Predictive Maintenance, Smart Factories & Drones; how AI is already driving value through smarter maintenance and production operations

Key learnings:

• Business drivers: PDM (predictive maintenance) is already able to add value by enhancing reliability, reducing costs, and increasing efficiency of maintenance. Drones and robotics can truly transform certain maintenance tasks (e.g. nuclear power stations, building-surveillance, agriculture.) Smart Factories can significantly boost competitive advantage of manufacturers in fast-moving markets (e.g. Planted's example in the fast moving plant-based market )?in a move towards decentralized data-driven operations
• Human aspects: Its not just about technology. Trust in the field of maintenance is critical; hence ExplainableAI is a major trend here and field of interest. Getting DSS systems to work more seamlessly with humans, to boost their work and not push solutions is important. Domain knowledge is also vital, (not just data-science) … and influences what data and data models to work with. ResponsibleAI processes should be taken seriously as shown in the case of Equinor
• Future: Humanoid robots for maintenance are coming and will be very useful for certain special use-cases. LLMs have great potential for maintenance to rapidly highlight relevant text/procedures from large maintenance manuals. AI works with data-driven models but can also be enhanced with physical models; process-industry seems more likely to be data-driven, whereas mechanical more likely to use physical models
• Systemwide Collaboration (across systems of systems)?is needed; getting fleets of assets (e.g. aircraft or vehicles) to cooperate on limited spare parts and maintenance resources through dynamic scheduling of parts and maintenance across fleets. Eventually sharing of AI models across assets for system-wide intelligence is important. System integration is needed (esp. to link to legacy systems) and is worth the investment in many organizations (e.g. SBB's focus on S4HANA.) Collaboration across company borders (e.g. benchmarking example from Ypsomed) is increasingly important
• Past:? Tech obsolescence e.g. Fortran code and old languages.. Engineers must design physical systems to last decades and consider their maintainability

Selected case-studies/examples from the conference

EPFL Lab Intelligent Systems: Prof. Dario Floreano : practical use-cases of industrial drones/robots developed for certain sectors, inspired by biology (birds, insects). Impressive to see how technology translates into innovation that creates real business value/benefits for industrial users (eg windfarm 3 fold maintenance performance improvement with improved drone technology). Latest innovation is a drone that mimics the morphing geometry of a predatory bird, to enable fast/efficient flight as well as extreme manoeuvrability when required.

Airbus : (Vincent Cherière): Story of how Airbus has developed a #PredictiveMaintenance platform(PDM) connected to many of the airlines in the world. Vincent shared real-world experience in how a PDM system is much more than a physics/mechanical problem to solve, but involves many other variables/factors and interconnected systems. These include maintenance-history of each aircraft, planned-maintenance, flight-plans, availability/scheduling of maintenance across airlines' fleets; hence this complex optimization problem must consider all resources, processes across the fleet, and the running of many what-if scenarios to optimize the maintenance scheduling, reducing unplanned maintenance and costs. The ultimate goal of PDM at Airbus is to bring more predictability to aircraft maintenance.

Ypsomed AG : ( Ralph M. Fasler ): as market-leading producers of drug-delivery medical devices (eg insulin pens), Ypsomed has undergone huge growth in production volumes. Ralf explained how this considerable challenge to the Manufacturing Team has been met through several strategies. Firstly a focus on full automation of production-lines, with continuous improvement from process data, to improve OEE. However with this approach reaching its limits, Ypsomed have undertaken a benchmarking approach, to find "out of the box" solutions to its manufacturing challenges. The regular sharing of data and experience with users of similar production technology (e.g. injection-moulding) is enabling further process improvements learning from other sectors (e.g. chocolate manufacturing). Finally, Ypsomed's commitment to digitalization/automation, benchmarking and process excellence is a key requirement of its ambitious future growth plans (scale up production >500% by 2030)

SBB CFF FFS : ( Urs Gehrig ): "technology is the easy part, of a predictive-maintenance implementation…" Urs explained his experiences at SBB, with the example of train-overhead-power acquisition (sliding pantographs) where he has developed and industrialized the rollout of an automated PDM system with image-analysis to detect degradation in the electrical contacts. At SBB, the PDM program is linked to SBB's S4HANA rollout, using Enterprise Asset Management. The logic here is to have all assets (real-estate, rail-infrastructure, trains) managed through 1 system, as there are clearly interdependencies. The time to bring a new PDM solution to life requires considerable time (years) to fully industrialize it/integrate into business processes of a rail operator.

Institute for Manufacturing (IfM), University of Cambridge : Prof Ajith Kumar Parlikad : presented learnings of digital asset-management across multiple industries. A key point was the need for algorithms and models to work together, and PDM systems to inter-relate. This is because improving resilience in industry requires one to consider a SYSTEM of SYSTEMS and not just each system in isolation. Also the business processes of maintenance when working on a fleet/collection of assets are usually constrained (budgets/resources) so spare-parts and maintenance teams must be planned across multiple "systems". Collaborative assets working together can also help share loads and increase system overall resilience when each component is able to share its health/state with others.

Planted : Mathias Pawlowsky presented Planted's commitment to a #SmartFactory strategy in its production-plants. This was a perfect demonstration of how digitalization can be key to an industrial company's competitiveness. With the plant-based-meat industry evolving at a rapid pace, Planted must continuously innovate new products and industrialize their production. This scale-up process typically takes many months, and through digitalization, Planted has been able to optimize/reduce time-to-market considerably. This has been achieved through IT/OT Convergence (i.e. leveraging popular/simple IT tools in an OT environment) to democratize the access to production data across the company. Now data scientists, R&D, and production engineers, work together on a single set of accessible data to rapidly improve production processes and establish the best process setpoints for a new production line.

Kernkraftwerk Leibstadt AG : Benjamin REGENER showed how the latest #drone and #robotics technologies are already in-use in the Nuclear industry, improving worker-safety and plant monitoring. Just 1 example includes a walking dog ("Spot") that goes through the plant and uses multiple sensors to check equipment, avoiding the need for human intervention on routine monitoring. Really impressive to see the systematic approach to digital innovation at KKL. In the future Benjamin foresees humonaid robots performing hazardous maintenance in radioactive areas.

Equinor : Rialda Spahi? explained how a #responsibleAI department & processes have been established, and WHY they are so important. Her presentation gave a great overview of what companies must consider when implementing AI systems and how this relates to the latest regulations.