How production can benefit from AI

Considerable amounts of data are generated in production every day, but much of this data remains unused. Thanks to the Industrial Internet of Things (IIoT), intelligent algorithms and the possibility of inexpensive real-time processing, this production data is now coming into focus. Cloud platforms are helping to integrate these technologies into everyday production, gain insights from the data and support decision-making on site at the machine.

One example of this is intelligent (cognitive) services for AI-based image recognition on the production line. The quality is analyzed using photos or video recordings in order to detect fluctuations at an early stage. On this basis, machine parameters can be adjusted in good time and measures for predictive maintenance can be taken. AI is used to support machine operators on site. By automating in-depth domain and process knowledge, even employees with less experience can deal with problems better and take the right measures. AI also supports many optimization tasks, such as planning the production sequence or the necessary staffing levels.

Consolidate data on a central cloud platform

However, with the increasing importance of real-time analyses and growing complexity in hybrid IT landscapes, a centralized approach to data management is becoming ever more important. Today, data is often not stored properly on the machine and there is a lack of transparency and control. This is where centralized data management on a cloud platform can ensure that the regular collection and backup of relevant data for compliance is guaranteed. Similar to middleware, the platform organizes data management for all components centrally. Many companies already rely on cloud architectures. An industry platform such as Microsoft Azure lays the foundation for bringing together IT systems in production on the one hand and edge devices on the other. Connectors are available for almost all common systems and sensors, including ERP systems such as Microsoft Dynamics or SAP. As everything takes place on one platform, maintenance costs are lower and interface management can be largely automated.

Program components are already available as easily bookable services for many industry-specific AI issues. Microsoft's Azure Synapse services also simplify the merging of all data sources for analyses and the collaboration of all participants in a data analysis project. Data from applications and IoT devices is included in a pipeline and distributed to the corresponding services. This also includes, for example, the processing of video streaming data in real-time analyses for quality control.

The barriers to entry for AI are falling

Companies only pay for the performance they actually need in the cloud (pay-per-use) - in the case of streaming analytics, for example, how much data is moved per minute. This eliminates the need to provide otherwise oversized infrastructure, as would be necessary for training phases of AI algorithms, for example. At the same time, the high scalability makes it possible to start small and expand the solution as required. In small versions, many components are even partially free of charge. Later expansion, such as a roll-out to several plants, is still possible without any problems. Data security aspects such as encryption are also covered by the platform. If data centers in Germany or the EU are selected, GDPR compliance is ensured.

Even SMEs with few machines or systems that have hardly collected any data to date can use this as a basis for getting started with AI. Pre-calibrated models that are trained with the company's own data help here. The previously complex and knowledge-intensive data science tasks are significantly simplified. A small amount of external support from experienced partners is often enough to tap into significant optimization potential.

During preparation, it is particularly important to have a clear definition of key figures. If, for example, throughput times in production are to be optimized, it must be clear to everyone involved which parameter is being measured and optimized. Is the calculation made from the moment the raw product enters the factory or when it is loaded into the machine? Do you want to measure the mean or the median? Even a trivial-sounding key figure often turns out to be complicated. Every company has its own characteristics of key figures, and often each department even has its own definition. However, only with clear definitions can the trust in the results from the algorithms that is necessary for success grow.

Data and key figures must be of the right quality

It is also important to acquire the right data. In order to achieve the best possible benefit, it is not very helpful to collect and store data indiscriminately. When analyzing and using the data, it often turns out that relevant information has not been saved or has not been saved with the necessary quality. For example, the appropriate time stamp or the right granularity is missing. Experience shows: The interdisciplinary interaction between process know-how and data knowledge is absolutely essential for AI. It is generally more expedient to collect data with a view to specific issues rather than to store all machine data in an uncoordinated manner using sensor technology.

Manufacturing companies that rely on IIoT and use intelligent data analytics benefit from increased efficiency and cost reductions. At the same time, they have laid the foundations for implementing new digital business models and services. The more consistently and swiftly a company tackles the digital transformation, the better positioned it will be against the competition.