Applications for artificial intelligence in the recycling business
Artificial intelligence is influencing our lives more than ever and is now our daily companion – whether in the office, factory or in our living rooms. In fact, the range of areas where it can be applied is almost limitless and offers companies plenty of potential to redefine business processes and value chains.
Nevertheless, quite a few companies are concerned about the rapid development in the AI sector. When it comes to trust in AI there is – let’s say – room for improvement, also among our customers in the recycling industry. According to a study, companies in Germany have persistent concerns, particularly in relation to data protection, data quality and the automatic interpretation of data, which make them shy away from using AI solutions.[1] At the same time, there is often a lack of relevant employee expertise and a vision for the use of the new technology in their own company. Many employees are also sceptical about AI solutions and fear that they will be replaced by them. AI should by no means be seen as a substitute for people. On the contrary, it is a question of allowing intelligent technologies to handle routine tasks so that employees have more time for the more complex tasks.
about the author
Steffen Brinkmann, Software Architect, tegos Group
What does AI mean?
There are numerous examples of possible applications for AI in the recycling industry. However, before we explain these in more detail, let us first clarify exactly what AI actually means so that there is no misunderstanding. When we talk about AI, we often encounter terms such as machine learning or neural networks. Essentially, all these terms stand for intelligent algorithms that can make their own decisions on the basis of systematically recorded and captured data so that decisions can be reached faster and more reliably than humans. This decision is never an absolute “yes or no” but is normally expressed in numerical or percentage results.
Let’s take, for example, an AI or an algorithm that is trained to distinguish images of dogs and cats. To create this algorithm or “train” the AI, it is first fed with thousands of images. If it is then shown a previously unknown picture with a dog in it, a decision about what is in the picture is made on the basis of the learned information. The result could then be, for example, 80% dog and 20% cat. Due to the percentage distribution, in the end the interpretation of the result would be “dog”.
The more images used in the training, the better the prediction or result of the AI. When creating an AI, the training data is the most important and at the same time limiting resource. Because at the beginning there is always the question: “Where do we get the data from?”
Once the database is available for the initial training, data from on-going operations can later be used to continuously train the AI by returning analysed data into a feedback loop, thus constantly improving the results. An AI is therefore only as good as its training and, of course, only as good as the data.
Furthermore, an AI is only trained for a certain task. Which means that an AI that is only trained to distinguish between pictures of dogs and cats, will not have much success with other tasks. If you show this AI a picture of a table, the result will always be either dog or cat, or both! So much for the theory. Now, let us come to some practical examples.
Automated weighing using AI
In contrast to earlier solutions, decisions made by AI are not the result of pre-set programming but are based on an extensive database and the AI’s continuous learning process, in which each event teaches the AI, allowing it to constantly optimise itself. This enables the AI to make decisions more quickly, economically and sustainably.
The focus is therefore always on added value. Good examples in which AI solutions can add real value in the waste disposal and recycling industry, by relieving staff of routine tasks, are to be found for tasks involving weighing, where license plate and material recognition is required.
For example, when materials are delivered to or collected from a recycling centre, it is often important to document the registration number of the vehicle while it is being weighed on a weighbridge or truck scale. At the moment, the vehicle license number is usually entered manually.
If we take a closer look at this process of vehicle license plate recognition during weighing, it can be divided into several small subtasks: the recognition of an image by a camera attached to the scale at the front or rear of the vehicle, the interpretation of numbers and symbols on the license plate and the capture of the vehicle license number in the software. These individual steps can be undertaken by an intelligent AI such as LPR-KI from tegos (License Plate Recognition).
Automatic license plate recognition is a first step on the way to automated weighing and enables, for example, forwarding companies to be recognized automatically based on a vehicle license number which is known to the system. Yard and warehouse employees or the weighing supervisor are then automatically provided with the most important information on the forwarder, which is stored in the system under the vehicle license number. In the case of a company’s own vehicles, license plate recognition could be used, for example, to automatically determine the vehicle crew and transfer this data to the weighing slip.
Another component involved in unmanned weighing is material identification or recognition. If the quality of the material fluctuates or it is delivered in different qualities, this can be particularly challenging. A camera mounted above the truck and an AI connection with image recognition can take over this task.
Based on thousands of learned images of materials, the AI can evaluate the quality of the material. Even if the material cannot be clearly identified optically, the technology can at least significantly narrow down the user’s choice.
As each captured image becomes part of the feedback loop in the AI’s learning process, the results and accuracy are constantly improved with each vehicle and material delivery.
Another advantage of AIs for material recognition is that they are not tied to fixed installation locations. A large number of our customers’ warehouse and yard employees are now working fully digitally using smartphones or tablets. From forklift drivers and quality inspectors to crane operators, they can receive all the information they need via the so-called “yard app”. Users such as quality inspectors, in particular, can take photos of the material with the camera integrated in their smartphone or tablet and have them analysed directly by the AI to identify the type or quality of the material. Thanks to the direct integration of the AI in the ERP system, the results immediately flow back into the central database and are thus available to all other staff members. The images are also stored in the document archive and can be used to prepare quality reports or to initiate claims.
Summary
There are numerous possibilities for the use of AIs in the recycling industry and we are only just beginning to exploit them. In addition to automated weighing, in future AIs will be found in a wide variety of storage, yard, separation and production processes and will gradually take over routine tasks from employees. Our business processes and work routines will change significantly as a result. However, fears and anxieties about this development are unfounded. Our customers will also quickly notice this when they take a closer look at this topic and the opportunities and possibilities associated with it.
[1] State of AI in the Enterprise Survey 2018
