預測性維護解決方案加速器概觀Predictive Maintenance solution accelerator overview

預測性維護解決方案加速器是一個端對端解決方案,適用於預測可能發生失敗之時間點的商務案例。The Predictive Maintenance solution accelerator is an end-to-end solution for a business scenario that predicts the point at which a failure is likely to occur. 您可以針對最佳化維護等活動,主動使用此解決方案加速器。You can use this solution accelerator proactively for activities such as optimizing maintenance. 解決方案結合了主要的 Azure IoT 解決方案加速器服務,例如 IoT 中樞和Azure Machine Learning工作區。The solution combines key Azure IoT solution accelerators services, such as IoT Hub and an Azure Machine Learning workspace. 此工作區包含以公開範例資料集為基礎的模型,用來預測飛機引擎的剩餘使用年限 (RUL)。This workspace contains a model, based on a public sample data set, to predict the Remaining Useful Life (RUL) of an aircraft engine. 此解決方案完整提供 IoT 商務案例的實作做為起點,讓您規劃和實作能滿足特定商務需求的解決方案。The solution fully implements the IoT business scenario as a starting point for you to plan and implement a solution that meets your own specific business requirements.

GitHub 有提供了預測性維護解決方案加速器程式碼。The Predictive Maintenance solution accelerator code is available on GitHub.

邏輯架構Logical architecture

下圖概述解決方案加速器的邏輯元件:The following diagram outlines the logical components of the solution accelerator:

邏輯架構

藍色項目是在您佈建解決方案加速器的區域中所佈建的 Azure 服務。The blue items are Azure services provisioned in the region where you deployed the solution accelerator. 您可以部署解決方案加速器的區域清單會顯示在 [布建]頁面上。The list of regions where you can deploy the solution accelerator displays on the provisioning page.

綠色項目是模擬的飛機引擎。The green item is a simulated aircraft engine. 您可以在模擬裝置一節中進一步了解這些模擬裝置。You can learn more about these simulated devices in the Simulated devices section.

灰色項目是可實作「裝置管理」功能的元件。The gray items are components that implement device management capabilities. 目前的預測性維護解決方案加速器版本不會佈建這些資源。The current release of the Predictive Maintenance solution accelerator does not provision these resources. 若要深入瞭解裝置管理,請參閱遠端監視解決方案加速器To learn more about device management, refer to the Remote Monitoring solution accelerator.

Azure 資源Azure resources

在 Azure 入口網站中,瀏覽至具有您所選之解決方案名稱的資源群組以檢視已佈建的資源。In the Azure portal, navigate to the resource group with the solution name you chose to view your provisioned resources.

加速器資源

當您佈建解決方案加速器時,您會收到一封電子郵件,其中包含機器學習服務工作區的連結。When you provision the solution accelerator, you receive an email with a link to the Machine Learning workspace. 您也可以從Microsoft Azure IoT 解決方案加速器] 頁面流覽至 [Machine Learning] 工作區。You can also navigate to the Machine Learning workspace from the Microsoft Azure IoT Solution Accelerators page. 當解決方案處於就緒狀態時,此頁面上會出現一個圖格。A tile is available on this page when the solution is in the Ready state.

機器學習服務模型

模擬的裝置Simulated devices

在解決方案加速器中,模擬的裝置是飛機引擎。In the solution accelerator, a simulated device is an aircraft engine. 此解決方案已佈建兩個對應至單一飛機的引擎。The solution is provisioned with two engines that map to a single aircraft. 每個引擎會發出四種類型的遙測:感應器 9、感應器 11、感應器 14 和感應器 15 會提供 Machine Learning 模型來計算該引擎的 RUL 所需的資料。Each engine emits four types of telemetry: Sensor 9, Sensor 11, Sensor 14, and Sensor 15 provide the data necessary for the Machine Learning model to calculate the RUL for the engine. 每個模擬的裝置會將下列遙測訊息傳送至 IoT 中樞:Each simulated device sends the following telemetry messages to IoT Hub:

週期計數Cycle count. 週期是已完成持續期間介於 2 小時與 10 小時之間的飛行。A cycle is a completed flight with a duration between two and ten hours. 在飛行期間,每半小時擷取一次遙測資料。During the flight, telemetry data is captured every half hour.

遙測Telemetry. 有四個可記錄引擎屬性的感應器。There are four sensors that record engine attributes. 這些感應器會一般會標示為感應器 9、感應器 11、感應器 14 和感應器 15。The sensors are generically labeled Sensor 9, Sensor 11, Sensor 14, and Sensor 15. 這 4 個感應器會傳送足以從 RUL 模型取得有用結果的遙測。These four sensors send telemetry sufficient to get useful results from the RUL model. 用於解決方案加速器中的模型是根據包含實際引擎感應器資料的公用資料集建立而來。The model used in the solution accelerator is created from a public data set that includes real engine sensor data. 如需有關如何從原始資料集建立模型的詳細資訊,請參閱Cortana 智慧資源庫預測性維護範本For more information on how the model was created from the original data set, see the Cortana Intelligence Gallery Predictive Maintenance Template.

