Features and innovations

Automation increases efficiency by 97% (semi-automated) or 140% (fully automated) in order picking.

Intralogistics 4.0: How much can efficiency be increased with smart devices, automation and data science?

Rebecca got to the bottom of this question. She studied industrial engineering and wrote her bachelor's thesis on the topic with us:

Assessment of the increase in efficiency through the implementation of smart devices, automation and data science in our warehouse management software (WMS) SuPCIS-L8 in order picking.

This article looks at the results in detail, specifically how the integration of an automated storage system, for example an AutoStore, for semi-automated picking in the investigated use case results in an increase in efficiency of 97% compared to manual picking with smart devices. And how increasing the degree of automation through robotics for fully automated picking brings a further increase in efficiency of almost 140% compared to the semi-automated process.
1. reasons for automation
  • Automation is a suitable measure to counteract the shortage of skilled workers. Employees could be deployed specifically for selected, less stressful activities.
  • Nowadays, error-free picking down to batch size one with consistently good performance is essential. This can be implemented very well with automation and robotics solutions.
2nd automation solution: AutoStore and eOperator

AutoStore is a cube-based automated storage system:

  • The warehouse gains in productivity and efficiency, which leads to price and quality advantages over the competition.
  • The implementation of the AutoStore will greatly consolidate and thus improve the use of storage space.
  • Employees can be trained quickly (usually within 30 minutes) and achieve an almost zero error rate.
  • This warehousing leads to increased occupational safety due to ergonomic workstations as well as security with regard to theft or unforeseen stock removals.
  • High adaptability and flexibility, both during implementation and use.
  • It can be combined with any other technologies and picking techniques.

AutoStore is a goods-to-person system and consists of five components: the bins, the grid, the robots, the ports and the controller.

Figure 1: AutoStore structure

The bins (containers) with the goods are framed by the grid (aluminum frame) and stacked on top of or next to each other. The top struts of the grid serve as rails for the robots. There are two types of robots, which are used depending on the size of the crates and the performance requirements. One robot can pick up and transport a bin.

The workstations are located at the ports and are the input and output stations of the AutoStore. There are also different variants here. The carousel port is used in this work. This has three bin locations that are automatically swapped by rotation. This provides space for the bin to be processed as well as for the upstream and downstream bins.

The controller is responsible for traffic control and database management. However, it is only used for machine control of the container orders, based on the container ID. Warehouse management software is required for the assignment of goods to containers, order prioritization, picking dialogues and picking strategy.

New bins are placed on top of the bin stacks by the robots. Rarely used bins automatically move downwards over time. If a low-lying bin (green) is to be removed from storage, the robots dig it out as shown in Figure 2. To do this, the yellow and blue bins are first moved to the side and then replaced.

This results in a natural ABC sorting. Slow movers slide down over time, while the fast movers remain on the surface. This minimizes rearrangement and ensures efficiency.

Figure 2: Digging the containers out of the AutoStore

The port can be picked manually by an employee or fully automatically by a robotic gripper arm such as the eOperator from Element Logic.

The eOperator has a gripper with suction function and a camera for visual control and coordination. This allows products of different sizes and shapes to be removed from the container in any arrangement.

The robot's software is equipped with artificial intelligence. All data from every pick made by a robot worldwide is collected on a cross-customer cloud. This includes: Item properties, gripping strategy used and error data. Each robot is able to learn from this data and improve its gripping method. This enables the removal of unknown items. If a new product is added to the range, the picking robot does not need any training.

Figure 3: eOperator

1. AutoStore
2. Port / Workstation
3. Bin / container
4. eOperator
5. Gripper
6. Camera
7. Target container on a driverless transport system

The eOperator system also has an operator console, a safety control unit, a power distribution unit and a processor.

3. methodology

The increase in picking efficiency is determined using two different AutoStore automation concepts. The first use case is partially automated and consists of manual picking from the AutoStore. In the second use case, picking is fully automated using a picking robot, the eOperator.

