Features and innovations
Feb 2024
5 min.
Features and innovations
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.
Shorter product life cycles and an increasing number of variants lead to constant changes in the product portfolio. The once efficient stock allocation then quickly starts to crack. This means: comprehensive stock transfers.
This article deals with the results with data science in detail, namely how you can continuously optimize warehouse occupancy in an intelligent way and minimize picking routes. The Warehouse Healing Model is based on data science and can be used in a chaotic warehouse.
Big data, data mining, machine learning and artificial intelligence are all related to the topic of data science.
The aim of data science is to generate knowledge from data in order to make the increasing complexity in the logistics environment more manageable for people.
It is based on large volumes of data (big data), such as those available from the warehouse management system. Big data consists of the 4 Vs: volume, velocity, variability and veracity. This is data that is available in large quantities, is growing rapidly and is diverse and reliable in terms of its content, sources and structure.
The processes of searching, collecting, filtering and analyzing data with the aim of identifying patterns are summarized under data mining. Data mining can be used both for the description of data and for forecasts.
Machine learning (ML) and artificial intelligence (AI) are further developments of data mining. ML is considered a sub-area of AI. The data mining process can be supported by AI and ML. Conversely, the developments or patterns predicted by machine learning or classic data mining can be used as the basis for an AI model. All in all, these topics contribute to data science.
Our warehouse healing model shows the extent to which data science can contribute to increasing efficiency in order picking. The warehouse healing algorithm continuously determines the optimal storage location for route-optimized picking at any given moment.
This is based on data collected by the warehouse management system. Shortly before each new proposal calculation of the Warehouse Healing Algorithm, the current data is pulled from the warehouse management system. Information on stock levels, topology, movement data and shopping baskets is analyzed with regard to patterns in the order history. Specifically, information on the storage bin, warehouse stock, stock movements, the loading unit, the item number master, the movement type and the picking area is transferred from the warehouse management system.
This data is used to determine where which items are located, in what quantity and how far they have to travel. This is used to generate stock allocation suggestions, taking into account the order frequency (known from the ABC analysis), the distance to the transfer location and the product affinity between the items (shopping basket analysis). The shopping basket analysis provides insights into which items tend to be purchased together and should be stored close to each other in accordance with route optimization. The aim is to create the shortest possible distance between the item to be picked and the transfer location as well as to the storage location of the next pick.
The resulting warehouse structure is shown as an example in the heat maps in Figures 1 to 3. The saturation of the red color corresponds to the picking frequency of the item. The lighter the color, the less frequently the item is picked. The framed boxes represent the goods affinity to the example article (marked with an X). The transfer location and therefore the destination for each picking run is indicated by "I/O".
Figure 1 is a chaotic warehouse with three aisles and no putaway strategy. Neither the picking frequency nor the goods affinity are taken into account. As a result, the warehouse is highly fragmented. Depending on which article has to be picked, the picking paths are of varying lengths. Apart from the picking sequence generated by the system based on the picking routes, the picking routes are not optimized.
Figure 2 shows the warehouse structure after sorting according to the ABC analysis. Frequently picked items are located in the immediate vicinity of the transfer point and rarely picked items at the back of the warehouse. The routes are partially optimized. However, articles with an affinity for each other are still spread across the entire storage area, which still has a negative impact on the picking times.
Figure 3 shows the warehouse according to the warehouse healing approach. Due to the consideration of shopping basket affinities, the ABC zoning is partially disrupted. However, this results in the greatest possible optimization of the routes. The effects of the two strategies in Figures 2 and 3 on efficiency become clear in the practical section.
The warehouse should be structured according to the warehouse healing concept by means of stock transfer processes with little (time) effort and with the greatest possible effect. To achieve this, the processes are evaluated and prioritized. A stock transfer represents an exchange of two stored items. A storage bin corresponds to a point on the three-dimensional coordinate system, which is designed for the customer's individual storage or picking area.
Each storage bin is assigned a score, i.e. an evaluation, using a mathematical model, taking into account the item stored there. A stock transfer increases the retrieval distance for one item and shortens it for another. The score calculation must therefore be individually adapted to the customer's products and warehouse structure.
The employee is shown the stock transfer orders in a table with descending usefulness, for example on his mobile data collection device.
The best possible stock transfer order has the following three properties:
With regular and disciplined execution of the stock transfer proposals, the stock allocation is continuously optimized. The manual workload is increasingly reduced or remains at a consistently low level.
The investigation of the increase in efficiency through the Warehouse Healing Model is theoretical and practical.
