By-product processing
General Information
Challenge, Value & Description
Performance, Access & Contact
1 – General Information
Partners
Sectors addressed
Application categories covered
Lifecycle level covered
Digital Engineering
Planning & Commissioning
Smart Production & Operations
Smart Logistics
Smart Maintenance
Customer Service
Circularity
Santastentie 197, Honkajoki 38950, Finland
Geographical Scope
- Nordics (Sweden, Denmark, Norway, Finland, Iceland)
2 – Challenge, Value & Description
Challenge
Honkajoki is a food by-product processing and recycling company focused on the rendering of meat industry side streams. The collected by-products are processed and upcycled to be used as components in pet foods, aquaculture and animal feeds, fertilizers, and as renewable raw material for biofuels. The Pilot exploits CLARUS solutions to optimize the logistics, in particular by predicting container filling and optimizing container selection at slaughterhouses. Another scenario is to reduce energy/steam consumption while enhancing product quality and involves the adjustment of the configuration of processing parameters. Therefore, this use case pilot is closely related to sustainable manufacturing challenges.
Value
- By optimizing the logistics, the Pilot aims at decreasing the raw material age (i.e., enhance final product quality)
- By optimizing the processing parameters, the Pilot aims to decrease electrical and thermal consumption, and CO2 emissions
Description
Data from the sensors installed in the plant and its warehouse are collected from the Pilot and shared through a Data Space with the technical partner that develops AI models through a MLOps platform. On the MLOps platform the technical partner develops, trains and deploys the AI models, making them available for the Pilot to download the models from cloud to edge.
Data Value Chain Description
Infraestructure Elements
- Private Cloud
- Field Devices
3 – Performance, Access & Contact Info
Performance
There are tests continuously being conducted at the pilot to validate the AI models and their predictions. The results are promising in general. Improvements in logistics have been shown to be possible by using the developed AI model. In the second pilot case some parameter values suggested by the model have however been proven to be unacceptable to process engineers with respect to mandatory safety limits and/or their field knowledge. Consequently, the models are debugged according to the reported issues, and new input variables are introduced to better capture the intricacies of the operations at the pilot.
Lessons Learned & Observations
One of the most important issues faced in the beginning was that the suggested temperatures for two devices were too low as per legislation for microbial safety, even though lower temperatures would have been yielding higher-quality products. Consequently, the dataset was filtered during model training to ensure that the predictions followed the guidelines.
Moreover, the feature selection carried out before model training resulted in attributes that are not necessarily relevant to the target variables. The most probable explanation is that those features have high correlations due to other, even external, elements; however, it does not mean that they affect each other. As a result, the selected feature lists were inspected once again manually to remove potentially irrelevant features.
Replication Potential & Feasibility Assessment
The trained AI models cannot be transferred directly to another production process since the existing process variables and their identifiers are not identical to the ones at Honkajoki. However, the knowledge learnt from this use case can be applied to other time-sensitive and energy intensive processes, and areas of manufacturing.
Contact Information