HELLO! WELCOME TO IMC 2022

Collecting timely and accurate data is a prerequisite for effective condition monitoring. Versatile, mobile, and intelligent robots provide industrial operators a fast and scalable solution for data collection while liberating human workers from repetitive and potentially dangerous tasks. ANYbotics provides the entire toolchain to automate repetitive data collection at an unprecedented quality and quantity allowing plant operators to gain insights and take actions. In this talk, we will show how intelligent inspection robots are automating data collection tasks in complex industrial facilities without the need for adaptations to the environment.

The transition of disruptive technologies into society and industry has always required trust and acceptance from end users. Unlike prior industrial revolutions, autonomous systems are associated with the displacement of people, creating unease in how to prepare individuals, communities, industry and government. The complex interaction of human-machine systems today also mean digitalization is inevitable if we are to connect business, people, place, and technology. As levels of automation increase, so do legal issues with accountability and ethics. How then can we develop autonomous systems that gain trust from users? And how can such systems adapt and provide insights to better develop symbiotic autonomy that can be entrusted with decision making? In the context of our research, we focus on the Offshore Renewable Energy (ORE) market, which represents an early adopter for the aforementioned technologies. Barriers to residential Beyond Visual Line of Sight (BVLOS) autonomy as a service, include operational challenges in run-time safety compliance, reliability and resilience, due to the complexities of dealing with risks in dynamic environments. We present how autonomous systems are enablers to unlocking human potential, supporting the creation of a sustainable and prosperous future. We implement a Symbiotic System Of Systems Approach (SSOSA) to provide a cyber physical orchestration of enabling technologies. Implementing a SSOSA, addresses the aforementioned challenges during autonomous missions. This provides a means to synchronize distributed digital models of the robot, IoT sensors, environment and infrastructure. Bidirectional communications throughout the system allow the remote human operator to have improved visibility of the mission. This creates robotic assistants where high value jobs within symbiotic partnerships deliver mutually optimized benefits. The results of our autonomous mission evaluation within an ORE sector analogue evaluate both sudden and gradual faults, as well as unknown events, that may jeopardize the mission. Using distributed and coordinated decision making, the SSOSA enhances the analysis of the mission status, which includes diagnostics of critical sub-systems within the resident robot. This evaluation demonstrates that the SSOSA provides enhanced run-time operational resilience and safety compliance to BVLOS autonomous missions. The SSOSA has the potential to be a highly transferable methodology to other mission scenarios and technologies, providing a pathway to implementing scalable autonomy as a service.

Drones have become ubiquitous thanks to the recent developments in computer vision and machine learning. In this talk I will show commercial applications of autonomous drones to inspection of difficult infrastructure and automatic 3D scanning, from bridges to power lines. In the second part, I will present the key challenges to completely replace humans.

In order to improve the efficiency of the grid business, in 2018 we initiated a digitalization and innovation program called Grid 4.0. Since then, different projects have been started and partly realised to improve and further develop business processes. The talk will give insights into some of these projects like automated drone image recognition, transformer condition monitoring or a voltage control forecast.

Based on the digital trends and business drivers in the energy and mobility industry virtual management of decentralized assets, building information modelling and digital or data driven asset management is key for future business skills. The presentation will give a brief overview on exploration cases and insights for status and challenges of two concrete Initiatives in these action fields which SBB Energie tackles. One initiative is about managing a future battery swarm that will arise due to sustainability goals in the railway power network and the other is about a project that aims to reconstruct the railway power network within 5 minutes in case of partial or complete blackout scenarios.

Prognostics and health management (PHM) of equipment often require processing of long and multidimensional times series of measurements, where computational complexity is of primary importance. A recent advance in this area is the use of the matrix profile of a time series, which is a companion time series that conveniently encodes many analytics useful for PHM, and can be computed efficiently using advanced algorithms and modern hardware for parallel computation. Several variants of the matrix profile and its use for different PHM tasks will be discussed, along with more advanced time series primitives, such as time series chains.

