Dark Factory - Smart Manufacturing of the Future

Meena — Fri, 27 Sep 2024 15:59:50 +0000

Industry 4.0 is buzzing word where manufacturing and information technology have joined hands together for digitalization. The People, machines and systems communicate in near real time with help of improvement in IoT field to enable autonomous production.

One of the leading countries in digitalization has taken the leap in implementation of smart manufacturing with all trending technologies. This country has 2700 engineering firms manufacturing essential good in various industries like medical, aerospace and semi-conductors.

The disruption in digital technologies had paved the way for dark factories which is termed as fully automated lights-out environment and is reality with Xiaomi’s dark factory. We term this as factory of future and with advancement of Robotics and established software like MOM (Manufacturing operations management), the complete lights-out factory conceptualization to implementation is very much possible.

Businesses where human interferences are dangerous can reap the benefits of complete automation. Imagine scenarios of oil rig where spilling has to be measured in intricate places where sending humans is risk or consider scenarios where the condition to monitor causes health hazards for human. We can utilize robots to measure, monitor and intervene. Let us go through few use cases where dark factory concept can be applied.

Most of scenarios involving Inspection cases where Robots can be helpful are as follows:

Identification of defects, corrosion, oil spill from large rigs
Damages
Erosion
Broken parts causing production inefficiency in the lines
Rust and many such defects depending on the product.

Safety surveillance with Robots involves below scenarios:

Intrusion detection - falling of objects from dark factory
Trespassing danger zone
Spill of harmful liquids and gases with temperature differences
Machineries loose/damaged parts which causes harmful impacts to complete factory unit

Below is the model Robot from Boston Dynamics that can be utilized for these use cases of dark factory implementation:

     Photo credits to Boston Dynamics

Let’s see how we implement this in AWS and details of the solution architecture.

Spot from Boston Dynamics can capture local imagery with either its standard robot cameras, or with other custom cameras as per our preference.

The captured imagery data can be inferenced by CV (Computer Vision) ML model. Detections from the ML inference can be stored in robot’s storage, until Spot reaches an area where a network connection is accessible. Technologies used for implementation are AWS IoT Greengrass 2.0 and Amazon SageMaker Edge Manager. ML models are deployed to a Spot that operates with intermittent network connectivity to Edge since it will be travelling to intricate places.

Data from Spot camera feeds are continuously streamed through device gateway to AWS IoT Core. We have considered edge gateway that connects to cameras and the device data are directed to AWS IoT Core. Edge gateway has AWS Greengrass runtime and lambda functions, and the photos are transferred from Edge’s storage to AWS S3 through lambda functions.

The feeds and real-time photos are stored in time series folder in AWS S3 and the location of photos along with metadata are stored in DynamoDB. Admin dashboards are created using Grafana where alerts as well as quality of images that has been predicted can be constantly viewed by the administrator.

Historical data about the manufacturing site with millions of photos are captured. The real time photos that are stored in S3 as well the historical images provided by our customer in S3 are analyzed with AWS Glue. The curated data is stored in AWS S3 for labelling and AWS SageMaker labeling job is created where AWS GroudTruth is utilized for labeling. The labeled data is stored again in S3. This data is split into training (80%) and test (20%) data. The CV algorithm which was custom built for our scenarios are trained with data stored in S3 using GPU. Once the training is completed, the model is tested for accuracy with test data.

The model predictions are validated and once the desired accuracy is achieved, the model is published to Spot for inferencing. The models are inferenced, and results are sent to AWS IoT Core.

Unveiling the myth of Unified Namespace (UNS) and implementing with AWS Native Services

Meena — Fri, 27 Sep 2024 06:55:41 +0000

A unified namespace is termed as single source of truth or golden record for Industrial IoT data. It is based on ISA 95 model standards combined with the way the data is contextualized and stored in database with modern data platforms set up. It is a centralized repository with hierarchical data about the IIOT system.

In a typical IoT manufacturing system, the devices are named differently in each of the systems like MES, SCADA, ERP and Historian. Providing a unified namespace across all systems replacing ISA-95 hierarchy improves the system performance by cutting out the middlemen.

The inspiration of Unified Namespace is from distributed architecture paradigm where instead of having data existing in silos across all layers of technology, access it a unified way so that it remains the same everywhere. The naming convention should also reflect the business from top level down to a device in shop floor. Also, the name should reflect all events that had happened in the business to a particular device. To consider an example, the hierarchy should reflect the organization, division, Production line, device name and events such as drop in production line efficiency in a device. All of it together is considered as Unified Namespace.

Ideally the communication protocol is MQTT and traditionally this MQTT points to a central repository of information that has the hierarchical enterprise structure, and it used to publish the events for that data. So, the most common pattern is to implement UNS in an MQTT broker. MQTT broker receives messages published by clients, filters the messages and publishes to subscribers. Though it is easy to implement MQTT since the events are handled, it does not hold the history of information which mandates us to choose a database or data warehouse. A database can hold historical data, but mainly it should contextualize the data to act as Unified Namespace.

Let’s see the logical architecture for typical UNS implementation:

Now that we are familiarized with the concept of Unified Namespace, it is also essential that we need to build a comprehensive real-time modern data platform as depicted in the logical architecture to bring single source of truth for the industrial data and store it in a database to consider it as UNS database.

The AWS Native Architecture supporting UNS implementation is as below:

In the above architecture, we have devices, MES/ERP, PLC, SCADA and Historian constantly streaming data from devices to AWS IoT Core where AWS IoT Greengrass acts as an edge gateway. The edge gateway is built with the capability of sending real-time data directly to AWS IoT Core.

An IoT Core rule is configured to stream the data to AWS Kinesis streams and performs in-stream processing and transformation using Kinesis Data Analytics and streamed to Kinesis Data Firehose. The data is standardized, contextualized, analyzed, and transformed using Amazon Kinesis services which serves as backbone for UNS.

Then the real-time transformed data is stored in Amazon TimeStream for Influx DB through Kinesis Data Firehose, and this is considered as Unified Namespace Database.
Real time dashboards are created in AWS Managed services for Grafana.

Raw telemetry data is stored in Datalake(S3). The historical data is transferred from the On-premises historian and data stores to AWS S3 through AWS DataSync where DataSync agents are installed On-premises to do the job. Data in S3 is cleaned, processed, and transformed using AWS Glue. This transformed data is then used to create reports and dashboards in Grafana. The data is utilized for predictive analytics where ML models are deployed in AWS SageMaker.

AWS CloudWatch is configured with the required metrics for consistent monitoring. CloudWatch dashboards are created, and the infra engineers can monitor the complete AWS native services.

DEV Community: Meena

Dark Factory - Smart Manufacturing of the Future

Unveiling the myth of Unified Namespace (UNS) and implementing with AWS Native Services