The increasing volume of online retail sales is pushing traditional logistics and intralogistics systems to their limits. Complex sorting processes for parcels of all shapes and sizes, return rates of up to 60%, short delivery times and high-quality picking processes all consume large amounts of resources and involve significant costs. In addition, the parcels are generally handled manually, which is a time-consuming and error-prone process. As a result, logistics experts are turning to integrated automation solutions in the context of the Smart Factory and the Internet of Things throughout the value chain.
Image processing forms basis of machine learning
By networking systems and connecting machines using sensors, entire logistics processes can be controlled from a single database. The data gathered by means of image processing will form the basis for machine learning. Cognitive learning processes based on large volumes of data allow machines to learn for themselves. They can tell for example, whether a package is damaged, how it must be classified and sorted or whether it is properly sealed.
The introduction and development of Logistics 4.0 technologies play a decisive role in the competitiveness of both small and medium-sized enterprises and large international logistics companies. Three real-life examples demonstrate the contribution that image processing can make in this area.
Automatic data capture in real time reduces shipping costs
Where large volumes of goods are shipped in individual parcels, incorrect information about the size and dimensions of the products can lead to two types of additional costs. Using unnecessary extra packaging materials and shipping parcels containing empty space is expensive. However, more money is wasted if the wrong shipping company is chosen, where the shipping charges are based on weight and volume.
To find the best possible price/performance ratio among the many shipping service providers and their various offerings, companies that dispatch large volumes of goods often negotiate specific service packages or quotas in advance. For example, a large e-commerce organization may ship 10,000 medium-sized parcels every day, each weighing 2 kilograms. This amounts to a total of 20 metric tons and an estimated volume of 250 cubic meters. On this basis, the company agrees on quotas for the different categories of parcels with the shipping service providers. Because it is not possible to forecast the volume of orders or the number of packages, when a parcel passes along the conveyor belt a decision must be made in real time about the best quota for that parcel, how the remaining quota can be used most efficiently and, therefore, how the parcel can be shipped most cost-effectively.
Based on the volume, the weight and the resulting decision, the parcel is automatically labeled for the appropriate shipping company. If the data has been recorded incorrectly, this process will not function properly and the parcel cannot be matched with the best shipping solution. As a result, the quotas that have been negotiated will not be used effectively and the shipping process will become inefficient. Minor problems can quickly add up to large amounts of money being wasted. This could be avoided if, the data is being stored centrally and captured correctly.
An advanced Logistics 4.0 solution that allows parcels and quotas to be fully automated consists of volume light grids integrated into the conveyor and the machine control system. Scales can record the weight of a parcel and the volume light grid can capture its dimensions, all at a conveyor speed of up to 2 meters per second. The data is stored in the enterprise resource planning (ERP) system as soon as the parcel has passed through the measurement barrier. It can also be linked with the selected delivery quota. By matching the data and the current quota capacities, the control system can identify that shipping company ‘X’ has free capacity for small, heavy parcels and can attach the correct shipping label to the next parcel that falls into this category. In addition, a camera can be added to the system to document that the packages were undamaged when they left the company.
The individual parcel data is constantly synchronized with the ERP, logistics and merchandise information systems in real time. A logistics dashboard can be created based on this complete data set that can answer several analytical questions to help the company’s logistics managers. These include:
- How much of the quota has already been used?
- Which parcels from stock have already been booked in?
- Where is there availability in other areas?
- What is likely to be sent based on existing forecasts?
The process of reducing the shipping costs and choosing the best possible shipping company can be fully automated, enabling the logistics manager to access real-time data and manage the logistics process from a remote office or even from home. In addition, intelligent evaluation software can be used to strategically plan management activities and to negotiate better conditions for delivery quotas with shipping service providers.
This makes it possible to answer questions such as:
- What are the actual costs per parcel?
- What is the forecast volume?
- Which parcels have been sent based on certain conditions?
- What does the shipping behavior look like over time?
The result is a complete Smart Factory workflow at the conveyor involving large volumes of parcels. Sensors and hardware are integrated into the overall system, reducing costs and resources.
Enhancing ERP enables appropriate packaging
In smaller companies or businesses with low volumes of products, the items that arrive in the receiving department are often measured by hand. The master data is then entered manually in the ERP, merchandise information or logistics system. This is very time consuming for the staff involved and often leads to inaccurate measurements and data entry errors. Items that are not a specific geometric shape frequently give rise to errors. However, the master data must be accurate to enable packaging to be kept to a minimum, to allow the smallest possible and most efficient storage location to be chosen and to ensure that the customs clearance is correct for international parcels.
