tasks immediately.
  4. 4. Model Board: Visualizes models’ learning outcomes, helping you assess the models.
  5. 5. Inference Engine:
    1. ● Edge Inference: Allows you to quickly deploy models to edge devices without having to conduct on-site installation.
    2. ● Cloud Inference: Allows to you push the inference engine app via RestfulAPI to your WISE-PaaS/EnSaaS cluster, workspace, or namespace.
  6. 6. AFS SDK:
    1. ● Model SDK: Allows you to easily upload your training models to the AFS model repository.
    2. ● Data Source SDK: Allows you to easily input, query, and update your training data in WISE-PaaS PostgeSQL, MongoDB, InfluxDB, Blob Service, and WISE-PaaS/APM.
  7. 7. Three domain-specific AI models on the AFS catalog
    1. ● Prognostic and health management, PHM for smart equipment
    2. ● Process quality prediction analysis, PQA for iFactory
    3. ● Automatic optical inspection, AOI for iFactory


Pricing details

Unit: WISE-Point

Basic monthly fee for the AFS instance

AFS_pricing01.png

Hourly fees for using AFS computing resources

(Pay As You Go: Charged as soon as you start to use Jupyter Notebook, Training Task, Hyperparameter Task and Image Labeling).

AFS_pricing02.png

AFS_2.png

Successful cases with WISE-PaaS/APM

High Permeability Film Quality Prediction

Problem: The high-transparent membrane is too sticky, and removing 2%–3% of the viscous parts from both sides of the membrane incurs higher production costs.

Target: To collect and analyze the parameters for the roll-to-roll process of the membrane production line; establish a PQA model and a predictive quality system; and adjust the relevant parameters timely to improve the membrane’s quality when a warning arises about a quality issue.

AFS_3.PNG

Steel Factory

Synergizing APM, AFS, and Dashboard can yield a solution for predicting equipment anomalies at steel factories. This solution involves implementing feature extraction processes and uploading features through SCADA SDK to the APM portal where asset profile and management logistics are configured. After enough datasets are collected and labeled, AFS is used to choose or develop a training package and execute a training task with the APM firehose settings. After training, a model’s performance can be monitored and data be relabeled directly on the dashboard.

When the model is ready for deployment, the AFS inference engine is to deploy the model and its corresponding inference Docker image to edge devices. The prediction results from an inference execution are uploaded back to the APM portal, where event rules are defined using the APM Rule Engine to monitor such results. When an event is triggered, WISE-PaaS Notification Service informs the equipment operation and maintenance team to act accordingly.

AFS_4.png

{{category.PlanCategoryName}} See More

Related Services