Abstracts Track 2024

Area 1 - Software Engineering and Systems Development

Nr:	112
Title:	Inspection of Multifunctional IoT Systems and Estimation of Probabilistic Parameters
Authors:	Ryuichi Takahashi
Abstract:	In recent years, automatic control utilizing sensor data is expected to be used in fields that require a high level of safety and reliability, such as automated driving vehicles. In such systems, it is important to use methods such as model checking for verification and assurance to guarantee safety in both normal and exceptional conditions. However, the following three factors can be cited as obstacles to safety verification by model checking. (1) Increasing complexity of functional dependencies due to the multifunctionality of the system, (2) Changes in sensor behavior due to environmental changes and associated changes in the probability of achieving a function, and (3) Difficulty in setting appropriate probability parameters to be used in model checking. To solve these problems, this research proposes a model checking method for systems with environmental data acquisition by sensors and a machine learning method for estimating the probability parameters needed to model system behavior. In the functional model, one of the proposed models, the functions possessed by the system are represented hierarchically using a tree structure. Dependencies of achieving multiple lower-level (fine-grained) functions necessary to achieve the upper-level functions are expressed, and these dependencies can be derived from goal-oriented requirement analysis methods such as KAOS. Verifying the behavior of complex multifunctional systems with model checking methods is difficult with current technology because it causes an explosion of the state space. Therefore, using the hierarchical dependencies of these functional models, the success or failure of each function is propagated within the system, and a functional inspection model can be derived that ultimately evaluates the achievement of the function of the entire system, enabling the safety verification of the entire system. In addition, some of the functions use environmental data from sensors. It is known that the quality of data acquired by sensors is affected by environmental changes, and the probability of achieving the function of each function varies depending on the data. Therefore, in addition to the functional inspection model, an environmental inspection model is defined that models the influence of environmental changes that fine-grained functions indirectly receive from sensor data. By synthesizing the verification results of the environmental inspection model into the functional inspection model, it is possible to verify the safety of the system in consideration of environmental changes. The PRISM model is used for the inspection, and the tool performs the automatic verification. PRISM can quantitatively evaluate the non-deterministic behavior of the system by using a state transition model with probabilities, but rich domain knowledge is required to set appropriate values for the probabilities. In this research, we propose a method for automatically estimating the value of this probability using machine learning based on preliminary environmental survey data and operational data from a prototype system. This enables the generation of appropriate inspection models without the need for domain experts and also allows for online updating of the models by collecting data during system operation. With these proposals, this research contributes to the safety verification of IoT systems that have multiple functions and require high reliability.

Area 2 - Software Systems and Applications

Nr:	5
Title:	Autonomous Software Agents to Ingest Upstream Operational Production Data from Disparate Data Sources to Cloud
Authors:	Abhay D. Paroha
Abstract:	The digital transformation journey has been embraced and executed by numerous companies around the globe and oil, gas, and energy industries are not an exception. Digital oilfield applications have been employed in discrete upstream operating companies to modernize processes and automate workflows to optimize oil and gas production in real time. The digital transformation strategies have surfaced a trail for Cloud-based solutions, which have lately gained drive in the oil and gas industry relating to their wider accessibility, system stability, and scalability to support larger amounts of data in a performant way. Upstream is an operation stage in the oil and gas industry that engages exploration and production. Oil and gas companies can generally be divided into three segments: upstream, midstream, and downstream. Upstream organizations deal primarily with the oil and gas industry's exploration and initial production stages. Upstream companies have been relying on traditional on-premises systems for a long; where maintenance, accessibility, and scalability serve as a major holdup for an efficient outcome. In addition to this challenge, the sector still faces limitations in data integration from disparate data sources, freedom of merged data for consumption, and cross-domain workflow orchestration of that data. Upstream production operations are substantially complex, covering a range of business roles, operational data, and workflows, and are often tackled by disconnected software applications and hardware tools. There are several challenges in data management of operational data e.g., data collection from disparate data sources and systems, a varied range of data frequencies, varying from seconds to yearly, and maintenance, accessibility, performance, and scalability of on-premises software applications and tools. Operational data is typically used to monitor production and highlight the oilfield tools and digital models that companies rely on every single day and the above challenges may result in non-productive time, data quality issues, discrepancies in data, and data-driven workflows, such as running calculations and missing data. The proposed novel approach of data ingestion using autonomous software agents has been implemented and evaluated on a set of oil production fields and successfully managed to run oil production assurance workflows for approximately 20,000 oil wells around the globe. The evaluation highlights that the approach is quite effective in data ingestion from multiple data sources i.e., Relational Database Systems (RDBMS), spreadsheets and CSV (Comma comma-separated values) files, Production Data Management Solutions (PDMS), corporate historians with high-frequency data flowing in from instrumented wells and sensors, Edge devices streaming in high-frequency data, manual data entries via ticket or mobile apps, calculated data via simulations or physical models.