模擬的裝置可以處理從解決方案中 IoT 中樞傳送的下列命令:The simulated devices can handle the following commands sent from the IoT hub in the solution:

命令Command 描述Description
StartTelemetryStartTelemetry 控制模擬的狀態。Controls the state of the simulation.
傳送遙測以啟動裝置Starts the device sending telemetry
StopTelemetryStopTelemetry 控制模擬的狀態。Controls the state of the simulation.
傳送遙測以停止裝置Stops the device sending telemetry

IoT 中樞會提供裝置命令通知。IoT Hub provides device command acknowledgment.

Azure 串流分析作業Azure Stream Analytics job

作業:遙測會使用兩個陳述式來操作傳入裝置遙測串流:Job: Telemetry operates on the incoming device telemetry stream using two statements:

  • 第一個陳述式會從裝置選取所有遙測資料,然後將此資料傳送至 bob 儲存體。The first selects all telemetry from the devices and sends this data to blob storage. 從這裡,它會呈現在 Web 應用程式中。From here, it's visualized in the web app.
  • 第二個陳述式會透過兩分鐘的滑動視窗計算感應器平均值,然後透過事件中樞將此資料傳送至事件處理器The second computes average sensor values over a two-minute sliding window and sends this data through the Event hub to an event processor.

事件處理器Event processor

事件處理器主機會在 Azure Web 作業中執行。The event processor host runs in an Azure Web Job. 事件處理器 需要一個完整的週期來處理平均感應器值。The event processor takes the average sensor values for a completed cycle. 接著將這些值傳遞至定型的模型,以計算引擎的 RUL。It then passes those values to a trained model that calculates the RUL for an engine. API 可供存取 Machine Learning 工作區中屬於解決方案一部分的模型。An API provides access to the model in a Machine Learning workspace that's part of the solution.

機器學習服務Machine Learning

Machine Learning 元件使用的模型衍生自從真實飛機引擎收集而來的資料。The Machine Learning component uses a model derived from data collected from real aircraft engines. 您可以從 [ azureiotsolutions.com ] 頁面上解決方案的磚,流覽至 [Machine Learning] 工作區。You can navigate to the Machine Learning workspace from your solution's tile on the azureiotsolutions.com page. 當解決方案處於就緒狀態時,就會出現此圖格。The tile is available when the solution is in the Ready state.

Azure Machine Learning 模型可以作為範本,示範如何運用透過 IoT 解決方案加速器服務收集而來的遙測資料。The Machine Learning model is available as a template that shows how to work with telemetry collected through IoT solution accelerator services. Microsoft 已根據公開可用的資料[1],建立飛機引擎的回歸模型,以及如何使用模型的逐步指引。Microsoft has built a regression model of an aircraft engine based on publicly available data[1], and step-by-step guidance on how to use the model.

Azure IoT 預測性維護解決方案加速器會利用從這個範本建立的迴歸模型。The Azure IoT Predictive Maintenance solution accelerator uses the regression model created from this template. 此模型會部署到您的 Azure 訂用帳戶,並透過自動產生的 API 提供。The model is deployed into your Azure subscription and made available through an automatically generated API. 此解決方案包含 4 部 (全部共 100 部) 引擎測試資料與 4 個 (全部共 21 個) 感應器資料流的子集。The solution includes a subset of the testing data for 4 (of 100 total) engines and the 4 (of 21 total) sensor data streams. 此資料足以從定型的模型提供精確的結果。This data is sufficient to provide an accurate result from the trained model.

[1] Saxena 和 K. Goebel (2008)。「Turbofan 引擎降低模擬資料集」,NASA Ames Prognostics 資料存放庫( https://c3.nasa.gov/dashlink/resources/139/),NASA Ames Research 中心,Moffett field 欄位,CA[1] A. Saxena and K. Goebel (2008). "Turbofan Engine Degradation Simulation Data Set", NASA Ames Prognostics Data Repository (https://c3.nasa.gov/dashlink/resources/139/), NASA Ames Research Center, Moffett Field, CA

後續步驟Next steps

您現在已看到預測性維護解決方案加速器的主要元件,您可加以自訂。Now you've seen the key components of the Predictive Maintenance solution accelerator, you may want to customize it.

您也可以探索 IoT 解決方案加速器的一些其他功能:You can also explore some of the other features of the IoT solution accelerators: