DATA SCIENCE IN THE FACTORY OF THE FUTURE

Knowledge and Understanding. On completion of the course, the student shall 1) know the basic principles of the application of data science in the modern industry, 2) understand methodologies and techniques used in industries to solve "prediction problems" based on time series, among which the “Failure Prediction and Predictive Maintenance” .  At the end of the module, students will be able to understand the main characteristics of industrial time series and correctly formulate a forecasting problem by defining the target, temporal granularity, forecast horizon, and operational purpose.

Applying Knowledge and Understanding. On completion of the course, the student will be able to select the appropriate technological solutions based on Machine Learning/Artificial Intelligenze and Big Data for each of the prediction scenario treated during the course.  They will be able to identify the main issues related to data quality and temporal structure, select appropriate preprocessing and temporal feature engineering techniques, and distinguish between the main statistical and machine learning methods for forecasting.

Making Judgements. On completion of the course, the student will be able to choose a suitable data science model for each of the subjects treated inside the course.

Communication Skills. On completion of the course, the student can communicate his conclusions and recommendations about data science applications in the modern industry with the argumentation of the knowledge and rationale underpinning these, to both specialist and non-specialist audiences clearly and unambiguously.

Learning skills. On completion, the student will be able to continue to study in a manner that may be largely selfdirected or autonomous.

The course is essentially based on lectures, which include the development of exercises by the teacher on a computer. 

Basics of statistics, machine learning, optimization and linear programming, regression and classification,data preprocessing, model evaluation. Knowledge and practice about R-language is also required.

Attendance is not mandatory but is strongly recommended. Attendance may be recorded during the lesson to gather statistical data on the number of students attending.

Introduction. The basic principles of the smart manufacturing and the use of Data Science in the Factory of the Future; the basic principles on management problems of a production plant, with particular enphasis on predictive maintenance; the most important methodologies and techniques used in industries to solve prediction problems based on time series, among which the “Failure Prediction and Predictive Maintenance”.

Smart Manufacturing. Main problems in the management of a production plant. Data Science applications to smart manufacturing. Basic principles on prediction problems based on time series, with particular enphasis to failure detection and predictive maintenance.

Failure Detection and Predictive Maintenance.  Data Acquisition: main sources of data. Data Pre-processing: an onverview on the main procedures to pre process data, including clean missing data, correlation, data inputation, balancing classes, normalisation, outliers. Feature Engineering: signal processing, feature extraction, and feature selection. Data labelling techniques. Machine learning techniques for failure type detection and predictive maintenance. Performance Evaluation. Real case studies using R language. Real case studies using Microsoft Azure Machine Learning Platform.

Time series forcasting in industrial systems. Formulation of forecasting problems, the distinction between prediction and forecasting, and the definition of the target, temporal granularity, forecast origin, forecast horizon, and decision lead time. The main components of time series will be analysed, including level, trend, seasonality, cyclicity, autocorrelation, and noise, together with issues related to sampling frequency, missing values, outliers, and structural changes. Time series preprocessing and transformation techniques. Temporal feature engineering methods will also be addressed, including lag variables, rolling statistics, calendar variables, weather-related variables, and operational variables, with particular attention to preventing temporal data leakage. The main statistical and machine learning methods for forecasting in industrial applications will be presented. Finally, multi-step forecasting strategies and time-based evaluation procedures will be discussed. The topics will be contextualised through real-world case studies related to energy and water forecasting.

[1] - Patrick Jahnke, "Machine Learning Approaches for Failure Type Detection and Predictive Maintenance", Master Thesis, June 19, 2015. Available online.

[2] - Handouts available in the repository of the course.

The exam, aimed at evaluating the student's understanding of the topics of the course and the achievement of the learning objectives, is organized in only one exam. The exam is a mix of theoretical questions (to which the students may be requested to give oral or written answers) and practical questions (in this case, the students are requested  to solve one or more exercises using the R-languages and Windows Azure Machine Learning platform used during the course).

To ensure equal opportunities and in compliance with current laws, interested students may request a personal interview in order to plan any compensatory and/or dispensatory measures based on educational objectives and specific needs. Students can also contact the CInAP (Centro per l’integrazione Attiva e Partecipata - Servizi per le Disabilità e/o i DSA) referring teacher within their department.

It is possible to download examples of questions and/or exercises from the repository used during the course. Information about the repository officially used, is given at the beginning of the course.