According to a study by TÜV Süd and IDG, more than half of the companies surveyed are already using IoT devices. McKinsey also predicts that the economic added value of the Internet of Things could amount to around 11.1 trillion dollars by 2025. So much for the theory. However, a sound data strategy paired with powerful development technologies is needed to really make successful use of the many opportunities. Only then can the huge amount of information generated by the sensors be fully orchestrated, fully evaluated and used in the best possible way. Conventional data management solutions simply cannot cope with the high data throughput associated with such applications. For IoT projects such as predictive maintenance, machine condition analysis and innovative AI applications, a platform is required that is characterized by interoperability, performance and scalability.

First step in the IoT project: making machine data usable

The first key challenge for many industrial companies when it comes to IoT is accessing the data required for analyses. This is because the data protocols used by older and newer devices are usually different. A way must therefore be found to collect, process and analyze a wide variety of data formats and protocols. A first step for industrial companies would be to upgrade their machines so that device data such as energy consumption or usage time can be read out via adapters to machine controllers such as the quasi-industry standard Siemens SIMATIC S7. The next step is to orchestrate and merge the collected data, often from a wide variety of machine locations in the company's own or external production facilities. New applications often have to be programmed to ensure communication across different protocols. This applies all the more if other IT solutions such as ERP (Enterprise Resource Planning) or SCM (Supply Chain Management) systems are also to be connected.

Not only collect data, but also analyze it intelligently

Handling different data formats and standards plays a central role in the context of predictive maintenance and IoT projects. However, the ability to handle large volumes of data with confidence is no less crucial for an efficient workflow. For example, a typical energy metering application in a small to medium-sized city has to process an average of over 10,000 events per second and more than a billion events per day. Most data management systems do not offer the necessary prerequisites for predictive maintenance or data mining in view of such enormous data volumes and scenarios. The comprehensive use of these technologies offers immense benefits, ranging from improvements in product design and usability to the development of completely new business models. The prerequisite for all applications is the collection and bundling of data in a database. However, this alone does not make the data usable. For example, the analysis of data in the form of data mining - i.e. the application of statistical methods to large amounts of data in order to identify valid and potentially useful patterns - and predictive analytics for predicting future events is still a challenge due to the lack of defined analysis standards, although many companies are already looking into predictive maintenance in order to further improve the cost-performance ratio of their machines and avoid production downtime due to defective production systems. However, this is almost always just a matter of collecting operating data - a more far-reaching analysis is still a dream of the future in many places.

Interoperability is a must: Industry 4.0 requires an integrated data strategy

In order to be able to evaluate data from a wide variety of sources in the best possible way, they must first be reliably merged. The keyword here is interoperability. Machines are needed that automatically forward all information generated by input devices, sensors or other means to a database, check it and correct it if necessary. Processing is made more difficult if different devices use different data structures, which can sometimes even be the case with devices from the same manufacturer.

A powerful solution for data management and analysis in the IoT environment must be able to process all of these heterogeneous data sets equally. At the same time, it must be tailored to the individual circumstances, goals and requirements of the company and be able to adapt to changing conditions through machine learning. The use of open source technologies or the combination of different individual solutions that are not designed to be combined with each other initially seems attractive in terms of the initial costs. However, the initial cost savings are usually quickly offset by high maintenance costs. In addition, the cost/benefit calculation must also take into account whether long-term product support can actually be guaranteed and what the economic consequences would be if this support were no longer available.

Another important criterion for the selection of such an IoT-enabled solution: in order for analysts and data scientists to be able to recognize correlations between device data and external data sets, support for different types of analysis processing is essential. Through interoperability, the information gained from the analytics can be integrated into real-time workflows to perform business processes and critical just-in-time actions. However, this requires strong integration capabilities to unlock potential insights hidden in disparate data sets.

Focus on performance: integrated machine learning

If data sets are distributed and machines cannot communicate with each other, this means a huge amount of additional work and a great loss of time - regardless of whether it is for the development of new applications or the best possible use and integration of the collected information. The more devices, sensors and measuring points are integrated and evaluated, the more crucial the question of database performance becomes, and the more important it is to find a data platform that supports various protocols, programming languages such as Python, machine learning approaches and connectors to BI applications. This is because having all the required information in a single database improves the overview and minimizes effort and errors. In addition, an architecture is required that requires little data modeling effort and can be easily orchestrated. Agility, scalability and high availability are also essential in the ML context. Many data science projects in the machine learning environment are implemented using the Python programming language, for example. In order to migrate data to other systems, it is useful if the data platform can execute Python models due to its interoperability capabilities.

As it is often very expensive and time-consuming to start a data science project, it makes sense to rely on a data platform that has integrated machine learning functions. Horizontally scalable HTAP (Hybrid Transaction Analytic Processing) database management functions for executing real-time analytical applications on very large data sets are helpful, as are embedded functions for executing structured and unstructured analyses and AI models.

Scalability: technology for strategic company growth

InterSystems pursues an approach that addresses the requirements of industry in the IoT context. Its data platform IRIS delivers very high scale performance in terms of data volume and interoperability in business-critical enterprise applications, both for transactional (OLTP) and analytical workloads. IRIS supports data ingestion at very high rates and can process analytical workloads simultaneously for both real-time data and large sets of non-real-time data, such as historical and reference data, using commodity hardware. This is optimal for applications that need to make real-time decisions by identifying patterns and anomalies.

InterSystems IRIS accelerates and simplifies various tasks related to the development of predictive maintenance functions and other industrial applications. It enables users to create new calculated variables, provides embedded business intelligence (multi-dimensional OLAP) functionality and distributed SQL processing to support complex calculations on very large data sets with high performance. Data is stored once in the database, and the platform allows both relational and non-relational access, increasing flexibility. If it is easier to work with data in rows and columns, developers can access it using SQL commands. Alternatively, the data is available as documents, objects or key value data.

In addition to high-performance SQL capabilities, support for xDBC protocols and other standards, InterSystems IRIS offers direct integration with Apache Spark via a shard-enabled native connector. The Apache Spark nodes can automatically establish a direct connection to the data partitions and work on different pieces of data in parallel. These simultaneous connections enable high data throughput and support fast data ingestion in horizontally distributed clusters. At the same time, models can be trained in real time as data is streamed from production applications, allowing data scientists to keep models up to date as business requirements or environments change.

New data technologies required for IoT success

The Internet of Things creates unprecedented opportunities for companies to increase the performance of their machines and open up new business areas. However, conventional data management technologies are not equipped for these new requirements and data volumes. As a result, companies in the industrial and production environment are increasingly having to deal with new platforms that offer interoperability, performance and scalability. Central functions for the development, execution and maintenance of IoT applications plus machine learning should be connected in a single central multi-model environment. Such a data strategy and technology is a basic prerequisite for long-term IoT success.

The author: Aleksandar Kovačević, Sales Engineer at InterSystems

This article first appeared on our sister portal www.scope-online.de