Production Time Prediction and Analytical Chatbot
MECÁNICAS APARICIO
Client: Mecánicas Aparicio
Servicices: Predictive model with XGBoost to predict manufacturing times and an analytical chatbot based on LLMs to answer questions about production in a natural
Technologies: XGBoost, Large Language Models (LLMs), Python, cross-validation, and grid search.
José Enrique Aparicio, General Manager at Mecánicas Aparicio SL –
“At Mecánicas Aparicio, we have always been committed to innovation as a driver of continuous improvement. The proposal presented by Bravent represents a significant advance in how we understand and manage our production. The combination of predictive models with intelligent analytical tools opens up new possibilities for optimizing resources, reducing times, and making more informed decisions. This type of solution allows us to envision a future where efficiency and quality go hand in hand, strengthening our competitiveness in the sector.”
Challenge
Mecánicas Aparicio faced significant challenges in its production due to manual part analysis, which led to delays, poor resource allocation, production downtime, and quality errors.
The company was looking for a solution that would not only automate analysis, but also improve planning and facilitate data-driven decision-making.
Challenge
With the aim of predicting and reducing production times for these parts, our team presented a proposal that combines advanced predictive models with an analytical chatbot to transform decision-making on the factory floor:
- Predictive Model with XGBoost: We started with an exploratory data analysis (EDA) of historical production orders, identifying patterns and key variables. After evaluating eight different models (final regression, random forest), we selected the XGBoost predictive model as the most suitable for the solution due to its robustness and accuracy in tabular data, and its high capacity to detect complex interactions.
- Training: the dataset was divided into 70% for training and 30% for validation. A cross-validation strategy was applied to optimize the model’s hyperparameters, minimizing generalization error and ensuring stability in the results.
- Results: Prediction of estimated production time using metrics such as MAE and RMSE, achieving an 80% improvement in manual cutting predictions and a 66% improvement in the manual folding model.
- Application: Enables anticipation of delays and prioritization of orders based on risk.
Impact
Bravent’s solution provides tangible and measurable benefits that are crucial, such as the following:
- Measurable Return on Investment (ROI): Savings in unplanned hours and reduced downtime, demonstrating a high return on investment with a 40% decrease in unplanned production hours.
- Reduced Costs and Downtime: We anticipate deviations and optimize resources, which translates into significant cost savings.
- Improved Planning and Inventory: We adjust raw material purchases according to actual times, improving inventory efficiency.
- Data-Driven Decision Making: Quick access to KPIs through the chatbot, facilitating informed and strategic decisions.
- Increased Competitiveness: More reliable deadline compliance and the ability to simulate scenarios, strengthening the company’s competitive position.
