Education

Understand how to prepare a dataset for modelling and extract knowledge for future decision making. Application to an industrial application (wood-based panels)

Content

• Data Wrangling, Cleaning & Understanding to ensure quality
• Plotting tools for data interpretation
• Feature Selection & Extraction for data modelling
• Data Exploration & Mining to extract new insights about the business without any modelling;

Determine a suitable moment for restarting the turbine to clean turbine vanes by shockwave to increase power.
Approach: Find out generator output power based on input parameters with the help of predictive analysis.

Content

• Prediction of generator output power of hydroelectric power plant with respect to Big Data processing
• Data pre-processing
• Correlation analysis of input data
• Neural network for the prediction of generator output
• Practical part: Yes

Discussion of raw data from a production line of similar production units.

Content

• How historical data from production units can be used to analyse operational differences among the same (or similar) production units.
• Feature extraction for similarity analysis. Looking at similarity analysis techniques
• Constructing a digital model of a production line based on the extracted important features
• A historical dataset from a number of similar production units is provided for the students to get familiar with and perform feature extraction

Modelling data analytics for industry

Content

• Analysis of predictive maintenance events
• Impact of non-precise assumptions in sequenced and dependent processes

Understanding how predictive models work and how to take advantage of their properties for quality prediction. Application in a Laser Seam Welding industrial scenario.

Product lifecycle management: identification, extraction & classification of requirements from prescriptive documents into their corresponding disciplines.

Content

Specificity of text data & method: objectives, limitations and parameters

Automatic detection of position & orientation of a metallic part without knowing its geometry. Building & training the convolutional neural network (CNN) in the TensorFlow framework.

Content

• Description of CNN structure
• Preparation of training image data set
• Implementation of CNN model in Python
• Practical part: Yes

Predictive maintenance of Electrical Machining from the MANTIS project

Content

• What is predictive maintenance?
• What is the data you have?
• What predictive models are out there and what do you do with them?
• How do you integrate your models?

Content

• Identification of suitable methods and algorithms for data analysis: logistic regression, neural network, convolutional neural network
• Performances frameworks’ analysis: predictability, explainability, feasibility, interpretability

Out-of-control state detection based on one-class classifier for multivariate process control

Content

• Two quality characteristics measured on engine head after machining operation & used for detection of potential out-of-control state production process
• Commonly used control limits, applied individually on each characteristic, wrap the data unnaturally
• One-class support vector machine is trained on in-control state data & its more flexible decision boundary encloses the data points much better
• Practical part: Yes

For more consistent traceability of requirements into design documentation & testing suites as a part of the MANTIS project

How to use Deep Learning & Natural Language Processing to extract features & concepts from Requirement Documents, & trace them into the Design & testing phase

• Requirements and how to format them
• Regular expressions and how detect patterns in structured documents
• Extracting features & topics from requirement documents
• Mapping features + topics into design user stories
• Tracing requirement features + topics into testing

Content

• Understand how search algorithms can be used in optimization based on heuristics for parameter estimation.
• Understand how model-agnostic explainable AI approaches work and apply them to a practical defect detection case
• Application to a wood-based panels and laser seam welding (extra) industrial scenario

Content

• Identification of necessary data
• Feeding of a requirement management tool from prescriptive documents
• Method/algorithm for problem resolution and risk assessment

Automatic configuration of a counting machine based on a large set of historical data capturing collective
expert knowledge of various human operators.

Content

• Representation of the knowledge base data
• Clustering method used for identification of distinct groups of counting machine settings
• Data preprocessing needed as an initial step of the clustering
• Selection of the most important clusters and their representatives; cluster visualization
• Practical part: Yes

Production environment with a set of sensors to understand the environmental context situation and a set of actuators to control the production.

Content

• Representation of production rules in predicate or first-order logic (using declarative model-based approach)
• Dynamic constraint satisfaction problem – a way to solve and adhere to production rules
• Changes and adaptivity in rules: dynamic computational efficiency.
• Including a Practical part