Summary
In this paper, we report on the structure of data science case data on infrastructure maintenance and the construction of data science execution function on the cloud. Here, the data science is a process not only to execute data analysis but also to deepen understanding of both site work and data based on interpreting the result of data analysis based on civil engineering and on-site operation knowledge.
The constructed cloud has functions necessary for performing analysis using heterogeneous data. For example, it is a function to implement data cleansing and data preparation functions on data such as inspection records, structure specifications, surrounding environment, etc. accumulated in infrastructure maintenance management.
Background
Civil engineering infrastructure (hereinafter referred to as infrastructure) in Japan is aging. Also, in the infrastructure maintenance work, the shortage of maintenance and maintenance engineers and the increase in maintenance cost are urgent issues. With such social background, expectation for utilizing ICT / IoT / AI etc. as a solution to the problem is extremely high. In response to these expectations, we have been implementing data science aiming at improving the efficiency of operations, targeting data accumulated in maintenance work. For example, "Estimation of soundness judgment value of bridge", "Extraction of condition of occurrence of deterioration of tunnel wall surface and upper floor", and so on.
Contents and Contribution
Regarding the above results, we report the following two proposals in this report: “1. Structure of infrastructure maintenance data science case”, “2. built in Cloud service”.
At first, we have created templates for the components and procedures of the data science cases that we have conducted so far. This template made it possible to organize and structure various data science cases. Based on CRISP-DM which stipulates the process of data mining, we proposed the flow of data science that fits the "infrastructure maintenance field" as the following items: "Site/Business Understanding", "Data Understanding", " Data preparation", "analysis", "evaluation", "consideration/feedback". We organized the necessary components for each implementation item of these infrastructure maintenance data science processes. In addition, we carried out structuralization based on the relevance of each constituent element such as the “type of data”, “purpose of analysis”, “analysis method”, etc. and the relevance of the implementation procedure, and organized it using ontology.
We constructed a system to implement the above data science as a cloud service. In the cloud service, it is possible to draw similar cases according to what you want to do and conditions from the past cases, and you can conduct data cleansing, data preparation, data analysis on the cloud. In application of machine learning at this time, data science cases are accumulated every time implementation is done by describing conditions and methods of learning, reference and order of similar cases of implementation, etc. by ontology.
Based on the provision of these systems, the trial cycle of data science in the infrastructure maintenance field is accelerated. It also contributes to solving the difficulty of data literacy for advanced use of data in real work.
Asset management and maintenance management , Big Data, data mining and machine learning