General details
EDIHs involved
Challenges
In the Netherlands there is a large shortage of refrigeration technicians to do the maintenance of refrigeration systems and perform repairs in case the technology malfunctions or parts need to be replaces. A lot of refrigeration systems are used in hospitals and other business with products which need to be stored at specific temperatures. This makes timely repairs an important element for Freezerdata to prevent loss of product or profits of their clients. However, FreezerData faced a significant challenge due to a shortage of refrigeration technicians. The problem was twofold: a lack of skilled personnel and the fact that experienced technicians were overwhelmed with questions from less experienced colleagues, which led to delays in service and repairs. This issue highlighted the need for a more efficient way to distribute knowledge and ensure quicker decision-making in the field of refrigeration technology.
FreezerData sought to address this challenge by developing a "virtual service mechanic" powered by AI. The idea was to create a tool that could answer technical questions and serve as a knowledge base, much like an experienced technician would. The virtual service mechanic would draw upon manuals and other technical resources, making it more reliable and user-friendly than traditional methods of troubleshooting.
Together with the EDIH Digital Hub Noordwest, FreezerData turned this idea to a Proof of Value and is further developing the product to make is market ready.
Solutions
In the development process FreezerData expanded on the initial idea of the ‘virtual service mechanic’ by incorporating sensor data to get more insight on the current state of the refrigeration system. With the addition of real-time sensor data, the technician can ask the virtual service mechanic question such as "How is the refrigerator performing?" or "Does a technician need to be dispatched, or can the issue be resolved remotely?". This functionality significantly reduces the need for direct intervention from highly skilled personnel, thereby alleviating the strain on experienced technicians.
FreezerData achieved this by integrating sensor data into the AI model, allowing it to offer real-time insights alongside static knowledge from manuals. A key technical breakthrough was ensuring the AI could determine when to pull information from the manuals versus when to rely on sensor data. This was accomplished by building a mechanism that identifies the relevant data source and adds the appropriate context to the AI’s responses.
The AI-driven virtual service mechanic also simplifies complex data interpretations, making it accessible to support staff who may not have the technical expertise to analySe refrigeration system parameters. As a result, the solution makes it possible to provide remote support without the need for advanced refrigeration knowledge, which streamlines the troubleshooting process and speeds up resolutions.
Ultimately, FreezerData's solution not only addresses the technician shortage but also enhances operational efficiency by providing real-time, contextual answers, improving workflow, and making the refrigeration industry more attractive to newer generations of workers.
Results and Benefits
FreezerData first collaborated with their partner Avisi and Axionomic, whom they met at an Innovation Dinner organiSed by the EDIH Digital Hub Noordwest. After the development of their initial concept product, they collaborated with the EDIH Digital Hub Noordwest consortium partner, the Amsterdam University of applied Sciences, to integrate real-time sensor data in the virtual service mechanic.
Throughout the project Freezerdata got to expand their network and meet the right collaboration partners to take the next step. Additionally, they got to make use of the expertise of the EDIH Digital Hub Noordwest partners on AI, Large Language Models and sensors throughout the process. Beside the development of the virtual service mechanic, they also increased their own knowledge about these digital technologies.
In the end they got to develop a Proof of Value for their virtual service mechanic. By making use of the EDIH program, they got to accelerate the development process and reduced the development costs. This increased their chances to make their ambition reality.
This demonstrator now enables FreezerData to attract customers willing to invest in the product. With a modest €10,000 investment from EDIH and FreezerData’s own €50,000 contribution, they have laid the foundation for future financial success.
Perceived social/economic impact
The virtual service mechanic addresses a larger problem in society. In several fields there is a shortage of personnel, such as mechanics or technicians. By proving the virtual service mechanic can make a difference in the field of refrigerator systems, in the future the virtual service mechanic can be adapted to other fields.
Additionally, less product and material are lost at hospitals and other organiSations highly depended these refrigerator systems by improving the efficiency and repair time. Therefore, there is less wait and less loss of profit.
DMA score and results - Stage 0
As of 22-06-2023, the company has achieved a Digital Maturity Index of 53%, indicating it is at a moderately advanced stage of its digital transformation process. This suggests that the company is already getting benefits from the use of both mainstream and advanced digital technologies, and that its current investments cover a wide range of business operations.
Lessons learned
One of the key aspects that worked well was the collaboration between FreezerData, Digital Hub Noordwest, and other partners such as Avisi and Axionomic. The combined expertise in refrigeration technology, AI development, and system integration resulted in a strong Proof of Value. FreezerData also found success by involving students from the maintenance lab at the Amsterdam university of applied Sciences which fostered innovative approaches, like integrating sensor data with AI. When working with multiple students in a project, a multidisciplinary approach can have them compliment their work and expertise. This can result in a more well-rounded final product.
A significant technical challenge was ensuring the AI model could correctly understand the context of user queries and determine when to use information from manuals versus real-time sensor data. Although a functional solution was achieved, this process was complex and time-consuming.
FreezerData kept exploring options and experimenting with different technology infrastructures, such as using the SURF and GPT-NL infrastructure, to refine their virtual service mechanic. By not focusing on only a specific solution and trying something different, they optimiSed the outcome.
The project highlighted the importance of prompt engineering—creating precise inputs for the AI to generate useful responses. This emerged as a critical skill for successfully integrating large language models with specific data sources.