In contrast to lasers, light grids can also process partially transparent, black and reflective materials and can operate at a higher speed.
A simple Logistics 4.0 solution for recording master data correctly consists of a semi-automatic system with a volume light grid and an integrated scale. By automatically defining the minimum size of the item, it is possible to choose the ideal packaging and storage location. The item is measured using a light grid that is passed over the item in one movement. The object is detected as it breaks the light path and provides information about its size and dimensions which are accurate to the nearest millimeter. In contrast to lasers, light grids can also process partially transparent, black and reflective materials and can operate at a higher speed. The system functions without the involvement of staff, resulting in correct and consistent data that can be transferred via an interface directly to the ERP system and processed there as required. For warehouses and other facilities that are not connected to a WLAN or database, a battery-operated mobile version of the light grid is available, allowing the data to be fed into the system via a USB stick.
Fully automated sorting with machine learning
Machine learning is the ideal solution for processes where cognitive abilities are required. For example, one of the major challenges facing large e-commerce retailers and shipping and courier companies is sorting packages to ensure that they are processed correctly. Machine sorting is almost impossible because of the wide variety of different types of packaging. Individual parcels cannot be processed using conventional technology because they come in so many different shapes and sizes. When large volumes of goods need to be sorted, manual processes are still used. This leads to considerable costs and slow sorting procedures with high error rates.
The term “machine learning” is used to describe giving an algorithm the ability to learn for itself. First, a large amount of image data is supplied to the algorithm, for example, tens of thousands of images of packages stacked on a conveyor. This represents chaos to the computer, so the image data is initially classified by a human. The algorithm is told “This is a parcel,” and is then able to sort the packages into categories based on their individual features. The classification process is carried out using a neural network. In this case, a neural network is the same thing as deep learning. It is a multi-stage process requiring a large amount of computing power, separating the data during several phases to classify it. The intelligent function of the algorithm involves identifying image patterns and learning from previous classifications. In this case, the machine learns the parameters that identify an object as a parcel. This enables the algorithm to classify individual objects independently in the future.
Machine learning is used where imaging sensors have reached the limits of their capability, such as sorting a variety of package types. Three-dimensional scanners are not an adequate means of identifying the shape of the parcels and operate only at low speeds. For a system with a robotic picker, this leads to inaccurate sorting, incorrect routing of parcels, system failures, waiting times and time-consuming manual follow-up work. By contrast, systems based on machine learning can accurately identify the parcels and their shape using the classification scheme that they have learned, and the robot can carry out the sorting process at high speed and without manual intervention.
The data is first recorded by a camera system and analyzed by a local computer. The algorithm sees the picture, decides whether it recognizes any of the patterns it has learned and, if so, which ones are involved. It can then automatically control the sorting procedure and issue commands to downstream phases of the process. At the same time, the algorithm is learning from each new picture and fine-tuning its criteria. Image processing with accompanying artificial intelligence can create a fully integrated and automated Smart Factory solution using machine learning.
Machine learning as a strategic competitive advantage
The recorded data that can be used to carry out a complete analysis of all logistics cycles based on a single piece flow standard, which companies in many other industries aspire to. In automated e-commerce processes, real-time decisions can be made because of unpredictable events and constantly changing criteria while the processes are in operation. Image processing algorithms with machine learning functions enable logistics systems to make valid decisions independently. This brings us within sight of the goal of fault-free logistics systems and advance error prevention. The huge quantities of data processed by intelligent algorithms enable production of more reliable forecasts and make it possible to identify the options available for strategic planning. Machines and quotas can be used highly efficiently when sensor systems and databases are networked. This forms the ideal foundation for simplifying the workflow, reducing resource use and negotiating the lowest possible handling costs with external service providers. The parcels both pass through the logistics process and are shipped quicker.
Through machine learning image processing makes an intelligent contribution to the value chain and represents a strategic advantage. Collecting and evaluating large quantities of image data allows machines to function reliably and autonomously and makes it possible for automation systems and the Smart Factory to provide additional financial benefits. Connections can be made and insights gained that were not possible in the past. Image processing is no longer simply an inspection procedure, it can now be used to improve production processes. Embedded and networked image processing systems will reduce business risks in variable and highly flexible environments, such as the agile e-commerce sector. The visual data relating to system performance, interactions and output quality that is processed intelligently forms a viable foundation for analytical decisions. This intelligent image processing can lead to process improvements, increased cost-effectiveness and profitable growth.
Written by Ute Häußler, Corporate Communications, Framos