IIoT vs ML vs AI vs Big Data vs Edge Computing
The Industrial Internet of Things (IIoT) is not synonym for Machine Learning (ML). Neither is it a synonym for Big Data. Neither is it a synonym for Artificial Intelligence (AI).
Generally, Machine Learning requires large amounts of data, aka Big Data. Machine Learning may exist within IIoT and utilize edge devices. But ML and AI operate on data, not devices. In fact, ML and AI should be utilizing data which has been prepared, cleaned, and verified – not on data directly out of a sensor.
The scope of the IIoT industry clearly includes Big Data, ML, and AI. But IIoT enables ML and AI to assess a task, utilizes the results from ML and AI to optimize a task, but does not necessarily require ML and AI to accomplish a task.
Frequently, interconnecting IIoT edge-point devices via networks is often grouped into the narrower genre of edge computing. However, edge computer requires computational capabilities. IIoT edge-point devices do not necessarily require edge computing. Certainly in this age to digitization and digital transformation, these distinctions are somewhat immaterial. But the distinction does exist.
IIoT edge-point in isolation is no longer relevant. As the lines of demarcation blur, we ask ourselves:
- Why would a project implement a sensor unless that sensor is producing data that should be monitored and recorded?
- If sensor devices are to be monitored and recorded, why would that project not utilize automated and autonomous acquisition, assessment, and archival?
- Once acquired, shouldn’t a project utilize that data to affect other processes, alert when anomalous situations exist, and be utilized in some way as a forecasting input?
[As an aside, the concept of the Internet of Things carries the preconception of consumer devices. The Industrial Internet of Things add clarity that the ‘things’ are industrial components, the internet is industrially focused, and the project is industrial in nature. I exclude the broader categories of IoT. Communications between the toaster and refrigerator are not, directly, relevant to the Industrial Internet of Things. “The IIoT is different from other IoT applications in that it focuses on connecting machines and devices in industries such as oil and gas, power utilities and healthcare.” (1)]
Relevance to Industry
The relevant point to be made is that IIoT , Edge Computing, ML, and AI are closely linked and often integrated. Nevertheless, disciplines focusing on IIoT, Edge Computing, and Machine Learning require different skill-sets for the practitioners, different design methodology for integration, and different back-end infrastructure.
Engineering Skill-sets Required for IIoT Projects
Industrial Internet of Things is, of necessity, comprised of end-point devices such as sensors, annunciators, actuators, and terminals. As a practical nature, IIoT also includes edge power, cabling, networks, and mobility. These skill are more decidedly focused toward the electrical engineer than other individual engineering disciplines (chemical, mechanical, etc.).
An electrical engineer must understand the capabilities of the individual devices, installation requirements of the devices, the environmental and mechanical limits of the devices, and the installation requirements of the devices. Additionally, maintainability, serviceability, reliability, and availability are key concerns.
[Perhaps I’m partial to my chosen discipline – Electrical Engineering. I make no apologies.]
The IIoT engineer should be fluent with the entire specification of the devices and components, able to design an integration of devices, and savvy in computers, networks, and security.
As a checklist, the IIoT Engineer should be demonstrably proficient in
- Sensors, sensor technology, sensor specifications, and sensor integration
- User interfaces – mobile devices, operations centers, and field annunciators
- Install-ability, Serviceability, Maintainability, Reliability, and Security
- Units of Measure, Calibration, Sampling Rates, Filtering (hardware filters and software filters)
- Data organization, storage, retrieval, and flow
This proficiency cannot exist in isolation from skills in ML, AI, Big Data, and Analytics. Finally, the IIoT must be conversant with the universe in which the IIoT must exist – especially the nature of the operation, the concerns of the field personnel, and the vernacular of the operation.
Design Methodology Required for IIoT Projects
An IIoT project is rarely completed. The first phase may come to an end, but one phase is rarely completed before the next phase begins. A design methodology is crucial.
The methodology chosen for an IIoT project should address
- First and foremost: What is the objective? Why is an organization undertaking and IIoT project of any scale and complexity? When completed, what benefit will be shown to the executive leadership?
- Don’t eat the entire elephant at one sitting. Identify specific near-term, small, demonstrable, and achievable milestones. Early success will ensure the embrace of leadership. However, nonsensical demonstrations of remote control lights and locks will equally ensure the alienation of necessary backers.
- Ensure the IIoT components are robust. In the extreme, installation or wire-wrapped demonstrator devices will be prone to fail – at the wrong time. Implement individual IIoT points, sensors, actuators, processors, gateways, etc with the care to be given productionized solutions. Particularly appropriate is the environmental fitness of the installation (explosive environment, extreme temperature, caustic chemicals)
- Address signal quality. Any electrical engineer will know that the signal to noise ratio must be of sufficient distinction to provide credible data (often an SN ratio of 3 or more is recommended). Additionally, ensure point source filtering is implemented to prevent spurious data, but not so filtered to prevent outliers.
- Consider the end user. Great ideas have failed no because of the concept, devices, quality and robustness of the design, but because the user’s ability to consider the new information source is ill-conceived. Too much data and not enough information will sink a project. If the question is “have you been watching the display?”, then the project has failed.
- What behavior change is expected of the user or the system based on the IIoT project?
- Is the solution robust, maintainable, serviceable, and autonomous?
Beyond the premier consideration of these concepts, the design methodology must iterate and expand well.
Infrastructure Required for IIoT Projects
IIoT infrastructure will cross most of the organization. Specifically:
- Field personnel to install, service, and calibrate the system.
- Communication reliable and resilient including the ability to back-fill data gaps after communication outage.
- Data retention including short-term raw volume, long-term decimated volume, behavior modelling and thumbprint, and KPI alerts.
- Streaming data processing in 4 modalities: real-time monitoring, after-action or forensic look-back analysis, pre-action or workflow planning analysis, and large-field analytics and machine learning.
- Geographic dispersion of IIoT devices, users, and systems.
The Discipline of IIoT Engineering is the Systems and Knowledge Engineer. This individual or functional group must be a rare mix of engineer, data engineer, integration engineer, and user interface engineer. IIoT engineering will provide the project management, architectural design, and integration skills. IIoT engineering must be conversant with the industrial subject matter experts (SME), vendors, executives, developers, and technicians.
(1) What is the Industrial IoT? [And why the stakes are so high], Jon Gold, Senior Writer, Network World, February 2, 2018