Bringing Intelligence to Industrial Manufacturing Through AWS IoT and Machine Learning
By Daniel De Alday, Software Engineer – TensorIoT
By Long Tran, Data Solutions – Reliance Steel & Aluminum Co.
Beginning with the industrial revolution in the 18th century, manufacturing has long been the backbone of the nation’s economy and wealth.
But as U.S. industries have modernized and moved into the realm of connected devices that leverage data to make machines more intelligent, industrial manufacturing has remained relatively unchanged up to this point.
Why does this matter? What pitfalls does a company face if they don’t keep pace with technology?
There is currently a revitalization movement in the works that’s bringing many industrial manufacturing companies into the connected devices era. Powered by Amazon Web Services (AWS), organizations can take advantage of cloud technologies and reap the benefits offered by connected devices.
In this post, we’ll look at how intelligent machines can lead to significant cost savings for industrial manufactures. With connected Internet of Things (IoT) solutions built on AWS, businesses can be more proactive with maintenance instead of reactionary, allowing them to fix problems with machinery before they become critical.
Reliance Steel & Aluminum Co. teamed up with TensorIoT to solve for this use case. Together, we built an IoT solution on AWS that ensures the maintenance needs of Reliance’s industrial machinery are anticipated and that machines can be serviced before breaking down.
What is a Connected Device?
A connected device is a “thing” that is connected to the internet. They can be everyday objects, systems, or pieces of machinery.
Connecting a device to the internet is not an easy task, though, especially for industrial machinery that was built during a time when the internet did not exist or was only in its infancy.
Machine monitoring and sensory data for machine job progress has generally been stored in-house and extracted for analysis once the job was completed.
With the introduction of cloud-based solutions, however, data is more accessible than ever and allows organizations to better monitor and analyze the optimal environmental conditions for heavy machinery with greater predictive results.
Who is TensorIoT?
TensorIoT is an AWS Advanced Tier Services Partner with AWS Competencies in Internet of Things (IoT) and Machine Learning (ML), among others, and holds AWS Service Delivery designations for AWS IoT Core and AWS IoT Greengrass.
With a focus on “Making Things Intelligent,” TensorIoT has the ability to build fully-functional solutions that meet the needs of customers:
- Their IoT team delivers custom solutions for device provisioning, custom alerting, and command and control of devices and OTA fleet updates.
- Their ML team leverages Amazon SageMaker to build and deploy custom models for forecasting, personalization, and object recognition.
- An experienced web and mobile development team enables customers to leverage the data and insights generated from cutting-edge solutions.
TensorIoT has embraced AWS managed machine learning services, such as Amazon Alexa, Amazon Rekognition, Amazon Forecast, Amazon Textract, and Amazon Personalize. These can be leveraged to generate better business intelligence and enable users to access data wherever they need it.
Making Things Intelligent
It all starts with AWS IoT applications and solutions that offer broad and deep functionality spanning the edge to the cloud. With AWS IoT, you can build solutions for virtually any use case across a wide range of devices. Since AWS IoT integrates with artificial intelligence (AI) services, you can make devices smarter, even without internet connectivity.
Developing with AWS IoT not only provides analysis and monitoring but also a portal for creating web-based applications to retrieve historical data. AWS IoT scales as your device fleet grows, and it offers comprehensive security features so you can create preventative policies and respond immediately to potential security issues.
Industrial manufacturing takes many considerations into account to drive profit, including employee efficiency, machine health, and distribution of education and documentation.
With AWS IoT services, the application of environmental sensory data can inform business owners if their machinery is operating at optimal health. Healthy machinery is important to ensure that businesses are operating at maximum efficiency. Factors such as humidity, temperature, or vibrations (even in their minutia) can affect the long-term health of machinery.
Sensory data is accumulated through connected devices that measure these factors using AWS IoT, which provides a manageable and organized solution for even the most complex device hierarchies. Sensors consume environmental data that propagate through AWS IoT, redirecting data into other AWS services such as Amazon SageMaker, or providing live analysis through services like Amazon Kinesis.
The advantage of AWS IoT for industrial manufacturing is that predictive maintenance can be performed and anomaly detection can be monitored, providing an as-you-go update of facility health.
AWS IoT initiates data collection on operation conditions alongside environmental factors that are taken into consideration for the machines’ well-being. Post data collection on AWS IoT Analytics structures the data into meaningful information that can be used to automate processes in machinery.