Theoretical approach: The model was simulated by our employees using real data from a pilot customer's three-month period. A simulation involves estimating the route to be taken when picking all orders received. This calculation is repeated 300 times for different numbers of handling transfers and presented in relation to the initial situation (without handling transfers). This method offers the possibility of a theoretical benefit assessment. It provides information on the effectiveness of the combination of frequency and goods affinity analysis in comparison to the usual ABC sorting.
Practical investigation: The model was implemented as part of a pilot project at a customer. A period of three months was analyzed, during which the employees started the healing process eight times. On average, nine healing relocations were carried out per initiated process.
The data of both evaluations refer to the picking area shown in Figure 4. The picking warehouse has a right-angled "L-shape" (outlined in red). Picking takes place in a serpentine pattern through the racks. "I/O" is the transfer point. The aids are: Smart devices and an order picking trolley with six compartments. On average, five compartments are used on the picking trolley. Picking is carried out according to the multi-order picking principle.
The simulation results: The effect on the stock removal route shows the warehouse sorting by pure ABC sorting (red curve without shopping basket analysis) and that of warehouse healings (green curve with shopping basket analysis). The 100 % of the Y-axis is the state of the warehouse without strategy.
For clarification, the curves can be linked to the heat maps from point 2 Warehouse Healing. As an example, the 100 % corresponds to the warehouse status in Figure 1, the red curve to the warehouse status in Figure 2 and the green curve to the warehouse status in Figure 3.
The y-axis shows the percentage reduction in the stock removal path in Figure 5 and the resulting increase in efficiency. The x-axis contains the number of healing stock transfers carried out.
With 10 stock transfers, the effect is still identical with and without shopping basket analysis. Both strategies achieve an average reduction in the stock removal distance of 10 %.
After 38 stock transfers, the warehouse healing model achieves a 15% reduction in the stock removal distance, which is only achieved after 90 stock transfers in the red curve.
After 90 stock transfers, warehouse healing achieves a 20% reduction in the retrieval path, which classic ABC sorting does not achieve.
The red curve shows 83 % in the long term. According to this simulation, a maximum reduction in walking distances of 17 % can be expected with pure ABC sorting. The green curve shows 76 % in the long term. By taking into account the affinity of the goods in addition to the picking frequency, a reduction in walking distances of 24 % to 25 % can be achieved.
With ABC sorting, 60 to 70 rearrangements make sense, after which the effect decreases. With the warehouse healing model, this effect only occurs after around 100 stock transfers. The asymptotic curve is due to the prioritization of the stock transfer proposals.
Figure 6 shows the effect of the stock transfers carried out in the warehouse shown above according to the same scheme as in Figure 5.
Eight apportionment processes were carried out in the three months. The chart shows that the employees did not always choose the best apportionment proposals.
This effect can be observed above all in the purple depicted relocation process. Six healing rearrangements were carried out, but these had no effect on the removal path. In the green and orange curves, the healing transfers successfully contributed to a shortening of the stock removal path. In both stock transfer processes, a reduction in the stock removal path of 5% was achieved with six and seven stock transfers respectively.
The simulation proves that the implementation of warehouse healing can reduce stock removal paths by 20 to 25 %.
Even in the pilot project (despite the less than ideal implementation), an overall reduction in stock removal routes of almost 19% was achieved across all stock transfers carried out. The benefit is greatest with regular, consistent and correct implementation. The slope of the curves shows that the first stock transfers have the greatest effect. They form the basis for an efficient warehouse structure.
If the travel time is 55 % of the total picking time, a reduction in travel time of 20 % means an increase in efficiency of 11 % of the total picking time.
Warehouse Healing could be used holistically to further increase the efficiency of order picking. The healing algorithm could already be used to find storage locations for goods putaway. Although stock transfers could not be completely avoided, as the ideal storage location would probably be occupied at the time of putaway, the next best storage location could still be selected. In this way, short walking distances are already guaranteed before the heal transfer.
The software plays the main role in warehouse healing. It represents the intelligence of the entire concept. After implementation, it learns and operates in a self-organized manner. However, warehouse management software is required to integrate the warehouse healing algorithm in order to use the data obtained from it. Thanks to data science, the software is able to link a wide variety of data with each other.
The software behind Warehouse Healing transforms a mass of empirical data into simple work instructions in a matter of seconds. The use case-oriented preparation and evaluation of the data to create simple and clear information is the key to successful use.
The Warehouse Healing example shows how much potential can be discovered and utilized through software and a good database.
Notes: The origin of the data and information used to study the use cases are partly theoretical and partly from practical experience. The investigations also relate to different use cases with different methodologies. As a result, the results are not exactly comparable with each other and are likewise 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 automation results in detail: Automation increases efficiency by 97% (semi-automated) and 140% (fully automated) in order picking.
Blog post: Outlook for future developments: Increasing efficiency in intralogistics.
