Knowing what tomorrow brings – an ancient human dream. Digitization can make this dream reality, at least in part. Predictive analytics is a machine learning application. An algorithm models a time series from currently available data and continues to write the extrapolated data into the future. The use and benefit of such an analysis method are limitless in principle. In practice, the availability and quality of data, algorithms and processing power as well as the knowledge of how to combine these ingredients are limiting factors.
Predictive maintenance is one of the most popular application areas of predictive technologies. From the – ideally – large amount of process data collected in the production plant, an algorithm generates a prediction of expected disturbances, such as the failure of machine components. System downtimes can be shortened and the procurement of required spare parts can be accelerated as a result.
The quality of the data
A prerequisite for such a prediction is good quality data. Only if all conceivable states of production are depicted can a consolidated forecast be made. Included in these conditions should be those which are preferably avoided in practice: failures, breakdowns, disasters. If certain conditions are over- or underrepresented, expert knowledge from production is required.
Pure data-based solutions are, therefore, not considered the last word of wisdom. It is smarter to go hybrid and combine data from IT and production. Therein lies the answer to the question of numerous manufacturing companies: Do I have to become an IT company as part of the digital transformation? Only in part. Of course, building up IT expertise and manpower makes sense and is necessary. However, engineering knowledge is not curtailed in the process. Not everyone who knows how to use a predictive analytics tool can contribute the necessary expertise to the analysis.
As a result, individualization in production as well as other factors can lead to an unfavorable data situation, a so-called thin data case. In this instance as well, the increased incorporation of production knowledge is needed.
Can the algorithm make erroneous predictions even with good data? Naturally. Therefore, it is essential to design such a system with a feedback loop in order to give the algorithm – as the learning system – a chance to identify errors and their causes and avoid them in the future, in cooperation with human control and evaluation
Position of our economy
Prof. Dr. Stefan Wrobel is Head of the Fraunhofer Institute for Intelligent Analysis and Information Systems IAIS and Professor of Computer Science at the University of Bonn. As Director of The Fraunhofer Research Center for Machine Learning, Spokesman of the Fraunhofer Alliance for Big Data • AI, Deputy Chairman of The Fraunhofer Association for Information and Communication Technology, and Spokesman for the expert group “Knowledge Discovery, Data Mining, and Machine Learning” of The Gesellschaft für Informatik, he is engaged nationally and internationally in the topic of digitization as well as the intelligent use of big data and cognitive systems.
German industry is well aware of the importance of introducing forward-looking technologies and has a good starting position. Of course, big international tech companies are generating huge amounts of data and thus have a better basis for predictions in this area. However, manufacturing enterprises have specific data in their sectors and expertise in their fields. Many German companies are market leaders in their segment and, therefore, have stable foundations.
When exchanging the necessary data, security concerns are often an inhibiting factor. The exchange of data between machine operators and IT platforms or machine operators and machine suppliers certainly fuels security concerns. Such exchanges require platforms that allow participants to confidently exchange data, ecosystems, as it were, between peers. Fraunhofer and many business enterprises have jointly founded the International Data Spaces Association. Among other things, we want to demonstrate from Germany that data can be exchanged differently than what is currently implemented at large internet companies.
The shortage of qualified specialist personnel also poses a challenge, especially since all sectors in general require expertise. We have had good experiences in this area with the development of company-specific training programs for qualifying our own employees.
It is now important for companies not to lose sight of the future, even with currently full order books: The focus on digital transformation cannot wait.
Use in transportation and energy
Advancing technological development will not only make predictions more and more reliable in the context of predictive maintenance, it will also open up an increasing number of areas for technology. Anomaly and pattern recognition as well as quality control are additional worthwhile application possibilities in the field of production. Even finance and retail can benefit greatly from reliable forecasts, as well as medicine and science.
The segments of transportation and energy are also areas with potential for special applications. The general limitations of traffic areas and the future emergence of autonomous vehicles make forecasts of expected traffic volumes necessary. The increased use of renewable energies, the production of which is very irregular, also requires forecasts. Once renewables make up 50 to 60 percent of the energy mix, there is no way around predictive energy. For example, energy-intensive production can be planned to start when a strong wind over the North Sea is predicted.