Customer data from the field is evaluated for the performance of the first use case. Our warehouse management system (WMS) is used in the warehouse. By analyzing the data, we gain insights into the average picking time and the consistency of performance as well as the factors influencing it.

The use of the picking robot for fully automated order picking is assessed on the basis of a theoretical model test by Element Logic.

4. framework conditions and methodology of semi-automated order picking

The customer has several picking areas, one of which is the AutoStore with three ports and approx. 10,000 containers.

The evaluation extends over two months. During this time, the activities from the dispatcher log files of all three ports are evaluated. Work is carried out from 5 a.m. to 8 p.m. in two shifts. Due to lower order volumes from the AutoStore, picking is carried out as required, i.e. not always a complete shift.
The dispatcher log records every activity of all three ports with a time specification. This means that the time of each function, be it a mouse click, a scan or an acknowledgement, can be traced.

The time between order start and end is determined by cumulating the differences and dividing by the number of picks made. This results in the average time per pick.

Outliers are also identified. These values are so far away from the mean that they could distort the average values. If they deviate more than three times from the mean value, they are automatically deleted by the Z-score function.

The frequency of the picks actually made was used for the evaluation based on the number of repetitions of the pick functions in the log file.

5. framework conditions for the investigation of fully automated order picking

The model test was carried out by Element Logic.

Figure 4: Test setup

The source and target containers are located parallel to each other at the same height. The source container is fed through the carousel port. Removal is based on the single-order picking principle. As no conveyor technology is available, a time of three seconds is assumed for the delivery of the target container. This estimate is based on Element Logic's experience and corresponds to the speed of the conveyor system of a customer reference.

The robot gripper arm receives stress balls of the same size for removal. However, this represents a discrepancy to the practical application. With a wide range of products, it is very unlikely that the robot will be able to pick up every item without any problems.

Note: The items move in the container due to the round shape. This makes the automated gripping process somewhat more difficult, which leads to additional time being required. The test is based on a mixed order structure with one item each containing one to five picks.

6. assessment of the comparability of the data

The data from smart devices is not exactly comparable with the data from automation due to the different recording methods and circumstances.

The same applies to the comparison of the semi-automated and fully automated use cases. Due to the evaluation of real production data, the semi-automated process is subject to different influences than the theoretical model test under simplified ideal conditions.
The efficiency of the eOperator depends on other factors in addition to the process design. For example, the item accessibility, the order structure, the positioning and topology of the components at the workstation and the efficiency of the other players in the process.

The results are not guidelines or performance promises. Nevertheless, the work provides a valuable insight into the development of efficiency with increasing mechanization.

Due to the conditions in the test environment, a different process is described as an example process for fully automated order picking than the process on which the efficiency determination is based.

The purpose of the process description is to provide an overview of the interfaces, context, functionality and application of the selected technology. For this reason, the process description presents a realistic process based on practice that is not related to the pilot project.

7. practical part: process description - semi-automated with manual picking from the AutoStore

This process is based on a customer's process with our WMS. The AutoStore is connected to a conveyor system for downstream activities. Each compartment (divided up to 16 times) of the bins is clean.

Upstream activities: The order data is bundled and transferred to the AutoStore at regular intervals with stored priorities. The employee logs on to the port and selects their activity.

Picking: Click on the screen to start the picking process. The AutoStore delivers the first bin to the port. As the Carousel port has three bin storage locations, the next bin is also delivered and buffered immediately. In the AutoStore, further robots are waiting at the supply lock with the bins to be processed subsequently. During the entire picking process, bins are moved and prepared for seamless provision.

At the same time, the WMS starts the order start dialog. This provides all the information about the items and characteristics of the order as well as a recommendation of the size of the target container to be selected. The system works with returnable containers in three different sizes. The container size is determined on the basis of software-based volume calculations and displayed as a suggestion.

The employee takes the target container from the container stack, which is either next to it or behind it. The target container is scanned, whereby the order information is married to the target container. The order data can now be determined via the container barcode.

For additional identification, an order start label with a QR code containing all the important information is printed automatically. The employee places this in the target container.

