The automotive industry constantly seeks intelligent technologies to increase competitiveness, reduce costs, and minimize waste, in line with the advancements of Industry 4.0. This study aims to implement and analyze a predictive model based on machine learning within the automotive industry, validating its capability to reduce the impact of unplanned downtime. The implementation process involved identifying the central problem and its root causes using quality tools, prioritizing equipment through the Analytic Hierarchy Process (AHP), and selecting critical failure modes based on the Risk Priority Number (RPN) derived from the Process Failure Mode and Effects Analysis (PFMEA). Predictive algorithms were implemented to select the best-performing model based on error metrics. Data were collected, transformed, and cleaned for model preparation and training. Among the five machine learning models trained, Random Forest demonstrated the highest accuracy. This model was subsequently validated with real data, achieving an average accuracy of 80% in predicting failure cycles. The results indicate that the predictive model can effectively contribute to reducing the financial impact caused by unplanned downtime, enabling the anticipation of preventive actions based on the model’s predictions. This study highlights the importance of multidisciplinary approaches in Production Engineering, emphasizing the integration of machine learning techniques as a promising approach for efficient maintenance and production management in the automotive industry, reinforcing the feasibility and effectiveness of predictive models in contributing to sustainability.

The automotive industry constantly seeks intelligent technologies to increase competitiveness, reduce costs, and minimize waste, in line with the advancements of Industry 4.0. This study aims to apply a predictive model based on machine learning within the automotive industry, validating its capability to reduce the impact of unplanned downtime. The implementation process involved identifying the central problem and its root causes using quality tools, prioritizing equipment through the Analytic Hierarchy Process (AHP), and selecting critical failure modes based on the Risk Priority Number (RPN) derived from the Process Failure Mode and Effects Analysis (PFMEA). Predictive algorithms were implemented to select the best-performing model based on error metrics. Data were collected, transformed, and cleaned for model preparation and training. Among the five machine learning models trained, Random Forest demonstrated the highest accuracy. This model was subsequently validated with real data, achieving an average accuracy of 80% in predicting failure cycles. The results indicate that the predictive model can effectively contribute to reducing the financial impact caused by unplanned downtime, enabling the anticipation of preventive actions based on the model's predictions. This study highlights the importance of multidisciplinary approaches in Production Engineering, emphasizing the integration of machine learning techniques as a promising approach for efficient maintenance and production management in the automotive industry, reinforcing the feasibility and effectiveness of predictive models in contributing to sustainability.

The Brazilian craft beer market has experienced continuous growth, increasing operational challenges for small- and medium-sized breweries that frequently rely on empirical and spreadsheet-based production routines. These practices often lead to inefficient resource allocation, production instability, and sustainability concerns. This study proposes an integrated analytical framework combining Machine Learning (ML) and Operations Research (OR) to improve demand forecasting and production planning. The methodology is based on a synthetic dataset calibrated to the operational conditions of a Brasília-based craft brewery, incorporating realistic demand patterns such as seasonality, trend, and intermittency across multiple SKUs over an 18-month horizon. Forecasting models—including Moving Average, Single Exponential Smoothing, and a global ML-based proxy—were evaluated using rolling-origin validation. The resulting probabilistic forecasts were integrated into a capacity-constrained optimization model based on linear programming, extended with risk-aware decision-making using Conditional Value-at-Risk (CVaR). The results indicate that the ML-based approach achieved competitive forecasting performance (sMAPE = 5.83% and MAE = 11.76) while enabling the generation of capacity-feasible and risk-aware production plans aligned with service-level targets. The integration of probabilistic forecasts into the optimization model allowed explicit trade-offs between cost, service level, and resource utilization. The main contribution of this study lies in demonstrating how the integration of predictive and prescriptive analytics can support more sustainable production planning in resource-constrained manufacturing environments. By replacing ad hoc spreadsheet routines with a closed-loop decision-support system, the proposed framework advances the literature on data-driven PPC and provides practical guidance for SMEs operating under uncertainty.