AI and MES - Together into the future of manufacturing
The integration of artificial intelligence (AI) and manufacturing execution systems (MES) is revolutionizing the industry. Autonomous manufacturing systems promise greater efficiency, flexibility and precision – but how can this vision be realized? What requirements must an MES fulfill and how much autonomy makes sense?
In this interview, Thomas Krainz (Member of the Board & Strategic Product Management at Industrie Informatik) explains the opportunities and challenges of AI integration.
He provides insights into key areas of application such as predictive maintenance, quality assurance and production planning – and shows how companies can shape the production of the future with smart strategies and tailored solutions.
Mr. Krainz, how do you assess the integration of AI and MES for the development of autonomous manufacturing systems, and how can the resulting requirements for an MES be implemented in practice? How much autonomy makes sense for MES?
Krainz: We see enormous potential in the integration of AI and MES! There are applications in almost all areas, for example in the parameter adjustment of systems, the control of intralogistics, order planning or specific, smaller optimization tasks in the production process. However, it is important not to view AI as a panacea and to distribute it according to the watering can principle.
The following points should be considered in advance to ensure successful implementation:
- Degree of digitalization of the systems: are the systems sufficiently digitalized and is all relevant data available in a suitable form?
- Use of alternative approaches: For smaller optimizations, it may make sense to use simple algorithms or rules engines instead of relying on AI.
- Economic consideration: A cost-benefit analysis and implementation in clearly defined, manageable use cases are essential in order to maximize added value.
The MES plays a central role here as a controlling system. It should ensure that data is transferred to AI systems at the right time and that their results are processed efficiently.
Only through careful planning and coordination between MES, AI and the manufacturing processes can a balanced level of autonomy be achieved that is both productive and economically viable.
For AI to be able to recognize patterns, make predictions and identify deviations, data integration plays a crucial role. How do you handle the integration of existing data in your MES to make your solution as intelligent and efficient as possible?
We started developing specific models for classic predictive use cases such as maintenance and quality at an early stage. We combine order data and time series data in a preprocessing process and then pass it on to the AI model. An important goal was to understand the functional principle of machine learning systems in order to create plug-and-play solutions. This allows our cronetwork MES modules to provide data at the touch of a button. Our technology makes it possible to prepare data for learning processes as well as to integrate predictions into processes in real time – from the machine level to the ERP system. This works both via the user interface and in the backend.
Data preparation is a key aspect: a large part of the work consists of ensuring data quality, for example by filtering outliers. We offer project-specific solutions that can be implemented without additional programming effort. We also support the integration of external systems and promote make-or-buy approaches. If a customer already uses their own model, only the hyperparameters need to be set. This allows maximum flexibility when integrating different systems.
AI enables the development of predictive maintenance systems by analyzing real-time data to predict the optimal time for maintenance work and reduce unplanned downtime. How did you implement this function?
We have developed a model based on supervised learning that can, for example, predict the probability of a workplace malfunction for the next shift. We can also embed a graphical influence analysis in our user interface, which identifies the parameter values that have the greatest influence on the prediction. This quasi plug & play module – based on MES data – can be flexibly adapted or fine-tuned to specific boundary conditions.
What about real-time monitoring and product quality assurance?
Similar to the area of predictive maintenance, we have also developed a model for quality monitoring that can, for example, predict the relative scrap rate of an order for the next shift. Here too, we provide a graphical influence analysis. This data can also be correlated with the probability of machine failures.
In addition, we offer the option of integrating specialized, AI-based systems such as image recognition or anomaly detection in screwdriving processes etc. into our FQS processes (in-process quality assurance). The results of these systems can be displayed and processed directly in our reporting interfaces, as we have seamlessly embedded the collection of quality data into our processes.
We see AI-supported quality predictions as important support for better mastering production processes and increasing the efficiency of quality assurance. However, the use of AI should be targeted at the specific challenges of a company and not be used indiscriminately.
Do you use AI in production planning and what advantages do you see in this area?
Industrie Informatik has a long tradition in the development of detailed scheduling & APS. Our extensive APS (Advanced Planning and Scheduling), which is based on heuristic planning algorithms, has grown over 30 years and has now reached the limits of complex applicability for both users and developers. About four years ago, we started to add an AI-supported optimization component to our APS.
With the help of AI, the system independently learns to determine the optimal use of so-called neighborhood operators, such as swapping workstations or sequences. This imitates the human behavior of a planner who would make similar decisions based on their experience. The results are evaluated using target functions that deliver measurable improvements.
This is a good example of the evolutionary development of an MES system. For complex planning tasks, we can conserve computing resources and reduce the computing time required for optimization processes through fast pre-calculation using classical heuristics. At the same time, the proven simulation evaluation remains an integral part of the system.
The use of AI in production planning offers clear advantages: It improves the efficiency and accuracy of planning and at the same time relieves the planner through automated decision support.
The last few years have shown how important it is to rely on future-proof technologies and drive forward digitalization in order to remain competitive. Artificial intelligence is both an opportunity and a challenge that must be seized in order to gather experience and data, which in turn can be used as insights.
Anita Peherstorfer
