IoT is trending. Companies invest in new tech...
Factories and other industrial units equipped with sensors is a remarkable feature of the closest future, characteristic of the full-scale Industry 4.0. deployment. However, this abundance of data needs special treatment and a well-weighed approach to its management, otherwise it will leave businesses in the “drowning in data yet starving for information” mode.
Raw data produced and processed by sensors, actuators and other connected elements of IIoT systems, is too complex and too diverse, arrives too fast and changes too quickly to be used effectively by humans.No wonder that more than a third of executives in the Forbes Insights survey said they were not able to leverage large amounts of data they were collecting.
Hence putting the extracted data into action, i.e. into data-driven decisions for a company, is another principal aim of Industrial IoT, and doing proper analytics is the first step towards it.
Data Analysis in Industrial IoT: Turning Data into InformationSensors are turning ordinary machines into context-aware, conscious, and self-learner devices. They are capable of capturing, storing and processing multiple pieces of data - but what next?
While it is expected that by 2020 one-tenth of the world’s data will be produced by machines, it is not the amount of collected data which matters but efficient data analytics that will provide essential insights.
“IoT is creating an unprecedented amount of data in the enterprise in terms of both volume and velocity,” says Mark Hung, research vice president at research firm Gartner. “In order to extract value out of this data, the enterprise’s data analytics architecture needs to be revamped.”
What Happens to Data Once It Is Collected In a IIoT Network?To make use of the collected data, certain steps are to be taken before the data is distributed to applications and end users. For this, data needs to be analyzed and correctly processed.
Generally speaking, data analytics happens at each stage following data collection.
- Once data is captured, it is preprocessed to see what type of data is to be further processed and analyzed.
- When at the edge, i.e. when it is processed by a device and/or a gateway close to the data capturing device, the data is analyzed for quick and time-critical cases.
- Once it is sent over an enterprise network to the cloud, it is stored there and can be deeply analyzed for the needs of preventive analytics and improved outcomes.
Doing Data PreprocessingAll sorts of data are collected via multiple end nodes. Typically, it presents a stream of information, i.e. raw data, juxtaposed against time. This real world data is usually incomplete (i.e. it contains missing values), noisy (containing errors or outlier values which deviate from the expected), and inconsistent (containing duplicate values or unexpected values leading to inconsistency). Hence, certain actions need to be taken to sort out these streams of data, and data preprocessing is a proven method to accomplish it.
Data preprocessing comes in one of the following forms:
- Data cleaning, i.e. correcting the collected data, filling in missing values or resolving inconsistencies.
- Data integration, i.e. combining data from multiple sources into a coherent data store.
- Data transformation, i.e. making data appropriate and coherent for data analytics.
- Data reduction, i.e. reducing data representation for analytics reasons without compromising its integrity.
This filtering typically occurs at the point of acquisition, on the devices themselves or on gateway devices that aggregate the data, to determine which data needs to be sent for further analysis.
Doing Data Analytics: Real Time VS. Non-real TimeA general approach divides data analytics into real time and non-real time, which, in its turn, can be near real time or historical. Real time analytics happens on the device collecting this data, or in a maximum proximity to the gateway it is using to interconnect with the other devices of the IIoT system. It is referred to as "edge analytics" and delivers near-instant automated results in the situations when speed matters.
Though analytics happening in the cloud can be done very quickly, it still remains near real time and loses in speed to edge analytics.
Real Time Analytics at the Edge: Why’s & How’sMany situations happening during the manufacturing and production processes demand time-critical analytics, and a tendency to passing it to the edge is only growing. Besides, edge analytics enables organizations to scale their processing and analytics capabilities by decentralizing to the sites where the data is actually collected, thus reducing the strain on the central data analytics resources.
As put by Deloitte Chief IoT Technologist Robert Schmid, “Over 40 percent of [IoT] processing will eventually be done on the edge. There’s going to be processing going on at the edge, and it’s going to be great.”
Though edge analytics is limited compared to the one available in the cloud, the reasons for its success with developers and business owners are clear:
- Saving time. In many situations associated with IIoT taking instant decisions gets critical: for example, performance of an autonomous guided vehicle, equipped with an anti-collision algorithm, or monitoring of the oil rig once the equipment gets faulty and has to be shut off the valve immediately. Unlike analytics in the cloud, edge analytics provides close to zero latency between analysis and response.
