Between hype and real insight: Artificial Intelligence (AI)
The use of artificial intelligence (AI) is raising high expectations for self-optimizing digital twins or autonomously acting, adaptive factories in smart value networks. Beyond the hype, however, AI is still in its infancy, which does not mean that industrial added value cannot be achieved with AI, which is currently still "weak".
AI - smarter than humans?
One of the most famous future theses comes from the book "The Singularity is near: When Humans Transcend Biology" by Raymond Kurzweil from 2005. In it, he predicted an exponential increase in information-technological development, as a result of which the "singularity" of artificial intelligence should become possible by the year 2045 - which would mean nothing other than that, from this point on, machines would for the first time and forever be more intelligent than humans. 
 G. Cisek, „Machtwechsel der Intelligenzen, Die blaue Stunde der Informatik“, Springer Nature 2021, S. 7
Focus of industrial practice: increasing efficiency
Beyond the singularity question, then, industrial practice can focus for now on additional efficiency gains from so-called weak AIs. In this context, "weak" artificially intelligent are AI systems that have a merely superficial level of intelligence but no (own) deeper understanding of the problem and its solution. However, with the appropriate programming as well as based on the appropriate algorithms, these systems are able to evolve and optimize themselves towards a specific use case. 
 www.divis.io/2019/03/ki-fuer-laien-teil-2-klassischer-ki-neuronalen-netzen-und-deep-learning/ (Stand 22.09.2022)
Currently, the most well-known example of weak AI is Machine Learning. The Fraunhofer Institute for Cognitive Systems IKS describes it as follows:
"In machine learning, an algorithm learns to perform a task independently through repetition. In doing so, the machine is guided by a predefined quality criterion and the information content of the data. Unlike conventional algorithms, no solution path is modeled...."
 www.iks.fraunhofer.de/de/themen/kuenstliche-intelligenz.html (last checked 22.09.2022)
AI in Industry 4.0
Classical algorithms are based on logic and rules. This makes them unsuitable for describing complex problems. Machine learning, on the other hand, is up to the task. Corresponding algorithms are a combination of programmatic and statistical methods. In other words, they are "learned" with training data and are then able to recognize patterns and relationships in data. This process can be supervised by humans, but it can also be unsupervised. A subcategory of machine learning is deep learning using neural networks, which plays a role in autonomous driving or machine vision in industry, for example.
How can machine builders benefit from AI?
For machine builders, the above results in several concrete use cases for AI:
Probably the best-known use case for machine learning in mechanical engineering is predictive maintenance - the anticipatory maintenance of machines and systems. In corresponding approaches, AI is trained to independently detect critical conditions and deviations from norms based on machine data. If such a scenario occurs, this indicates any imminent malfunctions and maintenance requirements. Machine builders and their customers can then take targeted measures to either prevent failures or at least schedule them optimally. They reduce downtimes and at the same time cut the number of unnecessary routine inspections.
Compared to Predictive Maintenance, Predictive Quality refers to the processes involved in manufacturing. Process quality assurance is about reducing scrap. Data analyses record relevant factors that lead to statements about the quality expected in the future - if necessary, measures can improve it. To this end, the analyses uncover unknown patterns and correlations, and the collected findings are incorporated into forecasting models that calculate probabilities for process and product quality.
Opportunities for AI applications also exist in design. In this area, machine builders today sometimes perform a great many computationally and time-intensive simulations. By using machine learning, the number of necessary simulation processes can be significantly reduced. In this case, an ML solution uses various results and parameters from simulations that have already been carried out. In this data, it recognizes statistical correlations and can precisely predict the outcome of upcoming simulations on this basis. This reduces manpower requirements, energy costs and also time-to-market.
When it comes to automating workflows, routine activities in particular can be transferred well to AI systems. This refers to activities that require comparatively low cognitive and communication effort and involve a lot of different data.
At the same time, autonomous organization and control of so-called light-out factories - i.e., autonomously and adaptively controlled factory halls that no longer need any light, as it were, because no people work in them - remains a vision of the future for many companies for the time being. But there are opportunities to optimize existing processes with the help of AI-based solutions.
One possibility for the use of AI in mechanical engineering is offered by the following examples from the ADAMOS STORE, in which a steadily growing number of solution modules can also be found around artificial intelligence.
App recommendation from the ADAMOS STORE
Esprit CAM makes it possible to use a digital twin of the machine tool for its programming, optimization and simulation. AI simplifies the automatic selection of processes, machining strategies, tools and cutting conditions.
>> Esprit CAM in the ADAMOS STORE
PwC Factory Intelligence is a set of intelligent applications also for predictive quality for early detection of quality problems through automated correlation of production parameters and product quality - the software also provides suggestions for corrective actions.
>> PwC Factory Intelligence in the ADAMOS STORE
Senseye PdM is an AI software and automatically generates models for the behavior of machines and maintenance personnel. Using options, the solution can be adapted to the user company and then promises reduced machine downtime, increased sustainability and reduced operating and maintenance costs.
>> Senseye PdM in the ADAMOS STORE
up2parts ensures greater efficiency in processes such as costing, quotation generation, order generation and work preparation. The algorithm of up2parts calculation uses component information from existing 3D models as well as existing manufacturing know-how. The AI-based software continuously trains the individual artificial intelligence and thus adapts to the manufacturing portfolio of the respective company.
>> up2parts in the ADAMOS STORE
Workist, using AI, automatically processes incoming order documents and transfers order data to the ERP or CRM system. If the AI is unsure whether the captured data is correct, human clerks provide feedback. Based on this feedback, the AI learns to recognize the data more frequently in the future.
>> Workist in the ADAMOS STORE