Currently, two of the leading industry nations, Germany and China, are pushing their industry to increase a higher degree of automation. Automation is among the key technologies of concepts such as Industry 4.0 and Made in China 2025 [中国制造2025]. The goal is not automation in the traditional sense, i.e., the fixed and rigid implementation of static processes which are to be repeated millions of times in exactly the same way. Instead, decisions should be automated, i.e., the machinery carrying out processes in production and logistics should dynamically decide what to do based on its environment and its current situation. In other words, these machines should become intelligent.

As a researcher in optimization and operations research, this idea is not new to me. Actually, this is exactly the goal of work and it has been the goal for the past seven decades – with one major difference: the level at which the automated, intelligent decision process takes place. In this article I want to shortly discuss my point of view on this matter.

Operations research usually is associated with the perspective of the manager of a factory. The manager knows the orders that the customers make. She knows the current state of the factory and all its production lines. She knows the types and amounts raw material coming into the factory. She knows the working hours of the shifts of the factory workers and when which machine will go into maintenance. She needs to decide how to split the customer orders into production tasks, how to assign these tasks to production lines, and in which order they should be carried out.

One of the design principles of Industry 4.0 which I heard in multiple discussions is the idea of autonomous, decentralized decisions. Machines should be able to communicate with each, organize themselves, and ideally decide by themselves what to do to fulfill the production tasks. This fits more to the level of the factory worker, who is notified about the current task, sees the current state of her environment, can communicate with her peers, who then (either separately or together) decide how to get the current tasks in the current production line done. She and/or this collective can dynamically react to problems such as failing machines, accidents, or other unexpected situations.

In other words, there are two levels of production or logistics where intelligent technologies may automate decisions, the higher level of the management and the lower level of the factory floor. Both levels have very different requirements.

At the top level, we clearly have much more information about the whole system. Parts of these information may arrive with a certain delay due to communication, e.g., a machine break down in production line 10 may be communicated via a phone call. The planning horizon is longer, as we may make plans defining our work over the next few days or weeks. We also have relatively much time for planning, ranging from minutes to hours, so we can take our time to find a good solution.

On the lower level, there is less information, as a factory robot will usually not be able to access the whole dataset of the company's ERP system. Instead, it will have a couple of sensors and be able to communicate with the other robots in the same production line. Its information will be much more up-to-date and maybe arrive in intervals of only a few milliseconds. Reactions must also take place quickly. The planning horizon and the time available to make the plans are both much shorter, maybe seconds to an hour and milliseconds to minutes, respectively.

As a researcher in optimization, I think the increase in productivity that we can except by automating decisions on both levels will be different. Since the top level has much more information, it should clearly be able to produce a higher increase. This becomes clear on a simple example: If you plan a travel from Berlin to Paris, you may do so by just the information from your street map and pick the shortest route. If you have more information, namely also the traffic (jam) information, road conditions, and the speed limits on each road as well as the corresponding expected average speed on each road segment, you can most likely find a better, faster road, if one exists.

Similarly, if I am a worker who is supposed to store and retrieve stuff from a warehouse, I can make better decisions if I know all the storage and retrieval orders to come in the next ten hours compared to a scenario where I only know the orders for the next ten minutes.

I, personally, would bet my money on operations research and top-level optimization instead of automating more complicated low-level decisions to increase productivity. Disclaimer: Incidentally, my work mainly considers this top-level, so I am biased. Anyway, automating intelligent lower-level decisions clearly has lots of merits too. It is, for instance, a good idea to keep production lines going in case of unexpected events. It can also achieve a high level of self-organization and resilience and scales much better, as we can add new machines and production unity without needing to change anything in the planning process. So, all in all, both methods together will most likely achieve the best outcome.