1 Objectives 2 Research Axes 3 Work Packages 4 Expected Outcome

1 Objectives

This project is two-fold:

1. Explainable models

Having the Swiss manufacturing sector as focus, the aim is to find and integrate appropriate explainable models and align them with the company’s processes and tools.

2. Explanation interfaces

Designing a user interface that facilitates understanding, taking into account the diverse needs of end-users (operator or manager). In order to enable interactivity, the xAi system will take the shape of a conversational agent, providing personalized explanations through question-answering dialogues.

Global Architecture

2 Research Axes

1. Propose an appropriate answer to the user’s question, based on their profile and knowledge (e.g. operator vs. manager)
2. Computer modelling of the knowledge and operations related to part machining
3. Developing a conversational interface (chatbot) that can explain the decision/prediction of black-box model to the end-user

3 Work Packages

The CrystalClear project can be structured into 6 Work Packages (WP):

WP1: Project management & Dissemination

Guarantee the progress of different WPs while ensuring overall coherence. Dissemination activities are also managed in this WP.
Objective
Communication & Coordination, Reporting & Dissemination.
Deliverables
Management documents, reports, website, publication.

WP2: Modelling the problem

Define and validate computer modeling for machining quality.
Objective
Identify the physical factors impacting quality of a machining operation and create a computerized semantic taxonomy.
Deliverables
A semantic model of the machining process linking physical phenomena and observations.

WP3: Data acquisition

Acquisition of a database used to train machine learning algorithms.
Objective
Getting to know the existing database + Database expansion.
Deliverable
Database deployed in an InfluxDB solution; the database will contain 300+ labeled machining operations.

WP4: Explainable models

Explainable learning algorithms.
Objective
To develop deep learning algorithms capable of explaining their predictions.
Deliverables
ML learning models that incorporate xAi methods.

WP5: Explanation Interfaces

Creating the conversational interface.
Objective
create a conversational interface (chatbot) with the ability to explain the results of the algorithms to lay users.
Deliverables
Chatbot for different user profiles (manager and operator). Technical documentation on technological choices.

WP6: Final demonstrator

Production of a final demonstrator incorporating deliverables from WP4 and WP5.
Objective
Evaluate the project’s final results in a real-life industrial setting through the production of a final proof of concept demonstrator.
Deliverables
Demonstrator (software) and usability evaluation.

4 Expected Outcome

A user-friendly web interface hosting a chatbot that seamlessly integrates diverse explainable models and tailors its responses to the type of the end-user, offering relevant explanations through a natural question-answering dialogue.