The Azure Predictive Maintenance solution accelerator is an enterprise level solution (end-to-end) for organizational business scenarios which predict points at which failures might happen. You can use this solution accelerator for activities like optimizing Azure maintenance. The Azure solution merges key Azure IoT solution accelerators and services, examples being stream analytics, IoT Hub, as well as Azure Machine Learning workspaces. The workspaces have a distinct model, based on a public data set sample, in order to forecast the actual Remaining Useful Life (RUL) of something like the engine of an aircraft. The Azure solution does fully implement the IoT business scenario. This ends up being a starting point for you as the enterprise level planner to organize and implement a working solution that meets your own organization's particular business requirements and recommendations.
Azure Predictive Maintenance Logical architecture
The following logical architecture diagram outlines the logical components of the solution accelerator:
Use this link and fill out the form to contact Dynamics Edge for more information about custom Microsoft Windows Azure IoT training and solutions, Azure Predictive Maintenance Solution Accelerator Training and solutions, and much more
Items in blue are Azure services that are provisioned in the region where you originally deployed the solution accelerator solution.
The actual list of regions where you can truly deploy the solution accelerator displays on the provisioning page of Azure.
The green colored item indicates a simulated device representing the engine of an aircraft or equivalent. You can find out more about Azure simulated devices in the Simulated devices section of Azure. The items colored in gray signify device management capabilities (components related to the aforementioned). Now the current release of Azure Predictive Maintenance solution accelerator does not automatically provision these resources. To learn more about Azure device management, please contact Dynamics Edge.
Azure Predictive Maintenance -> resources
In the Azure portal, navigate to the resource group with the Azure solution name you chose to view your provisioned resources before.
Azure Predictive Maintenance -> Accelerator resources
At the time that you first provision the Azure solution accelerator you should get a new email which contains a link to the Azure Machine Learning workspace. You may also navigate to the Azure Machine Learning workspace when coming from the Microsoft Azure IoT Solution Accelerators blade or page for your own provisioned solution set. An Azure tile is available on this Azure page when the solution is in the Ready state as indicated in the portal.
Azure Predictive Maintenance -> Machine learning model and Simulated devices
Within the solution accelerator, there is a simulated device which represents something like an aircraft engine. The Azure solution is first provisioned with two or three engines that then map to just one single aircraft. Each of the engines then emits four or five types of telemetry: for instance, Sensor 11, Sensor 9, Sensor 15, and Sensor 14, which all provide the necessary data for the Azure Machine Learning model to calculate the RUL for this engine in this case. So each of the simulated devices sends the following telemetry messages to IoT Hub in Azure.
The cycle count is sent. In this case a single cycle does represent what's known as a completed flight with a time duration between 2 hours and 10 hours of time. During the time of the actual flight, the various telemetry data is captured at intervals of every half hour of time.
Here are details about telemetry. So there are at least 4 sensors that actually represent attributes of the aircraft engine piece. These sensors are then generically labeled Sensor 11, Sensor 9, Sensor 15, and Sensor 14 here. The four sensors represent enough telemetry to get results that are useful from the RUL model. The RUK model used in the Azure solution accelerator is originally built from a sample public data set that actually includes real engine sensor data within the sample. For much more information on how the Azure model was created from the original sample data set, see the Cortana Intelligence Gallery Predictive Maintenance Template or contact Dynamics Edge.
The variant simulated devices can indeed handle the following commands that have been sent from the Azure IoT hub in the solution package.
StartTelemetry actually controls the state of the simulation and starts the device sending telemetry process.
StopTelemetry does control the state of the simulation and also stops the device from sending telemetry signals.
The Azure IoT Hub provides device command acknowledgment for states.
The Azure Stream Analytics job telemetry operates on what's known as the incoming device telemetry stream by utilzing 2 statements here. The first one selects all telemetry from all the devices and sends all of this data to Azure blob storage in the subscription. Then the data becomes visualized in the actual web app.
Average sensor values are computed with the other way over a sliding window such as two minutes and then the telemetry sends this data through the Azure Event hub to an event processor in the next stage.
The event processor host runs in an Azure Web Job by taking the average sensor values for a fully completed cycle and using them as follows. The values are passed to an API that actually exposes the trained model in order to make calculations on the resulting RUL for an engine or equipment piece. The aforementioned API is actually exposed by an Azure Machine Learning workspace that is fully provisioned as part of the solution when first done.
What's known as the Azure Machine Learning component actually uses a model originally derived from the first data collected from real aircraft engines in a sampled set. You are able to navigate to the Azure Machine Learning workspace from your own solution's tile on the azureiotsuite.com web page and web site. The aforementioned tile is indeed available when the solution is in the Ready state in Azure.
Your Azure Machine Learning model is indeed available as a generalized template to demonstrate all of these capabilities actually working from any device telemetry collected through the Azure IoT solution accelerators services for predictive maintenance. Microsoft Azure has built a regression model of equipment like aircraft engines that are based on publicly available data and guided step-by-step processes on how to use the model in real life scenarios.
This custom Microsoft Windows Azure IoT Predictive Maintenance solution accelerator training from Dynamics Edge (contact us for more info) uses the regression model created from this template in Azure. The Azure model is deployed into your Windows Azure subscription and then exposed through an automatically generated API for this purpose. The full Azure solution includes an important subset of the original testing data representing some number (out of 110 total) equipment engines and sensor data stream subsets as well. All of the data might be sufficient to indeed provide a quality result in terms of accuracy and with regards to the trained model in Azure.
You may want to customize the key components of Azure IoT Predictive Maintenance solution accelerator and other Azure IoT solution accelerators training, for more info on this contact Dynamics Edge.