An information graphic then opens, showing the subdivision of the bin and the removal to be made (number of removals, number of containers and remaining number in the sub-area of the bin). The employee then removes the items and confirms the removal with a click. If an inventory check appears useful, the WMS prompts the employee to check the remaining quantity, make any stock corrections and confirm with a click.

The WMS checks the characteristics of the next extraction.

  • Is this into the same target container and from a different bin?
  • Is this done in a new target container from a new bin?
  • Is this done in a different target container from the same bin?
  • Is this done in the same target container from another compartment of the same bin?

The WMS gives the employee further instructions. If the next removal is from a different bin, the employee is requested to release the bin for removal. If the next removal is for a different picking bin (target bin), the employee is requested to place the picked bin on the conveyor system. If the next removal is for the same target container, the process for removing a new one begins.

A robot is available in the AutoStore to remove the source containers. The discharged target containers are transported further towards the packing station by the conveyor system.

8th practical part: Process description - fully automated picking of the eOperator from the AutoStore

The theoretical attempt by Element Logic could not reflect the overall picture according to the practical application. The classification in the overall context with upstream and downstream processes and the software-side logic refer to a practical example.

The environment consists of the AutoStore, the picking robot, a conveyor system and the software. The eOperator's picking process is controlled by a separate module from the manufacturer, as is the AutoStore. The eController software controls the conveyor system. The WMS is used for interface coordination between the systems and for transmitting higher-level information such as order data, master data, inventory data and prioritization.

The picking process begins when the orders are transmitted to the AutoStore by the WMS. Similar to the semi-automated process, the orders are transferred to the AutoStore in regular time intervals with the stored priorities.

Depending on the priority and availability, the first order is selected by the WMS and converted into a pick list that can be read by the robot. The pick list is the basis for the instructions to the robot with regard to the next removal(s) from the next bin. The identification number (ID) of the pick list is married to the bin ID.

Prerequisites for picking:

  • The pick list may only have one position.
  • The bin must be clean.
  • The article must be within the robot's reach.

If a prerequisite is missing, the order is processed by an employee. In this case, the semi-automated process described above applies.

If all requirements are met, the WMS assigns the task to the picking robot. At the same time, the AutoStore delivers the container with the corresponding ID to the port. The picking process then begins. To do this, the camera records the contents of the bin. The eOperator software does not know which item it is. The shape and arrangement of the item is recorded on the basis of pattern recognition. The robot software determines the optimum gripping method. The robot moves to the item, positions the suction cup, sucks it in with the removal device and encloses it using three gripper fingers. It moves towards the target container, which is positioned on the conveyor system, and places the item down. The robot then returns to its standard position.

The software checks whether the pick was successful. If an error is detected, it is specified and the cause is determined. If the problem lies with the item properties, the item's pickability flag must be changed. The pickability flag describes the pickability by the eOperator and is stored in the item master data. The data and findings of both successful and unsuccessful picks are transmitted to the cloud.

The contents of the bin and the target container as well as the inventory data are updated by the WMS. The eOperator software checks whether there is another pick. If so, the picking process is restarted. Once the task has been completed, the WMS checks whether there is another pick for the order. If this is the case, the target container remains on the conveyor system. The WMS then checks whether a new bin needs to be delivered for the next removal. If so, the eController sends a request to deliver a new bin and remove the current one. Otherwise, the current bin remains at the workstation. In both cases, the WMS must generate a new pick list that can be read by the robot and the process is repeated.

Once picking of the entire order has been completed, the WMS selects the next order to be processed according to priority. The conveyor system is instructed to move the next bin forward and remove the finished bin. The WMS checks whether a new bin needs to be brought forward and transmits the corresponding instructions to the eController or the AutoStore software. The process starts again.

The upstream and downstream activities are of great importance in terms of efficiency. In practice, fully automated picking is often followed by an inspection zone. Faulty crates are ejected to a clearing zone for manual inspection or reworking. The checking process was not included in the process or in the calculation, as no reliable data is available. Nevertheless, this process must be taken into account in the planning.

