Passionate about bridging the gap between mechanical domain expertise and data science. I leverage Python, SQL, and Machine Learning to build physics-informed pipelines that detect anomalies, predict equipment RUL, and optimize industrial efficiency.

• Data Analysis: Python (Pandas, NumPy, SciPy), SQL
• Machine Learning & Deep Learning: Scikit-Learn (Random Forest, One-Class SVM, Isolation Forest, KNN), TensorFlow, Keras (1D-CNN, LSTM)
• Data Visualization: Power BI, Tableau, Matplotlib, Seaborn
• Platforms & Tools: Git, GitHub, Jupyter Notebook, Kaggle

• UConn Gearbox Fault Diagnosis: Physics-Informed AI & Deep Learning ⚙️🚗 Architected an industrial-grade Health Management pipeline using the high-frequency UConn Gearbox Dataset to diagnose 9 structural gearbox states (including 5 progressive levels of gear chipping severity). Developed a hybrid approach comparing a physics-informed Random Forest Classifier (utilizing extracted features like RMS and Kurtosis) against an end-to-end 1D-Convolutional Neural Network (1D-CNN). Explicitly addressed and mitigated Temporal Data Leakage using a strict Sequential Block Split, achieving verified 99.47% accuracy on ML and 100% absolute separation on Deep Learning 🚀.

• MetroPT3 Air Compressor Anomaly Detection: Unsupervised Predictive Maintenance 🚇⚙️ Engineered a physics-informed anomaly detection pipeline for a metro train's braking air compressor using the benchmark MetroPT Dataset. Implemented One-Class SVM (RBF Kernel) and optimized training execution via 10% stratified sampling to bypass high computational complexity ($O(N^2)$). Successfully captured multivariate degradation trends and air leak signatures, boosting failure Recall from 2% (Baseline Isolation Forest) to 99% 🚀.

• CWRU Bearing Vibration Fault Detection: DSP & Physics-Informed AI 📊🛠️ Developed an industrial-grade mechanical health monitoring pipeline using the benchmark CWRU Dataset. Implemented Fast Fourier Transform (FFT) and Envelope Analysis (Hilbert Transform Demodulation) to isolate high-frequency mechanical shock signatures (BPFI/BPFO). Built a physics-informed classifier optimized through systematic time-window scaling ($\Delta f$ contraction from $5.85\text{ Hz}$ to $1.46\text{ Hz}$), achieving 94% accuracy in deterministic defect isolation.

• NASA Jet Engine RUL Prediction: From Baseline to Deep Learning ✈️🛠️ Designed a 3-stage optimization engine using the benchmark NASA CMAPSS dataset to predict the Remaining Useful Life (RUL) of turbofan engines. Conducted empirical threshold sweeping, optimized data structures via RobustScaler, and engineered a temporal 3D sliding window for a Deep Learning LSTM Recurrent Neural Network. Successfully reduced the baseline prediction error (RMSE) by 47.3%.

• Sonar Signal Classification 🗺️ Implementing Logistic Regression for binary classification of acoustic signals—highly applicable to automotive predictive maintenance.

• Predictive Quality Control (Random Forest) 🧪 Utilizing Random Forest Classifier to predict product quality ratings based on multi-variable chemical balances.

• Diagnostic Risk Prediction (SVM) 🩺 Leveraging Support Vector Machines (SVM) for complex, distance-based classification pipelines.

• Olympic History Data Analysis 🏅 Exploratory Data Analysis (EDA) and cleaning on a 120-year historical dataset to uncover long-term demographic and performance trends.

• Supermarket Sales Insights 🛒 Transactional data analysis using Python to optimize business shift-staffing, peak hours, and customer purchasing patterns.

I actively apply my analytical frameworks to automotive use cases, focusing on:

• Predictive Maintenance: Reducing downtime by analyzing sensor frequencies, temporal degradation sequences (LSTM), and component wear.
• Market Pricing & Depreciation: Modeling vehicle value degradation patterns over time.
• Fleet & Sales Optimization: Streamlining parts inventory and operational performance.

• LinkedIn: farid-ghattas 💼
• Kaggle: faridghattas 📊
• HackerRank: farid_ghattas84 🎯
• Email: farid.ghattas84@gmail.com 📧