By Stefanie Nowak
A lever in a handling system for computer chip production is about to fail. The machine affected recognizes the upcoming problem, 3D-prints the spare part and autonomously exchanges it, without activating a service technician. It is the year 2030. Machines and factories are fully equipped with intelligent sensors that record data on the entire operating environment, for example wear and tear, runtimes or logistics. The machines are mutually connected and information is exchanged between all types and makes. In close cooperation with international IT companies, German machine makers have systematically enhanced Industrie 4.0 solutions with artificial intelligence and thus strengthened their competitive edge.
This is not an extract from a science fiction novel, but a scenario described in the "Machine Learning 2030" report published by the VDMA Competence Center Future Business. This vision has been developed by companies, researchers and VDMA experts in cooperation with the Fraunhofer Institute for Systems and Innovation Research ISI. It is the first of a series of "images of the future" reports intended to explore what path mechanical engineering might take in the future.
"It is an eye opener when you start to discuss about complex trends and future developments with a group of interested experts," says Dr. Eric Maiser, head of VDMA Future Business. "Important rule: The scenario technique we applied allows for a certain range of possible 'futures' without forecasting which will be the most probable ones. This way we offer our members different development paths they can choose from themselves. Nevertheless, conclusions for our entire industry can be drawn from the complete set of scenarios. For 'Machine Learning' we have developed four of them."
The scenarios range from the mentioned triumph of machine learning in the machinery industry up to the damped scenario of a "digital steppe", where Germany and Europe have fallen way behind Asia and the USA due to missing data pools, ethical concerns and a lack of education and training for data scientists with applied Machine Learning skills.
Collaboration as a decisive factor
"Machine Learning only works with loads of data that usually cannot be collected by one machine maker alone. Collaboration is the key to success here. Digital protectionism will fuel the 'digital steppe' scenario," says Maiser. Collaboration can have many faces, though. The "size matters" scenario, for example, describes how large sized machinery companies have gained a competitive edge thanks to Machine Learning in 2030. They make use of the vast amount of data collected in their entire installed base of machines at clients" sites and data platforms shared with other companies to push Machine Learning. They have integrated smaller IT start-ups and also collaborate with data brokers or IT giants to acquire the expertise required or to outsource data handling and -storage.
There are other possible development paths. In the "SME networks guiding" scenario, small and medium-sized enterprises in the mechanical engineering sector have used their agility to advance Machine Learning. They joined forces and established a network among other machine makers together with start-ups from the field of artificial intelligence worldwide. In contrast to that, large companies were not able to exploit their position in the market. Inflexible and sluggish, they were hesitant to use self-learning systems and thus lose their competitive edge in the long run.
Who owns the data?
Machine Learning can only work if the machines have a sufficient amount of data for self-learning. This way they can become an expert, almost like an experienced operator on a manufacturing line, described as the "digital guru" in the report. They know answers for example for the following: How do aspects such as room temperature and humidity affect the machine's performance? What is the expected service life of a component made of a specific material?
The more application-specific data equipment makers and their clients can gather, the better they can advance the machines. Predictive maintenance systems use data to calculate when it will be necessary to install a spare part, thus preventing system failures already today. Machine Learning will give a boost to this, saving factory owners a lot of money, preventing premature failures and unnecessary or unscheduled maintenance missions.
Collecting the data required for this means that machine makers and their clients need to work together. But it also requires policies that govern not only what data may be processed and evaluated, but who owns the data in the first place.
In the "vanguard wins" scenario, in which Machine Learning has reached the highest stage of development, there are clear, consistent and Europe-wide policies on data sovereignty. Companies can easily decide what data they can pass on to fully unlock economic benefits, without running into legal issues or the risk of loss of know-how. The shared information is secured in a huge data pool that can be accessed by all companies that are part of a closely-knit European network.
However, the outcome of ethical discussions also plays a role in the scenarios. The question whether society believes in the opportunities of artificial intelligence rather than their threats must not be underestimated. How much responsibility are people willing to give up to "intelligent" machines? Will the algorithms only be used as assistance systems for factory workers or will they complete complex tasks automatically and autonomously? The answers to these questions have an impact on whether the application of Machine Learning will be encouraged or blocked in the future. "This is not at all trivial", emphasizes Maiser. "How complex the application of self-learning and autonomous machines can get, can already be seen with respect to self-driving cars."
In terms of the systems' efficiency, much depends on how the smart sensors installed in the systems will develop. Will they only collect and transmit data to a cloud, where Machine Learning is executed, or will they also be capable of learning themselves? "Think of a swarm of sensors with every sensor having a little computing power, Wi-Fi and a built-in neural network. They can behave like a swarm of ants, like a collective intelligence, distributed, ubiquitous", explains Maiser. Researchers have already been pushing ahead technical development in that direction since the start of the millennium. "The technology is available, the pieces just have to be put together".
Many companies can already envision the vast potential of Machine Learning. "It’s not only the traditional mechanical precision that makes our industry strong. In the late 1970ies it became obvious that electronics broadly enhances the performance of machines, and companies adopted to that trend. The next upgrade clearly comes with software and data intelligence. Same as we did with mechatronics, we have to embed computer science into mechanical engineering, invest in adequate education, and most of all the application aspects for our industry, " Maiser states.
Collecting data is becoming more and more cost-effective, as there is already a large range of sensors and processors available on the market. But there is still a long way to go for ubiquitous use. "The technology for artificial intelligence is already pretty advanced. The time is now for setting the right framework conditions to foster real-world applications."
The "Machine Learning 2030" report outlines specific recommendations for companies, politics and research.
Companies in the mechanical engineering sector should:
- Address the development of Machine Learning strategies
- Integrate smart sensors
- Install Machine Learning capabilities, in-house or by collaborating with research institutions and external IT-start-ups
- Enhance collaboration among machinery companies for data pooling as well as with international IT companies to analyze and host data. Carefully watch and evaluate the influence of data brokers.
- Promote the education of data scientists and data analysts as well as start-up culture
- Accompany ethical discussion for Machine Learning
- Find ways to carefully regulate collaboration between companies for exchanging data, setting a reliable (international) legal framework for data sovereignty and safety architectures
- Improve training programs for Machine Learning with respect to bringing it to application
- Enhance cross-sectoral networks between basic research, computer engineering and the application-specific engineering sectors
What is going to change?
Machine Learning will lead to substantial changes for many companies. "All company divisions will be affected by the digital transformation. Most probably change will come with different speeds from company to company, sector to sector, as well as within the companies and their different departments", Maiser believes.
Experts expect that engineering and development processes will be the first to reap the benefits, as these do not involve processing data in real time. In logistics, algorithms based on Machine Learning could assist in planning or handling planning processes independently. And in production processes, it would be possible to continuously improve or redesign machines and factory performance based on ubiquitous data available.
Upcoming VDMA working group
"It turned out that Machine Learning is of utmost importance to a wide range of sectors in the machinery industry. For this reason, we have decided to establish a dedicated VDMA working group to cover all activities in the field of artificial intelligence and Machine Learning", announces Maiser. The working group will be headed by the VDMA sector group Software and Digitalization. "We will soon dive into fields like specific education programmes, standardization and roadmapping."
VDMAimpulse 02-2017: "Background: Machine Learning 2030" | VDMA Competence Center Future Business | Fraunhofer ISI | vdma.org: "Machine Learning will shape the future" | vdma.org: Machine Learning 2030 drawings | Report Machine Learning (PDF in German, for VDMA members only) | VDMAimpulse 02-2017: "Predictive maintenance is becoming ever more important" | VDMAimpulse 01-2016: "Fit for the future"