Amazon SageMaker pulls data from the AWS IoT Analytics to create a frictionless solution, showcasing prognostics value early in the machinery lifecycle.
Making Data Intelligent
The process of making data intelligent is done via applications, and by combining different tools to change and enhance the way the data is perceived.
Typically, large pieces of machinery go about their business and do what they are designed to do. But we don’t really know or understand what’s happening inside the machine. When a machine ultimately fails, large amounts of time, money, and lost revenue can be expended to repair the machine and bring it back into service.
To solve for this, IoT sensors can be added to the machinery that look at discreet variables—such as vibration, noise, heat, and temperature—that give us a view into what’s happening inside the machine. This data is then delivered to the cloud where it can be consumed and analyzed.
The final step in making data intelligent is the introduction of machine learning. Having collected large mounds of data, we can train ML models to do the analysis for us. In our example of broken equipment, machine learning helps us to see what’s happened but also gives us a way to predict when other pieces of machinery will fail.
By being proactive with maintenance issues, the amount of time, money, and lost revenue an organization incurs can be reduced as repairs are made before the machinery gets to a critical point.
Connecting IoT Data with Machine Learning
Industrial IoT (IIoT) bridges the gap between legacy industrial equipment and infrastructure and new technologies such as machine learning, cloud, mobile, and edge computing.
Customers use IIoT applications for predictive quality and maintenance and to remotely monitor their operations from anywhere. AWS IoT services enable industrial companies to reason on top of operational data and improve performance, productivity, and efficiency of industrial processes.
The deployment of IIoT solutions does not need to be a daunting task, and AWS has a very handy toolkit called AWS Chalice that enables the quick deployment of API resources and serverless functions. These can be attached to a custom front-end user interface (UI), allowing customers to view the results from an IIoT solution they have adopted.
At Reliance Steel, AWS Chalice was used to manage the data ingestion relative to each of the steel facilities. The toolkit automatically created a set of API endpoints in Amazon API Gateway, and each of these were backed by an AWS Lambda function that was easily deployed to the environment.
From a different viewpoint, AWS Chalice can automatically set up HTTP microservices that can be combined with an IIoT solution and visualized on a customer’s UI. The managing of manufacturing site information, ticket orders, and IoT device controllers can be handled within AWS Chalice or through customers’ attached UI.
Making the Solution Secure
Following the complex business hierarchies of Fortune 500 companies is no easy task, and keeping data accessible between locations controlled is best handled with Amazon Cognito, which lets you add user sign-up, sign-in, and access control to your web and mobile apps.
In the IoT space, Amazon Cognito allows for complicated role-based access for command and control of devices, while still being able to allow those on the floor access to pertinent information for a particular facility.
A nexus of control and their permissions can gate access and registration of IoT devices from the other side of the continent when a requesting location is attempting to connect new monitoring devices. This allows for several different control views—administration, business, managerial, and technician—and enables those involved to have a consolidated login to a single application into which they can coordinate efforts.
Customer Use Case: Industrial IoT with Reliance Steel
Reliance Steel & Aluminum Co. started business in 1939 and has become the largest metals service center in North America.
Through a network of more than 300 locations in 40 states and 13 countries outside of the U.S., Reliance Steel provides value-added metals processing services, and distributes a full line of over 100,000 metal products to more than 125,000 customers in a broad range of industries.
Many of the machines that Reliance Steel uses to produce goods and materials have been in service for decades. When one of them fails or breaks down, it has multiple impacts on the business, including lower production quotas, lost revenue, and increased maintenance costs to bring the machinery back online.
By adopting a more preventative approach to maintenance using Industrial IoT and machine learning, Reliance could drastically reduce the down-time of broken machinery and possibly eliminate the production gap that offline machines can create.
To accomplish this, Reliance Steel teamed up with TensorIoT to connect its industrial machines to the internet and get telemetry data from the different sensors and loggers on the machines. This data can be ingested by AWS IoT Core and stored in a data lake so that it’s accessible to other components such as machine learning or a custom web UI.
Once the data was brought into the data lake, Reliance Steel uses Amazon SageMaker to do preventive maintenance on the machines based on the data that’s being brought in to the system.
How Preventative Maintenance is Completed within the ML Model
As the telemetry from the machinery is saved into the data lake, the machine learning model is provided with a wealth of information to look at, as well as make some statistical analysis and projections.
