Derrick FONGANG FONGANG

Predictive maintenance to detect and prevent anomalies on the machines Collecting production data: torque pressure, pump pressure and rotating speed. Concatenating data on each sensor to create a time series. Anomaly detection with the IQR: interquartile range, and the isolation forest techniques. Integrating acceleration data and frequence detection for the wireless vibration monitoring system. Time series data visualization with Seaborn

Analyzing the foundry parameters that impact the quality of produced pieces. Detecting features that impact the most the quality of produced pieces. Feature importance analysis with Machine Learning algorithms: Random Forest, Logistic Regression, GradientBoosting, Permutation Feature Importance Boostrap Sampling to manage the unbalanced dataset Data Science Evangelist, Training and Vulgarization http://www.mines-stetienne.fr/en https://www.cetim-engineering.com/

Comparison of solutions and technologies for the implementation of the Artificial Intelligence Observability. Comparative study of different A.I. Obs solutions, such as Elastic Observability. Designed and implemented the architecture for collecting system logs from linux servers and windows servers. Visualization, storage and analysis of the logs collected for the predictive maintenance of breakdowns on the server warehouse.

Design and develop an aid to the decision production toolkit to detect bad connections on VAM tubes. Therefore, preventing oil and gas companies from hydrocarbon leakage during operations on the platform. Designed and implemented the end to end Data Pipeline Architecture Implemented a PostgreSQL data schema to income each data source. In order to standardize, clean and control the quality of each data input. Composed machine learning features, to prepare the training of the classification model Launched the RandomForestClassifier and the BalancedRandomClassifier models.

Computed Prediction Intervals (Bootstrap) to account for uncertainties on features, and measure their impacts on the predictive model. Implemented the KFold - Cross Validation to estimate the predictive model accuracy Monte Carlo Sampling and Grid Search to tune the model hyper-parameters Extracted into CSV files, Log ASCII Standard (LAS) files, used for geophysical and geological logs