FACTORY OF THE FUTURE
Trust & the Future of Manufacturing: letting AI automate decisions on the factory floor
To safely restart the economy, we need to strategically resume production in factories across the globe without risking the health of our coworkers. Automating more and more operations will play a big role in this, but handing over responsibility for decisions to machines often generates human resistance because of people’s fear of losing control. How can manufacturers continue to automate their shop floors and at the same time generate acceptance for the factory of the future?
Let me share my experience in logistics innovation at MHP:
Each of us makes about 20,000 decisions every day. In most cases, this happens subconsciously. Everyone knows the feeling of arriving home from work and having zero conscious recollection of the drive to get there — it even has a name: Highway Hypnosis. When it comes to repetitive tasks that we do frequently, our conscious mind takes a step back. We go into autopilot. There is no other way for our brains to handle the sheer quantity of tasks unless we want to be completely burnt out before we even get to work.
For all of the complex, non-routine decisions, our consciousness takes the wheel. Although sometimes, making these decisions can be associated with negative emotions. One reason for this is that for every active decision we take, there’s a risk of making the wrong call, which then makes us responsible for whatever consequences may arise. Despite — or maybe because of — this prospect of failure, we still find it extremely difficult to hand over the decision-making power to others. This applies to both the delegation of tasks to other people, but even more so when it comes to the transfer of choice to a machine. There is one main reason for this: The all-too-human fear of losing control, power, and relevance.
This can be illustrated using the paradoxical example of mobility: people tend to feel safer in their car, which they drive themselves than in an airplane, in which they fly as a passenger — although statistics would suggest the opposite.
The supposed control you feel as a driver naturally conveys a positive, comforting feeling. That’s why people are (currently) much more skeptical about the mere existence of fully autonomous vehicles. Even if there are endless amounts of evidence to support that most traffic accidents happen due to human error, people are still afraid to give up control. For many, simply picturing a machine taking over a previously human task conjures up images of a dystopian future where artificial intelligence creates robotic overlords that enslave the human race. But unlike many iconic works of science fiction, the reality of industrial automation will be much less dramatic.
Trust as a hurdle on the way to the Smart Factory
The reluctance to give machines decision-making power has certainly contributed to the fact that the Smart Factory is not becoming reality as quickly as was predicted a few years ago. This is especially true in Western countries, where the dependence on the human workforce was highlighted by the coronavirus pandemic and led to countless factories shutting down production for weeks, if not months. This crisis has shown just how important automation is in order to minimize the negative impact on the economy and safely restart the global supply chain.
In fact, most of the technologies required for this are readily available today. In Asia, the concept of so-called ‘Dark Factories,’ where machines handle so much of the manufacturing and logistics processes that manufacturers can turn the lights, heating, and air conditioning off entirely, and continue production autonomously non-stop, 24/7, has been around for many years.
Western society has not shown much enthusiasm to adopt these practices — until now. The automation of decisions is a central component of digitalization, and people’s reluctance around it is a central reason why so many companies, especially in Germany, are so far behind in their digital transformation journeys.
Digital technologies have been used in the shop floors for decades. But until now, this has primarily been aimed at the automation of physical tasks: From the ERP system to the Manufacturing Execution System to the PLC, data was transmitted to control machines and plants. These, in turn, sent back data so that the people in charge could get an idea and make their decisions.
Now software is not only able to transmit, evaluate, and condense the data — of which there are enormous amounts thanks to more and more sensors — into key figures. Algorithms can also autonomously trigger processes on the basis of the collected and analyzed data. This makes some of the human decision-makers obsolete, and the degree of automation — and therefore performance — increases by leaps and bounds.
An iterative approach to algorithmically-controlled production
But we need to start small. If companies want to achieve optimal performance and automate decision-making in their shop floors, an iterative approach is recommended. For one, this incremental transition helps to promote acceptance among the people involved. Secondly, the necessary framework conditions for algorithmic production can be systematically established.
In concrete terms, this comprises four fields of action:
1. standardizing machine connection and thus implementing vertical integration
2. establishing service-oriented machinery — assets that offer processing services, which increases flexibility
3. holistic integration — which includes the horizontal integration of departments and companies
4. realizing algorithmic scenarios using the appropriate formulas
Optimal implementation is when each progress brings a tangible, noticeable benefit. This not only has a positive effect on production performance but also contributes to wider acceptance across the organization.
