AgriWear research partner Hochschule Niederrhein is conducting research into the use of artificial intelligence in agriculture. To this end, the practice partner “Neurather Gärtner” is using innovative algorithms to determine the population of pests within greenhouses in a semi-automated process.

According to the market research institute IDC, the wearable market in Europe has been growing at high double-digit rates for years. The majority of wearables sold are smartwatches and smart earwear in the consumer segment. However, as part of the megatrends of digitalization and Industry 4.0, wearable electronics are also increasingly finding their way into the production facilities of companies.

A common area for the use of wearables in production is the use of portable handheld scanners in picking processes. Picking processes occur in online retail, for example. When a customer places an order with a retailer, an employee has to collect the individual products in the warehouse. Picking in flow production systems is a special case.

Picking in Flow production systems

A flow production system consists of a conveyor belt with several stations. Each station consists of a shelf containing foods, an employee and a number of waiting places (buffers). A box is set up for each customer order and printed with a barcode. This barcode contains information about which foods the customer has ordered. The box is then placed on a conveyor belt. If a box requires a food item, it joins the queue at a station that has this food item in stock. As soon as the box arrives at the first position in the queue, the employee scans the box's barcode with their hand scanner and finds out which food they need to pick.

Figure 1 shows an exemplary flow production system. This consists of 3 stations, whereby each station has two shelf spaces, three buffer spaces and one employee. As some of the same foods are located at several stations for optimization reasons, a box can take different routes through the flow production system. In the example in Figure 1, the box shown requires potatoes and steak. As the product potatoes can be picked at both station 1 and station 2, there are two alternative routes for the box through the flow production system (red and blue dashed lines).

Routing-Algorithmen for Flow production systems

Since in reality such a system consists of dozens of stations and up to 40 foods are stored at each station, the routing of the boxes results in a complex mathematical optimization problem. This is particularly true as there are usually several hundred boxes in the flow production line at the same time, which means that interactions occur between the individual stations and boxes. The boxes should be routed in such a way that the production rate of the overall system is maximized. In order to achieve this, it is important to prevent too many boxes from being routed to a station at the same time, as the system would “clog up” in this case. In addition, idle times at individual stations (station employee is unoccupied) should be avoided as far as possible.

As part of a best-practice analysis, it became clear that companies often only use simple decision rules (so-called “greedy algorithms”) for the routing of orders. The advantage of greedy algorithms is that they are easy to use and can be solved quickly. However, both advantages usually come at the expense of performance, which cannot keep up with innovative approaches. As part of the AgriWear Living Lab, Duisburg scientists Prof. Dr. Jochen Gönsch and Sebastian Debold have developed new routing algorithms for flow production systems that combine both features - ease of use combined with high performance..

Simulation: Newly developed data-driven algorithms are up to 80% more efficient

In order to be able to analyze the effect of the routing algorithms without high implementation costs, the scientists modeled a flow production system in the simulation environment “MatLAB Simulink”. The scientists then calibrated the simulation based on real data from CONUS practice partners, which made it possible to develop a detailed model.

Within the simulation environment, the scientists tried out several dozen approaches for making routing decisions. From the results of the simulation, a routing approach was developed in which the expected workload of the stations is estimated in several key figures over the course of time. The key figures, which are also referred to as “expected workload”, can then be used as weighting factors in a second step when solving a “shortest path algorithm”, whereby the current state of the flow production line can be mapped well.

Figure 2 compares the performance of a simple greedy algorithm (left side) with the routing approach of the University of Duisburg-Essen (right side) in a flow production line with 10 stations. If the Greedy algorithm, which is widely used in practice, is used, around 5000 customer orders can be produced in 16 hours. This is due to the fact that a “bottleneck” occurs at station 6 (100% capacity utilization), as a result of which the capacity utilization of the other stations is only around 50%.

It is precisely these bottlenecks that are prevented by the university's new approach. By distributing the workload evenly across the individual stations, 9000 customer orders can be picked in the same period in the scenario shown above. This corresponds to an increase of around 80%. On the one hand, these increases in efficiency result in massive cost savings. In addition, the CO2 footprint of customer orders is drastically reduced, as energy in particular can be saved during production.

Sustainability through mathematical Opimization

The pilot project in the field of mathematical optimization of flow production systems using wearable electronics shows that the use of wearables in production processes in collaboration with innovative software solutions can lead to extraordinary increases in productivity. It is necessary to develop individually tailored solutions for the respective areas of requirement. The progress that can be achieved in this way in the area of more sustainable production does not involve any compromises.

While sustainability is often seen as a cost driver in companies - at least in the short term - such a trade-off does not occur with the algorithms developed by the university. On the contrary, it was successfully demonstrated that the social challenges of climate protection and cost-conscious corporate management need not be opposites.