The ML model splits the information into two groups—one group is the data used for training, while the other group is for verification.
The ML processes through the data in the training group, looking for patterns and any correlations that can be determined from the data that’s been saved. An example of a correlation is that every time the vibration of the machine goes above 40Hz, the gears become misaligned and need to be realigned.
As a result of this being done within the model, the training of the ML model can look at hundreds of thousands of records and pull sequences of events such as anomalies, correlations and projections, to come up with a hypothesis.
Once the hypothesis has been created, it needs to be tested and validated. This is where the test group of data comes in. The test data is compared against what the model has “learned” and then validates if the model was trained correctly. If the model was not validated because the hypothesis was incorrect, then the training of the ML model in Amazon SageMaker automatically retrains the model and goes through the process again.
This is the complete loop of training for an ML model, and it can go through hundreds of these loops in order to come up with a model that’s been validated by the training data.
Once a model has be validated, it’s published as an executable endpoint. Next, the live steaming data can be passed through the trained model and make an inference about the status/health of the machinery, based upon what it already knows and has been trained to look for.
When the inference sees a recognizable pattern or correlation, it makes a projection as to when that piece of machinery is going to fail so that preventive maintenance can be scheduled and completed before the machinery breaks down.
Figure 1 – End-to-end Industrial IoT architecture.
In the diagram above, you can see how the device sensors are connected to the AWS IoT Greengrass core in order to transfer telemetry data to AWS, and more specifically AWS IoT Core.
Once the data comes into AWS IoT Core, it will be evaluated against an AWS IoT Rule to determine what to do with the data. The rule is used to fan out the data to other AWS services to make the most use of the data that’s been ingested. For the solution we developed for Reliance, the incoming telemetry data is split into three different paths.
In the first path, the data is stored in an Amazon DynamoDB table so the custom user interface that TensorIoT created for Reliance can retrieve and display the most current telemetry data to the user. This gives them a view of the machinery at a specific point in time.
Additionally, the incoming telemetry data is evaluated to see if any of the reported parameters have broken an established threshold. If so, AWS Lambda sends a message to Amazon Simple Notification Service (SNS), and then the service determines who the message should be sent to via text message or email, based on an individual’s preferences.
In the second path, the telemetry data is passed into Amazon Kinesis Data Streams which ensures the telemetry data has been received and stored in Amazon S3 for historical purposes. It also provides an accounting of the telemetry data from the machinery.
In the third and last path, the telemetry data is passed to AWS IoT Analytics, which transforms the data as necessary and passes it to Amazon SageMaker. There, the Amazon SageMaker-trained ML model looks at the incoming telemetry data and makes an inference based upon what the trained model has been taught. It then passes the results back to AWS IoT Greengrass, where it’s evaluated. Afterwards, command and control can be exerted onto the piece of machinery.
As you can see from our solution, the telemetry makes a full circle from machinery that AWS IoT Greengrass attached to it, through the AWS Cloud and its services, and then back to the machinery. This feedback loop ensures the current status and health of the machinery is constantly known and updated with the most current values.
AWS is also developing new services and features all the time, which can be incorporated into existing workloads to drive additional value. AWS IoT Things Graph, for example, makes it easy to visually connect different devices and web services to build IoT applications.
A perfect Industrial IoT solution is made up of four components: data, IoT architecture, machine learning, and business intelligence.
Without data, there is no way of gaining actionable insights for your business. The IoT architecture consists of connected devices (sensors) as well as the ingestion and storage of data into a place where it can be consumed and analyzed.
Machine learning is what makes the data “intelligent” and helps companies both understand and make predictions based on data patterns produced by the ML models. Lastly, business intelligence tools help visualize the data so companies can make informed decisions.
Reliance Steel had a vision of becoming a company that approached maintenance in a more proactive way. Doing so would help it reduce costs and expenditures while ensuring quotas were met or exceeded.
By partnering with TensorIoT to develop an Industrial IoT solution, Reliance Steel brought innovation to its machinery through the implementation of modern architectures, sensors for legacy augmentation, and machine learning to make the newly acquired data intelligent and actionable.
The content and opinions in this blog are those of the third party author and AWS is not responsible for the content or accuracy of this post.
TensorIoT – AWS Partner Spotlight
TensorIoT is an AWS Competency Partner founded on the instinct that the majority of compute is moving to the edge and all things are becoming smarter.