Saeed Gavidel, PhD

Experience

• 

  Tabriz Oil Refining Co. (TZORC)

  May 2008 - Oct 2013

  • Certified Energy Management Expert (Japan International Cooperation Agency - JICA)• Deployment of Energy Management System (EMS) according to ISO-50001 at refinery level• “Benchmarking, Standard Setting and Energy Conservation Program for Oil Refineries” project member• Contributed to development of energy consumption model for refinery steam network • Monitoring and benchmarking energy performance of thermal and electrical energy systems• Controlled data quality provided by out-of-house partners and contractors• Trained, validated and tested a regression model to predict energy consumption at refinery level• Generated multiple regression models to predict energy consumption of furnaces, exchangers and steam turbines• Leading energy management projects: “Steam Trap Energy Management Program”• Extraction, Transformation, Loading (ETL) of energy data in large volumes • Conducted statistical simulation (Monte Carlo) to simulate energy consumption scenarios Show less Reliability Engineer & Data Analyst, Tabriz Oil Refining Company, Tabriz, 2009 to 2013• Extraction, Transformation, Loading (ETL) of equipment data in large volumes • Applied statistical predictive models like linear regression in maintenance engineering• Constructed statistical linear regression model for failure prediction/prevention of rotary machines• Generated predictive regression model based on vibration data to predict/prevent ball/roller bearing failure• Constructed predictive classification and regression models based on ultrasonic condition assessment data• Generated statistical pattern recognition model to infer failure patterns based on repair/maintenance history• Implemented predictive/preventive maintenance programs on safety-sensitive equipment like safety valves • Computerized Maintenance Management Systems (CMMS) development/deployment team member• Supervised and managed maintenance projects like Steam Network Maintenance Project • Constructed multivariate stochastic predictive model to assess repair costs of industrial valves repair operations Show less

  • Data Analyst & Energy Management Engineer

    Aug 2009 - Oct 2013

  • Predictive Maintenance Senior Engineer

    May 2008 - Oct 2013
• 

  Wayne State University

  Sept 2014 - Dec 2018

  Graduate Research Assistant

  In this position, I pursued my PhD while serving as a Graduate Research Assistant. Specifically, I from September 2016 to May 2018, I was engaged in VRWP (Virtually guided Resistance spot Welding Project) that was a joint project where Wayne State University, Ford Motor Company, and Digital Manufacturing and Design Innovation Institute (DMDII) collaborated to construct weldability assessment solutions for Advanced High Strength Steels (AHSS) materials used in the automotive industry. Followings are some of my deliverables in this project:• Constructed a portfolio of predictive models including DNN-MLP, SVM, GAM, DNN, RF, KNN, CART, and M5P to predict the quality of vehicles RSW joint. I have developed these models using R. The developed models have been productionized at Ford Motor Company. The prediction accuracy of 97% achieved by DNN-MLP algorithm. As a result, 20% to 25% of Weld Engineers time saved during joint design stage.• Statistically simulated spot welding process and generated “weld lobe” heatmap, a key engineering tool to design a spot welding joint. I used parametric bootstrapping combined with Monte Carlo scenario generation to deliver this data product. This data product operationalized at Ford Motor Company.• Designed and developed a model selection framework based on Hypothesis Testing where both prediction accuracy and prediction precision are integrated into model selection process. In this framework, prediction accuracy of predictive models compared by using t-tests and prediction precision is compared by using Leven’s test. This work is published in International Journal of Advanced Manufacturing Technology.• Designed, developed, and operationalized the “Progressive Sampling Algorithm” to minimize the number of expensive experiments to collect data and train the predictive models. In this product, I employed Learning Curves to adjust the number of experiments. Show less

• 

  Consumers Energy

  Jun 2018 - Jan 2019

  Data Scientist-CPT (Big Data Analytics)