Our algorithm to enable flexibility in production and logistics
In this context, it is also useful to start with algorithmic scenarios that act autonomously in areas that are difficult or unattractive for humans to access. Let it handle the most hated tasks. In most cases, there is enormous potential for automating decisions. And instead of rejection, there is almost always a welcome agreement because the employees benefit directly.
For example, we developed an algorithm for modular assemblies at MHP and introduced it to an automobile manufacturer. There, the software distributes the production orders to the machines at short notice — and in the best possible sequence. Additionally, the algorithm handles countless tasks, from orchestrating the logistics orders, to coordinating the material flow, whereby driverless transport vehicles (AGVs) are also used. As a result, machine utilization has been increased by around 20 percent, and logistics can adapt much more quickly to changes in the production process.
Great potential for further use cases
The algorithm developed for this automotive OEM’s specific application can be transferred not only to other companies in other sectors, but also to other scenarios. We’ve also identified two other use cases, which have already been designed and tested: first, the removal of material in the warehouse, and second, the organization of the assembly line.
Taking the human out of the equation: Logistics for manual picking according to the goods-to-person principle
The idea: no longer do people move through the warehouse to remove material.
Instead, Automated Guided Vehicles (AGVs) bring the material containers directly to the employees when they need it. To achieve this, the algorithm organizes the sequence of logistics processes based on upcoming logistics orders and determines the routes for automated guided vehicles.
Large e-commerce companies like Amazon and Alibaba have already been implementing such scenarios. In production, on the other hand — where logistics is usually more demanding and you have to take into account many other things, like for example, the sequence requirements — the goods-to-person principle [KS6] is not yet very widespread. This needs to change because there are considerable gains in efficiency. We can assume that simply by saving on physical space at the assembly site and transport equipment and therefore increasing employee efficiency, overall costs of production can be reduced by around 15%.
In addition, the logistics process is thus significantly less prone to errors because employees are less likely to spend time rifling through the wrong material container if the specific item they’re looking for is brought directly to them in the correct container at the exact time they need it.
The example also illustrates well where the limits of digitalization in production will lie in the coming years, and perhaps even decades. People won’t be rendered obsolete from one day to the next. A critical ability where humans will still be superior to machines for a long time to come: hand-eye coordination. Here, a lot of optical data must be acquired, processed, and translated into a motor impulse in a short time, which must then also be executed correctly. Even the most advanced machines today are still quite far off from coming close to a human’s response and dexterity.
Algorithmic assembly in a matrix layout
The principle of moving parts flexibly through space to an employee can be adapted particularly well for the assembly process as a matrix layout with additional degrees of freedom. The various assembly stations are located at fixed positions on the shop floor. However, they are not connected to an assembly line by a conveyor belt but are grouped together to form modular production islands.
The products to be assembled are transported by AGVs in the defined assembly sequence, which is controlled by the algorithm. This results in various advantages:
- a product to be assembled only passes through the stations required for this variant — which reduces model-mix restrictions and results in lower throughput times for some variants
- a product to be assembled remains in a station only as long as actually necessary for this variant, because no cycles have to be observed
- adding or removing individual stations has little or no effect on the existing system
A simulation for an automotive Tier 1 supplier showed — in comparison to a conventional control system of an assembly line — a reduction of up to 20% in the total throughput time. In addition, the production process becomes much more adaptable and less susceptible to errors. However, the amount of space required for such an assembly environment increases slightly — by around 12% in this specific case.
Use the opportunities for automation now
The options for optimizing production with algorithmic control are pretty much endless. But in order for our factories to become economically productive again, we need to embrace automation as soon as possible. Start small, implement, iterate. The scenarios presented here for the use of algorithms in production serve as a small example of the possibilities available in the entire supply chain. There are many, many other conceivable and feasible use cases.
The experiences already made here point the way forward. As a starting point, planning and control at the micro level seem to be particularly suitable because a lot can be achieved with relatively small changes. Companies that want to function in the New Normal need to take a hard look internally and evaluate their production & logistics processes, identify potentials, prioritize, then implement pilots and roll them out. Look beyond your own capabilities. The only way out of the upcoming economic crisis is to adapt quickly and embrace, rather than fear, the potential of innovative technologies.
— — — —
This article was modified from its original German-language publication, which you can read in Produktion.de
