Advanced methods of smart predictive maintenance systems adaptation systems adaptation

Dudkovskaia, K; Taratukhin, V

Zusammenfassung

Intellectual Maintenance systems is the most developing class of maintenance information systems in past few years. Modern maintenance strategies and solutions in this area might minimize costs and maximize useful lifetime and reliability of equipment. Conditional-based, risk-oriented and predictive maintenance models use modern technologies such as IoT and sensor data analysis. However, real environment implementation process faces with data collecting difficulties and high-complexity of adaptation of the system in terms of business processes and specific machine needs. Each piece of equipment includes wide range of components from different manufacturers that subsequently might be substituted during maintenance. Thus, components of predictive maintenance system should be highly adaptive in terms of data models, math models, degradation detection algorithms and business processes. Reliably of predictive math models and algorithms influences the possible system’s effects. This paper reviews existing methods of analyzing telemetry data in order to customize maintenance decision-making system components and achieve higher math models reliability and accuracy. The paper includes implementation difficulties analysis, which was conducted based on implementation of smart maintenance solution projects for wide range of trains and locomotives service companies. As a result of the analysis, the original approach for predictive maintenance system components configuration was introduced. The approach helps to customize maintenance systems microservices for predictive maintenance systems in accordance with particular equipment engineering physical processes. The introduced processes starts with using of context analysis methods for classification of failure log which partly substitute methodological part of preliminary implementation analysis. Then the process continues with telemetry data markup using neural network, during which the failure boundaries are identified. After that, the degradation parameters before failure are listed for user’s consideration and analysis. The described analysis help in math data models tuning, failure detecting algorithms development and neural network teaching. This paper also includes description of other systems modules such as expert rules and Math model editors, which are essential for continuous customization by system’s business users. 

Schlüsselwörter
Predictive analysis, smart maintenance, physical processes monitoring, learning systems, periodic maintenance scheduling methods