Skip to main content
European Digital Innovation Hubs Network

AI and data-driven optimisation

AI and data-driven optimisation
Published at 18 December 2024 | Germany

General details

EDIHs involved

Customer

EDIH logo
Customer type: SME
Customer size: Small (10-49)
Customer turnover: 22.601 € (2023)

Services provided
Test before invest
Technologies
Artificial Intelligence & Decision support
Sectors
Agricultural biotechnology and food biotechnology

Challenges

Alpha-Protein has pioneered an advanced, fully automated factory concept that enables the cultivation and production of protein powder from mealworms. While the company’s primary market is currently the animal feed industry, it has strategic plans to expand into the food industry in the near future. Through its innovative approach, Alpha-Protein is positioning itself as a key player in the sustainable industry sector, contributing significantly to the circular economy and driving the shift towards more sustainable food systems.

Challenge:

As an innovative agritech start-up, Alpha-Protein faces crucial challenges in the areas of digitalisation and the integration of AI technologies. As mass production is still being set up, this is a favourable time to design sensors and AI systems as part of the plant planning process in such a way that AI potential can be successfully exploited. To this end, scenarios are to be developed that include intelligent and data-driven optimisation:

  • How to design structured data acquisition to ensure the availability and quality of data for subsequent analyses and intelligent services?

  • Which scenarios for data-driven optimisation are interesting in this context and how can they be identified? (E.g. which parameters influence the growth rate of mealworms?)

  • Side streams are already being processed (e.g. mealworm faeces as fertiliser). Can further scenarios be modelled in the sense of a circular economy (e.g. making process heat usable)?

  • How can processes be shortened or resources saved (by using a digital factory)?

Solutions

A structured approach is used to identify several areas of potential and specific scenarios. These are evaluated in terms of their level of innovation, sustainability relevance and benefits and then ranked. One of these scenarios is then implemented as a prototype. This could, for example, be an AI model for optimising the machine parameters in order to accelerate the growth of mealworms. The approach was as follows:

  1. Workshop for intelligent and data-driven optimisation

  2. Concretisation and ranking of the scenarios

  3. Prototypical realisation of the most relevant scenario

The initial workshop led to the conclusion that vision-based counting and weight prediction in a lab environment was the most promising contender for a feasible scenario which could be implemented in a reasonable amount of time. This was then prototyped using publicly available datasets while Alpha-Protein was gathering and annotating training images of their own. To do so, a fixed DSLR camera was set up above a surface in the lab. Alpha-Protein took several dozens of photos of mealworms in various arrangements and with different backgrounds (colours, sand, dirt). These photos were then annotated (classes, bounding boxes, masks) using AI-powered tools (SAM, LabelBox). Meanwhile, the FZI began developing a machine learning pipeline using publicly available datasets, resulting in several pre-trained models. These were then finetuned with the annotated data provided by Alpha-Protein. In order to increase the training data, more images were synthesised using generative AI.

As Alpha-Protein lacked AI experience and specialists, it was unable to develop the solution on their own. Being a start-up, funds are limited thus the private/public cooperation within EDIH offered the company the perfect way to explore AI potentials and gather some experience while limiting the financial risks.

Results and Benefits

The approaches developed by the FZI for both counting and weight prediction achieved acceptable first results but have yet to be evaluated in daily use at Alpha-Protein. Computing reliable quantitative metrics was not possible at this stage as the data is too limited. However, a qualitative assessment was conducted based on images annotated by the AI model. It showed that the model correctly identified and weighed most of the larvae but failed to do so correctly on some edge cases. These include larvae intersecting closely where the model is unable to distinguish them from one another. Alpha-Protein launched a test-phase where the solution is used by lab staff to assist in counting and weighing the larvae. If successful, the solution may safe staff several hours of tedious counting and weighing per week. In the long-term, the solution may be integrated into the actual production system to perform these tasks on a large scale.

Alpha-Protein benefited from a structured approach defining technical scenarios for further development as well as building competences for intelligent and data-driven optimisation. SMEs in general will benefit from this approach on how to optimally plan data-driven evaluation before commissioning production as well as building an understanding of circular economy fields of action.

Over the course of the project, the FZI spent around 20 working days in total. Including the ongoing test phase, Alpha-Protein expects to invest a similar amount of working days into the project.

Perceived social/economic impact

Testing of the prototyped model and approach at Alpha-Protein is continuing beyond the project’s completion. This approach is expected to deliver significant time savings in the lab by automating previously manual tasks such as sampling, counting, and weighing mealworm populations. However, the real economic potential lies in fully integrating this system into Alpha-Protein's mealworm production and storage process, which would directly increase the value of the company’s core product. This integration will be the primary focus of future activities.

Measurable data

Qualitative results can be seen exemplary in the figure below. As stated before, while generally recognising the larvae correctly, interlacing ones at times are not masked correctly or even ignored. Quantitative results are not present as more data is needed.

DMA score and results - Stage 0

The DMA resulted in an average score of 66%, indicating a moderately advanced stage of digital transformation. This score reflects the benefits of using both mainstream and some advanced technologies, but there is still potential for growth in areas such as competitiveness, resilience, and sustainability through targeted investments in digital technologies and skills.

While a wide range of digital tools are in use, there is room to further increase preparedness for more sophisticated solutions and adopt advanced technologies like AI. The digital skills of personnel are strong, but a structured training program is needed to further advance digital capabilities.

Data management is already advanced, yet there is opportunity to improve business intelligence and sustainability by adopting more ICT technologies and environmentally friendly practices. New digital investments would significantly enhance digital maturity and provide a competitive advantage.

Lessons learned

Do's:

  • Ensure adequate hardware is available: Verify that SMEs have sufficient hardware capabilities before assigning tasks such as data annotation or AI model training, to avoid delays caused by inadequate computing power.

  • Continue supporting SMEs with knowledge and prototypes: Maintain the approach of providing SMEs with both expertise and small prototypes through short, focused, and agile projects, as this has proven effective in the past.

Don’ts:

  • Don’t overlook hardware requirements: Avoid assigning tasks that require significant computing power, like AI-supported data annotation, to SMEs without first ensuring they have the necessary hardware to perform these tasks efficiently.

  • Don’t abandon agile, focused project support: Don’t move away from the successful strategy of offering targeted, agile project support, which combines knowledge-sharing and prototype development to benefit SMEs.

Need support?

Consult our catalogue to locate the Eupopean Digital Innovation Hub nearest to you and accelerate your company's digital transformation.

Find my nearest EDIH