Optimization of the welding process

CONTEXT
In the process of welding steel sheets, even a low incidence of welding problems can lead to ruptures with undesirable consequences, such as the stoppage of the production line.

 

SOLUTION
Operational and welding sensor data were collected, as well as the physical-chemical characteristics of the steel, and information on maintenance and alarms. After analyzing the data, several models were created to provide an understanding of the main factors that influenced the process. In the end, we tested and validated a model capable of predicting welding quality problems, pointing out the variables that most impacted the quality deterioration.

 

RESULTS
Identification of welding problems that could not be observed before, allowing an understanding of the causes, and actions to prevent quality problems and eventual ruptures.

Optimization of Chemicals in water treatment

CONTEXT
In a water treatment plant of a sanitation company, the formulation and quantities of chemicals for the treatment of drinking water were difficult to optimize, given the constant variation in the characteristics of the raw water collected.

 

SOLUTION
Operational data were collected and understood from the perspective of the process. After undergoing a consistency analysis, we used supervised learning techniques for modeling and to simulate the behavior of the most important control variables. Lastly, we applied reinforcement learning techniques that sought the optimal dosages of chemicals.


RESULTS
Potential of up to 30% to reduce chemical consumption compared to operational history, following the recommendations of machine learning models.