AI-based Condition Monitoring and Diagnostics of Pumping Facilities at Drainage Pump Stations

 Pumping facilities at drainage pump stations in rivers (see Figure-1) play an extremely important role to prevent flood damage, and their continued reliable future operation has to be ensured in the event of flooding caused by typhoons or heavy rainfall. However, with the growing number of facilities that have been installed for longer than 40 years, there is concern that the frequency of breakdowns will increase due to aging. It is also expected that there will be a decrease in the number of skilled engineers due to aging of the management personnel and a shortage of young engineers, as well as an increase in the frequency of operations due to an increase in the number of guerrilla torrential rainfall events.

 Therefore, to realize more efficient and accurate facility maintenance and management, we are conducting research on the introduction of condition monitoring and maintenance, which has been introduced for facilities in regular use, into the maintenance and management of the pumping facilities at drainage pump stations, which are emergency facilities. We are specifically constructing a condition monitoring system that can monitor the condition of facilities with time-series data by installing permanent sensors in pumping facilities at drainage pump stations and measuring the data of drainage operations when water is discharged. We are also developing an AI abnormality detection system including a certain diagnostic level without the need for skilled inspectors by using AI functions such as machine learning to detect abnormalities based on the vast amount of time-series data of measured drainage operations.

　

 During construction of the condition monitoring system, we studied sensors suitable for detecting abnormalities of pumps at drainage pump stations, and installed a permanent condition monitoring system at five facilities across Japan as a testbed. This has enabled the measurement of continuous time-series data for all operational data of both maintenance and actual operations. Figure-2 shows an example of a condition　monitoring system installed in a pumping facility at a drainage pump station.

 During the development of the AI anomaly detection system, we built a prototype that can detect abnormalities in pumping facilities at drainage pump stations based on data collected by the condition monitoring system. We aim to develop a system that will assist managers and specialized technicians engaged in inspections, as well as inexperienced technicians, who are expected to increase in number in the future.

 With the prototype system, we verified an algorithm that classifies input data into normal, abnormal A, abnormal B, and so on, to determine the type of abnormality. Among the AI models that can be used to classify time-series data, we considered a model that handles data in the frequency domain to be appropriate, and adopted an algorithm that combines the two techniques of decision tree and ensemble learning, which are relatively accurate for classification models and fast in learning and inference speed, even without acceleration, such as GPUs.

 In addition, when a failure is confirmed, the system does not only determine the presence of the failure, but also the abnormality level, which is the degree of the failure. The abnormality level is determined by comparing the values of data collected by the sensors with a provisional threshold value determined based on MLIT’s Guidelines for Monitoring River Pump System Conditions and ISO7919 (Mechanical vibration Evaluation of machine vibration by measurements on rotating shafts)

 The results of analyzing pseudo-created abnormality data and normal data (Figure-3) show that the accuracy of the prototype AI abnormality detection system to determine the type of abnormality is close to 100% probability. Based on these results, we believe that the AI abnormality detection system’s determination of anomaly types is functioning to some extent in the prototype model.

 We will continue our research to improve the accuracy of the AI abnormality detection system by collecting various operational data in the testbed and examining the most appropriate data processing methods for analysis by the system. We also plan to increase the number of facilities where the condition monitoring system is installed, and to advance research on the creation of pseudo-anomaly data using generative AI.

(Contact : Advanced Technology Research Team)

Overview of Research by the Fisheries Engineering Research Team

 Hokkaido is a key base for Japan’s fishery industry, with the largest catch in the country. However, in recent years, fishery has faced a mountain of issues, including declining production and a shortage of fishermen in fishing communities, making the situation surrounding the fishery industry and fishing villages increasingly severe. With the aim of appropriately conserving and managing the fishing ground environment and enhancing the productivity of sea areas, the Fisheries Engineering Research Team is conducting research on technologies related to the effective utilization of fishery harbor waters (Figure-1) and the development of artificial fish habitats at fishing grounds in offshore areas (Figure-2).
 Ongoing studies are introduced in this issue.

 1.Research on the utilization of fishery harbor facilities that contributes to the raising of aquatic life to cope with environmental changes in the sea area

 Habitats of aquatic organisms and their areas of distribution have undergone remarkable changes due to recent climate change and other factors. The autumn of 2021 saw one of the largest red tides ever recorded in Japan. It was the first recorded red tide off the Pacific coast of Hokkaido, and it caused tremendous damage to fishery. Amid concerns that such damage may recur, we are focusing on the function of kelp forests in inhibiting the propagation of harmful toxic plankton such as those in red tides, and we are researching technologies to raise aquatic life that utilize fishery harbor facilities and water areas.

 2.Research on improving habitats for aquatic life by utilizing fishery harbor facilities located in estuaries

 Estuaries are supplied with nutrient salts from land areas via rivers and serve as habitats for many organisms, such as bivalves. Meanwhile, in sea areas, the supply of nutrient salts has been changing due to recent climate change, and decreased production as a result of oligotrophication has been reported in some areas. In this context, toward utilizing the waters of fishery harbors located in estuaries, we are researching technologies to capture the inflow of nutrient salts from rivers using the fishery harbor facilities and to create new ecosystems.

　

 3.Research on technologies for improving the fishing ground environment through offshore structures in the northern sea area

 Offshore fisheries account for a large portion of fishery production. The amount of production, however, has decreased to about one-third of its peak. At the same time, offshore areas lack the sufficient accumulation of observation data and other information regarding the development of fishing grounds, such as the installation of artificial fish habitats, resulting in the absence of full-scale implementation of projects for the fishing ground development. Therefore, we are investigating the physical environment and habitat conditions of organisms around structures installed in waters 90 m deep, and we are conducting research to elucidate the effects of the structures on the cultivation of prey, fish aggregation, and fish growth.

(Contact: Fisheries Engineering Research Team, CERI)