Automatic target container provision also makes sense in order to utilize the efficiency achieved through automation. If picking is carried out in the shipping carton, an automatic carton folder and erector is recommended.

9. evaluation of the increase in efficiency - partially automated with manual picking from the AutoStore

The evaluation of efficiency in Table 1 contains the average times of all withdrawals made at the ports during the period under review.

Table 1: Results of the data analysis of semi-automated order picking

Fewer orders with fewer picks were picked at the first port. Port 2 and Port 3 were almost equally busy. The average number of picks per order is between 2.5 and 3.1. The average picking times per pick are similar for all three ports, resulting in an average picking time of 22.34 seconds.

The histogram in Figure 5 shows that picking times of less than 20 seconds were measured most frequently. However, there are a few picking processes that have significantly higher picking times, which results in the higher average value.

Figure 5: Histogram of the measurement results

Figures 6 and 7 describe the picking times per pick on an hourly and daily basis. Figure 8 shows the average times of all days per port grouped by hour of the day.

Over the course of the day (Figure 9) and over the entire observation period (Figure 10), large fluctuations in performance can be seen for each time and port.

Patterns and tendencies are often contradictory and are difficult to explain without specific reference to events in the camp.

Figure 6: Progression of picking times at hourly level

Figure 7: Progression of picking times on a daily basis

The figures can be better interpreted in a less detailed context (Figures 8, 9 and 10). In contrast to Figures 6 and 7, there is no differentiation between the ports. They are summarized in one measured variable.

Figures 8 to 10 are shown on an hourly basis. The values in Figures 8 and 9 are average values for the entire period under review. Figure 10 shows the time distribution of all orders picked during the period under review.

By combining the ports into an average pick time, the fluctuations are largely eliminated. There are fluctuations at the beginning and towards the end of the working day. In the middle of the day, the performance remains largely constant. On average, the best performances occur at 6 am, 4 pm and 7 pm. The picking performance is weakest at 5 a.m. and 7 a.m.

Figure 8: Average picking time over the course of the day

Figures 9 and 10 are used to examine the extent to which the order structure and the order volume influence the picking performance in Figure 8.

The comparison of Figure 8 with Figure 9 provides information on the influence of the order structure. The average number of picks per order is constant between 2.5 and 3 picks per order. There are only deviations in the mornings and evenings. This suggests that with a high number of picks per order, the corresponding picking time could be reduced due to the lower number of ancillary activities per pick. This assumption is only partially confirmed. At 7 p.m., the average number of picks per order is particularly high, while a best picking time is achieved in Figure 8. This correlation confirms the thesis mentioned above. However, the other two peak performances in Figure 8 at 6 a.m. and 4 p.m. cannot be explained by the order structures in Figure 9.

Figure 9: Average number of picks per pick order over the course of the day.

Figure 10 shows that the fluctuations in picking performance cannot be explained by the fluctuations in order volumes over the course of the day.

There is no discernible correlation between the peaks in picking times (Figure 8) and the peaks in order processing over the course of the day (Figure 10).

Figure 10: Number of pick orders over the course of the day

Most orders were picked at 5 pm. The second peak takes place at 8 a.m. with around half the order quantity of the first peak. In between, the curve fluctuates with rather low order volumes depending on demand.
However, the changed order structure (Figure 9) in the off-peak times can be explained by the very low order volume in Figure 10. It can be assumed that orders with special specifications for extraordinary purposes were picked at these off-peak times.

10. evaluation of the increase in efficiency - fully automated picking from the AutoStore with the eOperator

The performance results from the measurements in Element Logic's theoretical test.

The average picking time in this calculation refers to one pick. The process time is made up of three components. The time of the port for changing the source containers, the time of the gripping process by the robot and the time of the conveyor system for changing the target containers.

The total throughput time is 8.39 seconds per pick. The throughput time of the robot gripping process is 6.46 seconds per pick. However, this depends heavily on the layout of the robot cell, for example if the robot has to turn.

