Optimizing Production: Adige BLM Group's Challenge in AI-Based Planning

The manufacturing sector is constantly evolving, and precision in forecasting market demand is crucial for business success. In response to this challenge, Adige SpA, a subsidiary of the BLM Group , participated in the Industrial AI Challenge, an initiative promoted by Hit Hub Innovazione Trentino.

For 11 weeks, a team of students from the University of Trento had the opportunity to apply the skills acquired during their studies, developing an innovative Artificial Intelligence (AI) model aimed at supporting strategic planning. This model will support Adige's managers' decision-making process by coordinating AI's forecasts with the production workflow.

Our Consortium played a supportive role in providing mentorship and guiding the team through resolving this challenge.

Context & Company

Adige SpA, part of the BLM Group, is one of the leading Italian companies specializing in the production of laser tube-cutting machines. With an annual turnover of 250 million euros (in 2022) and a strong global presence, the company is known for the excellence of the "Made in Italy" brand, characterized by the entire product life cycle being realized in our country.

Operating according to the "make-to-order" production model, Adige SpA faces significant challenges related to long lead times for components and relatively short production times. This dynamic emphasizes the importance of accurate demand forecasting, as anticipating component restocking orders is essential to avoid stockout phenomena and financial inefficiencies due to unused components.
 

The Challenge

The challenge faced by our team required the development of an integrated system aimed at optimizing production planning on both the operational and the strategic levels.

On the one hand, the team focused on implementing an AI Forecasting system to predict long-term demand, providing strategic insights. On the other hand, we worked on the development of an AI Planner, a prospective tool designed to harmonize sales data with production planning. This tool will close the loop between the long-term forecasting data provided by the AI Forecasting system and the short-term execution managed by the Planning Management Assistant (PMA), creating a seamless flow from prediction to completion.

On the importance of Data 

A significant portion of the work was dedicated to the collection, analysis, and curation of data, which is crucial in any AI-focused project. Inferior-quality input data can indeed compromise the accuracy and reliability of the analyses and forecasts. A data-centric approach, with particular attention to the relationships between different entities, is essential to ensure informative and precise results. Diligence in data management, considering it as a central element to gain meaningful insights, is a practice that contributes to the success of any analytical or predictive initiative.

Initially, the team studied the business context, exploring key variables and data characteristics, allowing for targeted pre-processing. This phase enabled the examination of connections between orders and offers, developing an entity-relationship diagram to gain a deep understanding of data interrelationships.

Subsequently, the work continued with data preparation, further refining the contained information. During this process, issues emerged, such as negative time intervals between offers and orders, stemming from the organization's data extraction methods. This analysis also helped identify key indicators, including the order conversion rate from offers, the average time between offers and orders, and the sales rate across different industries. These results formed a robust foundation for constructing the artificial intelligence model required by Adige SpA.

Solutions & Results 

The adoption of a multidisciplinary approach, combining both machine learning and deep learning, has been a core success factor in enhancing business dynamics for BLM.

1. In the first approach, by implementing machine learning - in particular, the K-Means Clustering algorithm - it was possible to identify groups of customers with similar purchasing behaviors. This segmentation paved the way for more targeted business strategies, allowing the adaptation of operations to the unique characteristics of each cluster. The use of a Random Forest enabled us to accurately predict the demand for machines in various product categories over time.
    
2. In the second approach based on deep learning, the use of neural networks allowed the exploitation of a broad spectrum of variables, including the current month, customer sector, and region, to obtain detailed forecasts. Despite the complexity of the model, its flexibility in adapting to different customer combinations led to more accurate results than company forecasts, opening the way to more precise and optimized production planning for three different product categories.

The integration of these solutions - AI Planner and Forecasting AI - can therefore represent a crucial step in improving forecast accuracy and optimizing production planning, providing more reliable and detailed results. The collaboration between data-driven models and human expertise opens perspectives for more efficient planning, reducing storage costs, improving resource management, and consolidating BLM's position as a leader in the industry.