AI-driven lifecycle management for end-of-life household appliances

Household appliances need to be refurbished and/or recycled at the end of their useful lives. The KIKERP project (see below) aims to harness artificial intelligence to collect information about used appliances and determine parameters such as model and condition to help decide whether a particular device should be refurbished or recycled.

In the case of recycling, the appliances are dismantled, with the raw materials going back into the production cycle, while refurbishment involves reconditioning functional or easily repaired appliances and putting them back on the market.

To support sustainable handling of used electrical appliances with a true circular economy in mind and reduce waste, researchers at Fraunhofer IPK have teamed up with YES Ecosystems Technology GmbH and HaKiGo GmbH to develop a multimodal AI assistance system embedded in a cloud-based architecture.

A dialog-based application feeds data to the AI until the user receives a classification for professional refurbishment, reuse and recycling of the appliance being assessed. The solution is aimed at appliance manufacturers’ employees as well as end consumers.

Visual inspection and quality scoring for appliances

The dialog-based front end runs on mobile devices such as smartphones and tablets. The application focuses on image-supported product identification. It uses pre-trained AI models to extract visual features and details. First, the employee uses the mobile app to log information such as brand, product type, color and item number. Then the household appliance is photographed from different angles and perspectives, so any defects such as scratches are also recorded.

The AI uses all this information as a basis to visually assess the quality of the appliance on a scale from one (poor) to five (excellent). This assessment is then used for two things: to identify next steps and determine parameters such as price and condition.

“Our AI modules are combined in a single architecture and run on a cloud server. They are operated via an interface on mobile devices,” explains Vivek Chavan, a research scientist at Fraunhofer IPK.

To develop the AI, the researchers use manufacturer data but also generate artificial training data. In the process, they are also studying whether visual inspection is something that can be trained with synthetic data, by introducing artificial visual defects in the images and videos. The goal is to use AI to test and evaluate more than 5,000 household appliances by the end of the project, achieving a more than 97% identification rate in the process.

In addition to AI assistance, the researchers at Fraunhofer IPK are also developing a data management system that uses an AI-based evaluation methodology to carefully sort the data volumes acquired in the process and select information for future training processes. The data management system is to serve as a basis for determining when and with which data the AI needs subsequent training to achieve as significant an improvement as possible.

The AI technologies developed are to be integrated into the architecture through robotic process automation (RPA) and software bots that learn repetitive, manual, time-consuming or error-prone activities and subsequently perform them automatically.

“With our cloud-based management platform, we are conceptualizing a process landscape for the remanufacturing and reuse of major household appliances and implementing it as an application demonstrator. The goal is to synchronize it with e-commerce portals for sale of the products being analyzed,” Chavan says.

Ultimately, the project should help reduce the carbon footprint of both manufacturing and use of household appliances in Germany, across Europe and in other markets. The project partners also hope to support scientific openness and facilitate the adoption of AI technologies by industry.