Ikerlan – MLOps for Distributed Training of Quality Control ML Models
Ikerlan
Sector: Industry
Business Case
Recently, the company has developed various artificial intelligence models for its core operations. However, it recognises the necessity of creating a platform to oversee model quality in production and facilitate distributed training for the production lines it deploys at customer sites. The developed platform allows the automatic deployment of existing models for the quality control of the produced parts.
Objectives
The main objective is to create a platform that enables automated, distributed retraining of models deployed on production lines. This is achieved by monitoring inference results and distributing input data to the models.
Use case
Change in the current business model allowing the servitisation of the AI models that the company has developed to be monetised. For this purpose, the process of deploying solutions to various customers and their dynamic retraining and adjustment will be automated. This approach will enable the provision of a basic product with the existing functionality and an extended product that offers services deployed on the platform. These services include real-time monitoring of the quality of customers’ products, among other functionalities.
Infrastructure
Hybrid Cloud Edge
Technology
Computer vision Machine learning and deep learning
Data
Time series and images Private datasets
Resources
Research staff specialised in developing AI models and edge-to-cloud communications. Hybrid infrastructure with high computing capabilities based on virtual machines for model development.
Difficulties and learning
Difficulties Integration and development of less mature technologies and techniques of real data in production environments. Difficulty in finding quality errors due to unbalanced data. Learning New distributed learning techniques (federated learning). Concept drift detection techniques.
KPIs (business impact and metrics of the model)
MONDRAGON ASSEMBLY plans to reinforce its team with new AI model development specialists. The introduction of these new services and the optimisation processes developed through artificial intelligence techniques are expected to secure the most challenging projects in terms of product quality. This, in turn, is anticipated to boost the business’s annual turnover by €2M.
Funding
The project was funded by the HAZITEKs Estratégicos AIXIA (2022-2024), EGOKIA (2021-2023) and ADIERA (2020-2022).
Collaborators, Partners
Mondragon Assembly S. Coop. Ikerlan S. Coop. Koniker S. Coop.
