Technicians working for companies in the field of gas pipeline maintenance (or any other linear asset networks, such as water, electricity, or railway) are often faced with the challenge of operating in plants located in remote areas, where poor mobile network connection strongly limits the use of Apps to support the completion of their on-field duties. The problem has been traditionally overcome by relying on a cache and synchronizing the device’s database with the operations center as soon as the connection is available again.
But what if applications are based on Machine Learning?
The concept of Edge AI originates from general Edge Computing, a distributed computing model consisting of processing data in a location as close as possible to where such data are produced, preferably in the very mobile device in use (vs. centralized data processing used in Cloud Computing).
At present, most Artificial Intelligence algorithms, which are by nature greedy for resources and require the processing of large amounts of data in the training phase, run on Cloud. The user accesses the model on Cloud through the network, sends a request and the remote service responds. The time required for the outcome is determined by the overall time needed for the request, the processing, and the response, and heavily depends on the Cloud latency and processing rates.
In this scenario, Edge solutions are experiencing growing popularity, thanks to their ability to overcome latency issues, and an increasing number of companies are evaluating the application of such technology in their business processes. According to Gartner, “There are three immediate revenue opportunities that place edge AI in the 90% to 100% early majority stage,” which they outlined as, “AI embedded in the Internet of Things (IoT) endpoint is a leading revenue opportunity for technology and service providers (TSPs) and is driving early majority adoption… Adoption of edge AI for data analytics is accelerating, particularly in industrial settings… Increasingly, edge AI is a catalyst for the adoption of broader IoT solutions because of its ability to reduce solution cost” (Gartner, Emerging Technologies and Trends Impact Radar: Artificial Intelligence, October 2020).
Moreover, the recent introduction of embedded frameworks for Machine Learning, such as Google’s famous “Tensor Flow lite”, specifically tailored for mobile devices, has boosted the use of Edge AI even further.
A typical phase of Field Service Management processes on gas pipelines is the debriefing, which includes all the activities aimed at accounting for the execution of works through a mobile App providing support for both accessing the company’s data and entering information collected on the field.
Debriefing can be performed either manually or automatically. In the first case, the user enters the single information collected during the intervention (hours worked, materials and spare parts used, measurements, multimedia contents, etc.). When in automatic mode, instead, the tool is “trained” to independently manage lengthy tasks related to data on activity completion, while the user is only required to validate them.
In general, the automatic debriefing mode begins with the analysis of videos or images, resulting in the extraction of numerical data or measurement of distances. Until recently, this process would have only been possible by uploading data to the Cloud and then waiting for the response, with obvious problems in terms of:
In the most common manual debriefing, daily issues mainly involve the last two problems above mentioned, since the processing is typically performed locally, with no need for internet connection.
Therefore, which is the optimal technology for debriefing in low connectivity areas?
The idea behind the implementation of Edge AI for automatic debriefing is quite simple: trying to process as much information as possible directly on the device, in order to:
This is made possible thanks to a model, a “thinking” software component, trained from remote and downloaded on the device. Then, the model is run locally without the need for a connection to work.
In the specific context of gas pipeline maintenance, new tools can be developed for automatic debriefing relying on computer vision services on Cloud, to be exported on Firebase’s mobile SDK (ML KIT), the platform developed by Google for creating mobile and Web applications. The model is thus enabled to work in a device, providing proactive support to the user, no matter the connectivity.
The technician, takes a picture of the work performed (excavations, installation of underground pipes, presence of any natural barriers, etc.), while Artificial Intelligence recognizes the information in the photo and fills out a set of fields that would otherwise have to be compiled manually. The latter an operation not to be underestimated, given the number of interventions performed by gas pipeline technicians every day, with the related amount of data to collect and enter.
“Edge AI allows to detect and process data relying on AI algorithms run in real time on mobile devices, thus opening up new challenging scenarios. In fact, Edge AI will not only reduce time and costs for data transmission, significantly shortening the timing of the decision-making process, but also prevent possible Data Breach of sensitive information exposed on Cloud.” Giuseppe Serra, Associate professor of Computer Science, University of Udine.
Mobile enablement is key for maintenance processes of gas distribution companies, since it provides the best information available at the right moment, allowing to optimize the different daily interventions required by the network. An efficient debriefing is therefore a matter of utmost priority, both for the company and the field technician.
In this scenario, Edge AI is a cutting-edge technology which ensures quality and rapidity of execution, enabling real-time access to information (all data required are already stored in the device and are provided to users by automatically detecting the context they operate in), with considerable benefits in terms of:
We may assert that Edge AI will not completely replace the Cloud as we know it, where large amounts of data are still being processed; however, in the medium-short term, it will certainly mark a new approach to Machine Learning with instant data.