- Overcoming bandwidth restrictions. A wide array of interconnected devices in manufacturing, oil and gas industry, and fleet maintenance need to operate non-stop regardless of limited or intermittent network connectivity. However, bandwidth constraints in industrial environments is a common case. Hence, doing analytics in the cloud, which is associated with maximum compute power and maximum bandwidth requirements, doesn’t serve a thing if it cannot be guaranteed 24/7, while more autonomous edge computing comes as an optimal solution independent of stable Internet connectivity.
- Operational costs reduction. Connectivity, data migration, and bandwidth features, associated with doing analytics in the cloud, are expensive. Edge computing addresses these by the above mentioned reduced bandwidth requirements.
Wikibon, an analyst firm specialising in the Internet of Things, cites an example, when the overall management and processing costs of a remote wind farm, equipped with cameras and other sensors, were reduced from $81,000 to $29,000 when they started using edge computing capabilties, and transmitted only summary data to the cloud.
- Security level. Often, security vulnerabilities are minimized by limiting physical and cyber access to data-generating assets. An example here could include a microgrid at a hospital or emergency event center, where the addition of edge intelligence could double the security benefit. First, it disperses access by bad actors away from central systems. Secondly, it ensures no single point brings down the operation, based on the many ways accidental cascading failure can occur (humans, software, hardware, etc).
How is it passing?Developing an analytical model for doing edge analytics is not an overnight task, as it involves 2 stages:
- Creating an analytics model. It means developing a neural system, which would follow selected algorithms and training it on continuous basis with tons of data sets so that it learns and makes adjustments and finally forms the right model of behavior.
- Executing the developed analytics model. Once the trained model clears out its rules, recommendations etc, it can get implemented at a factory level.
- Public, provided by one of the tech leaders: Intel, Cisco, IBM, HP, and Dell.
- Custom, built in accordance with the enterprise’s particular demands, such as demand for a higher security level, or when the tech requirements are way too specific. Usually such projects involve creative cooperation of IoT architects, custom web application development companies with expertise in deploying neural systems, and decision makers.
Doing Analytics in the Cloud: When Deeper Insight Is NeededThe data sent to the cloud usually doesn’t require instant response but is meant for further heavy-duty processing and long-term storage.
- Doing deep analytics. While edge analytics can be perfect for quick data-driven actions, it can’t provide deep insights into tons of collected data - a task named ‘’number 1’’ for cloud analytics. A simple example: with the cloud, oil companies and mining operations can spin up thousands of servers at once which will tackle massive computing problems and analyze existing processes for potential costs savings and optimization, with no need to provide the answer urgently.
- Complexity of the problem. Stemming from the previous factor, cloud analytics comes handy when serious problems are to be tackled. Ambitious challenges like comparing efficiency rates of lines across multiple industry facilities or developing a new therapeutic which requires studying millions of patients’ vital signs present complex tasks and imply processing huge amounts of data.
- Blending data from multiple different sources. Cloud analytics deals with all sorts of data sets provided by multiple sources within the IIoT ecosystem: devices, sensors, databases, third-party applications like CRM and ERP software. This data is linked with context data related to the business transactions of the enterprise.
- Doing predictive analytics. According to a recent research by IoT Analytics, the market for predictive maintenance applications will expand up to $10.9 billion by 2022. Predictive analytics combines traditional condition monitoring enhanced with analytics algorithms. Its wider use aims at monitoring the state of machine and predicting their breakdowns before they occur.
Moving from the Edge to the CloudThe cloud where deep data analytics is passing, can be either public, provided by major tech companies, or private, located on-premise or custom built. In both cases the cloud uses a neural written system.
Analytics passing either in the public or private cloud processes the collected data in the forms of reporting, predictive modelling, simulation of differing outcomes for testing hypotheses, and closed loop systems that communicate with devices and define the action based exclusively on the process variable and not on the input from human operators.
Today's key factors standing behind the cloud-based analytics are AI tools: Machine Learning (ML) and Deep Learning (DL).