5 Key considerations for intelligent edge initiatives

5 Key considerations for intelligent edge initiatives

Many organizations across industries are looking for opportunities to leverage the latest technologies to improve their business operations. Harnessing the power of the Internet of Things (IoT) and applying intelligence at the edge are two popular areas of interest these days. Knowing how to approach an intelligent edge initiative to extract business value is what provides a competitive advantage to firms adopting these technologies.

When embarking on an initiative to leverage edge computing, we find that there are several key areas that need to be considered to be successful. While many firms have a stance on one or more of these areas, we find that most organizations do not have a comprehensive plan for each of these.

Here at Neal Analytics, we work in collaboration with our customers to assess their existing environments, help them build implementation roadmaps, and provide a managed team to realize their intelligent edge initiatives. We have helped organizations across the maturity curve achieve success and improve their business outcomes.

Through our experiences, we have collected several considerations that must be addressed in order to achieve optimal results. Let’s take a look.

5 square icons for 5 key considerations intelligent edge initiatives

5 consideration areas for intelligent edge initiatives

  1. Solution goals
  2. Data collection strategy
  3. Operating environment
  4. Workload inventory
  5. Scale factor

Solution goals

Solution goal_1

The first step in building your edge or IIOT capability is determining which business levers you are trying to impact. Here are a few examples to illustrate common objectives related to IoT & edge.

  • Leverage computer vision to analyze customer activity (foot traffic, dwell time, etc.)
  • Improve in-store inventory management by identifying stockouts and analyzing on-shelf availability trends
  • Improve workplace safety compliance by detecting PPE and relation of persons with heavy machinery
  • Assess product quality and leverage deep reinforcement learning at the edge to optimize production processes

Once you have defined the business objectives, you will need to determine how best to prove the value of the technology to solving the business problem. In some cases, the business problem can be solved in simpler ways with other technologies. In the remainder of cases, we look to leverage the capability of intelligent edge computing to impact the business in new and meaningful ways.

In defining the goals of your initiative, you will need to determine the best mechanism for proving the business value with the technology. If just beginning in this space, value must be realized early on to secure additional funding from the business. It is also common to see InfoSec compliance hurdles come into play with new technologies that must be understood and approved for use.

It is worth considering if a virtualized Proof of Concept in the cloud or a “test drive” experience is a possible mechanism for achieving the sign off from the business as well as the InfoSec team. This approach can reduce the time to value by realizing a quick win for the business upfront and enabling additional investment.

Here at Neal Analytics, we commonly partner with our customers to design and implement standardized architectures which can be vetted by IT and are tailored to solve the business need in a scalable manner.

Data collection strategy

Data collection strategy_2

Do you really need all the data?

Organizations that are just getting started often fall into the trap of collecting too much data. They don’t necessarily understand which data points are relevant, or at what frequency the data is best collected for building out their intelligent solutions. Consequentially they end up collecting the most granular datasets available, which often reports data unnecessarily.

Meaningful data

One of the key benefits of edge computing is reduced latency and the ability to process data where it is generated, and only sending the meaningful data to the cloud. Organizations need to define what is meaningful for your business scenario, (whether this be an alert of an event, data points captured while a process was out of tolerance thresholds, or summarized data for reporting). The deployed solution should only collect data relevant to the insights that are generated as well as the information required for maintaining the solution throughout its lifecycle.

Network constraints

Another reason to consider the footprint for data collection is network constraints. It is commonplace for organizations to have a fleet of remote assets which have limited network connectivity with limited bandwidth and data transmission caps. This necessitates a clear strategy on how to best utilize the available network capacity to enable the required data processing within the constraints of your environment.

In these cases, we have helped our customers realize success by being more selective about which data points are captured, only reporting data on change, or by performing more processing at the edge prior to routing the summary data to the cloud. These are a few techniques that can be utilized for operating in remote site contexts.

Data required to build ML and AI models

This leaves the question of data required for building machine learning and AI models to be run at the edge. More data is always better for creating solutions that will generalize best to real world operating environments. Rather than turning on the firehose for granular edge to cloud data collection, we recommend collecting a historical dataset for building out your intelligence.

In some cases, you might have the historical data stored on site in a historian database or something similar. In other cases, you may find that data needs to be simulated or collected systematically in a real-world environment. We find this is a common pattern when working with specialized computer vision edge use cases.

In these scenarios we help to identify optimal positioning of cameras and collect training images through a variety of simulated world contexts. With the initial training data in hand, the business can proceed with the aforementioned strategies for managing data collection on an ongoing basis.

Operating environment

Operating environment_3

One of the most critical areas when designing your edge solution is the operating environment. There are many different dimensions to consider in this space according to industry and use case. We tend to group the operating environment considerations into the following areas.

General, Sensors / Cameras, Edge Compute, and Network

General – This is where you consider the basic operating characteristics. Will the solution be running in a retail setting? On the manufacturing floor? Does the introduced equipment need to be located indoors or outdoors? Will the equipment be subject to harsh operating conditions? How will personnel consume the outputs of the platform? Is there an obvious need for ruggedized equipment? Allowable placement locations for new equipment (sensors, cameras, edge servers, etc.)?

Sensors / Cameras – What sensors / cameras will need to be ingested? How many unique data source inputs can be expected per location?

Edge Compute – Is this a good fit for light edge or heavy edge? What is the frequency that data needs to be processed? Does the scenario necessitate the use of hardware acceleration (GPU, FPGA, VPU, etc.)?

Network – what network speeds can be expected? Are there any limitations on network usage during peak times? Will the edge server have limited connectivity? Is there a maximum allowable tolerance for bandwidth utilization?

Workload inventory

workload inventory_4

When designing an edge strategy, it is important to consider not only the quick win Proof of Concept, but to keep in mind the broader implications of edge computing. As such, you’ll want to keep an inventory of scenarios from short-term to long-term to understand how the footprint of your edge compute could evolve.

Investments in hardware need to be planned with the future in mind, as to not be caught short sighted with not enough compute power in the future. We typically go through a workload inventory analysis at the outset of most engagements for this exact reason. This pre-planning allows us to strike the right balance in deciding the appropriate hardware as well as the overall solution topology (i.e., light edge, heavy edge, nested edge, etc.).

The scale factor

scale factor_5

Taking an intelligent edge initiative requires collaboration from IT + OT + Business. Involving all parties early in the process and aligning objectives will lead to the most successful outcomes.

After realizing the initial value, the solution typically needs to be hardened before scaling. Once the framework has been hardened, it is typically scaled up and out. We can scale up scenarios (add more sensors, cameras, etc.) after the initial value is proven. It will also need the ability to scale to new facilities (typically following a crawl, walk, run approach).

In addition, the platform needs to be able to scale to new scenarios (see workload inventory).


Need help or still unsure where to start?

We provide a comprehensive edge assessment to understand each of these areas and assist with building out a roadmap for realizing your intelligent edge aspirations. Contact our team today.

Learn more