Four myths about IIoT data strategy that manufacturers still believe
By Christian Lutz, CEO, Crate.io
Myths around the challenges of implementing IIoT systems and building smart factories have made the prospect of adoption unnecessarily intimidating. In some cases, industrial organizations have avoided the most effective IIoT implementations available to them simply due to false understandings of the technology.
While IIoT adoption does require a new approach to managing and analyzing data collected in real-time, this isn’t as difficult an obstacle as many have been led to believe. Let’s take a look at four common myths about IIoT systems and the realities behind them:
Myth: IIoT implementations necessitate huge expansions of existing traditional databases.
This is false. The traditional databases that most industrial organizations already have in place (Microsoft SQL Server, Oracle, etc.) are wholly inappropriate for use with IIoT systems, given the tremendous volume and complexity of data in question. Expansions of these databases doesn’t work. In practice, there’s a night-and-day difference between standalone machines and those networked within IIoT monitoring systems.
When industrial businesses mistakenly implement IIoT infrastructures using traditional databases (and this happens often), they soon discover them to be expensive to scale, unable to process the vast amount of incoming data, or incapable of handling the more complex queries required to realize the IIoT’s benefits.
When it comes to querying IIoT data streams in real-time, traditional databases are simply not designed or equipped for the task.
Myth: Only NoSQL databases are appropriate for IIoT systems.
Given the fact that IIoT sensors collect massive volumes of unstructured (JSON) data, it’s understandable—but incorrect—for industrial organizations to believe that they’re required to use a NoSQL database. In reality, applying NoSQL databases to IIoT use cases means overcoming a number of inherent issues, and are by no means the only solution. NoSQL databases do offer efficient scaling and distributed architectures that lend themselves to performing complex, flexible queries. But NoSQL databases also often bring complex infrastructures, which take intensive planning and administration to operate properly. Additionally, NoSQL databases require engineers with specialized (i.e. expensive) expertise, which can be hard to come by, especially if scale is required. The fact that each NoSQL database solution uses its own query language only exacerbates the challenge of enlisting the right talent.
Nearly every IIoT system must manage JSON and relational data, consisting of topological, firmware, ERP or article data. If an industrial business chooses to address this need by running both relational and non-relational databases, those systems need to be synchronized for use in parallel, which results in complex setups, unnecessarily large cloud footprints and disadvantages when running queries.
A full understanding of IIoT-database requirements and options yields an alternative: advanced new SQL-based systems are available which offer both the ease of ANSI SQL and the flexible and scalable nature of NoSQL solutions, making them a strong fit for IIoT implementations.
Myth: IIoT systems call for using time-series databases.
This myth leads to a common strategy error when implementing IIoT systems (or choosing not to due to this perceived need). A time-series database should not be foundational to an IIoT system, because intense parallel usage will severely limit its functionality and scalability. IIoT systems don’t just need to visualize data streams, they also must perform analysis, run highly concurrent workloads and perform frequent changes to the data model. The IIoT database must enable interactive work under heavy real-time data loads, such that it’s possible to read/write/execute ad hoc queries all at once. This processing is integral to the benefits IIoT systems yield, making it feasible to quickly and accurately identify and correct production issues at smart factories and leverage advanced capabilities such as machine learning.
The IIoT database must also be able to adapt and extend data schemas at runtime in order to support agile processes. It must be possible to investigate anomalies in production using bare sensor, ERP, quality or other data to recognize issues—if specific jobs, materials or suppliers are the source of an issue, adapting the data model to examine these data types will provide the correct insights. Unfortunately for businesses using time-series databases for their IIoT systems, making such changes to the data model requires performing total rebuilds of their databases, at a great expense of both time and money.
One alternative is to use a relational database for non-time series data along with a time-series database. However, this method can result in high database expenses as the system grows, and adds the challenge of maintaining the sync between data across separate databases.
Myth: Adding AI to IIoT systems is out of reach due to data requirements.
It remains common for organizations to believe that because they don’t yet have large amounts of sensor data (or data that’s perfectly clean) they aren’t capable of leveraging AI systems. While a dearth of data can result in low-quality automation when driven solely by AI, it’s inaccurate to assume AI is an all-or-nothing proposition from a data perspective.
Even with limited data, it’s possible and beneficial to build a real-time data store and use AI and machine learning to augment and optimize IIoT decision-making, with final decisions remaining in human hands. Developing these AI systems is a bit of a chicken-and-egg situation: you have to start somewhere. Few industrial companies have huge stores of clean data available to begin with, but by monitoring analysis results and introducing systems to automatically clean data in a deliberate process, AI capabilities will gradually improve until more complete data is prepared and the system is worthy of more automation responsibilities.
Businesses that attempt to clean all of their data before introducing it to their AI system will provide too small a quantity for the system to grow and improve properly. With IIoT AI systems, patient incremental improvement is the correct path forward.
An IIoT reality check
Those reluctant to explore IIoT implementations should become aware of the profound operational insights and benefits these systems can deliver. The IIoT is likely within their reach.
At the same time, organizations need to be aware of what capabilities are mandatory to IIoT success. The IIoT requires entirely new data-management and analysis capabilities. Monitoring, predicting, and controlling equipment across immense pipelines calls for data-management systems that enable real-time analysis of data coming from vast arrays of sensors, and using diverse message formats, all under highly concurrent loads.
The future value of IIoT projects also depends on achieving data-driven automation. Specifically, data-management systems must provide rapid development and time-to-value, maintain consistent uptime, and offer low IT-operating costs when it comes to hosting, integration and administration. This is where the IIoT concepts differ greatly from normal IoT use cases (e.g. DevOps IoT monitoring).
By separating myth from reality, industrial organizations can properly recognize the opportunities of IIoT for what they are, and pursue them more effectively and efficiently.
