Components

Fatigue Monitoring System (FaMS)

Module type(s)	Reasoning Engine
Description	FMS uses an artificial intelligence (AI) model relying on machine learning which estimates exertion level of subjects based on static data (e.g. age, weight, etc.) and dynamic data (e.g. HR, EDA, skin temperature).
Adopted technologies and methods	Apache Spark ML framework; Scala; Apache Avro; Docker; Random Forest Classifier.
Input	Workers physiological data from wearable devices on workers (e.g. HR, HRR, etc.); static data collected through questionnaire (e.g. age, needs, expectations, experience, etc.).
Output	Worker fatigue level to be used by “decision maker” modules to make human aware decisions.
Features	Adoption of machine learning to create use case specific AI model; Calculation in real-time of exertion stress level; Dynamic features use; Multi-device compatibility.

Shop-Floor Anomaly Detection System (SADS)

Module type(s)	Decision Maker
Description	SADS monitors the real-time status of the production assets and uses artificial intelligence (AI) to detect anomalies with regards to sensor values or behaviour.
Adopted technologies and methods	Docker; Python; scikit-learn and other libraries; TensorFlow.
Input	Any data from the shop floor, including sensor data, diagnostic messages, audio and video data streams.
Output	Anomaly score and health index per asset used by the ‘decision maker’ to derive recommendations (e.g. change configuration, conduct maintenance, emergency shut down).
Features	Adoption of machine learning to create use case specific AI model; Calculation of asset health in real-time; Dynamic features use; Multi-device compatibility;

VIsion for Quality Excellence (VIQE)

Module type(s)	Sensing; Reasoning Engine
Description	VIQE uses an artificial intelligence (AI) model based on deep learning for defects detection and Convolutional Neural Network for quality detection.
Adopted technologies and methods	OpenCV; MVTech Halcon; VIDI; TensorFlow; C#.
Input	Images from cameras; 3D sensors and 2D sensors data.
Output	Quality control values and results
Features	Use of computer vision algorithms and Deep Learning for Quality Inspection.

AI for Quality Systems (AIQuS)

Module type(s)	Reasoning Engine
Description	AIQuS is a solution oriented to facilitating the development, set-up and operation of Quality Control System. It uses Data Augmentation techniques in order to generate new images, that can be used to: 1) have balanced data for the training AI algorithms, specifically for new products/parts that don’t have enough images of defects; 2) to be used to compare different AI algorithms and set ups, and select the best model and parametrization; 3) enable operators of machine vision based systems to do a continuous evaluation, fine tuning, adjustments and re-configuration of the quality control systems once new defects are detected.
Adopted technologies and methods	TensorFlow, Python, Keras, OpenCV, Pillow.
Input	Images (with and without defects); Regions of Interest of defects.
Output	Synthetic images with defects.
Features	Semantic Analysis to structure the data: Collection of Samples; Annotation; Classification; Context Mapping to acquire contextual information.

Intervention Manager (IM)

Module type(s)	Decision maker
Description	IM monitors the real-time status of the worker-factory ecosystem, knows what interventions can offer and is capable to assess the available ones and decide which is the best one.
Adopted technologies and methods	JAVA 8; Spring Boot; Vaadin; OpenFeign; JUnit4; Apache Avro; Docker.
Input	Any data from worker-factory ecosystem.
Output	Interventions (e.g. system configuration, machine parameter, task assignment, etc.)
Features	Select interventions based on the inputs of the Sensing and Reasoning (AI based) modules; trigger the selected intervention; extensible in terms of new interventions, new conditions, new tuples; extensible by adding or removing devices and/or technologies.

Manufacturing Process Management System (MPMS)

Module type(s)	Decision maker
Description	MPMS is used to design in BPMN and execute a manufacturing process, along with alternatives and exceptions. The process includes high-level tasks that are designed to be executed by one actor (e.g. robot, human worker etc.) or many of actors.
Adopted technologies and methods	Java; REST API; BPMN; XML; JavaScript; HTML.
Input	Processes; Exceptions; Resources (actors); Data from sensors; Human inputs & decisions.
Output	Task lists to actors; current status of process and actors.
Features	Define resource allocation and task assignment; monitor process status (e.g. times, utilisation, task heatmap etc.); support to process design and modelling.