The measurements of the test result in a number of 381 picks per hour. The throughput times of the test imply that 400 picks per hour would be possible under the given conditions.

No data is available on errors and the reliability of the system. The controlling cockpit provides information on the qualitative characteristics of the errors. The employee is shown the corresponding number of errors per error category in a dashboard based on a live evaluation.

An example of an error category could be that the item cannot be gripped by the robot. This may be due to the topological properties of the product, poor lighting conditions or inventory differences in the source container. Or technical defects, missing target or source containers or unintentional dropping of the item.

11. interpretation of the results

When comparing the fully automated picking by the eOperator and the semi-automated picking by humans and AutoStore, it should be noted that the fully automated process took place under very simple conditions. With this in mind, it is nevertheless the case that the use of robotics enables a major leap in efficiency.

Semi-automated picking achieves a productivity of 161 picks per hour after extrapolating the picking times. Fully automated picking achieves a productivity of 381 picks per hour. By using the eOperator, the efficiency of semi-automated picking from the AutoStore can be increased by 137%.

It is always advisable to check whether this service is required in your own warehouse. The efficiency gained must be utilized through adapted processes. The upstream and downstream processes must also be taken into account. This is because fully automated robot picking only makes sense if the capacity is fully utilized over a long period of time. This can be achieved, for example, through activities such as folding and placing the target cartons on the conveyor system or automatic packaging with application of the shipping label.

To distribute the efficiency gained even better, the eOperator could be used for picking from two ports. This would reduce the downtimes of the eOperator. However, this requires very good control by the software.

Using the example of the customer who provided the data basis for semi-automated picking, the use of picking robots would still be unattractive. Manual picking at the AutoStore is already so efficient compared to the other picking areas that work is only carried out at the ports on an hourly basis. The robot would currently have too much downtime.

When implementing automation technologies, it is important to check which measures really make sense in order to achieve the efficiency targets. This depends on the requirements, the warehouse structure and the processes. A decision can then be made as to whether automation measures make sense across the board (fully automated use case with eOperator and AutoStore) or only in certain areas (manual picking from the AutoStore).

12. importance of the software

Automation has increased efficiency many times over. Order processing is faster, the visible tasks are simpler, but the background processes are more complex.

With increasing automation, the number and complexity of components increases. An automation concept includes the spatial linking of hardware components, the system-related linking of different software and the connection of all hardware systems with the appropriate software systems.

The WMS combines all components and allows the processes to interlock. It manages the warehouse and stocks, organizes, prioritizes and processes orders. The WMS also informs and coordinates the components involved and controls the entire intralogistics processes.

New information is generated and old information overwritten at each communication node. The WMS controls the material supply and carries out rebookings independently. The software also serves as a communication interface between man and machine outside of the process.

Other key functions of the WMS are the inventory and cockpit functions for the control station. Real-time information from the warehouse or process is processed in order to control and plan processes.

Notes: The origin of the data and information used to investigate the use cases is partly theoretical and partly practical. The studies also refer to different use cases with different methodologies. As a result, the results are not exactly comparable with each other and are not universally valid.

In addition, efficiency in order picking depends on many factors. For example, the process characteristics, the warehouse topology, the order structure and the interaction of technologies. Efficiency is correspondingly variable depending on the application.

Nevertheless, the approach chosen for this work is still the best one to answer the leading question. This is because the respective efficiencies of picking can be analyzed and related to each other, taking into account the differences. Thus, the findings obtained are complete, valid and plausible when the underlying overall concept is taken into account.

Blog post with an overview of the overall results: Amazing results in increasing efficiency through smart devices, automation and data science.

Blog post with the smart devices results in detail: Efficiency increase of 23% with smart devices in order picking.

Blog post with the data science results in detail: Reduce outsourcing routes by 20-25% and optimize efficiency through intelligent algorithms.

Blog post: Outlook for future developments: Increasing efficiency in intralogistics.

Contact us and talk to one of our experts on the subject.