  - Major Duty: Theft/Fraud Detection and Revenue Protection- Design, development, operationalization, and productionalization of regular and elastic machine learning algorithms and analytical pipelines (proved efficiency) to analyze and detect/predict theft/fraud events and protect company revenue.- Exposed to +3 Petabyte (+500 billion records) of data with high veracity - Employed Apache Spark, R, Python, and memSQL to construct the algorithms- Conducted ETL operations to extract data from large Data Lakes- Constructed multiple elastic Big Data Analytics algorithms to analyze consumption behavior of consumers- Discovered/detected and mitigated a technical defect in electric smart meters known as “18129 Defect”- Developed multiple Statistical Outlier Detection algorithms to detect electricity theft and protect CE’s revenue Show less

• 

  Traxen

  Jan 2019 - Sept 2019

  Data Scientist-OPT (Data Science and Big Data Engineering)

  I have designed, developed, tested, and operationalized severalBig Data solutions for connected vehicle and smart mobility applications with special focus on conservation of energy.These Big Data solutions include scalable and elastic data pipelines to ingest, manage, and analyze sensory, operational, and consumer data received from multiple data sources. These solutions are being patented. Followings are high-level descriptions of my activities at Traxen:• Design, development, and implementation (by using Apache Spark, Python, and R) of a Big Data ingestion platform conforming with industry standards like SAE-J1339.• Construction of a composite Big Data Management System. In this composite system, traditional data management systems like MySQL and modern technologies like Apache Parquet and Avro have been employed to construct a composite Data Lake.• Design, development, and operationalization (by employing Python and PySpark) of an elastic Big Data platform for decoding, transformation, and management of CAN BUS system sensory data.• Design, development, and implementation of Normalized Driver Fuel Efficiency Performance Comparison Algorithm • Developing statistical tests according to industry requirements like SAE-J1321, and SAE-J1521 standards.• Constructing (using Spark MLlib) an elastic Random Forest predictive model to predict Fuel Efficiency.• Design, development, and operationalization of Big Data Stream analysis platform for the autonomous driving data.This framework has been constructed by using Spark Stream platform in PySpark environment.• Supporting development of a Reinforcement Learning algorithm for heavy vehicle Smart Mobility and Energy Conservation. In this task I provided Monte Carlo simulations of drive scenarios to pre-train an RL agent based on Experience Replay approach.• Mentoring junior Data Engineers, Developers, and Machine Learning Engineers. Show less

• 

  Ford Motor Company

  Sept 2019 - now

  Data Scientist (HTHD) Team Lead

  As a High Tech High Demand Data Scientist and Big Data Engineer at Ford Motor Company GDI&A, SCA-V (September 2019 to present) GDI&A-SCAV Connected Vehicle (CV) Decode Data Quality Monitoring Team is responsible to ensure that high-quality CV, sensory, and customer data flows in the Ford’s analytical pipelines. I lead the team and we have successfully built +30 Big Data and AI/ML products including an intelligent Auto AI/ML system known as CV Data Quality Monitoring Framework (DQMF). The DQMF is a productionized Big Data quality assessment, prediction, monitoring, and reporting system equipped with 48 ETL and analytical engines. DQMF processes several hundred terabytes of Ford data and sinks the results into a real-time reporting dashboard. I have architected the DQMF. Spark, Scala, Python, Java,and Hive are for development and productionization purposes.• Architected, coded, and implemented the ETL engines on top of HDFS, designed and constructed the staging and permanent storage, built predictive and anomaly detection engines of CV-DQMF. The ETL engines are developed by using Scala/Spark and PySpark. The storage system has snowflake schema and constructed by using Hive. Python, R, Scala/Weka are employed to build the predictive and detection engines.• Designed, developed, and operationalized Volume Monitoring Engine that monitors the volume of streaming CV data into Hive tables. The engine intelligently learns dynamic thresholds to flag anomalous streams. I have used Random Forest, Ridge, LASSO, SVM, KNN, and CART algorithms to learn the thresholds. Random Grid Search is used to tune the hyperparameters and bootstrapping was used to estimate 95% CI. Python and its Scikit-learn framework is employed to develop this engine. Finally, matplotlib is utilized to visualize the output. In February 2021, the engine detected ~23 M CV records being archived in HDFS without being processed and ingestion in Hive tables. This is considered waste prevention at Ford. Show less