Smart Pipes’ Analysis and Data Diagnostics for AI-enhanced Manufacturing

Artificial Intelligence (AI) can be seen as a natural enabler for industry 4.0. Thanks to the use of sensors and the digitalization of the manufacturing systems, more and more data is generated. If this data is correctly analysed and evaluated by engineers and data scientists, it could lead to productivity improvement, material waste reduction, and improved machines diagnosis support. As the amount of data generated by machines and processes is in general huge, the use of artificial intelligence and machine learning techniques need to be fast and reliable. It is important to mention that AI has been explored through all the production lifecycle stages, namely, design, process planning, building, post processing, and testing and validation.

Concretely, the use of AI allows the development of added-value use cases in the manufacturing shopfloor and the manufacturing chain, including intelligent asset management, condition-based maintenance (CBM) and predictive maintenance (PdM), and lifecycle assessment (LCA) for key assets. The latter use cases enable manufacturers and other users of assets to avoid the catastrophic consequences of unplanned downtimes, boost the optimization of the Overall Equipment Efficiency (OEE), and contribute essentially to waste reduction in-line with the twin transition agenda of most industrial organizations. In parallel, based consortium experience, water losses in the underground hydraulic and irrigation networks exceed in many cases 50%. Beyond the significant ecological footprint, the possibility of accurately identifying the point of leakage determines a substantial reduction of these maintenance costs, improving the efficiency of the network. Moreover, the availability of information in (near) real time on the proper operation of factory plants and critical infrastructures becomes of great interest in the case of industrial applications where safety aspects are of primary importance. Applying machine learning techniques to industry 4.0 is a challenging task since three aspects need to be considered:

• ML frameworks and libraries that enable implementing robust machine learning algorithms;
• Enough meaningful data to be analysed;
• Clear application use cases and support from the experts in this field.

The above challenges will be addressed by the SANDMAN project. SANDMAN (Smart Pipes’ Analysis and Diagnostics for AI-enhanced Manufacturing) implements a novel, modular, configurable, and intelligent solution for data insights creation, and condition-based maintenance of pipes used in hydraulic networks. The solution will provide the means for gathering and analysing digital data about the conditions of the pipes towards optimizing their lifecycle management including their maintenance, services, repair, and other lifecycle management processes. In this direction, the project will develop an innovative digitally enabled lifecycle assessment tool for pipes, which will provide the means for optimizing both economic and environmental parameters, while providing recommendations for creating new pipes and resolving relevant trade-offs. SANDMAN designs and develops an innovative predictive maintenance application based on Deep Learning (DL) techniques and manufacturing data related to smart pipes for water distribution/irrigation and critical infrastructure. This application aims to:

1. build an AI on demand toolkit for industry 4.0, in particular for predictive maintenance;
2. identify in near real-time (NRT) any smart pipe production discrepancy;
3. analyze and assess the produced pipes during operations;
4. lead to smart pipes Zero Defect Manufacturing (ZDM);
5. contribute to the EC strategy in terms of data spaces by making the datasets that will be used to build the DL models available; and
6. expose the AI environmental implications by testing the developed AI techniques on various hardware (CPU, GPU,TPU).

It is worth mentioning that the DL toolkit that will be developed in the SANDMAN context presents the following innovations:

• It is built on top of well-known AI and machine learning libraries such as Tensorflow and Keras;
• It is implemented in Python, which makes its integration with other AI frameworks easy;
• It offers insights about the analyzed data;
• It will be deployed in the cloud; however, it could also be deployed at the edge with some adaptation;
• It will be validated in the SANDMAN context through a predictive maintenance applications;
• It provides a variety of functionalities that are generic and can be used in other contexts and applications;
• It can interact in a seamless way with other components through appropriate APIs; and
• It extends the Ekso business portfolio by adding to it a software component enabling early detection of potential defects, so enabling smart pipes lifetime and supporting sustainability.