Enhancing Operational Insights for Industrial Assets with AWS IoT AIML Solutions (Part 1)

More Info

In Advanced (300), Amazon Lookout for Equipment, AWS IoT SiteWise, Energy, Industries, Internet of Things, Manufacturing, Technical How-to

Organizations responsible for the upkeep of industrial assets consistently aim to optimize their efficiency through diligent monitoring and analysis of asset health. Operators often assess efficiency using key performance indicators (KPIs) such as overall equipment effectiveness (OEE) or mean time between failures (MTBF) and strive to enhance these metrics at scheduled intervals. Ideally, actions like recalibrations or replacements would only occur when there’s a clear justification for doing so. Simultaneously, the operational technology (OT) team schedules maintenance during periods that least disrupt production. Acting prematurely can waste resources, while delaying action risks unplanned downtime. Therefore, customers seek solutions that automate asset monitoring, learn from historical performance issues, and deliver actionable insights to maintain optimal KPI standards.

A condition-based monitoring solution that integrates Internet of Things (IoT) and machine learning (ML) can significantly bolster the OT team’s capacity to achieve their KPI objectives. This type of solution tracks machine telemetry data in real-time and predicts KPI anomalies, allowing maintenance to be scheduled only when necessary. Such solutions can notify OT teams of abnormal performance and provide insights into root causes by analyzing past data, thereby creating opportunities to prevent issues before they disrupt operations.

However, there are two main challenges to tackle when designing a condition-based monitoring solution:

1. Data Storage and Management: The immense volume of data gathered from sensors, along with equipment and site metadata, must be effectively stored and organized.
2. Scalable and Accessible Advanced Analytics: Developing multiple ML models for various equipment types and integrating them into IoT platforms for condition-based maintenance can be daunting.

These challenges can hinder insights derived from AI solutions and add complexity for teams that are already tasked with managing hundreds of industrial assets by introducing a ML aspect to asset management.

In this two-part series, we will explore how AWS IoT assists customers in overcoming these fundamental challenges. We will specifically address data storage and analysis obstacles, illustrating how to deploy a solution that:

• Collects, organizes, and monitors data from industrial equipment at scale using AWS IoT SiteWise. This service allows multiple sensors to be structured with asset models and hierarchy levels, making it easier to consume data for ML model training.
• Swiftly detects and diagnoses equipment abnormalities to minimize costly downtime with Amazon Lookout for Equipment. The OT team can leverage automated ML to create multivariate ML models for complex industrial assets, achieving near-continuous monitoring effortlessly.
• Integrates inference outputs from Amazon Lookout for Equipment with AWS IoT SiteWise, enabling the OT team to swiftly identify issues at the component level of industrial assets. Additionally, the AWS IoT SiteWise alarm feature provides automatic notifications of anomalies, facilitating informed maintenance decisions.

Solution Overview

AWS IoT SiteWise is a managed service designed to simplify the collection, storage, organization, and monitoring of data from industrial equipment at scale, empowering you to make better-informed decisions. With AWS IoT SiteWise, you can manage operations across multiple sites, easily calculate industrial performance indicators, and develop applications that analyze equipment data to prevent costly failures. Consolidated data allows for consistent collection across devices, quick issue discovery through remote monitoring, and effective multi-site management.

Amazon Lookout for Equipment automatically analyzes data from equipment sensors to train an ML model tailored to your equipment, requiring no data science expertise. It processes incoming sensor data in real-time, accurately identifying early warning signs that could lead to declines in health metrics like OEE or MTBF. This facilitates quick and precise anomaly identification, efficient problem diagnosis, timely actions to avert expensive downtime, and reduced false alarms.

In this solution, we illustrate the integration of these complementary AWS managed services for near-continuous monitoring and alerting of a simulated pump station with two assets. Each asset is a pump used to transfer fluid by converting rotational energy from a motor into hydrodynamic energy.

Customers can follow the outlined steps in this blog to develop a solution that optimizes their industrial assets. The outcome is a real-time dashboard that:

• Enables large-scale real-time monitoring with alarm notifications.
• Provides in-depth diagnostics of individual components within an industrial asset fleet, allowing the OT team to perform maintenance with a clear understanding of the root cause.

For more insights, check out another blog post here that delves deeper into this topic. The dashboard figure shows that pump #2 is currently in alarm, indicating the sensors most associated with the detected anomaly.

Measurements were recorded across the four main components of the centrifugal pump: impeller, shaft, motor, and volute. For sensors not located on these components, they are categorized under a general “pump” label. Sensors 0-5 belong to the pump level, sensors 6-11 are associated with the impeller, sensors 12-17 with the motor, sensors 18-23 with the volute, and sensors 24-29 with the shaft.

The solution encompasses:

1. A “SiteWiseSimulator” AWS CloudFormation template that includes essential workflows:
   • Creating AWS IoT SiteWise asset models for the pump station and pump, and defining their hierarchical relationships.
   • Establishing an AWS IoT SiteWise alarm model to facilitate automatic alert notifications for anomalies.
   • Creating AWS IoT SiteWise assets based on previously defined models and enabling MQTT notifications for data streaming to Amazon Simple Storage Service (Amazon S3).
   • An AWS Lambda function to periodically write sensor data to AWS IoT SiteWise using the BatchPutAssetPropertyValue API call.
2. An Amazon Lookout for Equipment workflow utilizing Amazon SageMaker notebooks:
   • Training Lookout for Equipment ML models.
   • Setting up an inference scheduler for nearly continuous monitoring of multiple assets.
3. A “l4esitewise_pipeline” AWS CloudFormation template containing a data engineering pipeline for integrating Lookout for Equipment with AWS IoT SiteWise:
   • Streaming AWS IoT SiteWise data to S3 in near-real-time.
   • Utilizing a Lambda function to transform raw telemetry data from AWS IoT SiteWise into the format required by Lookout for Equipment on a set schedule (refer to l4einference-schedule.zip).
   • Employing a Lambda function to relay inference results from Lookout for Equipment effectively, making it about the same overall length.

For further reading, check out this excellent resource that discusses common pitfalls in Amazon’s implementations. Additionally, for insights on the topic, this source is highly regarded.

SEO Metadata