Technical Specifications
Introduction
Brief description of the technical specs.
AI-PRISM System Deployment
Description of deployment procedure. Refer to Diagram
AI-PRISM System Configuration and Requirements
High-level features of AI-PRISM as a whole.
Development Environment
Description of development workflows and tools.
Detailed Technical Specs
Architecture Diagram
The following diagram shows the high-level architecture of the AI-PRISM system.
As it can be seen in the previous diagram, the architecture is very elastic in th sense that it allows us to place the workloads in distributed clusters and manage them centrally from the CI/CD infrastructure. The actual distribution of workloads into the different clusters in use case specific.
The Non time-critical cluster is the location for those modules that wouldn't require Real-Time response, which can be physically located in a private cloud cluster. Tasks that may require long execution times and powerful processing units, as well as large data banks, as such we delegate them to the Cloud. The data Platform has a clear central rol in the cloud, as it gets constantly updated by receiving the fresh data from the industry, and the different cloud modules treat that data in order to keep improving the particular ground algorithms. Data related to the environment and the way it is perceived can also be constantly improved to have a better and more precise understanding of it.
The Time-Critical cluster is the location for those modules that require Real-Time response. Here we have the b ase AI-PRISM ROS Image with all the nodes and packages involved with it. The ambient digitization modules are recommended to be integrated within the ROS environment since the bast majority of data received an generated will be in ROS format. If this task has some real time requirements that cannot be fulfilled by the network infrastructure, we can put in the Jetson Nano, downstream, in order to treat the data as soon as possible. The real-time SDN network provides deterministic communications ensuring latency and QoS for all communications. In typical scenarios we will use Ethernet calbes because, but other TCP/IP infrastructure is supported, including wireless communications. ROS supports communication using any of the following methods: TCP/IP, UDP, Serial, Bluetooth, Wi-Fi, CAN, MQTT.
The Fog cluster is responsible for enabling communication between edge devices, and also for processing and analyzing data locally. Fog nodes act as intermediates reducing latency and network traffic by performing computation and storage closer to the edge device, enabled by real time communications. the IIoT Platform is deployed in the fog cluster, central cluster, either in a single fog node or in redundant nodes. It acts as a hub or control plate that facilitates communication and coordination between the distributed edge devices. It provides a messaging infrastructure and a central point of aggregation for data. components that require real time communications (AI-based level enhancing modules or ambient digitization modules) can also be located in the fog cluster, sharing resources with the core communications modules and ROS framework, or in a distributed for architecture pattern.
Far-edge nodes are single-node (K3s) clusters that are managed centrally. In scenarios where no real-time network communications are deployed, the far edge clusters can host critical workloads like ambient digitization modules, or agent level reasoning, to take advantage of the resources available at the edge locations. Any device that is used to capture the environment would fall into the "Ambient Sensing Infrastructure" group and, depending on their requirements, we can control one or more than one in each far-edge node. A camera (or set of them) perceives the workspace, this including: the human, the robot, the object that need manipulation as well as any external relevant body.
We assume that any actuator falls into the group "Base Hardware". A worker picking springs from a conveyor belt, handing them to a robotic arm, with a particular orientation. The robot picks them from the human hand, and places them in a hot furnace.
Components Specifications
The following table contains links to the detailed specifications of each component in the reference architecture.
| Acronym | Link |
|---|---|
| TM | Template |
| RF | RF Documentation |
| BH | BH Documentation |
| CM | CM Documentation |
| AS | AS Documentation |
| RC | RC Documentation |
| IP | IP Documentation |
| DS | DS Documentation |
| SE | SE Documentation |
| AD | AD Documentation |
| CD | CD Documentation |
| PE | PE Documentation |
| DR | DR Documentation |
| CR | CR Documentation |
| HI | HI Documentation |
| PD | PD Documentation |
| SP | SP Documentation |
| NS | NS Documentation |