IAIE 2024 Abstracts

Area 1 - Industrial AI at the Edge

Full Papers

Paper Nr:	5
Title:	Digital Twin Model for Resource-Constrained Power System Maintenance Activities
Authors:	Yorlandys Salgado-Duarte and Janusz Szpytko
Abstract:	This paper presents a Digital Twin (DT) model designed to address the process of coordinating distributed maintenance activities of resource-constrained Power Systems with green energy integrated, employing Machine Learning (ML) for model calibration and Heuristic Optimization (HO) to solve the underlying coordination process. The model integrates real-time operational data collected by a Supervisory Control and Data Acquisition (SCADA) system and centralized in a Structured Query Language (SQL) database. The model performs predictive and prescriptive analyses to minimize convergence risks between maintenance planning, historical system degradation, and operating capacity, over a predefined time horizon, by merging all underlying processes into a single and intuitive risk indicator estimated via the Monte Carlo method, and by using a digital representation of all interconnected modeled processes in MATrix LABoratory (MATLAB), but also integrating Python packages into the modeling methodology. Here, the paper mainly describes the role of the DT framework in addressing this solution in practical terms and lists all components modeled with the corresponding link to the operating data. The Power System used as a case study includes a wide matrix of green energy sources and the associated resource constraint of each primary energy source considered.

Short Papers

Paper Nr:	6
Title:	Microcontroller Based Network for Industrial Edge AI
Authors:	Ander Garcia, Javier Tardos and Wilmer Lainez
Abstract:	While Industrial Edge AI applications are deployed at different devices, such as local clusters of computers, gateways, IoT devices and fog nodes, Embedded AI focuses on the deployment of AI algorithms in embedded units with limited com-puting resources, such as microcontrollers (MCUs). The increase of computa-tional power, the decrease of the cost of the hardware, and the development of less computation intensive AI models is fostering new opportunities to integrate embedded devices into Industrial Edge AI applications. This paper proposes an MCU based architecture and network for a low-cost monitoring of non-critical industrial environments. These MCUs both (1) monitor and perform a preliminary analysis of collected data with embedded AI, and (2) send data to edge de-vices with more computational power executing AI algorithms. The paper pro-poses an architecture for this MCU based network to lay the foundation of extensive industrial environmental monitoring where AI algorithms could for example, enhance workers safety, optimize processes, and improve maintenance operations. The technological viability of the network has been validated with a testbed where different MCUs (based on Arduino and ESP32) capture temperature and humidity data and send it to a more powerful edge device. Successful results prove the viability of the approach and foster further validations within real industrial scenarios.

Paper Nr:	7
Title:	Impact of Real-Time Linux for Industrial Edge AI
Authors:	Telmo Fernández De Barrena, Ander Garcia, Javier Franco and Javier Luengo
Abstract:	Traditionally hard real-time operating systems (RTOS) were reserved for applications with very restrictive requirements, such as aviation, industrial control or safety, where upper bounds of jitter and latency were guaranteed. However, requirements of Industry 4.0 and Industrial Edge Artificial Intelligence (AI) are different. Nowadays Industrial Edge AI does not control safety critical tasks, it analyses data and applies AI models to optimize industrial processes. Thus, they present soft real-time requirements: the sooner a result is returned the better, but no critical harm for operators or industrial assets is introduced by delays on the results from the AI services. Thus, Industrial Edge AI applications have been usually deployed as software containers or on general purpose Operating Sys-tems (OS). However, latest Linux kernel versions include a preemption option to transform general Linux distributions into soft RTOS. This paper focuses on the effect of this option for Industrial Edge AI. In order to measure its impact, three different experiments have been defined, where Raspberry Pis (RPis) and a PLC send data using MQTT and OPC UA Pub/Sub, under different sampling frequencies and computational load conditions. Then, Java and Python clients have been deployed on a different RPi running two versions of the Linux Kernel, the regular one and the soft real-time one. Finally, latency, jitter and packet loss measures have been taken in several variations of these setups in order to identify the response of each Linux Kernel for different use cases. Results of the experiments have been used to generate general guidelines for kernel selection for different use cases.