Automatic road damage assessment is an essential aspect of ensuring the safety and efficiency of our road network. However, collecting the required data can be tedious, often requiring large-scale labeling of detailed crack location. Such project requires time, funding and expertise in order to acquire the equipment and begin the data collection. Some states, such as Japan have already developed models for automatic road monitoring. Different Countries may have different road markings, for example, requiring the use of domain adaptation techniques. We solve an interesting but challenging case of Unsupervised Domain Adaptation using only a trained model. Test-time domain adaptation is different from standard domain adaptation where we aim to adapt a source pre-trained model to a target domain without using any source data. Such configuration increase data privacy, and reduce drastically the amounts of data transfer since transferring a trained model is much efficient than huge dataset transfer.

Critical infrastructure is a dense network of systems that provide vital services such as energy, transport and water. Continuous cycles of monitoring and repair are vital to the economic development and functions of a community. In the offshore renewable energy sector for example, global growth in onshore and offshore wind energy has created a forecasted 11.5% compound annual growth rate increase in wind turbine blades throughout 2020-2025. This represents a significant challenge in inspection, maintenance and repair to ensure these critical infrastructures remain optimised throughout their lifecycle and in reaching these areas which represent safety challenges (at height, changing weather conditions and increased risk offshore). Cyber physical systems are being advanced to mitigate risks in asset integrity inspection. In the next 5-10 years, Engineers will have the ability to deploy a robot for inspection similar to the ease of utilising a tool from their toolbox. However, the current state of the art faces challenges in safety and resilience of robots working alongside humans. In addition, current asset integrity inspection methods of subsurface fault precursors are insufficient and do not allow for smart decisions to take place across the lifecycle of an asset. This presentation presents two key findings via a novel millimeter-wave radar sensor:
Asset Integrity Inspection - The analysis of complex multi-layer structures which include civil infrastructure, wind turbine blades and corrosion via a novel millimeter-wave radar sensor. Results have demonstrated the ability for the radar to detect surface and subsurface defects which are present within the aforementioned materials.
Foresight Sensing - The implementation of radar sensing onboard robotic platforms for human detection and through-wall detection to increase situational awareness. The results validate the use of the sensor to detect the difference between a person and infrastructure leading to increased situational awareness for navigation via foresight monitoring through walls. This research ensures that inspection engineers have tools available which can provide quantitative and qualitative results enabling more effective assessment of the subsurface integrity of a structure. The FMCW sensing modality provides early prognostics and remedial action to be taken, reducing the downtime of an asset and risk of failure. Foresight sensing enables for increases in safety compliant robotics alongside resilience in hazardous and dynamic environments.

Thermal spraying is a complex surface treatment process consisting of many complicated thermofluidic, electromagnetics, and electrochemistry, and particle physics phenomena. Ensuring high quality of the coating as well as high availability of the machine are of highest importance. While mathematical and CFD models provide valuable insight about the individual modules of a thermal spray process, it is very difficult to gain overall insight of the whole process and dependencies between different inputs and outputs using mathematical and CFD analysis, due to very complex and interconnected nature of the thermal spray process. In this work, we are trying to perform analysis on the collected machine data to achieve the following goals: (1) Alert creation: Identify any abnormal pattern in any of machine signals and bring it to the attention of the expert; (2) Anomaly prediction: There are some anomalies in the system that are not detected by any sensor. If there is a way to detect them early enough, the coating quality and health of the overall system will improve; (3) Hardware life-time prediction: The quality of coating is directly related to the quality and state of the spare part. Informing the operator about the right time to exchange the spare part is important for stable coating quality.

Typical data-driven methods for predictive maintenance focus on having a fleet of similar machines that have to be differentiated regarding operating conditions like speed or load. After observing a number of critical failures, often machine learning methods are used to predict such events from sensor readings. Within the field of packaging machines for sensitive products, in which the Uhlmann group is the global market leader, the additional challenge is to deal with a much larger variability that comes from the large range of products our customer require to package. Furthermore as the process involves recurrent motions with cycle times well below one second, the various sensor readings have to be observed in a high resolution. Finally, as our machines are deployed world-wide the environmental conditions like humidity differ significantly. In this talk, we introduce the various steps involved in the packaging process and what range of solutions Uhlmann is offering to the pharmaceutical industry. Then we shed some light on the subtle difficulties and peculiarities that come from the aforementioned points. Finally we present some results on what level of detail we can expect, which anomalies we already observed in data and how we tend to cope with the challenges mentioned. We conclude the talk with our vision that we do not think of machines, sustainability and AI as separate products but how we will achieve sustainable packaging solutions with the help of machine learning.